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 #if IS_ENABLED(CONFIG_SUNRPC_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 #if IS_ENABLED(CONFIG_SUNRPC_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 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 220 {
 221         return (struct rpc_xprt *) sk->sk_user_data;
 222 }
 223
 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 225 {
 226         return (struct sockaddr *) &xprt->addr;
 227 }
 228
 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 230 {
 231         return (struct sockaddr_un *) &xprt->addr;
 232 }
 233
 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr_in *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_in6 *) &xprt->addr;
 242 }
 243
 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 245 {
 246         struct sockaddr *sap = xs_addr(xprt);
 247         struct sockaddr_in6 *sin6;
 248         struct sockaddr_in *sin;
 249         struct sockaddr_un *sun;
 250         char buf[128];
 251
 252         switch (sap->sa_family) {
 253         case AF_LOCAL:
 254                 sun = xs_addr_un(xprt);
 255                 strlcpy(buf, sun->sun_path, sizeof(buf));
 256                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 257                                                 kstrdup(buf, GFP_KERNEL);
 258                 break;
 259         case AF_INET:
 260                 (void)rpc_ntop(sap, buf, sizeof(buf));
 261                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 262                                                 kstrdup(buf, GFP_KERNEL);
 263                 sin = xs_addr_in(xprt);
 264                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 265                 break;
 266         case AF_INET6:
 267                 (void)rpc_ntop(sap, buf, sizeof(buf));
 268                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 269                                                 kstrdup(buf, GFP_KERNEL);
 270                 sin6 = xs_addr_in6(xprt);
 271                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 272                 break;
 273         default:
 274                 BUG();
 275         }
 276
 277         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 278 }
 279
 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 281 {
 282         struct sockaddr *sap = xs_addr(xprt);
 283         char buf[128];
 284
 285         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 286         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 287
 288         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 289         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 290 }
 291
 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 293                                      const char *protocol,
 294                                      const char *netid)
 295 {
 296         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 297         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 298         xs_format_common_peer_addresses(xprt);
 299         xs_format_common_peer_ports(xprt);
 300 }
 301
 302 static void xs_update_peer_port(struct rpc_xprt *xprt)
 303 {
 304         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 305         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 306
 307         xs_format_common_peer_ports(xprt);
 308 }
 309
 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 311 {
 312         unsigned int i;
 313
 314         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 315                 switch (i) {
 316                 case RPC_DISPLAY_PROTO:
 317                 case RPC_DISPLAY_NETID:
 318                         continue;
 319                 default:
 320                         kfree(xprt->address_strings[i]);
 321                 }
 322 }
 323
 324 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 325
 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 327 {
 328         struct msghdr msg = {
 329                 .msg_name       = addr,
 330                 .msg_namelen    = addrlen,
 331                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 332         };
 333         struct kvec iov = {
 334                 .iov_base       = vec->iov_base + base,
 335                 .iov_len        = vec->iov_len - base,
 336         };
 337
 338         if (iov.iov_len != 0)
 339                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 340         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 341 }
 342
 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 344 {
 345         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 346                         int offset, size_t size, int flags);
 347         struct page **ppage;
 348         unsigned int remainder;
 349         int err;
 350
 351         remainder = xdr->page_len - base;
 352         base += xdr->page_base;
 353         ppage = xdr->pages + (base >> PAGE_SHIFT);
 354         base &= ~PAGE_MASK;
 355         do_sendpage = sock->ops->sendpage;
 356         if (!zerocopy)
 357                 do_sendpage = sock_no_sendpage;
 358         for(;;) {
 359                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 360                 int flags = XS_SENDMSG_FLAGS;
 361
 362                 remainder -= len;
 363                 if (remainder != 0 || more)
 364                         flags |= MSG_MORE;
 365                 err = do_sendpage(sock, *ppage, base, len, flags);
 366                 if (remainder == 0 || err != len)
 367                         break;
 368                 *sent_p += err;
 369                 ppage++;
 370                 base = 0;
 371         }
 372         if (err > 0) {
 373                 *sent_p += err;
 374                 err = 0;
 375         }
 376         return err;
 377 }
 378
 379 /**
 380  * xs_sendpages - write pages directly to a socket
 381  * @sock: socket to send on
 382  * @addr: UDP only -- address of destination
 383  * @addrlen: UDP only -- length of destination address
 384  * @xdr: buffer containing this request
 385  * @base: starting position in the buffer
 386  * @zerocopy: true if it is safe to use sendpage()
 387  * @sent_p: return the total number of bytes successfully queued for sending
 388  *
 389  */
 390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 391 {
 392         unsigned int remainder = xdr->len - base;
 393         int err = 0;
 394         int sent = 0;
 395
 396         if (unlikely(!sock))
 397                 return -ENOTSOCK;
 398
 399         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 400         if (base != 0) {
 401                 addr = NULL;
 402                 addrlen = 0;
 403         }
 404
 405         if (base < xdr->head[0].iov_len || addr != NULL) {
 406                 unsigned int len = xdr->head[0].iov_len - base;
 407                 remainder -= len;
 408                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 409                 if (remainder == 0 || err != len)
 410                         goto out;
 411                 *sent_p += err;
 412                 base = 0;
 413         } else
 414                 base -= xdr->head[0].iov_len;
 415
 416         if (base < xdr->page_len) {
 417                 unsigned int len = xdr->page_len - base;
 418                 remainder -= len;
 419                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 420                 *sent_p += sent;
 421                 if (remainder == 0 || sent != len)
 422                         goto out;
 423                 base = 0;
 424         } else
 425                 base -= xdr->page_len;
 426
 427         if (base >= xdr->tail[0].iov_len)
 428                 return 0;
 429         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 430 out:
 431         if (err > 0) {
 432                 *sent_p += err;
 433                 err = 0;
 434         }
 435         return err;
 436 }
 437
 438 static void xs_nospace_callback(struct rpc_task *task)
 439 {
 440         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 441
 442         transport->inet->sk_write_pending--;
 443         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 444 }
 445
 446 /**
 447  * xs_nospace - place task on wait queue if transmit was incomplete
 448  * @task: task to put to sleep
 449  *
 450  */
 451 static int xs_nospace(struct rpc_task *task)
 452 {
 453         struct rpc_rqst *req = task->tk_rqstp;
 454         struct rpc_xprt *xprt = req->rq_xprt;
 455         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 456         struct sock *sk = transport->inet;
 457         int ret = -EAGAIN;
 458
 459         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 460                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 461                         req->rq_slen);
 462
 463         /* Protect against races with write_space */
 464         spin_lock_bh(&xprt->transport_lock);
 465
 466         /* Don't race with disconnect */
 467         if (xprt_connected(xprt)) {
 468                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 469                         /*
 470                          * Notify TCP that we're limited by the application
 471                          * window size
 472                          */
 473                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 474                         sk->sk_write_pending++;
 475                         /* ...and wait for more buffer space */
 476                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 477                 }
 478         } else {
 479                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 480                 ret = -ENOTCONN;
 481         }
 482
 483         spin_unlock_bh(&xprt->transport_lock);
 484
 485         /* Race breaker in case memory is freed before above code is called */
 486         sk->sk_write_space(sk);
 487         return ret;
 488 }
 489
 490 /*
 491  * Construct a stream transport record marker in @buf.
 492  */
 493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 494 {
 495         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 496         rpc_fraghdr *base = buf->head[0].iov_base;
 497         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 498 }
 499
 500 /**
 501  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 502  * @task: RPC task that manages the state of an RPC request
 503  *
 504  * Return values:
 505  *        0:    The request has been sent
 506  *   EAGAIN:    The socket was blocked, please call again later to
 507  *              complete the request
 508  * ENOTCONN:    Caller needs to invoke connect logic then call again
 509  *    other:    Some other error occured, the request was not sent
 510  */
 511 static int xs_local_send_request(struct rpc_task *task)
 512 {
 513         struct rpc_rqst *req = task->tk_rqstp;
 514         struct rpc_xprt *xprt = req->rq_xprt;
 515         struct sock_xprt *transport =
 516                                 container_of(xprt, struct sock_xprt, xprt);
 517         struct xdr_buf *xdr = &req->rq_snd_buf;
 518         int status;
 519         int sent = 0;
 520
 521         xs_encode_stream_record_marker(&req->rq_snd_buf);
 522
 523         xs_pktdump("packet data:",
 524                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 525
 526         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 527                               true, &sent);
 528         dprintk("RPC:       %s(%u) = %d\n",
 529                         __func__, xdr->len - req->rq_bytes_sent, status);
 530         if (likely(sent > 0) || status == 0) {
 531                 req->rq_bytes_sent += sent;
 532                 req->rq_xmit_bytes_sent += sent;
 533                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 534                         req->rq_bytes_sent = 0;
 535                         return 0;
 536                 }
 537                 status = -EAGAIN;
 538         }
 539
 540         switch (status) {
 541         case -ENOBUFS:
 542         case -EAGAIN:
 543                 status = xs_nospace(task);
 544                 break;
 545         default:
 546                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 547                         -status);
 548         case -EPIPE:
 549                 xs_close(xprt);
 550                 status = -ENOTCONN;
 551         }
 552
 553         return status;
 554 }
 555
 556 /**
 557  * xs_udp_send_request - write an RPC request to a UDP socket
 558  * @task: address of RPC task that manages the state of an RPC request
 559  *
 560  * Return values:
 561  *        0:    The request has been sent
 562  *   EAGAIN:    The socket was blocked, please call again later to
 563  *              complete the request
 564  * ENOTCONN:    Caller needs to invoke connect logic then call again
 565  *    other:    Some other error occurred, the request was not sent
 566  */
 567 static int xs_udp_send_request(struct rpc_task *task)
 568 {
 569         struct rpc_rqst *req = task->tk_rqstp;
 570         struct rpc_xprt *xprt = req->rq_xprt;
 571         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 572         struct xdr_buf *xdr = &req->rq_snd_buf;
 573         int sent = 0;
 574         int status;
 575
 576         xs_pktdump("packet data:",
 577                                 req->rq_svec->iov_base,
 578                                 req->rq_svec->iov_len);
 579
 580         if (!xprt_bound(xprt))
 581                 return -ENOTCONN;
 582         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 583                               xdr, req->rq_bytes_sent, true, &sent);
 584
 585         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 586                         xdr->len - req->rq_bytes_sent, status);
 587
 588         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 589         if (status == -EPERM)
 590                 goto process_status;
 591
 592         if (sent > 0 || status == 0) {
 593                 req->rq_xmit_bytes_sent += sent;
 594                 if (sent >= req->rq_slen)
 595                         return 0;
 596                 /* Still some bytes left; set up for a retry later. */
 597                 status = -EAGAIN;
 598         }
 599
 600 process_status:
 601         switch (status) {
 602         case -ENOTSOCK:
 603                 status = -ENOTCONN;
 604                 /* Should we call xs_close() here? */
 605                 break;
 606         case -EAGAIN:
 607                 status = xs_nospace(task);
 608                 break;
 609         default:
 610                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 611                         -status);
 612         case -ENETUNREACH:
 613         case -ENOBUFS:
 614         case -EPIPE:
 615         case -ECONNREFUSED:
 616         case -EPERM:
 617                 /* When the server has died, an ICMP port unreachable message
 618                  * prompts ECONNREFUSED. */
 619                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 620         }
 621
 622         return status;
 623 }
 624
 625 /**
 626  * xs_tcp_shutdown - gracefully shut down a TCP socket
 627  * @xprt: transport
 628  *
 629  * Initiates a graceful shutdown of the TCP socket by calling the
 630  * equivalent of shutdown(SHUT_WR);
 631  */
 632 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 633 {
 634         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 635         struct socket *sock = transport->sock;
 636
 637         if (sock != NULL) {
 638                 kernel_sock_shutdown(sock, SHUT_WR);
 639                 trace_rpc_socket_shutdown(xprt, sock);
 640         }
 641 }
 642
 643 /**
 644  * xs_tcp_send_request - write an RPC request to a TCP socket
 645  * @task: address of RPC task that manages the state of an RPC request
 646  *
 647  * Return values:
 648  *        0:    The request has been sent
 649  *   EAGAIN:    The socket was blocked, please call again later to
 650  *              complete the request
 651  * ENOTCONN:    Caller needs to invoke connect logic then call again
 652  *    other:    Some other error occurred, the request was not sent
 653  *
 654  * XXX: In the case of soft timeouts, should we eventually give up
 655  *      if sendmsg is not able to make progress?
 656  */
 657 static int xs_tcp_send_request(struct rpc_task *task)
 658 {
 659         struct rpc_rqst *req = task->tk_rqstp;
 660         struct rpc_xprt *xprt = req->rq_xprt;
 661         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 662         struct xdr_buf *xdr = &req->rq_snd_buf;
 663         bool zerocopy = true;
 664         int status;
 665         int sent;
 666
 667         xs_encode_stream_record_marker(&req->rq_snd_buf);
 668
 669         xs_pktdump("packet data:",
 670                                 req->rq_svec->iov_base,
 671                                 req->rq_svec->iov_len);
 672         /* Don't use zero copy if this is a resend. If the RPC call
 673          * completes while the socket holds a reference to the pages,
 674          * then we may end up resending corrupted data.
 675          */
 676         if (task->tk_flags & RPC_TASK_SENT)
 677                 zerocopy = false;
 678
 679         /* Continue transmitting the packet/record. We must be careful
 680          * to cope with writespace callbacks arriving _after_ we have
 681          * called sendmsg(). */
 682         while (1) {
 683                 sent = 0;
 684                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 685                                       req->rq_bytes_sent, zerocopy, &sent);
 686
 687                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 688                                 xdr->len - req->rq_bytes_sent, status);
 689
 690                 if (unlikely(sent == 0 && status < 0))
 691                         break;
 692
 693                 /* If we've sent the entire packet, immediately
 694                  * reset the count of bytes sent. */
 695                 req->rq_bytes_sent += sent;
 696                 req->rq_xmit_bytes_sent += sent;
 697                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 698                         req->rq_bytes_sent = 0;
 699                         return 0;
 700                 }
 701
 702                 if (sent != 0)
 703                         continue;
 704                 status = -EAGAIN;
 705                 break;
 706         }
 707
 708         switch (status) {
 709         case -ENOTSOCK:
 710                 status = -ENOTCONN;
 711                 /* Should we call xs_close() here? */
 712                 break;
 713         case -ENOBUFS:
 714         case -EAGAIN:
 715                 status = xs_nospace(task);
 716                 break;
 717         default:
 718                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 719                         -status);
 720         case -ECONNRESET:
 721                 xs_tcp_shutdown(xprt);
 722         case -ECONNREFUSED:
 723         case -ENOTCONN:
 724         case -EPIPE:
 725                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 726         }
 727
 728         return status;
 729 }
 730
 731 /**
 732  * xs_tcp_release_xprt - clean up after a tcp transmission
 733  * @xprt: transport
 734  * @task: rpc task
 735  *
 736  * This cleans up if an error causes us to abort the transmission of a request.
 737  * In this case, the socket may need to be reset in order to avoid confusing
 738  * the server.
 739  */
 740 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 741 {
 742         struct rpc_rqst *req;
 743
 744         if (task != xprt->snd_task)
 745                 return;
 746         if (task == NULL)
 747                 goto out_release;
 748         req = task->tk_rqstp;
 749         if (req == NULL)
 750                 goto out_release;
 751         if (req->rq_bytes_sent == 0)
 752                 goto out_release;
 753         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 754                 goto out_release;
 755         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 756 out_release:
 757         xprt_release_xprt(xprt, task);
 758 }
 759
 760 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 761 {
 762         transport->old_data_ready = sk->sk_data_ready;
 763         transport->old_state_change = sk->sk_state_change;
 764         transport->old_write_space = sk->sk_write_space;
 765         transport->old_error_report = sk->sk_error_report;
 766 }
 767
 768 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 769 {
 770         sk->sk_data_ready = transport->old_data_ready;
 771         sk->sk_state_change = transport->old_state_change;
 772         sk->sk_write_space = transport->old_write_space;
 773         sk->sk_error_report = transport->old_error_report;
 774 }
 775
 776 /**
 777  * xs_error_report - callback to handle TCP socket state errors
 778  * @sk: socket
 779  *
 780  * Note: we don't call sock_error() since there may be a rpc_task
 781  * using the socket, and so we don't want to clear sk->sk_err.
 782  */
 783 static void xs_error_report(struct sock *sk)
 784 {
 785         struct rpc_xprt *xprt;
 786         int err;
 787
 788         read_lock_bh(&sk->sk_callback_lock);
 789         if (!(xprt = xprt_from_sock(sk)))
 790                 goto out;
 791
 792         err = -sk->sk_err;
 793         if (err == 0)
 794                 goto out;
 795         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 796                         xprt, -err);
 797         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 798         if (test_bit(XPRT_CONNECTION_REUSE, &xprt->state))
 799                 goto out;
 800         xprt_wake_pending_tasks(xprt, err);
 801  out:
 802         read_unlock_bh(&sk->sk_callback_lock);
 803 }
 804
 805 static void xs_reset_transport(struct sock_xprt *transport)
 806 {
 807         struct socket *sock = transport->sock;
 808         struct sock *sk = transport->inet;
 809
 810         if (sk == NULL)
 811                 return;
 812
 813         transport->srcport = 0;
 814
 815         write_lock_bh(&sk->sk_callback_lock);
 816         transport->inet = NULL;
 817         transport->sock = NULL;
 818
 819         sk->sk_user_data = NULL;
 820
 821         xs_restore_old_callbacks(transport, sk);
 822         write_unlock_bh(&sk->sk_callback_lock);
 823
 824         trace_rpc_socket_close(&transport->xprt, sock);
 825         sock_release(sock);
 826 }
 827
 828 /**
 829  * xs_close - close a socket
 830  * @xprt: transport
 831  *
 832  * This is used when all requests are complete; ie, no DRC state remains
 833  * on the server we want to save.
 834  *
 835  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 836  * xs_reset_transport() zeroing the socket from underneath a writer.
 837  */
 838 static void xs_close(struct rpc_xprt *xprt)
 839 {
 840         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 841
 842         dprintk("RPC:       xs_close xprt %p\n", xprt);
 843
 844         cancel_delayed_work_sync(&transport->connect_worker);
 845
 846         xs_reset_transport(transport);
 847         xprt->reestablish_timeout = 0;
 848
 849         smp_mb__before_atomic();
 850         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 851         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 852         clear_bit(XPRT_CLOSING, &xprt->state);
 853         smp_mb__after_atomic();
 854         xprt_disconnect_done(xprt);
 855 }
 856
 857 static void xs_tcp_close(struct rpc_xprt *xprt)
 858 {
 859         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 860                 xs_close(xprt);
 861         else
 862                 xs_tcp_shutdown(xprt);
 863 }
 864
 865 static void xs_xprt_free(struct rpc_xprt *xprt)
 866 {
 867         xs_free_peer_addresses(xprt);
 868         xprt_free(xprt);
 869 }
 870
 871 /**
 872  * xs_destroy - prepare to shutdown a transport
 873  * @xprt: doomed transport
 874  *
 875  */
 876 static void xs_destroy(struct rpc_xprt *xprt)
 877 {
 878         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 879
 880         xs_close(xprt);
 881         xs_xprt_free(xprt);
 882         module_put(THIS_MODULE);
 883 }
 884
 885 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 886 {
 887         struct xdr_skb_reader desc = {
 888                 .skb            = skb,
 889                 .offset         = sizeof(rpc_fraghdr),
 890                 .count          = skb->len - sizeof(rpc_fraghdr),
 891         };
 892
 893         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 894                 return -1;
 895         if (desc.count)
 896                 return -1;
 897         return 0;
 898 }
 899
 900 /**
 901  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 902  * @sk: socket with data to read
 903  * @len: how much data to read
 904  *
 905  * Currently this assumes we can read the whole reply in a single gulp.
 906  */
 907 static void xs_local_data_ready(struct sock *sk)
 908 {
 909         struct rpc_task *task;
 910         struct rpc_xprt *xprt;
 911         struct rpc_rqst *rovr;
 912         struct sk_buff *skb;
 913         int err, repsize, copied;
 914         u32 _xid;
 915         __be32 *xp;
 916
 917         read_lock_bh(&sk->sk_callback_lock);
 918         dprintk("RPC:       %s...\n", __func__);
 919         xprt = xprt_from_sock(sk);
 920         if (xprt == NULL)
 921                 goto out;
 922
 923         skb = skb_recv_datagram(sk, 0, 1, &err);
 924         if (skb == NULL)
 925                 goto out;
 926
 927         repsize = skb->len - sizeof(rpc_fraghdr);
 928         if (repsize < 4) {
 929                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 930                 goto dropit;
 931         }
 932
 933         /* Copy the XID from the skb... */
 934         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 935         if (xp == NULL)
 936                 goto dropit;
 937
 938         /* Look up and lock the request corresponding to the given XID */
 939         spin_lock(&xprt->transport_lock);
 940         rovr = xprt_lookup_rqst(xprt, *xp);
 941         if (!rovr)
 942                 goto out_unlock;
 943         task = rovr->rq_task;
 944
 945         copied = rovr->rq_private_buf.buflen;
 946         if (copied > repsize)
 947                 copied = repsize;
 948
 949         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 950                 dprintk("RPC:       sk_buff copy failed\n");
 951                 goto out_unlock;
 952         }
 953
 954         xprt_complete_rqst(task, copied);
 955
 956  out_unlock:
 957         spin_unlock(&xprt->transport_lock);
 958  dropit:
 959         skb_free_datagram(sk, skb);
 960  out:
 961         read_unlock_bh(&sk->sk_callback_lock);
 962 }
 963
 964 /**
 965  * xs_udp_data_ready - "data ready" callback for UDP sockets
 966  * @sk: socket with data to read
 967  * @len: how much data to read
 968  *
 969  */
 970 static void xs_udp_data_ready(struct sock *sk)
 971 {
 972         struct rpc_task *task;
 973         struct rpc_xprt *xprt;
 974         struct rpc_rqst *rovr;
 975         struct sk_buff *skb;
 976         int err, repsize, copied;
 977         u32 _xid;
 978         __be32 *xp;
 979
 980         read_lock_bh(&sk->sk_callback_lock);
 981         dprintk("RPC:       xs_udp_data_ready...\n");
 982         if (!(xprt = xprt_from_sock(sk)))
 983                 goto out;
 984
 985         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 986                 goto out;
 987
 988         repsize = skb->len - sizeof(struct udphdr);
 989         if (repsize < 4) {
 990                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 991                 goto dropit;
 992         }
 993
 994         /* Copy the XID from the skb... */
 995         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 996                                 sizeof(_xid), &_xid);
 997         if (xp == NULL)
 998                 goto dropit;
 999
1000         /* Look up and lock the request corresponding to the given XID */
1001         spin_lock(&xprt->transport_lock);
1002         rovr = xprt_lookup_rqst(xprt, *xp);
1003         if (!rovr)
1004                 goto out_unlock;
1005         task = rovr->rq_task;
1006
1007         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1008                 copied = repsize;
1009
1010         /* Suck it into the iovec, verify checksum if not done by hw. */
1011         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1012                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1013                 goto out_unlock;
1014         }
1015
1016         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1017
1018         xprt_adjust_cwnd(xprt, task, copied);
1019         xprt_complete_rqst(task, copied);
1020
1021  out_unlock:
1022         spin_unlock(&xprt->transport_lock);
1023  dropit:
1024         skb_free_datagram(sk, skb);
1025  out:
1026         read_unlock_bh(&sk->sk_callback_lock);
1027 }
1028
1029 /*
1030  * Helper function to force a TCP close if the server is sending
1031  * junk and/or it has put us in CLOSE_WAIT
1032  */
1033 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1034 {
1035         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1036         xprt_force_disconnect(xprt);
1037 }
1038
1039 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1040 {
1041         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1042         size_t len, used;
1043         char *p;
1044
1045         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1046         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1047         used = xdr_skb_read_bits(desc, p, len);
1048         transport->tcp_offset += used;
1049         if (used != len)
1050                 return;
1051
1052         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1053         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1054                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1055         else
1056                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1057         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1058
1059         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1060         transport->tcp_offset = 0;
1061
1062         /* Sanity check of the record length */
1063         if (unlikely(transport->tcp_reclen < 8)) {
1064                 dprintk("RPC:       invalid TCP record fragment length\n");
1065                 xs_tcp_force_close(xprt);
1066                 return;
1067         }
1068         dprintk("RPC:       reading TCP record fragment of length %d\n",
1069                         transport->tcp_reclen);
1070 }
1071
1072 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1073 {
1074         if (transport->tcp_offset == transport->tcp_reclen) {
1075                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1076                 transport->tcp_offset = 0;
1077                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1078                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1079                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1080                         transport->tcp_copied = 0;
1081                 }
1082         }
1083 }
1084
1085 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1086 {
1087         size_t len, used;
1088         char *p;
1089
1090         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1091         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1092         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1093         used = xdr_skb_read_bits(desc, p, len);
1094         transport->tcp_offset += used;
1095         if (used != len)
1096                 return;
1097         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1098         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1099         transport->tcp_copied = 4;
1100         dprintk("RPC:       reading %s XID %08x\n",
1101                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1102                                                               : "request with",
1103                         ntohl(transport->tcp_xid));
1104         xs_tcp_check_fraghdr(transport);
1105 }
1106
1107 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1108                                        struct xdr_skb_reader *desc)
1109 {
1110         size_t len, used;
1111         u32 offset;
1112         char *p;
1113
1114         /*
1115          * We want transport->tcp_offset to be 8 at the end of this routine
1116          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1117          * When this function is called for the first time,
1118          * transport->tcp_offset is 4 (after having already read the xid).
1119          */
1120         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1121         len = sizeof(transport->tcp_calldir) - offset;
1122         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1123         p = ((char *) &transport->tcp_calldir) + offset;
1124         used = xdr_skb_read_bits(desc, p, len);
1125         transport->tcp_offset += used;
1126         if (used != len)
1127                 return;
1128         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1129         /*
1130          * We don't yet have the XDR buffer, so we will write the calldir
1131          * out after we get the buffer from the 'struct rpc_rqst'
1132          */
1133         switch (ntohl(transport->tcp_calldir)) {
1134         case RPC_REPLY:
1135                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1136                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1137                 transport->tcp_flags |= TCP_RPC_REPLY;
1138                 break;
1139         case RPC_CALL:
1140                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1141                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1142                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1143                 break;
1144         default:
1145                 dprintk("RPC:       invalid request message type\n");
1146                 xs_tcp_force_close(&transport->xprt);
1147         }
1148         xs_tcp_check_fraghdr(transport);
1149 }
1150
1151 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1152                                      struct xdr_skb_reader *desc,
1153                                      struct rpc_rqst *req)
1154 {
1155         struct sock_xprt *transport =
1156                                 container_of(xprt, struct sock_xprt, xprt);
1157         struct xdr_buf *rcvbuf;
1158         size_t len;
1159         ssize_t r;
1160
1161         rcvbuf = &req->rq_private_buf;
1162
1163         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1164                 /*
1165                  * Save the RPC direction in the XDR buffer
1166                  */
1167                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1168                         &transport->tcp_calldir,
1169                         sizeof(transport->tcp_calldir));
1170                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1171                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1172         }
1173
1174         len = desc->count;
1175         if (len > transport->tcp_reclen - transport->tcp_offset) {
1176                 struct xdr_skb_reader my_desc;
1177
1178                 len = transport->tcp_reclen - transport->tcp_offset;
1179                 memcpy(&my_desc, desc, sizeof(my_desc));
1180                 my_desc.count = len;
1181                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1182                                           &my_desc, xdr_skb_read_bits);
1183                 desc->count -= r;
1184                 desc->offset += r;
1185         } else
1186                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1187                                           desc, xdr_skb_read_bits);
1188
1189         if (r > 0) {
1190                 transport->tcp_copied += r;
1191                 transport->tcp_offset += r;
1192         }
1193         if (r != len) {
1194                 /* Error when copying to the receive buffer,
1195                  * usually because we weren't able to allocate
1196                  * additional buffer pages. All we can do now
1197                  * is turn off TCP_RCV_COPY_DATA, so the request
1198                  * will not receive any additional updates,
1199                  * and time out.
1200                  * Any remaining data from this record will
1201                  * be discarded.
1202                  */
1203                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1204                 dprintk("RPC:       XID %08x truncated request\n",
1205                                 ntohl(transport->tcp_xid));
1206                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1207                                 "tcp_offset = %u, tcp_reclen = %u\n",
1208                                 xprt, transport->tcp_copied,
1209                                 transport->tcp_offset, transport->tcp_reclen);
1210                 return;
1211         }
1212
1213         dprintk("RPC:       XID %08x read %Zd bytes\n",
1214                         ntohl(transport->tcp_xid), r);
1215         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1216                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1217                         transport->tcp_offset, transport->tcp_reclen);
1218
1219         if (transport->tcp_copied == req->rq_private_buf.buflen)
1220                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1221         else if (transport->tcp_offset == transport->tcp_reclen) {
1222                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1223                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1224         }
1225 }
1226
1227 /*
1228  * Finds the request corresponding to the RPC xid and invokes the common
1229  * tcp read code to read the data.
1230  */
1231 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1232                                     struct xdr_skb_reader *desc)
1233 {
1234         struct sock_xprt *transport =
1235                                 container_of(xprt, struct sock_xprt, xprt);
1236         struct rpc_rqst *req;
1237
1238         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1239
1240         /* Find and lock the request corresponding to this xid */
1241         spin_lock(&xprt->transport_lock);
1242         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1243         if (!req) {
1244                 dprintk("RPC:       XID %08x request not found!\n",
1245                                 ntohl(transport->tcp_xid));
1246                 spin_unlock(&xprt->transport_lock);
1247                 return -1;
1248         }
1249
1250         xs_tcp_read_common(xprt, desc, req);
1251
1252         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1253                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1254
1255         spin_unlock(&xprt->transport_lock);
1256         return 0;
1257 }
1258
1259 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1260 /*
1261  * Obtains an rpc_rqst previously allocated and invokes the common
1262  * tcp read code to read the data.  The result is placed in the callback
1263  * queue.
1264  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1265  * connection and return -1.
1266  */
1267 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1268                                        struct xdr_skb_reader *desc)
1269 {
1270         struct sock_xprt *transport =
1271                                 container_of(xprt, struct sock_xprt, xprt);
1272         struct rpc_rqst *req;
1273
1274         /* Look up and lock the request corresponding to the given XID */
1275         spin_lock(&xprt->transport_lock);
1276         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1277         if (req == NULL) {
1278                 spin_unlock(&xprt->transport_lock);
1279                 printk(KERN_WARNING "Callback slot table overflowed\n");
1280                 xprt_force_disconnect(xprt);
1281                 return -1;
1282         }
1283
1284         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1285         xs_tcp_read_common(xprt, desc, req);
1286
1287         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1288                 xprt_complete_bc_request(req, transport->tcp_copied);
1289         spin_unlock(&xprt->transport_lock);
1290
1291         return 0;
1292 }
1293
1294 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1295                                         struct xdr_skb_reader *desc)
1296 {
1297         struct sock_xprt *transport =
1298                                 container_of(xprt, struct sock_xprt, xprt);
1299
1300         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1301                 xs_tcp_read_reply(xprt, desc) :
1302                 xs_tcp_read_callback(xprt, desc);
1303 }
1304 #else
1305 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1306                                         struct xdr_skb_reader *desc)
1307 {
1308         return xs_tcp_read_reply(xprt, desc);
1309 }
1310 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1311
1312 /*
1313  * Read data off the transport.  This can be either an RPC_CALL or an
1314  * RPC_REPLY.  Relay the processing to helper functions.
1315  */
1316 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1317                                     struct xdr_skb_reader *desc)
1318 {
1319         struct sock_xprt *transport =
1320                                 container_of(xprt, struct sock_xprt, xprt);
1321
1322         if (_xs_tcp_read_data(xprt, desc) == 0)
1323                 xs_tcp_check_fraghdr(transport);
1324         else {
1325                 /*
1326                  * The transport_lock protects the request handling.
1327                  * There's no need to hold it to update the tcp_flags.
1328                  */
1329                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1330         }
1331 }
1332
1333 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1334 {
1335         size_t len;
1336
1337         len = transport->tcp_reclen - transport->tcp_offset;
1338         if (len > desc->count)
1339                 len = desc->count;
1340         desc->count -= len;
1341         desc->offset += len;
1342         transport->tcp_offset += len;
1343         dprintk("RPC:       discarded %Zu bytes\n", len);
1344         xs_tcp_check_fraghdr(transport);
1345 }
1346
1347 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1348 {
1349         struct rpc_xprt *xprt = rd_desc->arg.data;
1350         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1351         struct xdr_skb_reader desc = {
1352                 .skb    = skb,
1353                 .offset = offset,
1354                 .count  = len,
1355         };
1356
1357         dprintk("RPC:       xs_tcp_data_recv started\n");
1358         do {
1359                 trace_xs_tcp_data_recv(transport);
1360                 /* Read in a new fragment marker if necessary */
1361                 /* Can we ever really expect to get completely empty fragments? */
1362                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1363                         xs_tcp_read_fraghdr(xprt, &desc);
1364                         continue;
1365                 }
1366                 /* Read in the xid if necessary */
1367                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1368                         xs_tcp_read_xid(transport, &desc);
1369                         continue;
1370                 }
1371                 /* Read in the call/reply flag */
1372                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1373                         xs_tcp_read_calldir(transport, &desc);
1374                         continue;
1375                 }
1376                 /* Read in the request data */
1377                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1378                         xs_tcp_read_data(xprt, &desc);
1379                         continue;
1380                 }
1381                 /* Skip over any trailing bytes on short reads */
1382                 xs_tcp_read_discard(transport, &desc);
1383         } while (desc.count);
1384         trace_xs_tcp_data_recv(transport);
1385         dprintk("RPC:       xs_tcp_data_recv done\n");
1386         return len - desc.count;
1387 }
1388
1389 /**
1390  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1391  * @sk: socket with data to read
1392  * @bytes: how much data to read
1393  *
1394  */
1395 static void xs_tcp_data_ready(struct sock *sk)
1396 {
1397         struct rpc_xprt *xprt;
1398         read_descriptor_t rd_desc;
1399         int read;
1400         unsigned long total = 0;
1401
1402         dprintk("RPC:       xs_tcp_data_ready...\n");
1403
1404         read_lock_bh(&sk->sk_callback_lock);
1405         if (!(xprt = xprt_from_sock(sk))) {
1406                 read = 0;
1407                 goto out;
1408         }
1409         /* Any data means we had a useful conversation, so
1410          * the we don't need to delay the next reconnect
1411          */
1412         if (xprt->reestablish_timeout)
1413                 xprt->reestablish_timeout = 0;
1414
1415         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1416         rd_desc.arg.data = xprt;
1417         do {
1418                 rd_desc.count = 65536;
1419                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1420                 if (read > 0)
1421                         total += read;
1422         } while (read > 0);
1423 out:
1424         trace_xs_tcp_data_ready(xprt, read, total);
1425         read_unlock_bh(&sk->sk_callback_lock);
1426 }
1427
1428 /*
1429  * Do the equivalent of linger/linger2 handling for dealing with
1430  * broken servers that don't close the socket in a timely
1431  * fashion
1432  */
1433 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1434                 unsigned long timeout)
1435 {
1436         struct sock_xprt *transport;
1437
1438         if (xprt_test_and_set_connecting(xprt))
1439                 return;
1440         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1441         transport = container_of(xprt, struct sock_xprt, xprt);
1442         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1443                            timeout);
1444 }
1445
1446 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1447 {
1448         struct sock_xprt *transport;
1449
1450         transport = container_of(xprt, struct sock_xprt, xprt);
1451
1452         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1453             !cancel_delayed_work(&transport->connect_worker))
1454                 return;
1455         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1456         xprt_clear_connecting(xprt);
1457 }
1458
1459 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1460 {
1461         smp_mb__before_atomic();
1462         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1463         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1464         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1465         clear_bit(XPRT_CLOSING, &xprt->state);
1466         smp_mb__after_atomic();
1467 }
1468
1469 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1470 {
1471         xs_sock_reset_connection_flags(xprt);
1472         /* Mark transport as closed and wake up all pending tasks */
1473         xprt_disconnect_done(xprt);
1474 }
1475
1476 /**
1477  * xs_tcp_state_change - callback to handle TCP socket state changes
1478  * @sk: socket whose state has changed
1479  *
1480  */
1481 static void xs_tcp_state_change(struct sock *sk)
1482 {
1483         struct rpc_xprt *xprt;
1484
1485         read_lock_bh(&sk->sk_callback_lock);
1486         if (!(xprt = xprt_from_sock(sk)))
1487                 goto out;
1488         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1489         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1490                         sk->sk_state, xprt_connected(xprt),
1491                         sock_flag(sk, SOCK_DEAD),
1492                         sock_flag(sk, SOCK_ZAPPED),
1493                         sk->sk_shutdown);
1494
1495         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1496         switch (sk->sk_state) {
1497         case TCP_ESTABLISHED:
1498                 spin_lock(&xprt->transport_lock);
1499                 if (!xprt_test_and_set_connected(xprt)) {
1500                         struct sock_xprt *transport = container_of(xprt,
1501                                         struct sock_xprt, xprt);
1502
1503                         /* Reset TCP record info */
1504                         transport->tcp_offset = 0;
1505                         transport->tcp_reclen = 0;
1506                         transport->tcp_copied = 0;
1507                         transport->tcp_flags =
1508                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1509                         xprt->connect_cookie++;
1510
1511                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1512                 }
1513                 spin_unlock(&xprt->transport_lock);
1514                 break;
1515         case TCP_FIN_WAIT1:
1516                 /* The client initiated a shutdown of the socket */
1517                 xprt->connect_cookie++;
1518                 xprt->reestablish_timeout = 0;
1519                 set_bit(XPRT_CLOSING, &xprt->state);
1520                 smp_mb__before_atomic();
1521                 clear_bit(XPRT_CONNECTED, &xprt->state);
1522                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1523                 smp_mb__after_atomic();
1524                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1525                 break;
1526         case TCP_CLOSE_WAIT:
1527                 /* The server initiated a shutdown of the socket */
1528                 xprt->connect_cookie++;
1529                 clear_bit(XPRT_CONNECTED, &xprt->state);
1530                 xs_tcp_force_close(xprt);
1531         case TCP_CLOSING:
1532                 /*
1533                  * If the server closed down the connection, make sure that
1534                  * we back off before reconnecting
1535                  */
1536                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1537                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1538                 break;
1539         case TCP_LAST_ACK:
1540                 set_bit(XPRT_CLOSING, &xprt->state);
1541                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1542                 smp_mb__before_atomic();
1543                 clear_bit(XPRT_CONNECTED, &xprt->state);
1544                 smp_mb__after_atomic();
1545                 break;
1546         case TCP_CLOSE:
1547                 xs_tcp_cancel_linger_timeout(xprt);
1548                 xs_sock_mark_closed(xprt);
1549         }
1550  out:
1551         read_unlock_bh(&sk->sk_callback_lock);
1552 }
1553
1554 static void xs_write_space(struct sock *sk)
1555 {
1556         struct socket *sock;
1557         struct rpc_xprt *xprt;
1558
1559         if (unlikely(!(sock = sk->sk_socket)))
1560                 return;
1561         clear_bit(SOCK_NOSPACE, &sock->flags);
1562
1563         if (unlikely(!(xprt = xprt_from_sock(sk))))
1564                 return;
1565         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1566                 return;
1567
1568         xprt_write_space(xprt);
1569 }
1570
1571 /**
1572  * xs_udp_write_space - callback invoked when socket buffer space
1573  *                             becomes available
1574  * @sk: socket whose state has changed
1575  *
1576  * Called when more output buffer space is available for this socket.
1577  * We try not to wake our writers until they can make "significant"
1578  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1579  * with a bunch of small requests.
1580  */
1581 static void xs_udp_write_space(struct sock *sk)
1582 {
1583         read_lock_bh(&sk->sk_callback_lock);
1584
1585         /* from net/core/sock.c:sock_def_write_space */
1586         if (sock_writeable(sk))
1587                 xs_write_space(sk);
1588
1589         read_unlock_bh(&sk->sk_callback_lock);
1590 }
1591
1592 /**
1593  * xs_tcp_write_space - callback invoked when socket buffer space
1594  *                             becomes available
1595  * @sk: socket whose state has changed
1596  *
1597  * Called when more output buffer space is available for this socket.
1598  * We try not to wake our writers until they can make "significant"
1599  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1600  * with a bunch of small requests.
1601  */
1602 static void xs_tcp_write_space(struct sock *sk)
1603 {
1604         read_lock_bh(&sk->sk_callback_lock);
1605
1606         /* from net/core/stream.c:sk_stream_write_space */
1607         if (sk_stream_is_writeable(sk))
1608                 xs_write_space(sk);
1609
1610         read_unlock_bh(&sk->sk_callback_lock);
1611 }
1612
1613 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1614 {
1615         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1616         struct sock *sk = transport->inet;
1617
1618         if (transport->rcvsize) {
1619                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1620                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1621         }
1622         if (transport->sndsize) {
1623                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1624                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1625                 sk->sk_write_space(sk);
1626         }
1627 }
1628
1629 /**
1630  * xs_udp_set_buffer_size - set send and receive limits
1631  * @xprt: generic transport
1632  * @sndsize: requested size of send buffer, in bytes
1633  * @rcvsize: requested size of receive buffer, in bytes
1634  *
1635  * Set socket send and receive buffer size limits.
1636  */
1637 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1638 {
1639         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1640
1641         transport->sndsize = 0;
1642         if (sndsize)
1643                 transport->sndsize = sndsize + 1024;
1644         transport->rcvsize = 0;
1645         if (rcvsize)
1646                 transport->rcvsize = rcvsize + 1024;
1647
1648         xs_udp_do_set_buffer_size(xprt);
1649 }
1650
1651 /**
1652  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1653  * @task: task that timed out
1654  *
1655  * Adjust the congestion window after a retransmit timeout has occurred.
1656  */
1657 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1658 {
1659         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1660 }
1661
1662 static unsigned short xs_get_random_port(void)
1663 {
1664         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1665         unsigned short rand = (unsigned short) prandom_u32() % range;
1666         return rand + xprt_min_resvport;
1667 }
1668
1669 /**
1670  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1671  * @sock: socket
1672  *
1673  * Note that this function has to be called on all sockets that share the
1674  * same port, and it must be called before binding.
1675  */
1676 static void xs_sock_set_reuseport(struct socket *sock)
1677 {
1678         char opt = 1;
1679
1680         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt));
1681 }
1682
1683 static unsigned short xs_sock_getport(struct socket *sock)
1684 {
1685         struct sockaddr_storage buf;
1686         int buflen;
1687         unsigned short port = 0;
1688
1689         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1690                 goto out;
1691         switch (buf.ss_family) {
1692         case AF_INET6:
1693                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1694                 break;
1695         case AF_INET:
1696                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1697         }
1698 out:
1699         return port;
1700 }
1701
1702 /**
1703  * xs_set_port - reset the port number in the remote endpoint address
1704  * @xprt: generic transport
1705  * @port: new port number
1706  *
1707  */
1708 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1709 {
1710         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1711
1712         rpc_set_port(xs_addr(xprt), port);
1713         xs_update_peer_port(xprt);
1714 }
1715
1716 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1717 {
1718         if (transport->srcport == 0)
1719                 transport->srcport = xs_sock_getport(sock);
1720 }
1721
1722 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1723 {
1724         unsigned short port = transport->srcport;
1725
1726         if (port == 0 && transport->xprt.resvport)
1727                 port = xs_get_random_port();
1728         return port;
1729 }
1730
1731 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1732 {
1733         if (transport->srcport != 0)
1734                 transport->srcport = 0;
1735         if (!transport->xprt.resvport)
1736                 return 0;
1737         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1738                 return xprt_max_resvport;
1739         return --port;
1740 }
1741 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1742 {
1743         struct sockaddr_storage myaddr;
1744         int err, nloop = 0;
1745         unsigned short port = xs_get_srcport(transport);
1746         unsigned short last;
1747
1748         /*
1749          * If we are asking for any ephemeral port (i.e. port == 0 &&
1750          * transport->xprt.resvport == 0), don't bind.  Let the local
1751          * port selection happen implicitly when the socket is used
1752          * (for example at connect time).
1753          *
1754          * This ensures that we can continue to establish TCP
1755          * connections even when all local ephemeral ports are already
1756          * a part of some TCP connection.  This makes no difference
1757          * for UDP sockets, but also doens't harm them.
1758          *
1759          * If we're asking for any reserved port (i.e. port == 0 &&
1760          * transport->xprt.resvport == 1) xs_get_srcport above will
1761          * ensure that port is non-zero and we will bind as needed.
1762          */
1763         if (port == 0)
1764                 return 0;
1765
1766         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1767         do {
1768                 rpc_set_port((struct sockaddr *)&myaddr, port);
1769                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1770                                 transport->xprt.addrlen);
1771                 if (err == 0) {
1772                         transport->srcport = port;
1773                         break;
1774                 }
1775                 last = port;
1776                 port = xs_next_srcport(transport, port);
1777                 if (port > last)
1778                         nloop++;
1779         } while (err == -EADDRINUSE && nloop != 2);
1780
1781         if (myaddr.ss_family == AF_INET)
1782                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1783                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1784                                 port, err ? "failed" : "ok", err);
1785         else
1786                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1787                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1788                                 port, err ? "failed" : "ok", err);
1789         return err;
1790 }
1791
1792 /*
1793  * We don't support autobind on AF_LOCAL sockets
1794  */
1795 static void xs_local_rpcbind(struct rpc_task *task)
1796 {
1797         rcu_read_lock();
1798         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1799         rcu_read_unlock();
1800 }
1801
1802 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1803 {
1804 }
1805
1806 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1807 static struct lock_class_key xs_key[2];
1808 static struct lock_class_key xs_slock_key[2];
1809
1810 static inline void xs_reclassify_socketu(struct socket *sock)
1811 {
1812         struct sock *sk = sock->sk;
1813
1814         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1815                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1816 }
1817
1818 static inline void xs_reclassify_socket4(struct socket *sock)
1819 {
1820         struct sock *sk = sock->sk;
1821
1822         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1823                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1824 }
1825
1826 static inline void xs_reclassify_socket6(struct socket *sock)
1827 {
1828         struct sock *sk = sock->sk;
1829
1830         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1831                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1832 }
1833
1834 static inline void xs_reclassify_socket(int family, struct socket *sock)
1835 {
1836         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1837         if (sock_owned_by_user(sock->sk))
1838                 return;
1839
1840         switch (family) {
1841         case AF_LOCAL:
1842                 xs_reclassify_socketu(sock);
1843                 break;
1844         case AF_INET:
1845                 xs_reclassify_socket4(sock);
1846                 break;
1847         case AF_INET6:
1848                 xs_reclassify_socket6(sock);
1849                 break;
1850         }
1851 }
1852 #else
1853 static inline void xs_reclassify_socketu(struct socket *sock)
1854 {
1855 }
1856
1857 static inline void xs_reclassify_socket4(struct socket *sock)
1858 {
1859 }
1860
1861 static inline void xs_reclassify_socket6(struct socket *sock)
1862 {
1863 }
1864
1865 static inline void xs_reclassify_socket(int family, struct socket *sock)
1866 {
1867 }
1868 #endif
1869
1870 static void xs_dummy_setup_socket(struct work_struct *work)
1871 {
1872 }
1873
1874 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1875                 struct sock_xprt *transport, int family, int type,
1876                 int protocol, bool reuseport)
1877 {
1878         struct socket *sock;
1879         int err;
1880
1881         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1882         if (err < 0) {
1883                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1884                                 protocol, -err);
1885                 goto out;
1886         }
1887         xs_reclassify_socket(family, sock);
1888
1889         if (reuseport)
1890                 xs_sock_set_reuseport(sock);
1891
1892         err = xs_bind(transport, sock);
1893         if (err) {
1894                 sock_release(sock);
1895                 goto out;
1896         }
1897
1898         return sock;
1899 out:
1900         return ERR_PTR(err);
1901 }
1902
1903 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1904                                       struct socket *sock)
1905 {
1906         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1907                                                                         xprt);
1908
1909         if (!transport->inet) {
1910                 struct sock *sk = sock->sk;
1911
1912                 write_lock_bh(&sk->sk_callback_lock);
1913
1914                 xs_save_old_callbacks(transport, sk);
1915
1916                 sk->sk_user_data = xprt;
1917                 sk->sk_data_ready = xs_local_data_ready;
1918                 sk->sk_write_space = xs_udp_write_space;
1919                 sk->sk_error_report = xs_error_report;
1920                 sk->sk_allocation = GFP_ATOMIC;
1921
1922                 xprt_clear_connected(xprt);
1923
1924                 /* Reset to new socket */
1925                 transport->sock = sock;
1926                 transport->inet = sk;
1927
1928                 write_unlock_bh(&sk->sk_callback_lock);
1929         }
1930
1931         /* Tell the socket layer to start connecting... */
1932         xprt->stat.connect_count++;
1933         xprt->stat.connect_start = jiffies;
1934         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1935 }
1936
1937 /**
1938  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1939  * @xprt: RPC transport to connect
1940  * @transport: socket transport to connect
1941  * @create_sock: function to create a socket of the correct type
1942  */
1943 static int xs_local_setup_socket(struct sock_xprt *transport)
1944 {
1945         struct rpc_xprt *xprt = &transport->xprt;
1946         struct socket *sock;
1947         int status = -EIO;
1948
1949         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1950         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1951                                         SOCK_STREAM, 0, &sock, 1);
1952         if (status < 0) {
1953                 dprintk("RPC:       can't create AF_LOCAL "
1954                         "transport socket (%d).\n", -status);
1955                 goto out;
1956         }
1957         xs_reclassify_socketu(sock);
1958
1959         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1960                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1961
1962         status = xs_local_finish_connecting(xprt, sock);
1963         trace_rpc_socket_connect(xprt, sock, status);
1964         switch (status) {
1965         case 0:
1966                 dprintk("RPC:       xprt %p connected to %s\n",
1967                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1968                 xprt_set_connected(xprt);
1969         case -ENOBUFS:
1970                 break;
1971         case -ENOENT:
1972                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1973                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1974                 break;
1975         case -ECONNREFUSED:
1976                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1977                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1978                 break;
1979         default:
1980                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1981                                 __func__, -status,
1982                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1983         }
1984
1985 out:
1986         xprt_clear_connecting(xprt);
1987         xprt_wake_pending_tasks(xprt, status);
1988         return status;
1989 }
1990
1991 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1992 {
1993         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1994         int ret;
1995
1996          if (RPC_IS_ASYNC(task)) {
1997                 /*
1998                  * We want the AF_LOCAL connect to be resolved in the
1999                  * filesystem namespace of the process making the rpc
2000                  * call.  Thus we connect synchronously.
2001                  *
2002                  * If we want to support asynchronous AF_LOCAL calls,
2003                  * we'll need to figure out how to pass a namespace to
2004                  * connect.
2005                  */
2006                 rpc_exit(task, -ENOTCONN);
2007                 return;
2008         }
2009         ret = xs_local_setup_socket(transport);
2010         if (ret && !RPC_IS_SOFTCONN(task))
2011                 msleep_interruptible(15000);
2012 }
2013
2014 #ifdef CONFIG_SUNRPC_SWAP
2015 static void xs_set_memalloc(struct rpc_xprt *xprt)
2016 {
2017         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2018                         xprt);
2019
2020         if (xprt->swapper)
2021                 sk_set_memalloc(transport->inet);
2022 }
2023
2024 /**
2025  * xs_swapper - Tag this transport as being used for swap.
2026  * @xprt: transport to tag
2027  * @enable: enable/disable
2028  *
2029  */
2030 int xs_swapper(struct rpc_xprt *xprt, int enable)
2031 {
2032         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2033                         xprt);
2034         int err = 0;
2035
2036         if (enable) {
2037                 xprt->swapper++;
2038                 xs_set_memalloc(xprt);
2039         } else if (xprt->swapper) {
2040                 xprt->swapper--;
2041                 sk_clear_memalloc(transport->inet);
2042         }
2043
2044         return err;
2045 }
2046 EXPORT_SYMBOL_GPL(xs_swapper);
2047 #else
2048 static void xs_set_memalloc(struct rpc_xprt *xprt)
2049 {
2050 }
2051 #endif
2052
2053 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2054 {
2055         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2056
2057         if (!transport->inet) {
2058                 struct sock *sk = sock->sk;
2059
2060                 write_lock_bh(&sk->sk_callback_lock);
2061
2062                 xs_save_old_callbacks(transport, sk);
2063
2064                 sk->sk_user_data = xprt;
2065                 sk->sk_data_ready = xs_udp_data_ready;
2066                 sk->sk_write_space = xs_udp_write_space;
2067                 sk->sk_allocation = GFP_ATOMIC;
2068
2069                 xprt_set_connected(xprt);
2070
2071                 /* Reset to new socket */
2072                 transport->sock = sock;
2073                 transport->inet = sk;
2074
2075                 xs_set_memalloc(xprt);
2076
2077                 write_unlock_bh(&sk->sk_callback_lock);
2078         }
2079         xs_udp_do_set_buffer_size(xprt);
2080 }
2081
2082 static void xs_udp_setup_socket(struct work_struct *work)
2083 {
2084         struct sock_xprt *transport =
2085                 container_of(work, struct sock_xprt, connect_worker.work);
2086         struct rpc_xprt *xprt = &transport->xprt;
2087         struct socket *sock = transport->sock;
2088         int status = -EIO;
2089
2090         /* Start by resetting any existing state */
2091         xs_reset_transport(transport);
2092         sock = xs_create_sock(xprt, transport,
2093                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2094                         IPPROTO_UDP, false);
2095         if (IS_ERR(sock))
2096                 goto out;
2097
2098         dprintk("RPC:       worker connecting xprt %p via %s to "
2099                                 "%s (port %s)\n", xprt,
2100                         xprt->address_strings[RPC_DISPLAY_PROTO],
2101                         xprt->address_strings[RPC_DISPLAY_ADDR],
2102                         xprt->address_strings[RPC_DISPLAY_PORT]);
2103
2104         xs_udp_finish_connecting(xprt, sock);
2105         trace_rpc_socket_connect(xprt, sock, 0);
2106         status = 0;
2107 out:
2108         xprt_clear_connecting(xprt);
2109         xprt_wake_pending_tasks(xprt, status);
2110 }
2111
2112 /*
2113  * We need to preserve the port number so the reply cache on the server can
2114  * find our cached RPC replies when we get around to reconnecting.
2115  */
2116 static void xs_abort_connection(struct sock_xprt *transport)
2117 {
2118         int result;
2119         struct sockaddr any;
2120
2121         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2122
2123         /*
2124          * Disconnect the transport socket by doing a connect operation
2125          * with AF_UNSPEC.  This should return immediately...
2126          */
2127         memset(&any, 0, sizeof(any));
2128         any.sa_family = AF_UNSPEC;
2129         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2130         trace_rpc_socket_reset_connection(&transport->xprt,
2131                         transport->sock, result);
2132         if (!result)
2133                 xs_sock_reset_connection_flags(&transport->xprt);
2134         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2135 }
2136
2137 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2138 {
2139         unsigned int state = transport->inet->sk_state;
2140
2141         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2142                 /* we don't need to abort the connection if the socket
2143                  * hasn't undergone a shutdown
2144                  */
2145                 if (transport->inet->sk_shutdown == 0)
2146                         return;
2147                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2148                                 __func__, transport->inet->sk_shutdown);
2149         }
2150         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2151                 /* we don't need to abort the connection if the socket
2152                  * hasn't undergone a shutdown
2153                  */
2154                 if (transport->inet->sk_shutdown == 0)
2155                         return;
2156                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2157                                 "sk_shutdown set to %d\n",
2158                                 __func__, transport->inet->sk_shutdown);
2159         }
2160         xs_abort_connection(transport);
2161 }
2162
2163 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2164 {
2165         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2166         int ret = -ENOTCONN;
2167
2168         if (!transport->inet) {
2169                 struct sock *sk = sock->sk;
2170                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2171                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2172                 unsigned int opt_on = 1;
2173
2174                 /* TCP Keepalive options */
2175                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2176                                 (char *)&opt_on, sizeof(opt_on));
2177                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2178                                 (char *)&keepidle, sizeof(keepidle));
2179                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2180                                 (char *)&keepidle, sizeof(keepidle));
2181                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2182                                 (char *)&keepcnt, sizeof(keepcnt));
2183
2184                 write_lock_bh(&sk->sk_callback_lock);
2185
2186                 xs_save_old_callbacks(transport, sk);
2187
2188                 sk->sk_user_data = xprt;
2189                 sk->sk_data_ready = xs_tcp_data_ready;
2190                 sk->sk_state_change = xs_tcp_state_change;
2191                 sk->sk_write_space = xs_tcp_write_space;
2192                 sk->sk_error_report = xs_error_report;
2193                 sk->sk_allocation = GFP_ATOMIC;
2194
2195                 /* socket options */
2196                 sock_reset_flag(sk, SOCK_LINGER);
2197                 tcp_sk(sk)->linger2 = 0;
2198                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2199
2200                 xprt_clear_connected(xprt);
2201
2202                 /* Reset to new socket */
2203                 transport->sock = sock;
2204                 transport->inet = sk;
2205
2206                 write_unlock_bh(&sk->sk_callback_lock);
2207         }
2208
2209         if (!xprt_bound(xprt))
2210                 goto out;
2211
2212         xs_set_memalloc(xprt);
2213
2214         /* Tell the socket layer to start connecting... */
2215         xprt->stat.connect_count++;
2216         xprt->stat.connect_start = jiffies;
2217         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2218         switch (ret) {
2219         case 0:
2220                 xs_set_srcport(transport, sock);
2221         case -EINPROGRESS:
2222                 /* SYN_SENT! */
2223                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2224                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2225         }
2226 out:
2227         return ret;
2228 }
2229
2230 /**
2231  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2232  * @xprt: RPC transport to connect
2233  * @transport: socket transport to connect
2234  * @create_sock: function to create a socket of the correct type
2235  *
2236  * Invoked by a work queue tasklet.
2237  */
2238 static void xs_tcp_setup_socket(struct work_struct *work)
2239 {
2240         struct sock_xprt *transport =
2241                 container_of(work, struct sock_xprt, connect_worker.work);
2242         struct socket *sock = transport->sock;
2243         struct rpc_xprt *xprt = &transport->xprt;
2244         int status = -EIO;
2245
2246         if (!sock) {
2247                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2248                 sock = xs_create_sock(xprt, transport,
2249                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2250                                 IPPROTO_TCP, true);
2251                 if (IS_ERR(sock)) {
2252                         status = PTR_ERR(sock);
2253                         goto out;
2254                 }
2255         } else {
2256                 int abort_and_exit;
2257
2258                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2259                                 &xprt->state);
2260                 /* "close" the socket, preserving the local port */
2261                 set_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2262                 xs_tcp_reuse_connection(transport);
2263                 clear_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2264
2265                 if (abort_and_exit)
2266                         goto out_eagain;
2267         }
2268
2269         dprintk("RPC:       worker connecting xprt %p via %s to "
2270                                 "%s (port %s)\n", xprt,
2271                         xprt->address_strings[RPC_DISPLAY_PROTO],
2272                         xprt->address_strings[RPC_DISPLAY_ADDR],
2273                         xprt->address_strings[RPC_DISPLAY_PORT]);
2274
2275         status = xs_tcp_finish_connecting(xprt, sock);
2276         trace_rpc_socket_connect(xprt, sock, status);
2277         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2278                         xprt, -status, xprt_connected(xprt),
2279                         sock->sk->sk_state);
2280         switch (status) {
2281         default:
2282                 printk("%s: connect returned unhandled error %d\n",
2283                         __func__, status);
2284         case -EADDRNOTAVAIL:
2285                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2286                  * and retry
2287                  */
2288                 xs_tcp_force_close(xprt);
2289                 break;
2290         case 0:
2291         case -EINPROGRESS:
2292         case -EALREADY:
2293                 xprt_clear_connecting(xprt);
2294                 return;
2295         case -EINVAL:
2296                 /* Happens, for instance, if the user specified a link
2297                  * local IPv6 address without a scope-id.
2298                  */
2299         case -ECONNREFUSED:
2300         case -ECONNRESET:
2301         case -ENETUNREACH:
2302         case -ENOBUFS:
2303                 /* retry with existing socket, after a delay */
2304                 goto out;
2305         }
2306 out_eagain:
2307         status = -EAGAIN;
2308 out:
2309         xprt_clear_connecting(xprt);
2310         xprt_wake_pending_tasks(xprt, status);
2311 }
2312
2313 /**
2314  * xs_connect - connect a socket to a remote endpoint
2315  * @xprt: pointer to transport structure
2316  * @task: address of RPC task that manages state of connect request
2317  *
2318  * TCP: If the remote end dropped the connection, delay reconnecting.
2319  *
2320  * UDP socket connects are synchronous, but we use a work queue anyway
2321  * to guarantee that even unprivileged user processes can set up a
2322  * socket on a privileged port.
2323  *
2324  * If a UDP socket connect fails, the delay behavior here prevents
2325  * retry floods (hard mounts).
2326  */
2327 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2328 {
2329         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2330
2331         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2332                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2333                                 "seconds\n",
2334                                 xprt, xprt->reestablish_timeout / HZ);
2335                 queue_delayed_work(rpciod_workqueue,
2336                                    &transport->connect_worker,
2337                                    xprt->reestablish_timeout);
2338                 xprt->reestablish_timeout <<= 1;
2339                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2340                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2341                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2342                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2343         } else {
2344                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2345                 queue_delayed_work(rpciod_workqueue,
2346                                    &transport->connect_worker, 0);
2347         }
2348 }
2349
2350 /**
2351  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2352  * @xprt: rpc_xprt struct containing statistics
2353  * @seq: output file
2354  *
2355  */
2356 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2357 {
2358         long idle_time = 0;
2359
2360         if (xprt_connected(xprt))
2361                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2362
2363         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2364                         "%llu %llu %lu %llu %llu\n",
2365                         xprt->stat.bind_count,
2366                         xprt->stat.connect_count,
2367                         xprt->stat.connect_time,
2368                         idle_time,
2369                         xprt->stat.sends,
2370                         xprt->stat.recvs,
2371                         xprt->stat.bad_xids,
2372                         xprt->stat.req_u,
2373                         xprt->stat.bklog_u,
2374                         xprt->stat.max_slots,
2375                         xprt->stat.sending_u,
2376                         xprt->stat.pending_u);
2377 }
2378
2379 /**
2380  * xs_udp_print_stats - display UDP socket-specifc stats
2381  * @xprt: rpc_xprt struct containing statistics
2382  * @seq: output file
2383  *
2384  */
2385 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2386 {
2387         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2388
2389         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2390                         "%lu %llu %llu\n",
2391                         transport->srcport,
2392                         xprt->stat.bind_count,
2393                         xprt->stat.sends,
2394                         xprt->stat.recvs,
2395                         xprt->stat.bad_xids,
2396                         xprt->stat.req_u,
2397                         xprt->stat.bklog_u,
2398                         xprt->stat.max_slots,
2399                         xprt->stat.sending_u,
2400                         xprt->stat.pending_u);
2401 }
2402
2403 /**
2404  * xs_tcp_print_stats - display TCP socket-specifc stats
2405  * @xprt: rpc_xprt struct containing statistics
2406  * @seq: output file
2407  *
2408  */
2409 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2410 {
2411         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2412         long idle_time = 0;
2413
2414         if (xprt_connected(xprt))
2415                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2416
2417         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2418                         "%llu %llu %lu %llu %llu\n",
2419                         transport->srcport,
2420                         xprt->stat.bind_count,
2421                         xprt->stat.connect_count,
2422                         xprt->stat.connect_time,
2423                         idle_time,
2424                         xprt->stat.sends,
2425                         xprt->stat.recvs,
2426                         xprt->stat.bad_xids,
2427                         xprt->stat.req_u,
2428                         xprt->stat.bklog_u,
2429                         xprt->stat.max_slots,
2430                         xprt->stat.sending_u,
2431                         xprt->stat.pending_u);
2432 }
2433
2434 /*
2435  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2436  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2437  * to use the server side send routines.
2438  */
2439 static void *bc_malloc(struct rpc_task *task, size_t size)
2440 {
2441         struct page *page;
2442         struct rpc_buffer *buf;
2443
2444         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2445         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2446                 return NULL;
2447
2448         page = alloc_page(GFP_KERNEL);
2449         if (!page)
2450                 return NULL;
2451
2452         buf = page_address(page);
2453         buf->len = PAGE_SIZE;
2454
2455         return buf->data;
2456 }
2457
2458 /*
2459  * Free the space allocated in the bc_alloc routine
2460  */
2461 static void bc_free(void *buffer)
2462 {
2463         struct rpc_buffer *buf;
2464
2465         if (!buffer)
2466                 return;
2467
2468         buf = container_of(buffer, struct rpc_buffer, data);
2469         free_page((unsigned long)buf);
2470 }
2471
2472 /*
2473  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2474  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2475  */
2476 static int bc_sendto(struct rpc_rqst *req)
2477 {
2478         int len;
2479         struct xdr_buf *xbufp = &req->rq_snd_buf;
2480         struct rpc_xprt *xprt = req->rq_xprt;
2481         struct sock_xprt *transport =
2482                                 container_of(xprt, struct sock_xprt, xprt);
2483         struct socket *sock = transport->sock;
2484         unsigned long headoff;
2485         unsigned long tailoff;
2486
2487         xs_encode_stream_record_marker(xbufp);
2488
2489         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2490         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2491         len = svc_send_common(sock, xbufp,
2492                               virt_to_page(xbufp->head[0].iov_base), headoff,
2493                               xbufp->tail[0].iov_base, tailoff);
2494
2495         if (len != xbufp->len) {
2496                 printk(KERN_NOTICE "Error sending entire callback!\n");
2497                 len = -EAGAIN;
2498         }
2499
2500         return len;
2501 }
2502
2503 /*
2504  * The send routine. Borrows from svc_send
2505  */
2506 static int bc_send_request(struct rpc_task *task)
2507 {
2508         struct rpc_rqst *req = task->tk_rqstp;
2509         struct svc_xprt *xprt;
2510         u32                     len;
2511
2512         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2513         /*
2514          * Get the server socket associated with this callback xprt
2515          */
2516         xprt = req->rq_xprt->bc_xprt;
2517
2518         /*
2519          * Grab the mutex to serialize data as the connection is shared
2520          * with the fore channel
2521          */
2522         if (!mutex_trylock(&xprt->xpt_mutex)) {
2523                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2524                 if (!mutex_trylock(&xprt->xpt_mutex))
2525                         return -EAGAIN;
2526                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2527         }
2528         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2529                 len = -ENOTCONN;
2530         else
2531                 len = bc_sendto(req);
2532         mutex_unlock(&xprt->xpt_mutex);
2533
2534         if (len > 0)
2535                 len = 0;
2536
2537         return len;
2538 }
2539
2540 /*
2541  * The close routine. Since this is client initiated, we do nothing
2542  */
2543
2544 static void bc_close(struct rpc_xprt *xprt)
2545 {
2546 }
2547
2548 /*
2549  * The xprt destroy routine. Again, because this connection is client
2550  * initiated, we do nothing
2551  */
2552
2553 static void bc_destroy(struct rpc_xprt *xprt)
2554 {
2555         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2556
2557         xs_xprt_free(xprt);
2558         module_put(THIS_MODULE);
2559 }
2560
2561 static struct rpc_xprt_ops xs_local_ops = {
2562         .reserve_xprt           = xprt_reserve_xprt,
2563         .release_xprt           = xs_tcp_release_xprt,
2564         .alloc_slot             = xprt_alloc_slot,
2565         .rpcbind                = xs_local_rpcbind,
2566         .set_port               = xs_local_set_port,
2567         .connect                = xs_local_connect,
2568         .buf_alloc              = rpc_malloc,
2569         .buf_free               = rpc_free,
2570         .send_request           = xs_local_send_request,
2571         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2572         .close                  = xs_close,
2573         .destroy                = xs_destroy,
2574         .print_stats            = xs_local_print_stats,
2575 };
2576
2577 static struct rpc_xprt_ops xs_udp_ops = {
2578         .set_buffer_size        = xs_udp_set_buffer_size,
2579         .reserve_xprt           = xprt_reserve_xprt_cong,
2580         .release_xprt           = xprt_release_xprt_cong,
2581         .alloc_slot             = xprt_alloc_slot,
2582         .rpcbind                = rpcb_getport_async,
2583         .set_port               = xs_set_port,
2584         .connect                = xs_connect,
2585         .buf_alloc              = rpc_malloc,
2586         .buf_free               = rpc_free,
2587         .send_request           = xs_udp_send_request,
2588         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2589         .timer                  = xs_udp_timer,
2590         .release_request        = xprt_release_rqst_cong,
2591         .close                  = xs_close,
2592         .destroy                = xs_destroy,
2593         .print_stats            = xs_udp_print_stats,
2594 };
2595
2596 static struct rpc_xprt_ops xs_tcp_ops = {
2597         .reserve_xprt           = xprt_reserve_xprt,
2598         .release_xprt           = xs_tcp_release_xprt,
2599         .alloc_slot             = xprt_lock_and_alloc_slot,
2600         .rpcbind                = rpcb_getport_async,
2601         .set_port               = xs_set_port,
2602         .connect                = xs_connect,
2603         .buf_alloc              = rpc_malloc,
2604         .buf_free               = rpc_free,
2605         .send_request           = xs_tcp_send_request,
2606         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2607         .close                  = xs_tcp_close,
2608         .destroy                = xs_destroy,
2609         .print_stats            = xs_tcp_print_stats,
2610 };
2611
2612 /*
2613  * The rpc_xprt_ops for the server backchannel
2614  */
2615
2616 static struct rpc_xprt_ops bc_tcp_ops = {
2617         .reserve_xprt           = xprt_reserve_xprt,
2618         .release_xprt           = xprt_release_xprt,
2619         .alloc_slot             = xprt_alloc_slot,
2620         .buf_alloc              = bc_malloc,
2621         .buf_free               = bc_free,
2622         .send_request           = bc_send_request,
2623         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2624         .close                  = bc_close,
2625         .destroy                = bc_destroy,
2626         .print_stats            = xs_tcp_print_stats,
2627 };
2628
2629 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2630 {
2631         static const struct sockaddr_in sin = {
2632                 .sin_family             = AF_INET,
2633                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2634         };
2635         static const struct sockaddr_in6 sin6 = {
2636                 .sin6_family            = AF_INET6,
2637                 .sin6_addr              = IN6ADDR_ANY_INIT,
2638         };
2639
2640         switch (family) {
2641         case AF_LOCAL:
2642                 break;
2643         case AF_INET:
2644                 memcpy(sap, &sin, sizeof(sin));
2645                 break;
2646         case AF_INET6:
2647                 memcpy(sap, &sin6, sizeof(sin6));
2648                 break;
2649         default:
2650                 dprintk("RPC:       %s: Bad address family\n", __func__);
2651                 return -EAFNOSUPPORT;
2652         }
2653         return 0;
2654 }
2655
2656 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2657                                       unsigned int slot_table_size,
2658                                       unsigned int max_slot_table_size)
2659 {
2660         struct rpc_xprt *xprt;
2661         struct sock_xprt *new;
2662
2663         if (args->addrlen > sizeof(xprt->addr)) {
2664                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2665                 return ERR_PTR(-EBADF);
2666         }
2667
2668         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2669                         max_slot_table_size);
2670         if (xprt == NULL) {
2671                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2672                                 "rpc_xprt\n");
2673                 return ERR_PTR(-ENOMEM);
2674         }
2675
2676         new = container_of(xprt, struct sock_xprt, xprt);
2677         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2678         xprt->addrlen = args->addrlen;
2679         if (args->srcaddr)
2680                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2681         else {
2682                 int err;
2683                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2684                                         (struct sockaddr *)&new->srcaddr);
2685                 if (err != 0) {
2686                         xprt_free(xprt);
2687                         return ERR_PTR(err);
2688                 }
2689         }
2690
2691         return xprt;
2692 }
2693
2694 static const struct rpc_timeout xs_local_default_timeout = {
2695         .to_initval = 10 * HZ,
2696         .to_maxval = 10 * HZ,
2697         .to_retries = 2,
2698 };
2699
2700 /**
2701  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2702  * @args: rpc transport creation arguments
2703  *
2704  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2705  */
2706 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2707 {
2708         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2709         struct sock_xprt *transport;
2710         struct rpc_xprt *xprt;
2711         struct rpc_xprt *ret;
2712
2713         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2714                         xprt_max_tcp_slot_table_entries);
2715         if (IS_ERR(xprt))
2716                 return xprt;
2717         transport = container_of(xprt, struct sock_xprt, xprt);
2718
2719         xprt->prot = 0;
2720         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2721         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2722
2723         xprt->bind_timeout = XS_BIND_TO;
2724         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2725         xprt->idle_timeout = XS_IDLE_DISC_TO;
2726
2727         xprt->ops = &xs_local_ops;
2728         xprt->timeout = &xs_local_default_timeout;
2729
2730         INIT_DELAYED_WORK(&transport->connect_worker,
2731                         xs_dummy_setup_socket);
2732
2733         switch (sun->sun_family) {
2734         case AF_LOCAL:
2735                 if (sun->sun_path[0] != '/') {
2736                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2737                                         sun->sun_path);
2738                         ret = ERR_PTR(-EINVAL);
2739                         goto out_err;
2740                 }
2741                 xprt_set_bound(xprt);
2742                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2743                 ret = ERR_PTR(xs_local_setup_socket(transport));
2744                 if (ret)
2745                         goto out_err;
2746                 break;
2747         default:
2748                 ret = ERR_PTR(-EAFNOSUPPORT);
2749                 goto out_err;
2750         }
2751
2752         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2753                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2754
2755         if (try_module_get(THIS_MODULE))
2756                 return xprt;
2757         ret = ERR_PTR(-EINVAL);
2758 out_err:
2759         xs_xprt_free(xprt);
2760         return ret;
2761 }
2762
2763 static const struct rpc_timeout xs_udp_default_timeout = {
2764         .to_initval = 5 * HZ,
2765         .to_maxval = 30 * HZ,
2766         .to_increment = 5 * HZ,
2767         .to_retries = 5,
2768 };
2769
2770 /**
2771  * xs_setup_udp - Set up transport to use a UDP socket
2772  * @args: rpc transport creation arguments
2773  *
2774  */
2775 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2776 {
2777         struct sockaddr *addr = args->dstaddr;
2778         struct rpc_xprt *xprt;
2779         struct sock_xprt *transport;
2780         struct rpc_xprt *ret;
2781
2782         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2783                         xprt_udp_slot_table_entries);
2784         if (IS_ERR(xprt))
2785                 return xprt;
2786         transport = container_of(xprt, struct sock_xprt, xprt);
2787
2788         xprt->prot = IPPROTO_UDP;
2789         xprt->tsh_size = 0;
2790         /* XXX: header size can vary due to auth type, IPv6, etc. */
2791         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2792
2793         xprt->bind_timeout = XS_BIND_TO;
2794         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2795         xprt->idle_timeout = XS_IDLE_DISC_TO;
2796
2797         xprt->ops = &xs_udp_ops;
2798
2799         xprt->timeout = &xs_udp_default_timeout;
2800
2801         switch (addr->sa_family) {
2802         case AF_INET:
2803                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2804                         xprt_set_bound(xprt);
2805
2806                 INIT_DELAYED_WORK(&transport->connect_worker,
2807                                         xs_udp_setup_socket);
2808                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2809                 break;
2810         case AF_INET6:
2811                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2812                         xprt_set_bound(xprt);
2813
2814                 INIT_DELAYED_WORK(&transport->connect_worker,
2815                                         xs_udp_setup_socket);
2816                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2817                 break;
2818         default:
2819                 ret = ERR_PTR(-EAFNOSUPPORT);
2820                 goto out_err;
2821         }
2822
2823         if (xprt_bound(xprt))
2824                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2825                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2826                                 xprt->address_strings[RPC_DISPLAY_PORT],
2827                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2828         else
2829                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2830                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2831                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2832
2833         if (try_module_get(THIS_MODULE))
2834                 return xprt;
2835         ret = ERR_PTR(-EINVAL);
2836 out_err:
2837         xs_xprt_free(xprt);
2838         return ret;
2839 }
2840
2841 static const struct rpc_timeout xs_tcp_default_timeout = {
2842         .to_initval = 60 * HZ,
2843         .to_maxval = 60 * HZ,
2844         .to_retries = 2,
2845 };
2846
2847 /**
2848  * xs_setup_tcp - Set up transport to use a TCP socket
2849  * @args: rpc transport creation arguments
2850  *
2851  */
2852 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2853 {
2854         struct sockaddr *addr = args->dstaddr;
2855         struct rpc_xprt *xprt;
2856         struct sock_xprt *transport;
2857         struct rpc_xprt *ret;
2858         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2859
2860         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2861                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2862
2863         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2864                         max_slot_table_size);
2865         if (IS_ERR(xprt))
2866                 return xprt;
2867         transport = container_of(xprt, struct sock_xprt, xprt);
2868
2869         xprt->prot = IPPROTO_TCP;
2870         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2871         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2872
2873         xprt->bind_timeout = XS_BIND_TO;
2874         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2875         xprt->idle_timeout = XS_IDLE_DISC_TO;
2876
2877         xprt->ops = &xs_tcp_ops;
2878         xprt->timeout = &xs_tcp_default_timeout;
2879
2880         switch (addr->sa_family) {
2881         case AF_INET:
2882                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2883                         xprt_set_bound(xprt);
2884
2885                 INIT_DELAYED_WORK(&transport->connect_worker,
2886                                         xs_tcp_setup_socket);
2887                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2888                 break;
2889         case AF_INET6:
2890                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2891                         xprt_set_bound(xprt);
2892
2893                 INIT_DELAYED_WORK(&transport->connect_worker,
2894                                         xs_tcp_setup_socket);
2895                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2896                 break;
2897         default:
2898                 ret = ERR_PTR(-EAFNOSUPPORT);
2899                 goto out_err;
2900         }
2901
2902         if (xprt_bound(xprt))
2903                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2904                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2905                                 xprt->address_strings[RPC_DISPLAY_PORT],
2906                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2907         else
2908                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2909                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2910                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2911
2912         if (try_module_get(THIS_MODULE))
2913                 return xprt;
2914         ret = ERR_PTR(-EINVAL);
2915 out_err:
2916         xs_xprt_free(xprt);
2917         return ret;
2918 }
2919
2920 /**
2921  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2922  * @args: rpc transport creation arguments
2923  *
2924  */
2925 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2926 {
2927         struct sockaddr *addr = args->dstaddr;
2928         struct rpc_xprt *xprt;
2929         struct sock_xprt *transport;
2930         struct svc_sock *bc_sock;
2931         struct rpc_xprt *ret;
2932
2933         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2934                         xprt_tcp_slot_table_entries);
2935         if (IS_ERR(xprt))
2936                 return xprt;
2937         transport = container_of(xprt, struct sock_xprt, xprt);
2938
2939         xprt->prot = IPPROTO_TCP;
2940         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2941         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2942         xprt->timeout = &xs_tcp_default_timeout;
2943
2944         /* backchannel */
2945         xprt_set_bound(xprt);
2946         xprt->bind_timeout = 0;
2947         xprt->reestablish_timeout = 0;
2948         xprt->idle_timeout = 0;
2949
2950         xprt->ops = &bc_tcp_ops;
2951
2952         switch (addr->sa_family) {
2953         case AF_INET:
2954                 xs_format_peer_addresses(xprt, "tcp",
2955                                          RPCBIND_NETID_TCP);
2956                 break;
2957         case AF_INET6:
2958                 xs_format_peer_addresses(xprt, "tcp",
2959                                    RPCBIND_NETID_TCP6);
2960                 break;
2961         default:
2962                 ret = ERR_PTR(-EAFNOSUPPORT);
2963                 goto out_err;
2964         }
2965
2966         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2967                         xprt->address_strings[RPC_DISPLAY_ADDR],
2968                         xprt->address_strings[RPC_DISPLAY_PORT],
2969                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2970
2971         /*
2972          * Once we've associated a backchannel xprt with a connection,
2973          * we want to keep it around as long as the connection lasts,
2974          * in case we need to start using it for a backchannel again;
2975          * this reference won't be dropped until bc_xprt is destroyed.
2976          */
2977         xprt_get(xprt);
2978         args->bc_xprt->xpt_bc_xprt = xprt;
2979         xprt->bc_xprt = args->bc_xprt;
2980         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2981         transport->sock = bc_sock->sk_sock;
2982         transport->inet = bc_sock->sk_sk;
2983
2984         /*
2985          * Since we don't want connections for the backchannel, we set
2986          * the xprt status to connected
2987          */
2988         xprt_set_connected(xprt);
2989
2990         if (try_module_get(THIS_MODULE))
2991                 return xprt;
2992
2993         args->bc_xprt->xpt_bc_xprt = NULL;
2994         xprt_put(xprt);
2995         ret = ERR_PTR(-EINVAL);
2996 out_err:
2997         xs_xprt_free(xprt);
2998         return ret;
2999 }
3000
3001 static struct xprt_class        xs_local_transport = {
3002         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3003         .name           = "named UNIX socket",
3004         .owner          = THIS_MODULE,
3005         .ident          = XPRT_TRANSPORT_LOCAL,
3006         .setup          = xs_setup_local,
3007 };
3008
3009 static struct xprt_class        xs_udp_transport = {
3010         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3011         .name           = "udp",
3012         .owner          = THIS_MODULE,
3013         .ident          = XPRT_TRANSPORT_UDP,
3014         .setup          = xs_setup_udp,
3015 };
3016
3017 static struct xprt_class        xs_tcp_transport = {
3018         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3019         .name           = "tcp",
3020         .owner          = THIS_MODULE,
3021         .ident          = XPRT_TRANSPORT_TCP,
3022         .setup          = xs_setup_tcp,
3023 };
3024
3025 static struct xprt_class        xs_bc_tcp_transport = {
3026         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3027         .name           = "tcp NFSv4.1 backchannel",
3028         .owner          = THIS_MODULE,
3029         .ident          = XPRT_TRANSPORT_BC_TCP,
3030         .setup          = xs_setup_bc_tcp,
3031 };
3032
3033 /**
3034  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3035  *
3036  */
3037 int init_socket_xprt(void)
3038 {
3039 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3040         if (!sunrpc_table_header)
3041                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3042 #endif
3043
3044         xprt_register_transport(&xs_local_transport);
3045         xprt_register_transport(&xs_udp_transport);
3046         xprt_register_transport(&xs_tcp_transport);
3047         xprt_register_transport(&xs_bc_tcp_transport);
3048
3049         return 0;
3050 }
3051
3052 /**
3053  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3054  *
3055  */
3056 void cleanup_socket_xprt(void)
3057 {
3058 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3059         if (sunrpc_table_header) {
3060                 unregister_sysctl_table(sunrpc_table_header);
3061                 sunrpc_table_header = NULL;
3062         }
3063 #endif
3064
3065         xprt_unregister_transport(&xs_local_transport);
3066         xprt_unregister_transport(&xs_udp_transport);
3067         xprt_unregister_transport(&xs_tcp_transport);
3068         xprt_unregister_transport(&xs_bc_tcp_transport);
3069 }
3070
3071 static int param_set_uint_minmax(const char *val,
3072                 const struct kernel_param *kp,
3073                 unsigned int min, unsigned int max)
3074 {
3075         unsigned int num;
3076         int ret;
3077
3078         if (!val)
3079                 return -EINVAL;
3080         ret = kstrtouint(val, 0, &num);
3081         if (ret == -EINVAL || num < min || num > max)
3082                 return -EINVAL;
3083         *((unsigned int *)kp->arg) = num;
3084         return 0;
3085 }
3086
3087 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3088 {
3089         return param_set_uint_minmax(val, kp,
3090                         RPC_MIN_RESVPORT,
3091                         RPC_MAX_RESVPORT);
3092 }
3093
3094 static struct kernel_param_ops param_ops_portnr = {
3095         .set = param_set_portnr,
3096         .get = param_get_uint,
3097 };
3098
3099 #define param_check_portnr(name, p) \
3100         __param_check(name, p, unsigned int);
3101
3102 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3103 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3104
3105 static int param_set_slot_table_size(const char *val,
3106                                      const struct kernel_param *kp)
3107 {
3108         return param_set_uint_minmax(val, kp,
3109                         RPC_MIN_SLOT_TABLE,
3110                         RPC_MAX_SLOT_TABLE);
3111 }
3112
3113 static struct kernel_param_ops param_ops_slot_table_size = {
3114         .set = param_set_slot_table_size,
3115         .get = param_get_uint,
3116 };
3117
3118 #define param_check_slot_table_size(name, p) \
3119         __param_check(name, p, unsigned int);
3120
3121 static int param_set_max_slot_table_size(const char *val,
3122                                      const struct kernel_param *kp)
3123 {
3124         return param_set_uint_minmax(val, kp,
3125                         RPC_MIN_SLOT_TABLE,
3126                         RPC_MAX_SLOT_TABLE_LIMIT);
3127 }
3128
3129 static struct kernel_param_ops param_ops_max_slot_table_size = {
3130         .set = param_set_max_slot_table_size,
3131         .get = param_get_uint,
3132 };
3133
3134 #define param_check_max_slot_table_size(name, p) \
3135         __param_check(name, p, unsigned int);
3136
3137 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3138                    slot_table_size, 0644);
3139 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3140                    max_slot_table_size, 0644);
3141 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3142                    slot_table_size, 0644);
3143