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 -EADDRINUSE:
 725         case -EPIPE:
 726                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 727         }
 728
 729         return status;
 730 }
 731
 732 /**
 733  * xs_tcp_release_xprt - clean up after a tcp transmission
 734  * @xprt: transport
 735  * @task: rpc task
 736  *
 737  * This cleans up if an error causes us to abort the transmission of a request.
 738  * In this case, the socket may need to be reset in order to avoid confusing
 739  * the server.
 740  */
 741 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 742 {
 743         struct rpc_rqst *req;
 744
 745         if (task != xprt->snd_task)
 746                 return;
 747         if (task == NULL)
 748                 goto out_release;
 749         req = task->tk_rqstp;
 750         if (req == NULL)
 751                 goto out_release;
 752         if (req->rq_bytes_sent == 0)
 753                 goto out_release;
 754         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 755                 goto out_release;
 756         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 757 out_release:
 758         xprt_release_xprt(xprt, task);
 759 }
 760
 761 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 762 {
 763         transport->old_data_ready = sk->sk_data_ready;
 764         transport->old_state_change = sk->sk_state_change;
 765         transport->old_write_space = sk->sk_write_space;
 766         transport->old_error_report = sk->sk_error_report;
 767 }
 768
 769 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 770 {
 771         sk->sk_data_ready = transport->old_data_ready;
 772         sk->sk_state_change = transport->old_state_change;
 773         sk->sk_write_space = transport->old_write_space;
 774         sk->sk_error_report = transport->old_error_report;
 775 }
 776
 777 /**
 778  * xs_error_report - callback to handle TCP socket state errors
 779  * @sk: socket
 780  *
 781  * Note: we don't call sock_error() since there may be a rpc_task
 782  * using the socket, and so we don't want to clear sk->sk_err.
 783  */
 784 static void xs_error_report(struct sock *sk)
 785 {
 786         struct rpc_xprt *xprt;
 787         int err;
 788
 789         read_lock_bh(&sk->sk_callback_lock);
 790         if (!(xprt = xprt_from_sock(sk)))
 791                 goto out;
 792
 793         err = -sk->sk_err;
 794         if (err == 0)
 795                 goto out;
 796         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 797                         xprt, -err);
 798         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 799         if (test_bit(XPRT_CONNECTION_REUSE, &xprt->state))
 800                 goto out;
 801         xprt_wake_pending_tasks(xprt, err);
 802  out:
 803         read_unlock_bh(&sk->sk_callback_lock);
 804 }
 805
 806 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 807 {
 808         smp_mb__before_atomic();
 809         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 810         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
 811         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 812         clear_bit(XPRT_CLOSING, &xprt->state);
 813         smp_mb__after_atomic();
 814 }
 815
 816 static void xs_reset_transport(struct sock_xprt *transport)
 817 {
 818         struct socket *sock = transport->sock;
 819         struct sock *sk = transport->inet;
 820         struct rpc_xprt *xprt = &transport->xprt;
 821
 822         if (sk == NULL)
 823                 return;
 824
 825         write_lock_bh(&sk->sk_callback_lock);
 826         transport->inet = NULL;
 827         transport->sock = NULL;
 828
 829         sk->sk_user_data = NULL;
 830
 831         xs_restore_old_callbacks(transport, sk);
 832         write_unlock_bh(&sk->sk_callback_lock);
 833         xs_sock_reset_connection_flags(xprt);
 834
 835         trace_rpc_socket_close(xprt, sock);
 836         sock_release(sock);
 837 }
 838
 839 /**
 840  * xs_close - close a socket
 841  * @xprt: transport
 842  *
 843  * This is used when all requests are complete; ie, no DRC state remains
 844  * on the server we want to save.
 845  *
 846  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 847  * xs_reset_transport() zeroing the socket from underneath a writer.
 848  */
 849 static void xs_close(struct rpc_xprt *xprt)
 850 {
 851         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 852
 853         dprintk("RPC:       xs_close xprt %p\n", xprt);
 854
 855         xs_reset_transport(transport);
 856         xprt->reestablish_timeout = 0;
 857
 858         xprt_disconnect_done(xprt);
 859 }
 860
 861 static void xs_tcp_close(struct rpc_xprt *xprt)
 862 {
 863         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 864                 xs_close(xprt);
 865         else
 866                 xs_tcp_shutdown(xprt);
 867 }
 868
 869 static void xs_xprt_free(struct rpc_xprt *xprt)
 870 {
 871         xs_free_peer_addresses(xprt);
 872         xprt_free(xprt);
 873 }
 874
 875 /**
 876  * xs_destroy - prepare to shutdown a transport
 877  * @xprt: doomed transport
 878  *
 879  */
 880 static void xs_destroy(struct rpc_xprt *xprt)
 881 {
 882         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 883
 884         xs_close(xprt);
 885         xs_xprt_free(xprt);
 886         module_put(THIS_MODULE);
 887 }
 888
 889 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 890 {
 891         struct xdr_skb_reader desc = {
 892                 .skb            = skb,
 893                 .offset         = sizeof(rpc_fraghdr),
 894                 .count          = skb->len - sizeof(rpc_fraghdr),
 895         };
 896
 897         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 898                 return -1;
 899         if (desc.count)
 900                 return -1;
 901         return 0;
 902 }
 903
 904 /**
 905  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 906  * @sk: socket with data to read
 907  * @len: how much data to read
 908  *
 909  * Currently this assumes we can read the whole reply in a single gulp.
 910  */
 911 static void xs_local_data_ready(struct sock *sk)
 912 {
 913         struct rpc_task *task;
 914         struct rpc_xprt *xprt;
 915         struct rpc_rqst *rovr;
 916         struct sk_buff *skb;
 917         int err, repsize, copied;
 918         u32 _xid;
 919         __be32 *xp;
 920
 921         read_lock_bh(&sk->sk_callback_lock);
 922         dprintk("RPC:       %s...\n", __func__);
 923         xprt = xprt_from_sock(sk);
 924         if (xprt == NULL)
 925                 goto out;
 926
 927         skb = skb_recv_datagram(sk, 0, 1, &err);
 928         if (skb == NULL)
 929                 goto out;
 930
 931         repsize = skb->len - sizeof(rpc_fraghdr);
 932         if (repsize < 4) {
 933                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 934                 goto dropit;
 935         }
 936
 937         /* Copy the XID from the skb... */
 938         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 939         if (xp == NULL)
 940                 goto dropit;
 941
 942         /* Look up and lock the request corresponding to the given XID */
 943         spin_lock(&xprt->transport_lock);
 944         rovr = xprt_lookup_rqst(xprt, *xp);
 945         if (!rovr)
 946                 goto out_unlock;
 947         task = rovr->rq_task;
 948
 949         copied = rovr->rq_private_buf.buflen;
 950         if (copied > repsize)
 951                 copied = repsize;
 952
 953         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 954                 dprintk("RPC:       sk_buff copy failed\n");
 955                 goto out_unlock;
 956         }
 957
 958         xprt_complete_rqst(task, copied);
 959
 960  out_unlock:
 961         spin_unlock(&xprt->transport_lock);
 962  dropit:
 963         skb_free_datagram(sk, skb);
 964  out:
 965         read_unlock_bh(&sk->sk_callback_lock);
 966 }
 967
 968 /**
 969  * xs_udp_data_ready - "data ready" callback for UDP sockets
 970  * @sk: socket with data to read
 971  * @len: how much data to read
 972  *
 973  */
 974 static void xs_udp_data_ready(struct sock *sk)
 975 {
 976         struct rpc_task *task;
 977         struct rpc_xprt *xprt;
 978         struct rpc_rqst *rovr;
 979         struct sk_buff *skb;
 980         int err, repsize, copied;
 981         u32 _xid;
 982         __be32 *xp;
 983
 984         read_lock_bh(&sk->sk_callback_lock);
 985         dprintk("RPC:       xs_udp_data_ready...\n");
 986         if (!(xprt = xprt_from_sock(sk)))
 987                 goto out;
 988
 989         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 990                 goto out;
 991
 992         repsize = skb->len - sizeof(struct udphdr);
 993         if (repsize < 4) {
 994                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 995                 goto dropit;
 996         }
 997
 998         /* Copy the XID from the skb... */
 999         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1000                                 sizeof(_xid), &_xid);
1001         if (xp == NULL)
1002                 goto dropit;
1003
1004         /* Look up and lock the request corresponding to the given XID */
1005         spin_lock(&xprt->transport_lock);
1006         rovr = xprt_lookup_rqst(xprt, *xp);
1007         if (!rovr)
1008                 goto out_unlock;
1009         task = rovr->rq_task;
1010
1011         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1012                 copied = repsize;
1013
1014         /* Suck it into the iovec, verify checksum if not done by hw. */
1015         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1016                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1017                 goto out_unlock;
1018         }
1019
1020         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1021
1022         xprt_adjust_cwnd(xprt, task, copied);
1023         xprt_complete_rqst(task, copied);
1024
1025  out_unlock:
1026         spin_unlock(&xprt->transport_lock);
1027  dropit:
1028         skb_free_datagram(sk, skb);
1029  out:
1030         read_unlock_bh(&sk->sk_callback_lock);
1031 }
1032
1033 /*
1034  * Helper function to force a TCP close if the server is sending
1035  * junk and/or it has put us in CLOSE_WAIT
1036  */
1037 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1038 {
1039         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1040         xprt_force_disconnect(xprt);
1041 }
1042
1043 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1044 {
1045         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1046         size_t len, used;
1047         char *p;
1048
1049         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1050         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1051         used = xdr_skb_read_bits(desc, p, len);
1052         transport->tcp_offset += used;
1053         if (used != len)
1054                 return;
1055
1056         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1057         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1058                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1059         else
1060                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1061         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1062
1063         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1064         transport->tcp_offset = 0;
1065
1066         /* Sanity check of the record length */
1067         if (unlikely(transport->tcp_reclen < 8)) {
1068                 dprintk("RPC:       invalid TCP record fragment length\n");
1069                 xs_tcp_force_close(xprt);
1070                 return;
1071         }
1072         dprintk("RPC:       reading TCP record fragment of length %d\n",
1073                         transport->tcp_reclen);
1074 }
1075
1076 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1077 {
1078         if (transport->tcp_offset == transport->tcp_reclen) {
1079                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1080                 transport->tcp_offset = 0;
1081                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1082                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1083                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1084                         transport->tcp_copied = 0;
1085                 }
1086         }
1087 }
1088
1089 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1090 {
1091         size_t len, used;
1092         char *p;
1093
1094         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1095         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1096         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1097         used = xdr_skb_read_bits(desc, p, len);
1098         transport->tcp_offset += used;
1099         if (used != len)
1100                 return;
1101         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1102         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1103         transport->tcp_copied = 4;
1104         dprintk("RPC:       reading %s XID %08x\n",
1105                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1106                                                               : "request with",
1107                         ntohl(transport->tcp_xid));
1108         xs_tcp_check_fraghdr(transport);
1109 }
1110
1111 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1112                                        struct xdr_skb_reader *desc)
1113 {
1114         size_t len, used;
1115         u32 offset;
1116         char *p;
1117
1118         /*
1119          * We want transport->tcp_offset to be 8 at the end of this routine
1120          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1121          * When this function is called for the first time,
1122          * transport->tcp_offset is 4 (after having already read the xid).
1123          */
1124         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1125         len = sizeof(transport->tcp_calldir) - offset;
1126         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1127         p = ((char *) &transport->tcp_calldir) + offset;
1128         used = xdr_skb_read_bits(desc, p, len);
1129         transport->tcp_offset += used;
1130         if (used != len)
1131                 return;
1132         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1133         /*
1134          * We don't yet have the XDR buffer, so we will write the calldir
1135          * out after we get the buffer from the 'struct rpc_rqst'
1136          */
1137         switch (ntohl(transport->tcp_calldir)) {
1138         case RPC_REPLY:
1139                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1140                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1141                 transport->tcp_flags |= TCP_RPC_REPLY;
1142                 break;
1143         case RPC_CALL:
1144                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1145                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1146                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1147                 break;
1148         default:
1149                 dprintk("RPC:       invalid request message type\n");
1150                 xs_tcp_force_close(&transport->xprt);
1151         }
1152         xs_tcp_check_fraghdr(transport);
1153 }
1154
1155 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1156                                      struct xdr_skb_reader *desc,
1157                                      struct rpc_rqst *req)
1158 {
1159         struct sock_xprt *transport =
1160                                 container_of(xprt, struct sock_xprt, xprt);
1161         struct xdr_buf *rcvbuf;
1162         size_t len;
1163         ssize_t r;
1164
1165         rcvbuf = &req->rq_private_buf;
1166
1167         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1168                 /*
1169                  * Save the RPC direction in the XDR buffer
1170                  */
1171                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1172                         &transport->tcp_calldir,
1173                         sizeof(transport->tcp_calldir));
1174                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1175                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1176         }
1177
1178         len = desc->count;
1179         if (len > transport->tcp_reclen - transport->tcp_offset) {
1180                 struct xdr_skb_reader my_desc;
1181
1182                 len = transport->tcp_reclen - transport->tcp_offset;
1183                 memcpy(&my_desc, desc, sizeof(my_desc));
1184                 my_desc.count = len;
1185                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1186                                           &my_desc, xdr_skb_read_bits);
1187                 desc->count -= r;
1188                 desc->offset += r;
1189         } else
1190                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1191                                           desc, xdr_skb_read_bits);
1192
1193         if (r > 0) {
1194                 transport->tcp_copied += r;
1195                 transport->tcp_offset += r;
1196         }
1197         if (r != len) {
1198                 /* Error when copying to the receive buffer,
1199                  * usually because we weren't able to allocate
1200                  * additional buffer pages. All we can do now
1201                  * is turn off TCP_RCV_COPY_DATA, so the request
1202                  * will not receive any additional updates,
1203                  * and time out.
1204                  * Any remaining data from this record will
1205                  * be discarded.
1206                  */
1207                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1208                 dprintk("RPC:       XID %08x truncated request\n",
1209                                 ntohl(transport->tcp_xid));
1210                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1211                                 "tcp_offset = %u, tcp_reclen = %u\n",
1212                                 xprt, transport->tcp_copied,
1213                                 transport->tcp_offset, transport->tcp_reclen);
1214                 return;
1215         }
1216
1217         dprintk("RPC:       XID %08x read %Zd bytes\n",
1218                         ntohl(transport->tcp_xid), r);
1219         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1220                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1221                         transport->tcp_offset, transport->tcp_reclen);
1222
1223         if (transport->tcp_copied == req->rq_private_buf.buflen)
1224                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1225         else if (transport->tcp_offset == transport->tcp_reclen) {
1226                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1227                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1228         }
1229 }
1230
1231 /*
1232  * Finds the request corresponding to the RPC xid and invokes the common
1233  * tcp read code to read the data.
1234  */
1235 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1236                                     struct xdr_skb_reader *desc)
1237 {
1238         struct sock_xprt *transport =
1239                                 container_of(xprt, struct sock_xprt, xprt);
1240         struct rpc_rqst *req;
1241
1242         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1243
1244         /* Find and lock the request corresponding to this xid */
1245         spin_lock(&xprt->transport_lock);
1246         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1247         if (!req) {
1248                 dprintk("RPC:       XID %08x request not found!\n",
1249                                 ntohl(transport->tcp_xid));
1250                 spin_unlock(&xprt->transport_lock);
1251                 return -1;
1252         }
1253
1254         xs_tcp_read_common(xprt, desc, req);
1255
1256         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1257                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1258
1259         spin_unlock(&xprt->transport_lock);
1260         return 0;
1261 }
1262
1263 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1264 /*
1265  * Obtains an rpc_rqst previously allocated and invokes the common
1266  * tcp read code to read the data.  The result is placed in the callback
1267  * queue.
1268  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1269  * connection and return -1.
1270  */
1271 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1272                                        struct xdr_skb_reader *desc)
1273 {
1274         struct sock_xprt *transport =
1275                                 container_of(xprt, struct sock_xprt, xprt);
1276         struct rpc_rqst *req;
1277
1278         /* Look up and lock the request corresponding to the given XID */
1279         spin_lock(&xprt->transport_lock);
1280         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1281         if (req == NULL) {
1282                 spin_unlock(&xprt->transport_lock);
1283                 printk(KERN_WARNING "Callback slot table overflowed\n");
1284                 xprt_force_disconnect(xprt);
1285                 return -1;
1286         }
1287
1288         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1289         xs_tcp_read_common(xprt, desc, req);
1290
1291         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1292                 xprt_complete_bc_request(req, transport->tcp_copied);
1293         spin_unlock(&xprt->transport_lock);
1294
1295         return 0;
1296 }
1297
1298 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1299                                         struct xdr_skb_reader *desc)
1300 {
1301         struct sock_xprt *transport =
1302                                 container_of(xprt, struct sock_xprt, xprt);
1303
1304         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1305                 xs_tcp_read_reply(xprt, desc) :
1306                 xs_tcp_read_callback(xprt, desc);
1307 }
1308 #else
1309 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1310                                         struct xdr_skb_reader *desc)
1311 {
1312         return xs_tcp_read_reply(xprt, desc);
1313 }
1314 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1315
1316 /*
1317  * Read data off the transport.  This can be either an RPC_CALL or an
1318  * RPC_REPLY.  Relay the processing to helper functions.
1319  */
1320 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1321                                     struct xdr_skb_reader *desc)
1322 {
1323         struct sock_xprt *transport =
1324                                 container_of(xprt, struct sock_xprt, xprt);
1325
1326         if (_xs_tcp_read_data(xprt, desc) == 0)
1327                 xs_tcp_check_fraghdr(transport);
1328         else {
1329                 /*
1330                  * The transport_lock protects the request handling.
1331                  * There's no need to hold it to update the tcp_flags.
1332                  */
1333                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1334         }
1335 }
1336
1337 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1338 {
1339         size_t len;
1340
1341         len = transport->tcp_reclen - transport->tcp_offset;
1342         if (len > desc->count)
1343                 len = desc->count;
1344         desc->count -= len;
1345         desc->offset += len;
1346         transport->tcp_offset += len;
1347         dprintk("RPC:       discarded %Zu bytes\n", len);
1348         xs_tcp_check_fraghdr(transport);
1349 }
1350
1351 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1352 {
1353         struct rpc_xprt *xprt = rd_desc->arg.data;
1354         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1355         struct xdr_skb_reader desc = {
1356                 .skb    = skb,
1357                 .offset = offset,
1358                 .count  = len,
1359         };
1360
1361         dprintk("RPC:       xs_tcp_data_recv started\n");
1362         do {
1363                 trace_xs_tcp_data_recv(transport);
1364                 /* Read in a new fragment marker if necessary */
1365                 /* Can we ever really expect to get completely empty fragments? */
1366                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1367                         xs_tcp_read_fraghdr(xprt, &desc);
1368                         continue;
1369                 }
1370                 /* Read in the xid if necessary */
1371                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1372                         xs_tcp_read_xid(transport, &desc);
1373                         continue;
1374                 }
1375                 /* Read in the call/reply flag */
1376                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1377                         xs_tcp_read_calldir(transport, &desc);
1378                         continue;
1379                 }
1380                 /* Read in the request data */
1381                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1382                         xs_tcp_read_data(xprt, &desc);
1383                         continue;
1384                 }
1385                 /* Skip over any trailing bytes on short reads */
1386                 xs_tcp_read_discard(transport, &desc);
1387         } while (desc.count);
1388         trace_xs_tcp_data_recv(transport);
1389         dprintk("RPC:       xs_tcp_data_recv done\n");
1390         return len - desc.count;
1391 }
1392
1393 /**
1394  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1395  * @sk: socket with data to read
1396  * @bytes: how much data to read
1397  *
1398  */
1399 static void xs_tcp_data_ready(struct sock *sk)
1400 {
1401         struct rpc_xprt *xprt;
1402         read_descriptor_t rd_desc;
1403         int read;
1404         unsigned long total = 0;
1405
1406         dprintk("RPC:       xs_tcp_data_ready...\n");
1407
1408         read_lock_bh(&sk->sk_callback_lock);
1409         if (!(xprt = xprt_from_sock(sk))) {
1410                 read = 0;
1411                 goto out;
1412         }
1413         /* Any data means we had a useful conversation, so
1414          * the we don't need to delay the next reconnect
1415          */
1416         if (xprt->reestablish_timeout)
1417                 xprt->reestablish_timeout = 0;
1418
1419         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1420         rd_desc.arg.data = xprt;
1421         do {
1422                 rd_desc.count = 65536;
1423                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1424                 if (read > 0)
1425                         total += read;
1426         } while (read > 0);
1427 out:
1428         trace_xs_tcp_data_ready(xprt, read, total);
1429         read_unlock_bh(&sk->sk_callback_lock);
1430 }
1431
1432 /*
1433  * Do the equivalent of linger/linger2 handling for dealing with
1434  * broken servers that don't close the socket in a timely
1435  * fashion
1436  */
1437 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1438                 unsigned long timeout)
1439 {
1440         struct sock_xprt *transport;
1441
1442         if (xprt_test_and_set_connecting(xprt))
1443                 return;
1444         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1445         transport = container_of(xprt, struct sock_xprt, xprt);
1446         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1447                            timeout);
1448 }
1449
1450 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1451 {
1452         struct sock_xprt *transport;
1453
1454         transport = container_of(xprt, struct sock_xprt, xprt);
1455
1456         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1457             !cancel_delayed_work(&transport->connect_worker))
1458                 return;
1459         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1460         xprt_clear_connecting(xprt);
1461 }
1462
1463 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1464 {
1465         xs_sock_reset_connection_flags(xprt);
1466         /* Mark transport as closed and wake up all pending tasks */
1467         xprt_disconnect_done(xprt);
1468 }
1469
1470 /**
1471  * xs_tcp_state_change - callback to handle TCP socket state changes
1472  * @sk: socket whose state has changed
1473  *
1474  */
1475 static void xs_tcp_state_change(struct sock *sk)
1476 {
1477         struct rpc_xprt *xprt;
1478
1479         read_lock_bh(&sk->sk_callback_lock);
1480         if (!(xprt = xprt_from_sock(sk)))
1481                 goto out;
1482         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1483         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1484                         sk->sk_state, xprt_connected(xprt),
1485                         sock_flag(sk, SOCK_DEAD),
1486                         sock_flag(sk, SOCK_ZAPPED),
1487                         sk->sk_shutdown);
1488
1489         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1490         switch (sk->sk_state) {
1491         case TCP_ESTABLISHED:
1492                 spin_lock(&xprt->transport_lock);
1493                 if (!xprt_test_and_set_connected(xprt)) {
1494                         struct sock_xprt *transport = container_of(xprt,
1495                                         struct sock_xprt, xprt);
1496
1497                         /* Reset TCP record info */
1498                         transport->tcp_offset = 0;
1499                         transport->tcp_reclen = 0;
1500                         transport->tcp_copied = 0;
1501                         transport->tcp_flags =
1502                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1503                         xprt->connect_cookie++;
1504
1505                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1506                 }
1507                 spin_unlock(&xprt->transport_lock);
1508                 break;
1509         case TCP_FIN_WAIT1:
1510                 /* The client initiated a shutdown of the socket */
1511                 xprt->connect_cookie++;
1512                 xprt->reestablish_timeout = 0;
1513                 set_bit(XPRT_CLOSING, &xprt->state);
1514                 smp_mb__before_atomic();
1515                 clear_bit(XPRT_CONNECTED, &xprt->state);
1516                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1517                 smp_mb__after_atomic();
1518                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1519                 break;
1520         case TCP_CLOSE_WAIT:
1521                 /* The server initiated a shutdown of the socket */
1522                 xprt->connect_cookie++;
1523                 clear_bit(XPRT_CONNECTED, &xprt->state);
1524                 xs_tcp_force_close(xprt);
1525         case TCP_CLOSING:
1526                 /*
1527                  * If the server closed down the connection, make sure that
1528                  * we back off before reconnecting
1529                  */
1530                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1531                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1532                 break;
1533         case TCP_LAST_ACK:
1534                 set_bit(XPRT_CLOSING, &xprt->state);
1535                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1536                 smp_mb__before_atomic();
1537                 clear_bit(XPRT_CONNECTED, &xprt->state);
1538                 smp_mb__after_atomic();
1539                 break;
1540         case TCP_CLOSE:
1541                 xs_tcp_cancel_linger_timeout(xprt);
1542                 xs_sock_mark_closed(xprt);
1543         }
1544  out:
1545         read_unlock_bh(&sk->sk_callback_lock);
1546 }
1547
1548 static void xs_write_space(struct sock *sk)
1549 {
1550         struct socket *sock;
1551         struct rpc_xprt *xprt;
1552
1553         if (unlikely(!(sock = sk->sk_socket)))
1554                 return;
1555         clear_bit(SOCK_NOSPACE, &sock->flags);
1556
1557         if (unlikely(!(xprt = xprt_from_sock(sk))))
1558                 return;
1559         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1560                 return;
1561
1562         xprt_write_space(xprt);
1563 }
1564
1565 /**
1566  * xs_udp_write_space - callback invoked when socket buffer space
1567  *                             becomes available
1568  * @sk: socket whose state has changed
1569  *
1570  * Called when more output buffer space is available for this socket.
1571  * We try not to wake our writers until they can make "significant"
1572  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1573  * with a bunch of small requests.
1574  */
1575 static void xs_udp_write_space(struct sock *sk)
1576 {
1577         read_lock_bh(&sk->sk_callback_lock);
1578
1579         /* from net/core/sock.c:sock_def_write_space */
1580         if (sock_writeable(sk))
1581                 xs_write_space(sk);
1582
1583         read_unlock_bh(&sk->sk_callback_lock);
1584 }
1585
1586 /**
1587  * xs_tcp_write_space - callback invoked when socket buffer space
1588  *                             becomes available
1589  * @sk: socket whose state has changed
1590  *
1591  * Called when more output buffer space is available for this socket.
1592  * We try not to wake our writers until they can make "significant"
1593  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1594  * with a bunch of small requests.
1595  */
1596 static void xs_tcp_write_space(struct sock *sk)
1597 {
1598         read_lock_bh(&sk->sk_callback_lock);
1599
1600         /* from net/core/stream.c:sk_stream_write_space */
1601         if (sk_stream_is_writeable(sk))
1602                 xs_write_space(sk);
1603
1604         read_unlock_bh(&sk->sk_callback_lock);
1605 }
1606
1607 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1608 {
1609         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1610         struct sock *sk = transport->inet;
1611
1612         if (transport->rcvsize) {
1613                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1614                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1615         }
1616         if (transport->sndsize) {
1617                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1618                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1619                 sk->sk_write_space(sk);
1620         }
1621 }
1622
1623 /**
1624  * xs_udp_set_buffer_size - set send and receive limits
1625  * @xprt: generic transport
1626  * @sndsize: requested size of send buffer, in bytes
1627  * @rcvsize: requested size of receive buffer, in bytes
1628  *
1629  * Set socket send and receive buffer size limits.
1630  */
1631 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1632 {
1633         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1634
1635         transport->sndsize = 0;
1636         if (sndsize)
1637                 transport->sndsize = sndsize + 1024;
1638         transport->rcvsize = 0;
1639         if (rcvsize)
1640                 transport->rcvsize = rcvsize + 1024;
1641
1642         xs_udp_do_set_buffer_size(xprt);
1643 }
1644
1645 /**
1646  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1647  * @task: task that timed out
1648  *
1649  * Adjust the congestion window after a retransmit timeout has occurred.
1650  */
1651 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1652 {
1653         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1654 }
1655
1656 static unsigned short xs_get_random_port(void)
1657 {
1658         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1659         unsigned short rand = (unsigned short) prandom_u32() % range;
1660         return rand + xprt_min_resvport;
1661 }
1662
1663 /**
1664  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1665  * @sock: socket
1666  *
1667  * Note that this function has to be called on all sockets that share the
1668  * same port, and it must be called before binding.
1669  */
1670 static void xs_sock_set_reuseport(struct socket *sock)
1671 {
1672         char opt = 1;
1673
1674         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt));
1675 }
1676
1677 static unsigned short xs_sock_getport(struct socket *sock)
1678 {
1679         struct sockaddr_storage buf;
1680         int buflen;
1681         unsigned short port = 0;
1682
1683         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1684                 goto out;
1685         switch (buf.ss_family) {
1686         case AF_INET6:
1687                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1688                 break;
1689         case AF_INET:
1690                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1691         }
1692 out:
1693         return port;
1694 }
1695
1696 /**
1697  * xs_set_port - reset the port number in the remote endpoint address
1698  * @xprt: generic transport
1699  * @port: new port number
1700  *
1701  */
1702 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1703 {
1704         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1705
1706         rpc_set_port(xs_addr(xprt), port);
1707         xs_update_peer_port(xprt);
1708 }
1709
1710 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1711 {
1712         if (transport->srcport == 0)
1713                 transport->srcport = xs_sock_getport(sock);
1714 }
1715
1716 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1717 {
1718         unsigned short port = transport->srcport;
1719
1720         if (port == 0 && transport->xprt.resvport)
1721                 port = xs_get_random_port();
1722         return port;
1723 }
1724
1725 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1726 {
1727         if (transport->srcport != 0)
1728                 transport->srcport = 0;
1729         if (!transport->xprt.resvport)
1730                 return 0;
1731         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1732                 return xprt_max_resvport;
1733         return --port;
1734 }
1735 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1736 {
1737         struct sockaddr_storage myaddr;
1738         int err, nloop = 0;
1739         unsigned short port = xs_get_srcport(transport);
1740         unsigned short last;
1741
1742         /*
1743          * If we are asking for any ephemeral port (i.e. port == 0 &&
1744          * transport->xprt.resvport == 0), don't bind.  Let the local
1745          * port selection happen implicitly when the socket is used
1746          * (for example at connect time).
1747          *
1748          * This ensures that we can continue to establish TCP
1749          * connections even when all local ephemeral ports are already
1750          * a part of some TCP connection.  This makes no difference
1751          * for UDP sockets, but also doens't harm them.
1752          *
1753          * If we're asking for any reserved port (i.e. port == 0 &&
1754          * transport->xprt.resvport == 1) xs_get_srcport above will
1755          * ensure that port is non-zero and we will bind as needed.
1756          */
1757         if (port == 0)
1758                 return 0;
1759
1760         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1761         do {
1762                 rpc_set_port((struct sockaddr *)&myaddr, port);
1763                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1764                                 transport->xprt.addrlen);
1765                 if (err == 0) {
1766                         transport->srcport = port;
1767                         break;
1768                 }
1769                 last = port;
1770                 port = xs_next_srcport(transport, port);
1771                 if (port > last)
1772                         nloop++;
1773         } while (err == -EADDRINUSE && nloop != 2);
1774
1775         if (myaddr.ss_family == AF_INET)
1776                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1777                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1778                                 port, err ? "failed" : "ok", err);
1779         else
1780                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1781                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1782                                 port, err ? "failed" : "ok", err);
1783         return err;
1784 }
1785
1786 /*
1787  * We don't support autobind on AF_LOCAL sockets
1788  */
1789 static void xs_local_rpcbind(struct rpc_task *task)
1790 {
1791         rcu_read_lock();
1792         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1793         rcu_read_unlock();
1794 }
1795
1796 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1797 {
1798 }
1799
1800 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1801 static struct lock_class_key xs_key[2];
1802 static struct lock_class_key xs_slock_key[2];
1803
1804 static inline void xs_reclassify_socketu(struct socket *sock)
1805 {
1806         struct sock *sk = sock->sk;
1807
1808         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1809                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1810 }
1811
1812 static inline void xs_reclassify_socket4(struct socket *sock)
1813 {
1814         struct sock *sk = sock->sk;
1815
1816         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1817                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1818 }
1819
1820 static inline void xs_reclassify_socket6(struct socket *sock)
1821 {
1822         struct sock *sk = sock->sk;
1823
1824         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1825                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1826 }
1827
1828 static inline void xs_reclassify_socket(int family, struct socket *sock)
1829 {
1830         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1831         if (sock_owned_by_user(sock->sk))
1832                 return;
1833
1834         switch (family) {
1835         case AF_LOCAL:
1836                 xs_reclassify_socketu(sock);
1837                 break;
1838         case AF_INET:
1839                 xs_reclassify_socket4(sock);
1840                 break;
1841         case AF_INET6:
1842                 xs_reclassify_socket6(sock);
1843                 break;
1844         }
1845 }
1846 #else
1847 static inline void xs_reclassify_socketu(struct socket *sock)
1848 {
1849 }
1850
1851 static inline void xs_reclassify_socket4(struct socket *sock)
1852 {
1853 }
1854
1855 static inline void xs_reclassify_socket6(struct socket *sock)
1856 {
1857 }
1858
1859 static inline void xs_reclassify_socket(int family, struct socket *sock)
1860 {
1861 }
1862 #endif
1863
1864 static void xs_dummy_setup_socket(struct work_struct *work)
1865 {
1866 }
1867
1868 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1869                 struct sock_xprt *transport, int family, int type,
1870                 int protocol, bool reuseport)
1871 {
1872         struct socket *sock;
1873         int err;
1874
1875         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1876         if (err < 0) {
1877                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1878                                 protocol, -err);
1879                 goto out;
1880         }
1881         xs_reclassify_socket(family, sock);
1882
1883         if (reuseport)
1884                 xs_sock_set_reuseport(sock);
1885
1886         err = xs_bind(transport, sock);
1887         if (err) {
1888                 sock_release(sock);
1889                 goto out;
1890         }
1891
1892         return sock;
1893 out:
1894         return ERR_PTR(err);
1895 }
1896
1897 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1898                                       struct socket *sock)
1899 {
1900         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1901                                                                         xprt);
1902
1903         if (!transport->inet) {
1904                 struct sock *sk = sock->sk;
1905
1906                 write_lock_bh(&sk->sk_callback_lock);
1907
1908                 xs_save_old_callbacks(transport, sk);
1909
1910                 sk->sk_user_data = xprt;
1911                 sk->sk_data_ready = xs_local_data_ready;
1912                 sk->sk_write_space = xs_udp_write_space;
1913                 sk->sk_error_report = xs_error_report;
1914                 sk->sk_allocation = GFP_ATOMIC;
1915
1916                 xprt_clear_connected(xprt);
1917
1918                 /* Reset to new socket */
1919                 transport->sock = sock;
1920                 transport->inet = sk;
1921
1922                 write_unlock_bh(&sk->sk_callback_lock);
1923         }
1924
1925         /* Tell the socket layer to start connecting... */
1926         xprt->stat.connect_count++;
1927         xprt->stat.connect_start = jiffies;
1928         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1929 }
1930
1931 /**
1932  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1933  * @xprt: RPC transport to connect
1934  * @transport: socket transport to connect
1935  * @create_sock: function to create a socket of the correct type
1936  */
1937 static int xs_local_setup_socket(struct sock_xprt *transport)
1938 {
1939         struct rpc_xprt *xprt = &transport->xprt;
1940         struct socket *sock;
1941         int status = -EIO;
1942
1943         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1944         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1945                                         SOCK_STREAM, 0, &sock, 1);
1946         if (status < 0) {
1947                 dprintk("RPC:       can't create AF_LOCAL "
1948                         "transport socket (%d).\n", -status);
1949                 goto out;
1950         }
1951         xs_reclassify_socketu(sock);
1952
1953         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1954                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1955
1956         status = xs_local_finish_connecting(xprt, sock);
1957         trace_rpc_socket_connect(xprt, sock, status);
1958         switch (status) {
1959         case 0:
1960                 dprintk("RPC:       xprt %p connected to %s\n",
1961                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1962                 xprt_set_connected(xprt);
1963         case -ENOBUFS:
1964                 break;
1965         case -ENOENT:
1966                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1967                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1968                 break;
1969         case -ECONNREFUSED:
1970                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1971                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1972                 break;
1973         default:
1974                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1975                                 __func__, -status,
1976                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1977         }
1978
1979 out:
1980         xprt_clear_connecting(xprt);
1981         xprt_wake_pending_tasks(xprt, status);
1982         return status;
1983 }
1984
1985 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1986 {
1987         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1988         int ret;
1989
1990          if (RPC_IS_ASYNC(task)) {
1991                 /*
1992                  * We want the AF_LOCAL connect to be resolved in the
1993                  * filesystem namespace of the process making the rpc
1994                  * call.  Thus we connect synchronously.
1995                  *
1996                  * If we want to support asynchronous AF_LOCAL calls,
1997                  * we'll need to figure out how to pass a namespace to
1998                  * connect.
1999                  */
2000                 rpc_exit(task, -ENOTCONN);
2001                 return;
2002         }
2003         ret = xs_local_setup_socket(transport);
2004         if (ret && !RPC_IS_SOFTCONN(task))
2005                 msleep_interruptible(15000);
2006 }
2007
2008 #ifdef CONFIG_SUNRPC_SWAP
2009 static void xs_set_memalloc(struct rpc_xprt *xprt)
2010 {
2011         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2012                         xprt);
2013
2014         if (xprt->swapper)
2015                 sk_set_memalloc(transport->inet);
2016 }
2017
2018 /**
2019  * xs_swapper - Tag this transport as being used for swap.
2020  * @xprt: transport to tag
2021  * @enable: enable/disable
2022  *
2023  */
2024 int xs_swapper(struct rpc_xprt *xprt, int enable)
2025 {
2026         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2027                         xprt);
2028         int err = 0;
2029
2030         if (enable) {
2031                 xprt->swapper++;
2032                 xs_set_memalloc(xprt);
2033         } else if (xprt->swapper) {
2034                 xprt->swapper--;
2035                 sk_clear_memalloc(transport->inet);
2036         }
2037
2038         return err;
2039 }
2040 EXPORT_SYMBOL_GPL(xs_swapper);
2041 #else
2042 static void xs_set_memalloc(struct rpc_xprt *xprt)
2043 {
2044 }
2045 #endif
2046
2047 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2048 {
2049         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2050
2051         if (!transport->inet) {
2052                 struct sock *sk = sock->sk;
2053
2054                 write_lock_bh(&sk->sk_callback_lock);
2055
2056                 xs_save_old_callbacks(transport, sk);
2057
2058                 sk->sk_user_data = xprt;
2059                 sk->sk_data_ready = xs_udp_data_ready;
2060                 sk->sk_write_space = xs_udp_write_space;
2061                 sk->sk_allocation = GFP_ATOMIC;
2062
2063                 xprt_set_connected(xprt);
2064
2065                 /* Reset to new socket */
2066                 transport->sock = sock;
2067                 transport->inet = sk;
2068
2069                 xs_set_memalloc(xprt);
2070
2071                 write_unlock_bh(&sk->sk_callback_lock);
2072         }
2073         xs_udp_do_set_buffer_size(xprt);
2074 }
2075
2076 static void xs_udp_setup_socket(struct work_struct *work)
2077 {
2078         struct sock_xprt *transport =
2079                 container_of(work, struct sock_xprt, connect_worker.work);
2080         struct rpc_xprt *xprt = &transport->xprt;
2081         struct socket *sock = transport->sock;
2082         int status = -EIO;
2083
2084         sock = xs_create_sock(xprt, transport,
2085                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2086                         IPPROTO_UDP, false);
2087         if (IS_ERR(sock))
2088                 goto out;
2089
2090         dprintk("RPC:       worker connecting xprt %p via %s to "
2091                                 "%s (port %s)\n", xprt,
2092                         xprt->address_strings[RPC_DISPLAY_PROTO],
2093                         xprt->address_strings[RPC_DISPLAY_ADDR],
2094                         xprt->address_strings[RPC_DISPLAY_PORT]);
2095
2096         xs_udp_finish_connecting(xprt, sock);
2097         trace_rpc_socket_connect(xprt, sock, 0);
2098         status = 0;
2099 out:
2100         xprt_unlock_connect(xprt, transport);
2101         xprt_clear_connecting(xprt);
2102         xprt_wake_pending_tasks(xprt, status);
2103 }
2104
2105 /*
2106  * We need to preserve the port number so the reply cache on the server can
2107  * find our cached RPC replies when we get around to reconnecting.
2108  */
2109 static void xs_abort_connection(struct sock_xprt *transport)
2110 {
2111         int result;
2112         struct sockaddr any;
2113
2114         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2115
2116         /*
2117          * Disconnect the transport socket by doing a connect operation
2118          * with AF_UNSPEC.  This should return immediately...
2119          */
2120         memset(&any, 0, sizeof(any));
2121         any.sa_family = AF_UNSPEC;
2122         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2123         trace_rpc_socket_reset_connection(&transport->xprt,
2124                         transport->sock, result);
2125         if (!result)
2126                 xs_sock_reset_connection_flags(&transport->xprt);
2127         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2128 }
2129
2130 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2131 {
2132         unsigned int state = transport->inet->sk_state;
2133
2134         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2135                 /* we don't need to abort the connection if the socket
2136                  * hasn't undergone a shutdown
2137                  */
2138                 if (transport->inet->sk_shutdown == 0)
2139                         return;
2140                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2141                                 __func__, transport->inet->sk_shutdown);
2142         }
2143         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2144                 /* we don't need to abort the connection if the socket
2145                  * hasn't undergone a shutdown
2146                  */
2147                 if (transport->inet->sk_shutdown == 0)
2148                         return;
2149                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2150                                 "sk_shutdown set to %d\n",
2151                                 __func__, transport->inet->sk_shutdown);
2152         }
2153         xs_abort_connection(transport);
2154 }
2155
2156 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2157 {
2158         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2159         int ret = -ENOTCONN;
2160
2161         if (!transport->inet) {
2162                 struct sock *sk = sock->sk;
2163                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2164                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2165                 unsigned int opt_on = 1;
2166
2167                 /* TCP Keepalive options */
2168                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2169                                 (char *)&opt_on, sizeof(opt_on));
2170                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2171                                 (char *)&keepidle, sizeof(keepidle));
2172                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2173                                 (char *)&keepidle, sizeof(keepidle));
2174                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2175                                 (char *)&keepcnt, sizeof(keepcnt));
2176
2177                 write_lock_bh(&sk->sk_callback_lock);
2178
2179                 xs_save_old_callbacks(transport, sk);
2180
2181                 sk->sk_user_data = xprt;
2182                 sk->sk_data_ready = xs_tcp_data_ready;
2183                 sk->sk_state_change = xs_tcp_state_change;
2184                 sk->sk_write_space = xs_tcp_write_space;
2185                 sk->sk_error_report = xs_error_report;
2186                 sk->sk_allocation = GFP_ATOMIC;
2187
2188                 /* socket options */
2189                 sock_reset_flag(sk, SOCK_LINGER);
2190                 tcp_sk(sk)->linger2 = 0;
2191                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2192
2193                 xprt_clear_connected(xprt);
2194
2195                 /* Reset to new socket */
2196                 transport->sock = sock;
2197                 transport->inet = sk;
2198
2199                 write_unlock_bh(&sk->sk_callback_lock);
2200         }
2201
2202         if (!xprt_bound(xprt))
2203                 goto out;
2204
2205         xs_set_memalloc(xprt);
2206
2207         /* Tell the socket layer to start connecting... */
2208         xprt->stat.connect_count++;
2209         xprt->stat.connect_start = jiffies;
2210         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2211         switch (ret) {
2212         case 0:
2213                 xs_set_srcport(transport, sock);
2214         case -EINPROGRESS:
2215                 /* SYN_SENT! */
2216                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2217                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2218         }
2219 out:
2220         return ret;
2221 }
2222
2223 /**
2224  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2225  * @xprt: RPC transport to connect
2226  * @transport: socket transport to connect
2227  * @create_sock: function to create a socket of the correct type
2228  *
2229  * Invoked by a work queue tasklet.
2230  */
2231 static void xs_tcp_setup_socket(struct work_struct *work)
2232 {
2233         struct sock_xprt *transport =
2234                 container_of(work, struct sock_xprt, connect_worker.work);
2235         struct socket *sock = transport->sock;
2236         struct rpc_xprt *xprt = &transport->xprt;
2237         int status = -EIO;
2238
2239         if (!sock) {
2240                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2241                 sock = xs_create_sock(xprt, transport,
2242                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2243                                 IPPROTO_TCP, true);
2244                 if (IS_ERR(sock)) {
2245                         status = PTR_ERR(sock);
2246                         goto out;
2247                 }
2248         } else {
2249                 int abort_and_exit;
2250
2251                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2252                                 &xprt->state);
2253                 /* "close" the socket, preserving the local port */
2254                 set_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2255                 xs_tcp_reuse_connection(transport);
2256                 clear_bit(XPRT_CONNECTION_REUSE, &xprt->state);
2257
2258                 if (abort_and_exit)
2259                         goto out_eagain;
2260         }
2261
2262         dprintk("RPC:       worker connecting xprt %p via %s to "
2263                                 "%s (port %s)\n", xprt,
2264                         xprt->address_strings[RPC_DISPLAY_PROTO],
2265                         xprt->address_strings[RPC_DISPLAY_ADDR],
2266                         xprt->address_strings[RPC_DISPLAY_PORT]);
2267
2268         status = xs_tcp_finish_connecting(xprt, sock);
2269         trace_rpc_socket_connect(xprt, sock, status);
2270         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2271                         xprt, -status, xprt_connected(xprt),
2272                         sock->sk->sk_state);
2273         switch (status) {
2274         default:
2275                 printk("%s: connect returned unhandled error %d\n",
2276                         __func__, status);
2277         case -EADDRNOTAVAIL:
2278                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2279                  * and retry
2280                  */
2281                 xs_tcp_force_close(xprt);
2282                 break;
2283         case 0:
2284         case -EINPROGRESS:
2285         case -EALREADY:
2286                 xprt_unlock_connect(xprt, transport);
2287                 xprt_clear_connecting(xprt);
2288                 return;
2289         case -EINVAL:
2290                 /* Happens, for instance, if the user specified a link
2291                  * local IPv6 address without a scope-id.
2292                  */
2293         case -ECONNREFUSED:
2294         case -ECONNRESET:
2295         case -ENETUNREACH:
2296         case -EADDRINUSE:
2297         case -ENOBUFS:
2298                 /* retry with existing socket, after a delay */
2299                 goto out;
2300         }
2301 out_eagain:
2302         status = -EAGAIN;
2303 out:
2304         xprt_unlock_connect(xprt, transport);
2305         xprt_clear_connecting(xprt);
2306         xprt_wake_pending_tasks(xprt, status);
2307 }
2308
2309 /**
2310  * xs_connect - connect a socket to a remote endpoint
2311  * @xprt: pointer to transport structure
2312  * @task: address of RPC task that manages state of connect request
2313  *
2314  * TCP: If the remote end dropped the connection, delay reconnecting.
2315  *
2316  * UDP socket connects are synchronous, but we use a work queue anyway
2317  * to guarantee that even unprivileged user processes can set up a
2318  * socket on a privileged port.
2319  *
2320  * If a UDP socket connect fails, the delay behavior here prevents
2321  * retry floods (hard mounts).
2322  */
2323 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2324 {
2325         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2326
2327         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2328
2329         /* Start by resetting any existing state */
2330         xs_reset_transport(transport);
2331
2332         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2333                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2334                                 "seconds\n",
2335                                 xprt, xprt->reestablish_timeout / HZ);
2336                 queue_delayed_work(rpciod_workqueue,
2337                                    &transport->connect_worker,
2338                                    xprt->reestablish_timeout);
2339                 xprt->reestablish_timeout <<= 1;
2340                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2341                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2342                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2343                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2344         } else {
2345                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2346                 queue_delayed_work(rpciod_workqueue,
2347                                    &transport->connect_worker, 0);
2348         }
2349 }
2350
2351 /**
2352  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2353  * @xprt: rpc_xprt struct containing statistics
2354  * @seq: output file
2355  *
2356  */
2357 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2358 {
2359         long idle_time = 0;
2360
2361         if (xprt_connected(xprt))
2362                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2363
2364         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2365                         "%llu %llu %lu %llu %llu\n",
2366                         xprt->stat.bind_count,
2367                         xprt->stat.connect_count,
2368                         xprt->stat.connect_time,
2369                         idle_time,
2370                         xprt->stat.sends,
2371                         xprt->stat.recvs,
2372                         xprt->stat.bad_xids,
2373                         xprt->stat.req_u,
2374                         xprt->stat.bklog_u,
2375                         xprt->stat.max_slots,
2376                         xprt->stat.sending_u,
2377                         xprt->stat.pending_u);
2378 }
2379
2380 /**
2381  * xs_udp_print_stats - display UDP socket-specifc stats
2382  * @xprt: rpc_xprt struct containing statistics
2383  * @seq: output file
2384  *
2385  */
2386 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2387 {
2388         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2389
2390         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2391                         "%lu %llu %llu\n",
2392                         transport->srcport,
2393                         xprt->stat.bind_count,
2394                         xprt->stat.sends,
2395                         xprt->stat.recvs,
2396                         xprt->stat.bad_xids,
2397                         xprt->stat.req_u,
2398                         xprt->stat.bklog_u,
2399                         xprt->stat.max_slots,
2400                         xprt->stat.sending_u,
2401                         xprt->stat.pending_u);
2402 }
2403
2404 /**
2405  * xs_tcp_print_stats - display TCP socket-specifc stats
2406  * @xprt: rpc_xprt struct containing statistics
2407  * @seq: output file
2408  *
2409  */
2410 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2411 {
2412         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2413         long idle_time = 0;
2414
2415         if (xprt_connected(xprt))
2416                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2417
2418         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2419                         "%llu %llu %lu %llu %llu\n",
2420                         transport->srcport,
2421                         xprt->stat.bind_count,
2422                         xprt->stat.connect_count,
2423                         xprt->stat.connect_time,
2424                         idle_time,
2425                         xprt->stat.sends,
2426                         xprt->stat.recvs,
2427                         xprt->stat.bad_xids,
2428                         xprt->stat.req_u,
2429                         xprt->stat.bklog_u,
2430                         xprt->stat.max_slots,
2431                         xprt->stat.sending_u,
2432                         xprt->stat.pending_u);
2433 }
2434
2435 /*
2436  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2437  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2438  * to use the server side send routines.
2439  */
2440 static void *bc_malloc(struct rpc_task *task, size_t size)
2441 {
2442         struct page *page;
2443         struct rpc_buffer *buf;
2444
2445         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2446         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2447                 return NULL;
2448
2449         page = alloc_page(GFP_KERNEL);
2450         if (!page)
2451                 return NULL;
2452
2453         buf = page_address(page);
2454         buf->len = PAGE_SIZE;
2455
2456         return buf->data;
2457 }
2458
2459 /*
2460  * Free the space allocated in the bc_alloc routine
2461  */
2462 static void bc_free(void *buffer)
2463 {
2464         struct rpc_buffer *buf;
2465
2466         if (!buffer)
2467                 return;
2468
2469         buf = container_of(buffer, struct rpc_buffer, data);
2470         free_page((unsigned long)buf);
2471 }
2472
2473 /*
2474  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2475  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2476  */
2477 static int bc_sendto(struct rpc_rqst *req)
2478 {
2479         int len;
2480         struct xdr_buf *xbufp = &req->rq_snd_buf;
2481         struct rpc_xprt *xprt = req->rq_xprt;
2482         struct sock_xprt *transport =
2483                                 container_of(xprt, struct sock_xprt, xprt);
2484         struct socket *sock = transport->sock;
2485         unsigned long headoff;
2486         unsigned long tailoff;
2487
2488         xs_encode_stream_record_marker(xbufp);
2489
2490         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2491         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2492         len = svc_send_common(sock, xbufp,
2493                               virt_to_page(xbufp->head[0].iov_base), headoff,
2494                               xbufp->tail[0].iov_base, tailoff);
2495
2496         if (len != xbufp->len) {
2497                 printk(KERN_NOTICE "Error sending entire callback!\n");
2498                 len = -EAGAIN;
2499         }
2500
2501         return len;
2502 }
2503
2504 /*
2505  * The send routine. Borrows from svc_send
2506  */
2507 static int bc_send_request(struct rpc_task *task)
2508 {
2509         struct rpc_rqst *req = task->tk_rqstp;
2510         struct svc_xprt *xprt;
2511         u32                     len;
2512
2513         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2514         /*
2515          * Get the server socket associated with this callback xprt
2516          */
2517         xprt = req->rq_xprt->bc_xprt;
2518
2519         /*
2520          * Grab the mutex to serialize data as the connection is shared
2521          * with the fore channel
2522          */
2523         if (!mutex_trylock(&xprt->xpt_mutex)) {
2524                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2525                 if (!mutex_trylock(&xprt->xpt_mutex))
2526                         return -EAGAIN;
2527                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2528         }
2529         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2530                 len = -ENOTCONN;
2531         else
2532                 len = bc_sendto(req);
2533         mutex_unlock(&xprt->xpt_mutex);
2534
2535         if (len > 0)
2536                 len = 0;
2537
2538         return len;
2539 }
2540
2541 /*
2542  * The close routine. Since this is client initiated, we do nothing
2543  */
2544
2545 static void bc_close(struct rpc_xprt *xprt)
2546 {
2547 }
2548
2549 /*
2550  * The xprt destroy routine. Again, because this connection is client
2551  * initiated, we do nothing
2552  */
2553
2554 static void bc_destroy(struct rpc_xprt *xprt)
2555 {
2556         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2557
2558         xs_xprt_free(xprt);
2559         module_put(THIS_MODULE);
2560 }
2561
2562 static struct rpc_xprt_ops xs_local_ops = {
2563         .reserve_xprt           = xprt_reserve_xprt,
2564         .release_xprt           = xs_tcp_release_xprt,
2565         .alloc_slot             = xprt_alloc_slot,
2566         .rpcbind                = xs_local_rpcbind,
2567         .set_port               = xs_local_set_port,
2568         .connect                = xs_local_connect,
2569         .buf_alloc              = rpc_malloc,
2570         .buf_free               = rpc_free,
2571         .send_request           = xs_local_send_request,
2572         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2573         .close                  = xs_close,
2574         .destroy                = xs_destroy,
2575         .print_stats            = xs_local_print_stats,
2576 };
2577
2578 static struct rpc_xprt_ops xs_udp_ops = {
2579         .set_buffer_size        = xs_udp_set_buffer_size,
2580         .reserve_xprt           = xprt_reserve_xprt_cong,
2581         .release_xprt           = xprt_release_xprt_cong,
2582         .alloc_slot             = xprt_alloc_slot,
2583         .rpcbind                = rpcb_getport_async,
2584         .set_port               = xs_set_port,
2585         .connect                = xs_connect,
2586         .buf_alloc              = rpc_malloc,
2587         .buf_free               = rpc_free,
2588         .send_request           = xs_udp_send_request,
2589         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2590         .timer                  = xs_udp_timer,
2591         .release_request        = xprt_release_rqst_cong,
2592         .close                  = xs_close,
2593         .destroy                = xs_destroy,
2594         .print_stats            = xs_udp_print_stats,
2595 };
2596
2597 static struct rpc_xprt_ops xs_tcp_ops = {
2598         .reserve_xprt           = xprt_reserve_xprt,
2599         .release_xprt           = xs_tcp_release_xprt,
2600         .alloc_slot             = xprt_lock_and_alloc_slot,
2601         .rpcbind                = rpcb_getport_async,
2602         .set_port               = xs_set_port,
2603         .connect                = xs_connect,
2604         .buf_alloc              = rpc_malloc,
2605         .buf_free               = rpc_free,
2606         .send_request           = xs_tcp_send_request,
2607         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2608         .close                  = xs_tcp_close,
2609         .destroy                = xs_destroy,
2610         .print_stats            = xs_tcp_print_stats,
2611 };
2612
2613 /*
2614  * The rpc_xprt_ops for the server backchannel
2615  */
2616
2617 static struct rpc_xprt_ops bc_tcp_ops = {
2618         .reserve_xprt           = xprt_reserve_xprt,
2619         .release_xprt           = xprt_release_xprt,
2620         .alloc_slot             = xprt_alloc_slot,
2621         .buf_alloc              = bc_malloc,
2622         .buf_free               = bc_free,
2623         .send_request           = bc_send_request,
2624         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2625         .close                  = bc_close,
2626         .destroy                = bc_destroy,
2627         .print_stats            = xs_tcp_print_stats,
2628 };
2629
2630 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2631 {
2632         static const struct sockaddr_in sin = {
2633                 .sin_family             = AF_INET,
2634                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2635         };
2636         static const struct sockaddr_in6 sin6 = {
2637                 .sin6_family            = AF_INET6,
2638                 .sin6_addr              = IN6ADDR_ANY_INIT,
2639         };
2640
2641         switch (family) {
2642         case AF_LOCAL:
2643                 break;
2644         case AF_INET:
2645                 memcpy(sap, &sin, sizeof(sin));
2646                 break;
2647         case AF_INET6:
2648                 memcpy(sap, &sin6, sizeof(sin6));
2649                 break;
2650         default:
2651                 dprintk("RPC:       %s: Bad address family\n", __func__);
2652                 return -EAFNOSUPPORT;
2653         }
2654         return 0;
2655 }
2656
2657 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2658                                       unsigned int slot_table_size,
2659                                       unsigned int max_slot_table_size)
2660 {
2661         struct rpc_xprt *xprt;
2662         struct sock_xprt *new;
2663
2664         if (args->addrlen > sizeof(xprt->addr)) {
2665                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2666                 return ERR_PTR(-EBADF);
2667         }
2668
2669         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2670                         max_slot_table_size);
2671         if (xprt == NULL) {
2672                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2673                                 "rpc_xprt\n");
2674                 return ERR_PTR(-ENOMEM);
2675         }
2676
2677         new = container_of(xprt, struct sock_xprt, xprt);
2678         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2679         xprt->addrlen = args->addrlen;
2680         if (args->srcaddr)
2681                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2682         else {
2683                 int err;
2684                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2685                                         (struct sockaddr *)&new->srcaddr);
2686                 if (err != 0) {
2687                         xprt_free(xprt);
2688                         return ERR_PTR(err);
2689                 }
2690         }
2691
2692         return xprt;
2693 }
2694
2695 static const struct rpc_timeout xs_local_default_timeout = {
2696         .to_initval = 10 * HZ,
2697         .to_maxval = 10 * HZ,
2698         .to_retries = 2,
2699 };
2700
2701 /**
2702  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2703  * @args: rpc transport creation arguments
2704  *
2705  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2706  */
2707 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2708 {
2709         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2710         struct sock_xprt *transport;
2711         struct rpc_xprt *xprt;
2712         struct rpc_xprt *ret;
2713
2714         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2715                         xprt_max_tcp_slot_table_entries);
2716         if (IS_ERR(xprt))
2717                 return xprt;
2718         transport = container_of(xprt, struct sock_xprt, xprt);
2719
2720         xprt->prot = 0;
2721         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2722         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2723
2724         xprt->bind_timeout = XS_BIND_TO;
2725         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2726         xprt->idle_timeout = XS_IDLE_DISC_TO;
2727
2728         xprt->ops = &xs_local_ops;
2729         xprt->timeout = &xs_local_default_timeout;
2730
2731         INIT_DELAYED_WORK(&transport->connect_worker,
2732                         xs_dummy_setup_socket);
2733
2734         switch (sun->sun_family) {
2735         case AF_LOCAL:
2736                 if (sun->sun_path[0] != '/') {
2737                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2738                                         sun->sun_path);
2739                         ret = ERR_PTR(-EINVAL);
2740                         goto out_err;
2741                 }
2742                 xprt_set_bound(xprt);
2743                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2744                 ret = ERR_PTR(xs_local_setup_socket(transport));
2745                 if (ret)
2746                         goto out_err;
2747                 break;
2748         default:
2749                 ret = ERR_PTR(-EAFNOSUPPORT);
2750                 goto out_err;
2751         }
2752
2753         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2754                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2755
2756         if (try_module_get(THIS_MODULE))
2757                 return xprt;
2758         ret = ERR_PTR(-EINVAL);
2759 out_err:
2760         xs_xprt_free(xprt);
2761         return ret;
2762 }
2763
2764 static const struct rpc_timeout xs_udp_default_timeout = {
2765         .to_initval = 5 * HZ,
2766         .to_maxval = 30 * HZ,
2767         .to_increment = 5 * HZ,
2768         .to_retries = 5,
2769 };
2770
2771 /**
2772  * xs_setup_udp - Set up transport to use a UDP socket
2773  * @args: rpc transport creation arguments
2774  *
2775  */
2776 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2777 {
2778         struct sockaddr *addr = args->dstaddr;
2779         struct rpc_xprt *xprt;
2780         struct sock_xprt *transport;
2781         struct rpc_xprt *ret;
2782
2783         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2784                         xprt_udp_slot_table_entries);
2785         if (IS_ERR(xprt))
2786                 return xprt;
2787         transport = container_of(xprt, struct sock_xprt, xprt);
2788
2789         xprt->prot = IPPROTO_UDP;
2790         xprt->tsh_size = 0;
2791         /* XXX: header size can vary due to auth type, IPv6, etc. */
2792         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2793
2794         xprt->bind_timeout = XS_BIND_TO;
2795         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2796         xprt->idle_timeout = XS_IDLE_DISC_TO;
2797
2798         xprt->ops = &xs_udp_ops;
2799
2800         xprt->timeout = &xs_udp_default_timeout;
2801
2802         switch (addr->sa_family) {
2803         case AF_INET:
2804                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2805                         xprt_set_bound(xprt);
2806
2807                 INIT_DELAYED_WORK(&transport->connect_worker,
2808                                         xs_udp_setup_socket);
2809                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2810                 break;
2811         case AF_INET6:
2812                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2813                         xprt_set_bound(xprt);
2814
2815                 INIT_DELAYED_WORK(&transport->connect_worker,
2816                                         xs_udp_setup_socket);
2817                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2818                 break;
2819         default:
2820                 ret = ERR_PTR(-EAFNOSUPPORT);
2821                 goto out_err;
2822         }
2823
2824         if (xprt_bound(xprt))
2825                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2826                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2827                                 xprt->address_strings[RPC_DISPLAY_PORT],
2828                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2829         else
2830                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2831                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2832                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2833
2834         if (try_module_get(THIS_MODULE))
2835                 return xprt;
2836         ret = ERR_PTR(-EINVAL);
2837 out_err:
2838         xs_xprt_free(xprt);
2839         return ret;
2840 }
2841
2842 static const struct rpc_timeout xs_tcp_default_timeout = {
2843         .to_initval = 60 * HZ,
2844         .to_maxval = 60 * HZ,
2845         .to_retries = 2,
2846 };
2847
2848 /**
2849  * xs_setup_tcp - Set up transport to use a TCP socket
2850  * @args: rpc transport creation arguments
2851  *
2852  */
2853 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2854 {
2855         struct sockaddr *addr = args->dstaddr;
2856         struct rpc_xprt *xprt;
2857         struct sock_xprt *transport;
2858         struct rpc_xprt *ret;
2859         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2860
2861         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2862                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2863
2864         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2865                         max_slot_table_size);
2866         if (IS_ERR(xprt))
2867                 return xprt;
2868         transport = container_of(xprt, struct sock_xprt, xprt);
2869
2870         xprt->prot = IPPROTO_TCP;
2871         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2872         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2873
2874         xprt->bind_timeout = XS_BIND_TO;
2875         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2876         xprt->idle_timeout = XS_IDLE_DISC_TO;
2877
2878         xprt->ops = &xs_tcp_ops;
2879         xprt->timeout = &xs_tcp_default_timeout;
2880
2881         switch (addr->sa_family) {
2882         case AF_INET:
2883                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2884                         xprt_set_bound(xprt);
2885
2886                 INIT_DELAYED_WORK(&transport->connect_worker,
2887                                         xs_tcp_setup_socket);
2888                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2889                 break;
2890         case AF_INET6:
2891                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2892                         xprt_set_bound(xprt);
2893
2894                 INIT_DELAYED_WORK(&transport->connect_worker,
2895                                         xs_tcp_setup_socket);
2896                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2897                 break;
2898         default:
2899                 ret = ERR_PTR(-EAFNOSUPPORT);
2900                 goto out_err;
2901         }
2902
2903         if (xprt_bound(xprt))
2904                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2905                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2906                                 xprt->address_strings[RPC_DISPLAY_PORT],
2907                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2908         else
2909                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2910                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2911                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2912
2913         if (try_module_get(THIS_MODULE))
2914                 return xprt;
2915         ret = ERR_PTR(-EINVAL);
2916 out_err:
2917         xs_xprt_free(xprt);
2918         return ret;
2919 }
2920
2921 /**
2922  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2923  * @args: rpc transport creation arguments
2924  *
2925  */
2926 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2927 {
2928         struct sockaddr *addr = args->dstaddr;
2929         struct rpc_xprt *xprt;
2930         struct sock_xprt *transport;
2931         struct svc_sock *bc_sock;
2932         struct rpc_xprt *ret;
2933
2934         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2935                         xprt_tcp_slot_table_entries);
2936         if (IS_ERR(xprt))
2937                 return xprt;
2938         transport = container_of(xprt, struct sock_xprt, xprt);
2939
2940         xprt->prot = IPPROTO_TCP;
2941         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2942         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2943         xprt->timeout = &xs_tcp_default_timeout;
2944
2945         /* backchannel */
2946         xprt_set_bound(xprt);
2947         xprt->bind_timeout = 0;
2948         xprt->reestablish_timeout = 0;
2949         xprt->idle_timeout = 0;
2950
2951         xprt->ops = &bc_tcp_ops;
2952
2953         switch (addr->sa_family) {
2954         case AF_INET:
2955                 xs_format_peer_addresses(xprt, "tcp",
2956                                          RPCBIND_NETID_TCP);
2957                 break;
2958         case AF_INET6:
2959                 xs_format_peer_addresses(xprt, "tcp",
2960                                    RPCBIND_NETID_TCP6);
2961                 break;
2962         default:
2963                 ret = ERR_PTR(-EAFNOSUPPORT);
2964                 goto out_err;
2965         }
2966
2967         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2968                         xprt->address_strings[RPC_DISPLAY_ADDR],
2969                         xprt->address_strings[RPC_DISPLAY_PORT],
2970                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2971
2972         /*
2973          * Once we've associated a backchannel xprt with a connection,
2974          * we want to keep it around as long as the connection lasts,
2975          * in case we need to start using it for a backchannel again;
2976          * this reference won't be dropped until bc_xprt is destroyed.
2977          */
2978         xprt_get(xprt);
2979         args->bc_xprt->xpt_bc_xprt = xprt;
2980         xprt->bc_xprt = args->bc_xprt;
2981         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2982         transport->sock = bc_sock->sk_sock;
2983         transport->inet = bc_sock->sk_sk;
2984
2985         /*
2986          * Since we don't want connections for the backchannel, we set
2987          * the xprt status to connected
2988          */
2989         xprt_set_connected(xprt);
2990
2991         if (try_module_get(THIS_MODULE))
2992                 return xprt;
2993
2994         args->bc_xprt->xpt_bc_xprt = NULL;
2995         xprt_put(xprt);
2996         ret = ERR_PTR(-EINVAL);
2997 out_err:
2998         xs_xprt_free(xprt);
2999         return ret;
3000 }
3001
3002 static struct xprt_class        xs_local_transport = {
3003         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3004         .name           = "named UNIX socket",
3005         .owner          = THIS_MODULE,
3006         .ident          = XPRT_TRANSPORT_LOCAL,
3007         .setup          = xs_setup_local,
3008 };
3009
3010 static struct xprt_class        xs_udp_transport = {
3011         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3012         .name           = "udp",
3013         .owner          = THIS_MODULE,
3014         .ident          = XPRT_TRANSPORT_UDP,
3015         .setup          = xs_setup_udp,
3016 };
3017
3018 static struct xprt_class        xs_tcp_transport = {
3019         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3020         .name           = "tcp",
3021         .owner          = THIS_MODULE,
3022         .ident          = XPRT_TRANSPORT_TCP,
3023         .setup          = xs_setup_tcp,
3024 };
3025
3026 static struct xprt_class        xs_bc_tcp_transport = {
3027         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3028         .name           = "tcp NFSv4.1 backchannel",
3029         .owner          = THIS_MODULE,
3030         .ident          = XPRT_TRANSPORT_BC_TCP,
3031         .setup          = xs_setup_bc_tcp,
3032 };
3033
3034 /**
3035  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3036  *
3037  */
3038 int init_socket_xprt(void)
3039 {
3040 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3041         if (!sunrpc_table_header)
3042                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3043 #endif
3044
3045         xprt_register_transport(&xs_local_transport);
3046         xprt_register_transport(&xs_udp_transport);
3047         xprt_register_transport(&xs_tcp_transport);
3048         xprt_register_transport(&xs_bc_tcp_transport);
3049
3050         return 0;
3051 }
3052
3053 /**
3054  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3055  *
3056  */
3057 void cleanup_socket_xprt(void)
3058 {
3059 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3060         if (sunrpc_table_header) {
3061                 unregister_sysctl_table(sunrpc_table_header);
3062                 sunrpc_table_header = NULL;
3063         }
3064 #endif
3065
3066         xprt_unregister_transport(&xs_local_transport);
3067         xprt_unregister_transport(&xs_udp_transport);
3068         xprt_unregister_transport(&xs_tcp_transport);
3069         xprt_unregister_transport(&xs_bc_tcp_transport);
3070 }
3071
3072 static int param_set_uint_minmax(const char *val,
3073                 const struct kernel_param *kp,
3074                 unsigned int min, unsigned int max)
3075 {
3076         unsigned int num;
3077         int ret;
3078
3079         if (!val)
3080                 return -EINVAL;
3081         ret = kstrtouint(val, 0, &num);
3082         if (ret == -EINVAL || num < min || num > max)
3083                 return -EINVAL;
3084         *((unsigned int *)kp->arg) = num;
3085         return 0;
3086 }
3087
3088 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3089 {
3090         return param_set_uint_minmax(val, kp,
3091                         RPC_MIN_RESVPORT,
3092                         RPC_MAX_RESVPORT);
3093 }
3094
3095 static struct kernel_param_ops param_ops_portnr = {
3096         .set = param_set_portnr,
3097         .get = param_get_uint,
3098 };
3099
3100 #define param_check_portnr(name, p) \
3101         __param_check(name, p, unsigned int);
3102
3103 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3104 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3105
3106 static int param_set_slot_table_size(const char *val,
3107                                      const struct kernel_param *kp)
3108 {
3109         return param_set_uint_minmax(val, kp,
3110                         RPC_MIN_SLOT_TABLE,
3111                         RPC_MAX_SLOT_TABLE);
3112 }
3113
3114 static struct kernel_param_ops param_ops_slot_table_size = {
3115         .set = param_set_slot_table_size,
3116         .get = param_get_uint,
3117 };
3118
3119 #define param_check_slot_table_size(name, p) \
3120         __param_check(name, p, unsigned int);
3121
3122 static int param_set_max_slot_table_size(const char *val,
3123                                      const struct kernel_param *kp)
3124 {
3125         return param_set_uint_minmax(val, kp,
3126                         RPC_MIN_SLOT_TABLE,
3127                         RPC_MAX_SLOT_TABLE_LIMIT);
3128 }
3129
3130 static struct kernel_param_ops param_ops_max_slot_table_size = {
3131         .set = param_set_max_slot_table_size,
3132         .get = param_get_uint,
3133 };
3134
3135 #define param_check_max_slot_table_size(name, p) \
3136         __param_check(name, p, unsigned int);
3137
3138 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3139                    slot_table_size, 0644);
3140 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3141                    max_slot_table_size, 0644);
3142 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3143                    slot_table_size, 0644);
3144