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 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68 #define XS_TCP_LINGER_TO        (15U * HZ)
  69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71 /*
  72  * We can register our own files under /proc/sys/sunrpc by
  73  * calling register_sysctl_table() again.  The files in that
  74  * directory become the union of all files registered there.
  75  *
  76  * We simply need to make sure that we don't collide with
  77  * someone else's file names!
  78  */
  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_RDWR);
 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_RDWR);
 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         case -ECONNREFUSED:
 722         case -ENOTCONN:
 723         case -EADDRINUSE:
 724         case -EPIPE:
 725                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 726         }
 727
 728         return status;
 729 }
 730
 731 /**
 732  * xs_tcp_release_xprt - clean up after a tcp transmission
 733  * @xprt: transport
 734  * @task: rpc task
 735  *
 736  * This cleans up if an error causes us to abort the transmission of a request.
 737  * In this case, the socket may need to be reset in order to avoid confusing
 738  * the server.
 739  */
 740 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 741 {
 742         struct rpc_rqst *req;
 743
 744         if (task != xprt->snd_task)
 745                 return;
 746         if (task == NULL)
 747                 goto out_release;
 748         req = task->tk_rqstp;
 749         if (req == NULL)
 750                 goto out_release;
 751         if (req->rq_bytes_sent == 0)
 752                 goto out_release;
 753         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 754                 goto out_release;
 755         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 756 out_release:
 757         xprt_release_xprt(xprt, task);
 758 }
 759
 760 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 761 {
 762         transport->old_data_ready = sk->sk_data_ready;
 763         transport->old_state_change = sk->sk_state_change;
 764         transport->old_write_space = sk->sk_write_space;
 765         transport->old_error_report = sk->sk_error_report;
 766 }
 767
 768 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 769 {
 770         sk->sk_data_ready = transport->old_data_ready;
 771         sk->sk_state_change = transport->old_state_change;
 772         sk->sk_write_space = transport->old_write_space;
 773         sk->sk_error_report = transport->old_error_report;
 774 }
 775
 776 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 777 {
 778         smp_mb__before_atomic();
 779         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 780         clear_bit(XPRT_CLOSING, &xprt->state);
 781         smp_mb__after_atomic();
 782 }
 783
 784 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 785 {
 786         xs_sock_reset_connection_flags(xprt);
 787         /* Mark transport as closed and wake up all pending tasks */
 788         xprt_disconnect_done(xprt);
 789 }
 790
 791 /**
 792  * xs_error_report - callback to handle TCP socket state errors
 793  * @sk: socket
 794  *
 795  * Note: we don't call sock_error() since there may be a rpc_task
 796  * using the socket, and so we don't want to clear sk->sk_err.
 797  */
 798 static void xs_error_report(struct sock *sk)
 799 {
 800         struct rpc_xprt *xprt;
 801         int err;
 802
 803         read_lock_bh(&sk->sk_callback_lock);
 804         if (!(xprt = xprt_from_sock(sk)))
 805                 goto out;
 806
 807         err = -sk->sk_err;
 808         if (err == 0)
 809                 goto out;
 810         /* Is this a reset event? */
 811         if (sk->sk_state == TCP_CLOSE)
 812                 xs_sock_mark_closed(xprt);
 813         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 814                         xprt, -err);
 815         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 816         xprt_wake_pending_tasks(xprt, err);
 817  out:
 818         read_unlock_bh(&sk->sk_callback_lock);
 819 }
 820
 821 static void xs_reset_transport(struct sock_xprt *transport)
 822 {
 823         struct socket *sock = transport->sock;
 824         struct sock *sk = transport->inet;
 825         struct rpc_xprt *xprt = &transport->xprt;
 826
 827         if (sk == NULL)
 828                 return;
 829
 830         write_lock_bh(&sk->sk_callback_lock);
 831         transport->inet = NULL;
 832         transport->sock = NULL;
 833
 834         sk->sk_user_data = NULL;
 835
 836         xs_restore_old_callbacks(transport, sk);
 837         write_unlock_bh(&sk->sk_callback_lock);
 838         xs_sock_reset_connection_flags(xprt);
 839
 840         trace_rpc_socket_close(xprt, sock);
 841         sock_release(sock);
 842 }
 843
 844 /**
 845  * xs_close - close a socket
 846  * @xprt: transport
 847  *
 848  * This is used when all requests are complete; ie, no DRC state remains
 849  * on the server we want to save.
 850  *
 851  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 852  * xs_reset_transport() zeroing the socket from underneath a writer.
 853  */
 854 static void xs_close(struct rpc_xprt *xprt)
 855 {
 856         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 857
 858         dprintk("RPC:       xs_close xprt %p\n", xprt);
 859
 860         xs_reset_transport(transport);
 861         xprt->reestablish_timeout = 0;
 862
 863         xprt_disconnect_done(xprt);
 864 }
 865
 866 static void xs_tcp_close(struct rpc_xprt *xprt)
 867 {
 868         xs_tcp_shutdown(xprt);
 869 }
 870
 871 static void xs_xprt_free(struct rpc_xprt *xprt)
 872 {
 873         xs_free_peer_addresses(xprt);
 874         xprt_free(xprt);
 875 }
 876
 877 /**
 878  * xs_destroy - prepare to shutdown a transport
 879  * @xprt: doomed transport
 880  *
 881  */
 882 static void xs_destroy(struct rpc_xprt *xprt)
 883 {
 884         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 885
 886         xs_close(xprt);
 887         xs_xprt_free(xprt);
 888         module_put(THIS_MODULE);
 889 }
 890
 891 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 892 {
 893         struct xdr_skb_reader desc = {
 894                 .skb            = skb,
 895                 .offset         = sizeof(rpc_fraghdr),
 896                 .count          = skb->len - sizeof(rpc_fraghdr),
 897         };
 898
 899         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 900                 return -1;
 901         if (desc.count)
 902                 return -1;
 903         return 0;
 904 }
 905
 906 /**
 907  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 908  * @sk: socket with data to read
 909  * @len: how much data to read
 910  *
 911  * Currently this assumes we can read the whole reply in a single gulp.
 912  */
 913 static void xs_local_data_ready(struct sock *sk)
 914 {
 915         struct rpc_task *task;
 916         struct rpc_xprt *xprt;
 917         struct rpc_rqst *rovr;
 918         struct sk_buff *skb;
 919         int err, repsize, copied;
 920         u32 _xid;
 921         __be32 *xp;
 922
 923         read_lock_bh(&sk->sk_callback_lock);
 924         dprintk("RPC:       %s...\n", __func__);
 925         xprt = xprt_from_sock(sk);
 926         if (xprt == NULL)
 927                 goto out;
 928
 929         skb = skb_recv_datagram(sk, 0, 1, &err);
 930         if (skb == NULL)
 931                 goto out;
 932
 933         repsize = skb->len - sizeof(rpc_fraghdr);
 934         if (repsize < 4) {
 935                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 936                 goto dropit;
 937         }
 938
 939         /* Copy the XID from the skb... */
 940         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 941         if (xp == NULL)
 942                 goto dropit;
 943
 944         /* Look up and lock the request corresponding to the given XID */
 945         spin_lock(&xprt->transport_lock);
 946         rovr = xprt_lookup_rqst(xprt, *xp);
 947         if (!rovr)
 948                 goto out_unlock;
 949         task = rovr->rq_task;
 950
 951         copied = rovr->rq_private_buf.buflen;
 952         if (copied > repsize)
 953                 copied = repsize;
 954
 955         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 956                 dprintk("RPC:       sk_buff copy failed\n");
 957                 goto out_unlock;
 958         }
 959
 960         xprt_complete_rqst(task, copied);
 961
 962  out_unlock:
 963         spin_unlock(&xprt->transport_lock);
 964  dropit:
 965         skb_free_datagram(sk, skb);
 966  out:
 967         read_unlock_bh(&sk->sk_callback_lock);
 968 }
 969
 970 /**
 971  * xs_udp_data_ready - "data ready" callback for UDP sockets
 972  * @sk: socket with data to read
 973  * @len: how much data to read
 974  *
 975  */
 976 static void xs_udp_data_ready(struct sock *sk)
 977 {
 978         struct rpc_task *task;
 979         struct rpc_xprt *xprt;
 980         struct rpc_rqst *rovr;
 981         struct sk_buff *skb;
 982         int err, repsize, copied;
 983         u32 _xid;
 984         __be32 *xp;
 985
 986         read_lock_bh(&sk->sk_callback_lock);
 987         dprintk("RPC:       xs_udp_data_ready...\n");
 988         if (!(xprt = xprt_from_sock(sk)))
 989                 goto out;
 990
 991         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 992                 goto out;
 993
 994         repsize = skb->len - sizeof(struct udphdr);
 995         if (repsize < 4) {
 996                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 997                 goto dropit;
 998         }
 999
1000         /* Copy the XID from the skb... */
1001         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1002                                 sizeof(_xid), &_xid);
1003         if (xp == NULL)
1004                 goto dropit;
1005
1006         /* Look up and lock the request corresponding to the given XID */
1007         spin_lock(&xprt->transport_lock);
1008         rovr = xprt_lookup_rqst(xprt, *xp);
1009         if (!rovr)
1010                 goto out_unlock;
1011         task = rovr->rq_task;
1012
1013         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1014                 copied = repsize;
1015
1016         /* Suck it into the iovec, verify checksum if not done by hw. */
1017         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1018                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1019                 goto out_unlock;
1020         }
1021
1022         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1023
1024         xprt_adjust_cwnd(xprt, task, copied);
1025         xprt_complete_rqst(task, copied);
1026
1027  out_unlock:
1028         spin_unlock(&xprt->transport_lock);
1029  dropit:
1030         skb_free_datagram(sk, skb);
1031  out:
1032         read_unlock_bh(&sk->sk_callback_lock);
1033 }
1034
1035 /*
1036  * Helper function to force a TCP close if the server is sending
1037  * junk and/or it has put us in CLOSE_WAIT
1038  */
1039 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1040 {
1041         xprt_force_disconnect(xprt);
1042 }
1043
1044 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1045 {
1046         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1047         size_t len, used;
1048         char *p;
1049
1050         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1051         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1052         used = xdr_skb_read_bits(desc, p, len);
1053         transport->tcp_offset += used;
1054         if (used != len)
1055                 return;
1056
1057         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1058         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1059                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1060         else
1061                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1062         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1063
1064         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1065         transport->tcp_offset = 0;
1066
1067         /* Sanity check of the record length */
1068         if (unlikely(transport->tcp_reclen < 8)) {
1069                 dprintk("RPC:       invalid TCP record fragment length\n");
1070                 xs_tcp_force_close(xprt);
1071                 return;
1072         }
1073         dprintk("RPC:       reading TCP record fragment of length %d\n",
1074                         transport->tcp_reclen);
1075 }
1076
1077 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1078 {
1079         if (transport->tcp_offset == transport->tcp_reclen) {
1080                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1081                 transport->tcp_offset = 0;
1082                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1083                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1084                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1085                         transport->tcp_copied = 0;
1086                 }
1087         }
1088 }
1089
1090 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1091 {
1092         size_t len, used;
1093         char *p;
1094
1095         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1096         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1097         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1098         used = xdr_skb_read_bits(desc, p, len);
1099         transport->tcp_offset += used;
1100         if (used != len)
1101                 return;
1102         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1103         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1104         transport->tcp_copied = 4;
1105         dprintk("RPC:       reading %s XID %08x\n",
1106                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1107                                                               : "request with",
1108                         ntohl(transport->tcp_xid));
1109         xs_tcp_check_fraghdr(transport);
1110 }
1111
1112 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1113                                        struct xdr_skb_reader *desc)
1114 {
1115         size_t len, used;
1116         u32 offset;
1117         char *p;
1118
1119         /*
1120          * We want transport->tcp_offset to be 8 at the end of this routine
1121          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1122          * When this function is called for the first time,
1123          * transport->tcp_offset is 4 (after having already read the xid).
1124          */
1125         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1126         len = sizeof(transport->tcp_calldir) - offset;
1127         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1128         p = ((char *) &transport->tcp_calldir) + offset;
1129         used = xdr_skb_read_bits(desc, p, len);
1130         transport->tcp_offset += used;
1131         if (used != len)
1132                 return;
1133         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1134         /*
1135          * We don't yet have the XDR buffer, so we will write the calldir
1136          * out after we get the buffer from the 'struct rpc_rqst'
1137          */
1138         switch (ntohl(transport->tcp_calldir)) {
1139         case RPC_REPLY:
1140                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1141                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1142                 transport->tcp_flags |= TCP_RPC_REPLY;
1143                 break;
1144         case RPC_CALL:
1145                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1146                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1147                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1148                 break;
1149         default:
1150                 dprintk("RPC:       invalid request message type\n");
1151                 xs_tcp_force_close(&transport->xprt);
1152         }
1153         xs_tcp_check_fraghdr(transport);
1154 }
1155
1156 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1157                                      struct xdr_skb_reader *desc,
1158                                      struct rpc_rqst *req)
1159 {
1160         struct sock_xprt *transport =
1161                                 container_of(xprt, struct sock_xprt, xprt);
1162         struct xdr_buf *rcvbuf;
1163         size_t len;
1164         ssize_t r;
1165
1166         rcvbuf = &req->rq_private_buf;
1167
1168         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1169                 /*
1170                  * Save the RPC direction in the XDR buffer
1171                  */
1172                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1173                         &transport->tcp_calldir,
1174                         sizeof(transport->tcp_calldir));
1175                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1176                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1177         }
1178
1179         len = desc->count;
1180         if (len > transport->tcp_reclen - transport->tcp_offset) {
1181                 struct xdr_skb_reader my_desc;
1182
1183                 len = transport->tcp_reclen - transport->tcp_offset;
1184                 memcpy(&my_desc, desc, sizeof(my_desc));
1185                 my_desc.count = len;
1186                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1187                                           &my_desc, xdr_skb_read_bits);
1188                 desc->count -= r;
1189                 desc->offset += r;
1190         } else
1191                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1192                                           desc, xdr_skb_read_bits);
1193
1194         if (r > 0) {
1195                 transport->tcp_copied += r;
1196                 transport->tcp_offset += r;
1197         }
1198         if (r != len) {
1199                 /* Error when copying to the receive buffer,
1200                  * usually because we weren't able to allocate
1201                  * additional buffer pages. All we can do now
1202                  * is turn off TCP_RCV_COPY_DATA, so the request
1203                  * will not receive any additional updates,
1204                  * and time out.
1205                  * Any remaining data from this record will
1206                  * be discarded.
1207                  */
1208                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1209                 dprintk("RPC:       XID %08x truncated request\n",
1210                                 ntohl(transport->tcp_xid));
1211                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1212                                 "tcp_offset = %u, tcp_reclen = %u\n",
1213                                 xprt, transport->tcp_copied,
1214                                 transport->tcp_offset, transport->tcp_reclen);
1215                 return;
1216         }
1217
1218         dprintk("RPC:       XID %08x read %Zd bytes\n",
1219                         ntohl(transport->tcp_xid), r);
1220         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1221                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1222                         transport->tcp_offset, transport->tcp_reclen);
1223
1224         if (transport->tcp_copied == req->rq_private_buf.buflen)
1225                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1226         else if (transport->tcp_offset == transport->tcp_reclen) {
1227                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1228                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1229         }
1230 }
1231
1232 /*
1233  * Finds the request corresponding to the RPC xid and invokes the common
1234  * tcp read code to read the data.
1235  */
1236 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1237                                     struct xdr_skb_reader *desc)
1238 {
1239         struct sock_xprt *transport =
1240                                 container_of(xprt, struct sock_xprt, xprt);
1241         struct rpc_rqst *req;
1242
1243         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1244
1245         /* Find and lock the request corresponding to this xid */
1246         spin_lock(&xprt->transport_lock);
1247         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1248         if (!req) {
1249                 dprintk("RPC:       XID %08x request not found!\n",
1250                                 ntohl(transport->tcp_xid));
1251                 spin_unlock(&xprt->transport_lock);
1252                 return -1;
1253         }
1254
1255         xs_tcp_read_common(xprt, desc, req);
1256
1257         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1258                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1259
1260         spin_unlock(&xprt->transport_lock);
1261         return 0;
1262 }
1263
1264 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1265 /*
1266  * Obtains an rpc_rqst previously allocated and invokes the common
1267  * tcp read code to read the data.  The result is placed in the callback
1268  * queue.
1269  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1270  * connection and return -1.
1271  */
1272 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1273                                        struct xdr_skb_reader *desc)
1274 {
1275         struct sock_xprt *transport =
1276                                 container_of(xprt, struct sock_xprt, xprt);
1277         struct rpc_rqst *req;
1278
1279         /* Look up and lock the request corresponding to the given XID */
1280         spin_lock(&xprt->transport_lock);
1281         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1282         if (req == NULL) {
1283                 spin_unlock(&xprt->transport_lock);
1284                 printk(KERN_WARNING "Callback slot table overflowed\n");
1285                 xprt_force_disconnect(xprt);
1286                 return -1;
1287         }
1288
1289         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1290         xs_tcp_read_common(xprt, desc, req);
1291
1292         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1293                 xprt_complete_bc_request(req, transport->tcp_copied);
1294         spin_unlock(&xprt->transport_lock);
1295
1296         return 0;
1297 }
1298
1299 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1300                                         struct xdr_skb_reader *desc)
1301 {
1302         struct sock_xprt *transport =
1303                                 container_of(xprt, struct sock_xprt, xprt);
1304
1305         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1306                 xs_tcp_read_reply(xprt, desc) :
1307                 xs_tcp_read_callback(xprt, desc);
1308 }
1309 #else
1310 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1311                                         struct xdr_skb_reader *desc)
1312 {
1313         return xs_tcp_read_reply(xprt, desc);
1314 }
1315 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1316
1317 /*
1318  * Read data off the transport.  This can be either an RPC_CALL or an
1319  * RPC_REPLY.  Relay the processing to helper functions.
1320  */
1321 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1322                                     struct xdr_skb_reader *desc)
1323 {
1324         struct sock_xprt *transport =
1325                                 container_of(xprt, struct sock_xprt, xprt);
1326
1327         if (_xs_tcp_read_data(xprt, desc) == 0)
1328                 xs_tcp_check_fraghdr(transport);
1329         else {
1330                 /*
1331                  * The transport_lock protects the request handling.
1332                  * There's no need to hold it to update the tcp_flags.
1333                  */
1334                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1335         }
1336 }
1337
1338 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1339 {
1340         size_t len;
1341
1342         len = transport->tcp_reclen - transport->tcp_offset;
1343         if (len > desc->count)
1344                 len = desc->count;
1345         desc->count -= len;
1346         desc->offset += len;
1347         transport->tcp_offset += len;
1348         dprintk("RPC:       discarded %Zu bytes\n", len);
1349         xs_tcp_check_fraghdr(transport);
1350 }
1351
1352 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1353 {
1354         struct rpc_xprt *xprt = rd_desc->arg.data;
1355         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1356         struct xdr_skb_reader desc = {
1357                 .skb    = skb,
1358                 .offset = offset,
1359                 .count  = len,
1360         };
1361
1362         dprintk("RPC:       xs_tcp_data_recv started\n");
1363         do {
1364                 trace_xs_tcp_data_recv(transport);
1365                 /* Read in a new fragment marker if necessary */
1366                 /* Can we ever really expect to get completely empty fragments? */
1367                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1368                         xs_tcp_read_fraghdr(xprt, &desc);
1369                         continue;
1370                 }
1371                 /* Read in the xid if necessary */
1372                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1373                         xs_tcp_read_xid(transport, &desc);
1374                         continue;
1375                 }
1376                 /* Read in the call/reply flag */
1377                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1378                         xs_tcp_read_calldir(transport, &desc);
1379                         continue;
1380                 }
1381                 /* Read in the request data */
1382                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1383                         xs_tcp_read_data(xprt, &desc);
1384                         continue;
1385                 }
1386                 /* Skip over any trailing bytes on short reads */
1387                 xs_tcp_read_discard(transport, &desc);
1388         } while (desc.count);
1389         trace_xs_tcp_data_recv(transport);
1390         dprintk("RPC:       xs_tcp_data_recv done\n");
1391         return len - desc.count;
1392 }
1393
1394 /**
1395  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1396  * @sk: socket with data to read
1397  * @bytes: how much data to read
1398  *
1399  */
1400 static void xs_tcp_data_ready(struct sock *sk)
1401 {
1402         struct rpc_xprt *xprt;
1403         read_descriptor_t rd_desc;
1404         int read;
1405         unsigned long total = 0;
1406
1407         dprintk("RPC:       xs_tcp_data_ready...\n");
1408
1409         read_lock_bh(&sk->sk_callback_lock);
1410         if (!(xprt = xprt_from_sock(sk))) {
1411                 read = 0;
1412                 goto out;
1413         }
1414         /* Any data means we had a useful conversation, so
1415          * the we don't need to delay the next reconnect
1416          */
1417         if (xprt->reestablish_timeout)
1418                 xprt->reestablish_timeout = 0;
1419
1420         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1421         rd_desc.arg.data = xprt;
1422         do {
1423                 rd_desc.count = 65536;
1424                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1425                 if (read > 0)
1426                         total += read;
1427         } while (read > 0);
1428 out:
1429         trace_xs_tcp_data_ready(xprt, read, total);
1430         read_unlock_bh(&sk->sk_callback_lock);
1431 }
1432
1433 /**
1434  * xs_tcp_state_change - callback to handle TCP socket state changes
1435  * @sk: socket whose state has changed
1436  *
1437  */
1438 static void xs_tcp_state_change(struct sock *sk)
1439 {
1440         struct rpc_xprt *xprt;
1441
1442         read_lock_bh(&sk->sk_callback_lock);
1443         if (!(xprt = xprt_from_sock(sk)))
1444                 goto out;
1445         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1446         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1447                         sk->sk_state, xprt_connected(xprt),
1448                         sock_flag(sk, SOCK_DEAD),
1449                         sock_flag(sk, SOCK_ZAPPED),
1450                         sk->sk_shutdown);
1451
1452         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1453         switch (sk->sk_state) {
1454         case TCP_ESTABLISHED:
1455                 spin_lock(&xprt->transport_lock);
1456                 if (!xprt_test_and_set_connected(xprt)) {
1457                         struct sock_xprt *transport = container_of(xprt,
1458                                         struct sock_xprt, xprt);
1459
1460                         /* Reset TCP record info */
1461                         transport->tcp_offset = 0;
1462                         transport->tcp_reclen = 0;
1463                         transport->tcp_copied = 0;
1464                         transport->tcp_flags =
1465                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1466                         xprt->connect_cookie++;
1467
1468                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1469                 }
1470                 spin_unlock(&xprt->transport_lock);
1471                 break;
1472         case TCP_FIN_WAIT1:
1473                 /* The client initiated a shutdown of the socket */
1474                 xprt->connect_cookie++;
1475                 xprt->reestablish_timeout = 0;
1476                 set_bit(XPRT_CLOSING, &xprt->state);
1477                 smp_mb__before_atomic();
1478                 clear_bit(XPRT_CONNECTED, &xprt->state);
1479                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1480                 smp_mb__after_atomic();
1481                 break;
1482         case TCP_CLOSE_WAIT:
1483                 /* The server initiated a shutdown of the socket */
1484                 xprt->connect_cookie++;
1485                 clear_bit(XPRT_CONNECTED, &xprt->state);
1486                 xs_tcp_force_close(xprt);
1487         case TCP_CLOSING:
1488                 /*
1489                  * If the server closed down the connection, make sure that
1490                  * we back off before reconnecting
1491                  */
1492                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1493                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1494                 break;
1495         case TCP_LAST_ACK:
1496                 set_bit(XPRT_CLOSING, &xprt->state);
1497                 smp_mb__before_atomic();
1498                 clear_bit(XPRT_CONNECTED, &xprt->state);
1499                 smp_mb__after_atomic();
1500                 break;
1501         case TCP_CLOSE:
1502                 xs_sock_mark_closed(xprt);
1503         }
1504  out:
1505         read_unlock_bh(&sk->sk_callback_lock);
1506 }
1507
1508 static void xs_write_space(struct sock *sk)
1509 {
1510         struct socket *sock;
1511         struct rpc_xprt *xprt;
1512
1513         if (unlikely(!(sock = sk->sk_socket)))
1514                 return;
1515         clear_bit(SOCK_NOSPACE, &sock->flags);
1516
1517         if (unlikely(!(xprt = xprt_from_sock(sk))))
1518                 return;
1519         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1520                 return;
1521
1522         xprt_write_space(xprt);
1523 }
1524
1525 /**
1526  * xs_udp_write_space - callback invoked when socket buffer space
1527  *                             becomes available
1528  * @sk: socket whose state has changed
1529  *
1530  * Called when more output buffer space is available for this socket.
1531  * We try not to wake our writers until they can make "significant"
1532  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1533  * with a bunch of small requests.
1534  */
1535 static void xs_udp_write_space(struct sock *sk)
1536 {
1537         read_lock_bh(&sk->sk_callback_lock);
1538
1539         /* from net/core/sock.c:sock_def_write_space */
1540         if (sock_writeable(sk))
1541                 xs_write_space(sk);
1542
1543         read_unlock_bh(&sk->sk_callback_lock);
1544 }
1545
1546 /**
1547  * xs_tcp_write_space - callback invoked when socket buffer space
1548  *                             becomes available
1549  * @sk: socket whose state has changed
1550  *
1551  * Called when more output buffer space is available for this socket.
1552  * We try not to wake our writers until they can make "significant"
1553  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1554  * with a bunch of small requests.
1555  */
1556 static void xs_tcp_write_space(struct sock *sk)
1557 {
1558         read_lock_bh(&sk->sk_callback_lock);
1559
1560         /* from net/core/stream.c:sk_stream_write_space */
1561         if (sk_stream_is_writeable(sk))
1562                 xs_write_space(sk);
1563
1564         read_unlock_bh(&sk->sk_callback_lock);
1565 }
1566
1567 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1568 {
1569         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1570         struct sock *sk = transport->inet;
1571
1572         if (transport->rcvsize) {
1573                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1574                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1575         }
1576         if (transport->sndsize) {
1577                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1578                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1579                 sk->sk_write_space(sk);
1580         }
1581 }
1582
1583 /**
1584  * xs_udp_set_buffer_size - set send and receive limits
1585  * @xprt: generic transport
1586  * @sndsize: requested size of send buffer, in bytes
1587  * @rcvsize: requested size of receive buffer, in bytes
1588  *
1589  * Set socket send and receive buffer size limits.
1590  */
1591 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1592 {
1593         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1594
1595         transport->sndsize = 0;
1596         if (sndsize)
1597                 transport->sndsize = sndsize + 1024;
1598         transport->rcvsize = 0;
1599         if (rcvsize)
1600                 transport->rcvsize = rcvsize + 1024;
1601
1602         xs_udp_do_set_buffer_size(xprt);
1603 }
1604
1605 /**
1606  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1607  * @task: task that timed out
1608  *
1609  * Adjust the congestion window after a retransmit timeout has occurred.
1610  */
1611 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1612 {
1613         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1614 }
1615
1616 static unsigned short xs_get_random_port(void)
1617 {
1618         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1619         unsigned short rand = (unsigned short) prandom_u32() % range;
1620         return rand + xprt_min_resvport;
1621 }
1622
1623 /**
1624  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1625  * @sock: socket
1626  *
1627  * Note that this function has to be called on all sockets that share the
1628  * same port, and it must be called before binding.
1629  */
1630 static void xs_sock_set_reuseport(struct socket *sock)
1631 {
1632         int opt = 1;
1633
1634         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1635                         (char *)&opt, sizeof(opt));
1636 }
1637
1638 static unsigned short xs_sock_getport(struct socket *sock)
1639 {
1640         struct sockaddr_storage buf;
1641         int buflen;
1642         unsigned short port = 0;
1643
1644         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1645                 goto out;
1646         switch (buf.ss_family) {
1647         case AF_INET6:
1648                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1649                 break;
1650         case AF_INET:
1651                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1652         }
1653 out:
1654         return port;
1655 }
1656
1657 /**
1658  * xs_set_port - reset the port number in the remote endpoint address
1659  * @xprt: generic transport
1660  * @port: new port number
1661  *
1662  */
1663 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1664 {
1665         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1666
1667         rpc_set_port(xs_addr(xprt), port);
1668         xs_update_peer_port(xprt);
1669 }
1670
1671 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1672 {
1673         if (transport->srcport == 0)
1674                 transport->srcport = xs_sock_getport(sock);
1675 }
1676
1677 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1678 {
1679         unsigned short port = transport->srcport;
1680
1681         if (port == 0 && transport->xprt.resvport)
1682                 port = xs_get_random_port();
1683         return port;
1684 }
1685
1686 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1687 {
1688         if (transport->srcport != 0)
1689                 transport->srcport = 0;
1690         if (!transport->xprt.resvport)
1691                 return 0;
1692         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1693                 return xprt_max_resvport;
1694         return --port;
1695 }
1696 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1697 {
1698         struct sockaddr_storage myaddr;
1699         int err, nloop = 0;
1700         unsigned short port = xs_get_srcport(transport);
1701         unsigned short last;
1702
1703         /*
1704          * If we are asking for any ephemeral port (i.e. port == 0 &&
1705          * transport->xprt.resvport == 0), don't bind.  Let the local
1706          * port selection happen implicitly when the socket is used
1707          * (for example at connect time).
1708          *
1709          * This ensures that we can continue to establish TCP
1710          * connections even when all local ephemeral ports are already
1711          * a part of some TCP connection.  This makes no difference
1712          * for UDP sockets, but also doens't harm them.
1713          *
1714          * If we're asking for any reserved port (i.e. port == 0 &&
1715          * transport->xprt.resvport == 1) xs_get_srcport above will
1716          * ensure that port is non-zero and we will bind as needed.
1717          */
1718         if (port == 0)
1719                 return 0;
1720
1721         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1722         do {
1723                 rpc_set_port((struct sockaddr *)&myaddr, port);
1724                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1725                                 transport->xprt.addrlen);
1726                 if (err == 0) {
1727                         transport->srcport = port;
1728                         break;
1729                 }
1730                 last = port;
1731                 port = xs_next_srcport(transport, port);
1732                 if (port > last)
1733                         nloop++;
1734         } while (err == -EADDRINUSE && nloop != 2);
1735
1736         if (myaddr.ss_family == AF_INET)
1737                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1738                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1739                                 port, err ? "failed" : "ok", err);
1740         else
1741                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1742                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1743                                 port, err ? "failed" : "ok", err);
1744         return err;
1745 }
1746
1747 /*
1748  * We don't support autobind on AF_LOCAL sockets
1749  */
1750 static void xs_local_rpcbind(struct rpc_task *task)
1751 {
1752         rcu_read_lock();
1753         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1754         rcu_read_unlock();
1755 }
1756
1757 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1758 {
1759 }
1760
1761 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1762 static struct lock_class_key xs_key[2];
1763 static struct lock_class_key xs_slock_key[2];
1764
1765 static inline void xs_reclassify_socketu(struct socket *sock)
1766 {
1767         struct sock *sk = sock->sk;
1768
1769         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1770                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1771 }
1772
1773 static inline void xs_reclassify_socket4(struct socket *sock)
1774 {
1775         struct sock *sk = sock->sk;
1776
1777         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1778                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1779 }
1780
1781 static inline void xs_reclassify_socket6(struct socket *sock)
1782 {
1783         struct sock *sk = sock->sk;
1784
1785         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1786                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1787 }
1788
1789 static inline void xs_reclassify_socket(int family, struct socket *sock)
1790 {
1791         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1792         if (sock_owned_by_user(sock->sk))
1793                 return;
1794
1795         switch (family) {
1796         case AF_LOCAL:
1797                 xs_reclassify_socketu(sock);
1798                 break;
1799         case AF_INET:
1800                 xs_reclassify_socket4(sock);
1801                 break;
1802         case AF_INET6:
1803                 xs_reclassify_socket6(sock);
1804                 break;
1805         }
1806 }
1807 #else
1808 static inline void xs_reclassify_socketu(struct socket *sock)
1809 {
1810 }
1811
1812 static inline void xs_reclassify_socket4(struct socket *sock)
1813 {
1814 }
1815
1816 static inline void xs_reclassify_socket6(struct socket *sock)
1817 {
1818 }
1819
1820 static inline void xs_reclassify_socket(int family, struct socket *sock)
1821 {
1822 }
1823 #endif
1824
1825 static void xs_dummy_setup_socket(struct work_struct *work)
1826 {
1827 }
1828
1829 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1830                 struct sock_xprt *transport, int family, int type,
1831                 int protocol, bool reuseport)
1832 {
1833         struct socket *sock;
1834         int err;
1835
1836         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1837         if (err < 0) {
1838                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1839                                 protocol, -err);
1840                 goto out;
1841         }
1842         xs_reclassify_socket(family, sock);
1843
1844         if (reuseport)
1845                 xs_sock_set_reuseport(sock);
1846
1847         err = xs_bind(transport, sock);
1848         if (err) {
1849                 sock_release(sock);
1850                 goto out;
1851         }
1852
1853         return sock;
1854 out:
1855         return ERR_PTR(err);
1856 }
1857
1858 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1859                                       struct socket *sock)
1860 {
1861         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1862                                                                         xprt);
1863
1864         if (!transport->inet) {
1865                 struct sock *sk = sock->sk;
1866
1867                 write_lock_bh(&sk->sk_callback_lock);
1868
1869                 xs_save_old_callbacks(transport, sk);
1870
1871                 sk->sk_user_data = xprt;
1872                 sk->sk_data_ready = xs_local_data_ready;
1873                 sk->sk_write_space = xs_udp_write_space;
1874                 sk->sk_error_report = xs_error_report;
1875                 sk->sk_allocation = GFP_ATOMIC;
1876
1877                 xprt_clear_connected(xprt);
1878
1879                 /* Reset to new socket */
1880                 transport->sock = sock;
1881                 transport->inet = sk;
1882
1883                 write_unlock_bh(&sk->sk_callback_lock);
1884         }
1885
1886         /* Tell the socket layer to start connecting... */
1887         xprt->stat.connect_count++;
1888         xprt->stat.connect_start = jiffies;
1889         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1890 }
1891
1892 /**
1893  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1894  * @xprt: RPC transport to connect
1895  * @transport: socket transport to connect
1896  * @create_sock: function to create a socket of the correct type
1897  */
1898 static int xs_local_setup_socket(struct sock_xprt *transport)
1899 {
1900         struct rpc_xprt *xprt = &transport->xprt;
1901         struct socket *sock;
1902         int status = -EIO;
1903
1904         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1905                                         SOCK_STREAM, 0, &sock, 1);
1906         if (status < 0) {
1907                 dprintk("RPC:       can't create AF_LOCAL "
1908                         "transport socket (%d).\n", -status);
1909                 goto out;
1910         }
1911         xs_reclassify_socketu(sock);
1912
1913         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1914                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915
1916         status = xs_local_finish_connecting(xprt, sock);
1917         trace_rpc_socket_connect(xprt, sock, status);
1918         switch (status) {
1919         case 0:
1920                 dprintk("RPC:       xprt %p connected to %s\n",
1921                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1922                 xprt_set_connected(xprt);
1923         case -ENOBUFS:
1924                 break;
1925         case -ENOENT:
1926                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1927                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1928                 break;
1929         case -ECONNREFUSED:
1930                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1931                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1932                 break;
1933         default:
1934                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1935                                 __func__, -status,
1936                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1937         }
1938
1939 out:
1940         xprt_clear_connecting(xprt);
1941         xprt_wake_pending_tasks(xprt, status);
1942         return status;
1943 }
1944
1945 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1946 {
1947         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1948         int ret;
1949
1950          if (RPC_IS_ASYNC(task)) {
1951                 /*
1952                  * We want the AF_LOCAL connect to be resolved in the
1953                  * filesystem namespace of the process making the rpc
1954                  * call.  Thus we connect synchronously.
1955                  *
1956                  * If we want to support asynchronous AF_LOCAL calls,
1957                  * we'll need to figure out how to pass a namespace to
1958                  * connect.
1959                  */
1960                 rpc_exit(task, -ENOTCONN);
1961                 return;
1962         }
1963         ret = xs_local_setup_socket(transport);
1964         if (ret && !RPC_IS_SOFTCONN(task))
1965                 msleep_interruptible(15000);
1966 }
1967
1968 #ifdef CONFIG_SUNRPC_SWAP
1969 static void xs_set_memalloc(struct rpc_xprt *xprt)
1970 {
1971         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1972                         xprt);
1973
1974         if (xprt->swapper)
1975                 sk_set_memalloc(transport->inet);
1976 }
1977
1978 /**
1979  * xs_swapper - Tag this transport as being used for swap.
1980  * @xprt: transport to tag
1981  * @enable: enable/disable
1982  *
1983  */
1984 int xs_swapper(struct rpc_xprt *xprt, int enable)
1985 {
1986         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1987                         xprt);
1988         int err = 0;
1989
1990         if (enable) {
1991                 xprt->swapper++;
1992                 xs_set_memalloc(xprt);
1993         } else if (xprt->swapper) {
1994                 xprt->swapper--;
1995                 sk_clear_memalloc(transport->inet);
1996         }
1997
1998         return err;
1999 }
2000 EXPORT_SYMBOL_GPL(xs_swapper);
2001 #else
2002 static void xs_set_memalloc(struct rpc_xprt *xprt)
2003 {
2004 }
2005 #endif
2006
2007 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2008 {
2009         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2010
2011         if (!transport->inet) {
2012                 struct sock *sk = sock->sk;
2013
2014                 write_lock_bh(&sk->sk_callback_lock);
2015
2016                 xs_save_old_callbacks(transport, sk);
2017
2018                 sk->sk_user_data = xprt;
2019                 sk->sk_data_ready = xs_udp_data_ready;
2020                 sk->sk_write_space = xs_udp_write_space;
2021                 sk->sk_allocation = GFP_ATOMIC;
2022
2023                 xprt_set_connected(xprt);
2024
2025                 /* Reset to new socket */
2026                 transport->sock = sock;
2027                 transport->inet = sk;
2028
2029                 xs_set_memalloc(xprt);
2030
2031                 write_unlock_bh(&sk->sk_callback_lock);
2032         }
2033         xs_udp_do_set_buffer_size(xprt);
2034 }
2035
2036 static void xs_udp_setup_socket(struct work_struct *work)
2037 {
2038         struct sock_xprt *transport =
2039                 container_of(work, struct sock_xprt, connect_worker.work);
2040         struct rpc_xprt *xprt = &transport->xprt;
2041         struct socket *sock = transport->sock;
2042         int status = -EIO;
2043
2044         sock = xs_create_sock(xprt, transport,
2045                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2046                         IPPROTO_UDP, false);
2047         if (IS_ERR(sock))
2048                 goto out;
2049
2050         dprintk("RPC:       worker connecting xprt %p via %s to "
2051                                 "%s (port %s)\n", xprt,
2052                         xprt->address_strings[RPC_DISPLAY_PROTO],
2053                         xprt->address_strings[RPC_DISPLAY_ADDR],
2054                         xprt->address_strings[RPC_DISPLAY_PORT]);
2055
2056         xs_udp_finish_connecting(xprt, sock);
2057         trace_rpc_socket_connect(xprt, sock, 0);
2058         status = 0;
2059 out:
2060         xprt_unlock_connect(xprt, transport);
2061         xprt_clear_connecting(xprt);
2062         xprt_wake_pending_tasks(xprt, status);
2063 }
2064
2065 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2066 {
2067         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2068         int ret = -ENOTCONN;
2069
2070         if (!transport->inet) {
2071                 struct sock *sk = sock->sk;
2072                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2073                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2074                 unsigned int opt_on = 1;
2075
2076                 /* TCP Keepalive options */
2077                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2078                                 (char *)&opt_on, sizeof(opt_on));
2079                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2080                                 (char *)&keepidle, sizeof(keepidle));
2081                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2082                                 (char *)&keepidle, sizeof(keepidle));
2083                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2084                                 (char *)&keepcnt, sizeof(keepcnt));
2085
2086                 write_lock_bh(&sk->sk_callback_lock);
2087
2088                 xs_save_old_callbacks(transport, sk);
2089
2090                 sk->sk_user_data = xprt;
2091                 sk->sk_data_ready = xs_tcp_data_ready;
2092                 sk->sk_state_change = xs_tcp_state_change;
2093                 sk->sk_write_space = xs_tcp_write_space;
2094                 sk->sk_error_report = xs_error_report;
2095                 sk->sk_allocation = GFP_ATOMIC;
2096
2097                 /* socket options */
2098                 sock_reset_flag(sk, SOCK_LINGER);
2099                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2100
2101                 xprt_clear_connected(xprt);
2102
2103                 /* Reset to new socket */
2104                 transport->sock = sock;
2105                 transport->inet = sk;
2106
2107                 write_unlock_bh(&sk->sk_callback_lock);
2108         }
2109
2110         if (!xprt_bound(xprt))
2111                 goto out;
2112
2113         xs_set_memalloc(xprt);
2114
2115         /* Tell the socket layer to start connecting... */
2116         xprt->stat.connect_count++;
2117         xprt->stat.connect_start = jiffies;
2118         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2119         switch (ret) {
2120         case 0:
2121                 xs_set_srcport(transport, sock);
2122         case -EINPROGRESS:
2123                 /* SYN_SENT! */
2124                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2125                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2126         }
2127 out:
2128         return ret;
2129 }
2130
2131 /**
2132  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2133  * @xprt: RPC transport to connect
2134  * @transport: socket transport to connect
2135  * @create_sock: function to create a socket of the correct type
2136  *
2137  * Invoked by a work queue tasklet.
2138  */
2139 static void xs_tcp_setup_socket(struct work_struct *work)
2140 {
2141         struct sock_xprt *transport =
2142                 container_of(work, struct sock_xprt, connect_worker.work);
2143         struct socket *sock = transport->sock;
2144         struct rpc_xprt *xprt = &transport->xprt;
2145         int status = -EIO;
2146
2147         if (!sock) {
2148                 sock = xs_create_sock(xprt, transport,
2149                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2150                                 IPPROTO_TCP, true);
2151                 if (IS_ERR(sock)) {
2152                         status = PTR_ERR(sock);
2153                         goto out;
2154                 }
2155         }
2156
2157         dprintk("RPC:       worker connecting xprt %p via %s to "
2158                                 "%s (port %s)\n", xprt,
2159                         xprt->address_strings[RPC_DISPLAY_PROTO],
2160                         xprt->address_strings[RPC_DISPLAY_ADDR],
2161                         xprt->address_strings[RPC_DISPLAY_PORT]);
2162
2163         status = xs_tcp_finish_connecting(xprt, sock);
2164         trace_rpc_socket_connect(xprt, sock, status);
2165         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2166                         xprt, -status, xprt_connected(xprt),
2167                         sock->sk->sk_state);
2168         switch (status) {
2169         default:
2170                 printk("%s: connect returned unhandled error %d\n",
2171                         __func__, status);
2172         case -EADDRNOTAVAIL:
2173                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2174                  * and retry
2175                  */
2176                 xs_tcp_force_close(xprt);
2177                 break;
2178         case 0:
2179         case -EINPROGRESS:
2180         case -EALREADY:
2181                 xprt_unlock_connect(xprt, transport);
2182                 xprt_clear_connecting(xprt);
2183                 return;
2184         case -EINVAL:
2185                 /* Happens, for instance, if the user specified a link
2186                  * local IPv6 address without a scope-id.
2187                  */
2188         case -ECONNREFUSED:
2189         case -ECONNRESET:
2190         case -ENETUNREACH:
2191         case -EADDRINUSE:
2192         case -ENOBUFS:
2193                 /* retry with existing socket, after a delay */
2194                 xs_tcp_force_close(xprt);
2195                 goto out;
2196         }
2197         status = -EAGAIN;
2198 out:
2199         xprt_unlock_connect(xprt, transport);
2200         xprt_clear_connecting(xprt);
2201         xprt_wake_pending_tasks(xprt, status);
2202 }
2203
2204 /**
2205  * xs_connect - connect a socket to a remote endpoint
2206  * @xprt: pointer to transport structure
2207  * @task: address of RPC task that manages state of connect request
2208  *
2209  * TCP: If the remote end dropped the connection, delay reconnecting.
2210  *
2211  * UDP socket connects are synchronous, but we use a work queue anyway
2212  * to guarantee that even unprivileged user processes can set up a
2213  * socket on a privileged port.
2214  *
2215  * If a UDP socket connect fails, the delay behavior here prevents
2216  * retry floods (hard mounts).
2217  */
2218 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2219 {
2220         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2221
2222         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2223
2224         /* Start by resetting any existing state */
2225         xs_reset_transport(transport);
2226
2227         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2228                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2229                                 "seconds\n",
2230                                 xprt, xprt->reestablish_timeout / HZ);
2231                 queue_delayed_work(rpciod_workqueue,
2232                                    &transport->connect_worker,
2233                                    xprt->reestablish_timeout);
2234                 xprt->reestablish_timeout <<= 1;
2235                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2236                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2237                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2238                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2239         } else {
2240                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2241                 queue_delayed_work(rpciod_workqueue,
2242                                    &transport->connect_worker, 0);
2243         }
2244 }
2245
2246 /**
2247  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2248  * @xprt: rpc_xprt struct containing statistics
2249  * @seq: output file
2250  *
2251  */
2252 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2253 {
2254         long idle_time = 0;
2255
2256         if (xprt_connected(xprt))
2257                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2258
2259         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2260                         "%llu %llu %lu %llu %llu\n",
2261                         xprt->stat.bind_count,
2262                         xprt->stat.connect_count,
2263                         xprt->stat.connect_time,
2264                         idle_time,
2265                         xprt->stat.sends,
2266                         xprt->stat.recvs,
2267                         xprt->stat.bad_xids,
2268                         xprt->stat.req_u,
2269                         xprt->stat.bklog_u,
2270                         xprt->stat.max_slots,
2271                         xprt->stat.sending_u,
2272                         xprt->stat.pending_u);
2273 }
2274
2275 /**
2276  * xs_udp_print_stats - display UDP socket-specifc stats
2277  * @xprt: rpc_xprt struct containing statistics
2278  * @seq: output file
2279  *
2280  */
2281 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2282 {
2283         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2284
2285         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2286                         "%lu %llu %llu\n",
2287                         transport->srcport,
2288                         xprt->stat.bind_count,
2289                         xprt->stat.sends,
2290                         xprt->stat.recvs,
2291                         xprt->stat.bad_xids,
2292                         xprt->stat.req_u,
2293                         xprt->stat.bklog_u,
2294                         xprt->stat.max_slots,
2295                         xprt->stat.sending_u,
2296                         xprt->stat.pending_u);
2297 }
2298
2299 /**
2300  * xs_tcp_print_stats - display TCP socket-specifc stats
2301  * @xprt: rpc_xprt struct containing statistics
2302  * @seq: output file
2303  *
2304  */
2305 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2306 {
2307         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2308         long idle_time = 0;
2309
2310         if (xprt_connected(xprt))
2311                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2312
2313         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2314                         "%llu %llu %lu %llu %llu\n",
2315                         transport->srcport,
2316                         xprt->stat.bind_count,
2317                         xprt->stat.connect_count,
2318                         xprt->stat.connect_time,
2319                         idle_time,
2320                         xprt->stat.sends,
2321                         xprt->stat.recvs,
2322                         xprt->stat.bad_xids,
2323                         xprt->stat.req_u,
2324                         xprt->stat.bklog_u,
2325                         xprt->stat.max_slots,
2326                         xprt->stat.sending_u,
2327                         xprt->stat.pending_u);
2328 }
2329
2330 /*
2331  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2332  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2333  * to use the server side send routines.
2334  */
2335 static void *bc_malloc(struct rpc_task *task, size_t size)
2336 {
2337         struct page *page;
2338         struct rpc_buffer *buf;
2339
2340         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2341         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2342                 return NULL;
2343
2344         page = alloc_page(GFP_KERNEL);
2345         if (!page)
2346                 return NULL;
2347
2348         buf = page_address(page);
2349         buf->len = PAGE_SIZE;
2350
2351         return buf->data;
2352 }
2353
2354 /*
2355  * Free the space allocated in the bc_alloc routine
2356  */
2357 static void bc_free(void *buffer)
2358 {
2359         struct rpc_buffer *buf;
2360
2361         if (!buffer)
2362                 return;
2363
2364         buf = container_of(buffer, struct rpc_buffer, data);
2365         free_page((unsigned long)buf);
2366 }
2367
2368 /*
2369  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2370  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2371  */
2372 static int bc_sendto(struct rpc_rqst *req)
2373 {
2374         int len;
2375         struct xdr_buf *xbufp = &req->rq_snd_buf;
2376         struct rpc_xprt *xprt = req->rq_xprt;
2377         struct sock_xprt *transport =
2378                                 container_of(xprt, struct sock_xprt, xprt);
2379         struct socket *sock = transport->sock;
2380         unsigned long headoff;
2381         unsigned long tailoff;
2382
2383         xs_encode_stream_record_marker(xbufp);
2384
2385         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2386         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2387         len = svc_send_common(sock, xbufp,
2388                               virt_to_page(xbufp->head[0].iov_base), headoff,
2389                               xbufp->tail[0].iov_base, tailoff);
2390
2391         if (len != xbufp->len) {
2392                 printk(KERN_NOTICE "Error sending entire callback!\n");
2393                 len = -EAGAIN;
2394         }
2395
2396         return len;
2397 }
2398
2399 /*
2400  * The send routine. Borrows from svc_send
2401  */
2402 static int bc_send_request(struct rpc_task *task)
2403 {
2404         struct rpc_rqst *req = task->tk_rqstp;
2405         struct svc_xprt *xprt;
2406         u32                     len;
2407
2408         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2409         /*
2410          * Get the server socket associated with this callback xprt
2411          */
2412         xprt = req->rq_xprt->bc_xprt;
2413
2414         /*
2415          * Grab the mutex to serialize data as the connection is shared
2416          * with the fore channel
2417          */
2418         if (!mutex_trylock(&xprt->xpt_mutex)) {
2419                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2420                 if (!mutex_trylock(&xprt->xpt_mutex))
2421                         return -EAGAIN;
2422                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2423         }
2424         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2425                 len = -ENOTCONN;
2426         else
2427                 len = bc_sendto(req);
2428         mutex_unlock(&xprt->xpt_mutex);
2429
2430         if (len > 0)
2431                 len = 0;
2432
2433         return len;
2434 }
2435
2436 /*
2437  * The close routine. Since this is client initiated, we do nothing
2438  */
2439
2440 static void bc_close(struct rpc_xprt *xprt)
2441 {
2442 }
2443
2444 /*
2445  * The xprt destroy routine. Again, because this connection is client
2446  * initiated, we do nothing
2447  */
2448
2449 static void bc_destroy(struct rpc_xprt *xprt)
2450 {
2451         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2452
2453         xs_xprt_free(xprt);
2454         module_put(THIS_MODULE);
2455 }
2456
2457 static struct rpc_xprt_ops xs_local_ops = {
2458         .reserve_xprt           = xprt_reserve_xprt,
2459         .release_xprt           = xs_tcp_release_xprt,
2460         .alloc_slot             = xprt_alloc_slot,
2461         .rpcbind                = xs_local_rpcbind,
2462         .set_port               = xs_local_set_port,
2463         .connect                = xs_local_connect,
2464         .buf_alloc              = rpc_malloc,
2465         .buf_free               = rpc_free,
2466         .send_request           = xs_local_send_request,
2467         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2468         .close                  = xs_close,
2469         .destroy                = xs_destroy,
2470         .print_stats            = xs_local_print_stats,
2471 };
2472
2473 static struct rpc_xprt_ops xs_udp_ops = {
2474         .set_buffer_size        = xs_udp_set_buffer_size,
2475         .reserve_xprt           = xprt_reserve_xprt_cong,
2476         .release_xprt           = xprt_release_xprt_cong,
2477         .alloc_slot             = xprt_alloc_slot,
2478         .rpcbind                = rpcb_getport_async,
2479         .set_port               = xs_set_port,
2480         .connect                = xs_connect,
2481         .buf_alloc              = rpc_malloc,
2482         .buf_free               = rpc_free,
2483         .send_request           = xs_udp_send_request,
2484         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2485         .timer                  = xs_udp_timer,
2486         .release_request        = xprt_release_rqst_cong,
2487         .close                  = xs_close,
2488         .destroy                = xs_destroy,
2489         .print_stats            = xs_udp_print_stats,
2490 };
2491
2492 static struct rpc_xprt_ops xs_tcp_ops = {
2493         .reserve_xprt           = xprt_reserve_xprt,
2494         .release_xprt           = xs_tcp_release_xprt,
2495         .alloc_slot             = xprt_lock_and_alloc_slot,
2496         .rpcbind                = rpcb_getport_async,
2497         .set_port               = xs_set_port,
2498         .connect                = xs_connect,
2499         .buf_alloc              = rpc_malloc,
2500         .buf_free               = rpc_free,
2501         .send_request           = xs_tcp_send_request,
2502         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2503         .close                  = xs_tcp_close,
2504         .destroy                = xs_destroy,
2505         .print_stats            = xs_tcp_print_stats,
2506 };
2507
2508 /*
2509  * The rpc_xprt_ops for the server backchannel
2510  */
2511
2512 static struct rpc_xprt_ops bc_tcp_ops = {
2513         .reserve_xprt           = xprt_reserve_xprt,
2514         .release_xprt           = xprt_release_xprt,
2515         .alloc_slot             = xprt_alloc_slot,
2516         .buf_alloc              = bc_malloc,
2517         .buf_free               = bc_free,
2518         .send_request           = bc_send_request,
2519         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2520         .close                  = bc_close,
2521         .destroy                = bc_destroy,
2522         .print_stats            = xs_tcp_print_stats,
2523 };
2524
2525 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2526 {
2527         static const struct sockaddr_in sin = {
2528                 .sin_family             = AF_INET,
2529                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2530         };
2531         static const struct sockaddr_in6 sin6 = {
2532                 .sin6_family            = AF_INET6,
2533                 .sin6_addr              = IN6ADDR_ANY_INIT,
2534         };
2535
2536         switch (family) {
2537         case AF_LOCAL:
2538                 break;
2539         case AF_INET:
2540                 memcpy(sap, &sin, sizeof(sin));
2541                 break;
2542         case AF_INET6:
2543                 memcpy(sap, &sin6, sizeof(sin6));
2544                 break;
2545         default:
2546                 dprintk("RPC:       %s: Bad address family\n", __func__);
2547                 return -EAFNOSUPPORT;
2548         }
2549         return 0;
2550 }
2551
2552 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2553                                       unsigned int slot_table_size,
2554                                       unsigned int max_slot_table_size)
2555 {
2556         struct rpc_xprt *xprt;
2557         struct sock_xprt *new;
2558
2559         if (args->addrlen > sizeof(xprt->addr)) {
2560                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2561                 return ERR_PTR(-EBADF);
2562         }
2563
2564         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2565                         max_slot_table_size);
2566         if (xprt == NULL) {
2567                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2568                                 "rpc_xprt\n");
2569                 return ERR_PTR(-ENOMEM);
2570         }
2571
2572         new = container_of(xprt, struct sock_xprt, xprt);
2573         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2574         xprt->addrlen = args->addrlen;
2575         if (args->srcaddr)
2576                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2577         else {
2578                 int err;
2579                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2580                                         (struct sockaddr *)&new->srcaddr);
2581                 if (err != 0) {
2582                         xprt_free(xprt);
2583                         return ERR_PTR(err);
2584                 }
2585         }
2586
2587         return xprt;
2588 }
2589
2590 static const struct rpc_timeout xs_local_default_timeout = {
2591         .to_initval = 10 * HZ,
2592         .to_maxval = 10 * HZ,
2593         .to_retries = 2,
2594 };
2595
2596 /**
2597  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2598  * @args: rpc transport creation arguments
2599  *
2600  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2601  */
2602 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2603 {
2604         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2605         struct sock_xprt *transport;
2606         struct rpc_xprt *xprt;
2607         struct rpc_xprt *ret;
2608
2609         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2610                         xprt_max_tcp_slot_table_entries);
2611         if (IS_ERR(xprt))
2612                 return xprt;
2613         transport = container_of(xprt, struct sock_xprt, xprt);
2614
2615         xprt->prot = 0;
2616         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2617         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2618
2619         xprt->bind_timeout = XS_BIND_TO;
2620         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2621         xprt->idle_timeout = XS_IDLE_DISC_TO;
2622
2623         xprt->ops = &xs_local_ops;
2624         xprt->timeout = &xs_local_default_timeout;
2625
2626         INIT_DELAYED_WORK(&transport->connect_worker,
2627                         xs_dummy_setup_socket);
2628
2629         switch (sun->sun_family) {
2630         case AF_LOCAL:
2631                 if (sun->sun_path[0] != '/') {
2632                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2633                                         sun->sun_path);
2634                         ret = ERR_PTR(-EINVAL);
2635                         goto out_err;
2636                 }
2637                 xprt_set_bound(xprt);
2638                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2639                 ret = ERR_PTR(xs_local_setup_socket(transport));
2640                 if (ret)
2641                         goto out_err;
2642                 break;
2643         default:
2644                 ret = ERR_PTR(-EAFNOSUPPORT);
2645                 goto out_err;
2646         }
2647
2648         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2649                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2650
2651         if (try_module_get(THIS_MODULE))
2652                 return xprt;
2653         ret = ERR_PTR(-EINVAL);
2654 out_err:
2655         xs_xprt_free(xprt);
2656         return ret;
2657 }
2658
2659 static const struct rpc_timeout xs_udp_default_timeout = {
2660         .to_initval = 5 * HZ,
2661         .to_maxval = 30 * HZ,
2662         .to_increment = 5 * HZ,
2663         .to_retries = 5,
2664 };
2665
2666 /**
2667  * xs_setup_udp - Set up transport to use a UDP socket
2668  * @args: rpc transport creation arguments
2669  *
2670  */
2671 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2672 {
2673         struct sockaddr *addr = args->dstaddr;
2674         struct rpc_xprt *xprt;
2675         struct sock_xprt *transport;
2676         struct rpc_xprt *ret;
2677
2678         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2679                         xprt_udp_slot_table_entries);
2680         if (IS_ERR(xprt))
2681                 return xprt;
2682         transport = container_of(xprt, struct sock_xprt, xprt);
2683
2684         xprt->prot = IPPROTO_UDP;
2685         xprt->tsh_size = 0;
2686         /* XXX: header size can vary due to auth type, IPv6, etc. */
2687         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2688
2689         xprt->bind_timeout = XS_BIND_TO;
2690         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2691         xprt->idle_timeout = XS_IDLE_DISC_TO;
2692
2693         xprt->ops = &xs_udp_ops;
2694
2695         xprt->timeout = &xs_udp_default_timeout;
2696
2697         switch (addr->sa_family) {
2698         case AF_INET:
2699                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2700                         xprt_set_bound(xprt);
2701
2702                 INIT_DELAYED_WORK(&transport->connect_worker,
2703                                         xs_udp_setup_socket);
2704                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2705                 break;
2706         case AF_INET6:
2707                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2708                         xprt_set_bound(xprt);
2709
2710                 INIT_DELAYED_WORK(&transport->connect_worker,
2711                                         xs_udp_setup_socket);
2712                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2713                 break;
2714         default:
2715                 ret = ERR_PTR(-EAFNOSUPPORT);
2716                 goto out_err;
2717         }
2718
2719         if (xprt_bound(xprt))
2720                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2721                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2722                                 xprt->address_strings[RPC_DISPLAY_PORT],
2723                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2724         else
2725                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2726                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2727                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2728
2729         if (try_module_get(THIS_MODULE))
2730                 return xprt;
2731         ret = ERR_PTR(-EINVAL);
2732 out_err:
2733         xs_xprt_free(xprt);
2734         return ret;
2735 }
2736
2737 static const struct rpc_timeout xs_tcp_default_timeout = {
2738         .to_initval = 60 * HZ,
2739         .to_maxval = 60 * HZ,
2740         .to_retries = 2,
2741 };
2742
2743 /**
2744  * xs_setup_tcp - Set up transport to use a TCP socket
2745  * @args: rpc transport creation arguments
2746  *
2747  */
2748 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2749 {
2750         struct sockaddr *addr = args->dstaddr;
2751         struct rpc_xprt *xprt;
2752         struct sock_xprt *transport;
2753         struct rpc_xprt *ret;
2754         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2755
2756         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2757                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2758
2759         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2760                         max_slot_table_size);
2761         if (IS_ERR(xprt))
2762                 return xprt;
2763         transport = container_of(xprt, struct sock_xprt, xprt);
2764
2765         xprt->prot = IPPROTO_TCP;
2766         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2767         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2768
2769         xprt->bind_timeout = XS_BIND_TO;
2770         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2771         xprt->idle_timeout = XS_IDLE_DISC_TO;
2772
2773         xprt->ops = &xs_tcp_ops;
2774         xprt->timeout = &xs_tcp_default_timeout;
2775
2776         switch (addr->sa_family) {
2777         case AF_INET:
2778                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2779                         xprt_set_bound(xprt);
2780
2781                 INIT_DELAYED_WORK(&transport->connect_worker,
2782                                         xs_tcp_setup_socket);
2783                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2784                 break;
2785         case AF_INET6:
2786                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2787                         xprt_set_bound(xprt);
2788
2789                 INIT_DELAYED_WORK(&transport->connect_worker,
2790                                         xs_tcp_setup_socket);
2791                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2792                 break;
2793         default:
2794                 ret = ERR_PTR(-EAFNOSUPPORT);
2795                 goto out_err;
2796         }
2797
2798         if (xprt_bound(xprt))
2799                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2800                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2801                                 xprt->address_strings[RPC_DISPLAY_PORT],
2802                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2803         else
2804                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2805                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2806                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2807
2808         if (try_module_get(THIS_MODULE))
2809                 return xprt;
2810         ret = ERR_PTR(-EINVAL);
2811 out_err:
2812         xs_xprt_free(xprt);
2813         return ret;
2814 }
2815
2816 /**
2817  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2818  * @args: rpc transport creation arguments
2819  *
2820  */
2821 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2822 {
2823         struct sockaddr *addr = args->dstaddr;
2824         struct rpc_xprt *xprt;
2825         struct sock_xprt *transport;
2826         struct svc_sock *bc_sock;
2827         struct rpc_xprt *ret;
2828
2829         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2830                         xprt_tcp_slot_table_entries);
2831         if (IS_ERR(xprt))
2832                 return xprt;
2833         transport = container_of(xprt, struct sock_xprt, xprt);
2834
2835         xprt->prot = IPPROTO_TCP;
2836         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2837         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2838         xprt->timeout = &xs_tcp_default_timeout;
2839
2840         /* backchannel */
2841         xprt_set_bound(xprt);
2842         xprt->bind_timeout = 0;
2843         xprt->reestablish_timeout = 0;
2844         xprt->idle_timeout = 0;
2845
2846         xprt->ops = &bc_tcp_ops;
2847
2848         switch (addr->sa_family) {
2849         case AF_INET:
2850                 xs_format_peer_addresses(xprt, "tcp",
2851                                          RPCBIND_NETID_TCP);
2852                 break;
2853         case AF_INET6:
2854                 xs_format_peer_addresses(xprt, "tcp",
2855                                    RPCBIND_NETID_TCP6);
2856                 break;
2857         default:
2858                 ret = ERR_PTR(-EAFNOSUPPORT);
2859                 goto out_err;
2860         }
2861
2862         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2863                         xprt->address_strings[RPC_DISPLAY_ADDR],
2864                         xprt->address_strings[RPC_DISPLAY_PORT],
2865                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2866
2867         /*
2868          * Once we've associated a backchannel xprt with a connection,
2869          * we want to keep it around as long as the connection lasts,
2870          * in case we need to start using it for a backchannel again;
2871          * this reference won't be dropped until bc_xprt is destroyed.
2872          */
2873         xprt_get(xprt);
2874         args->bc_xprt->xpt_bc_xprt = xprt;
2875         xprt->bc_xprt = args->bc_xprt;
2876         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2877         transport->sock = bc_sock->sk_sock;
2878         transport->inet = bc_sock->sk_sk;
2879
2880         /*
2881          * Since we don't want connections for the backchannel, we set
2882          * the xprt status to connected
2883          */
2884         xprt_set_connected(xprt);
2885
2886         if (try_module_get(THIS_MODULE))
2887                 return xprt;
2888
2889         args->bc_xprt->xpt_bc_xprt = NULL;
2890         xprt_put(xprt);
2891         ret = ERR_PTR(-EINVAL);
2892 out_err:
2893         xs_xprt_free(xprt);
2894         return ret;
2895 }
2896
2897 static struct xprt_class        xs_local_transport = {
2898         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2899         .name           = "named UNIX socket",
2900         .owner          = THIS_MODULE,
2901         .ident          = XPRT_TRANSPORT_LOCAL,
2902         .setup          = xs_setup_local,
2903 };
2904
2905 static struct xprt_class        xs_udp_transport = {
2906         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2907         .name           = "udp",
2908         .owner          = THIS_MODULE,
2909         .ident          = XPRT_TRANSPORT_UDP,
2910         .setup          = xs_setup_udp,
2911 };
2912
2913 static struct xprt_class        xs_tcp_transport = {
2914         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2915         .name           = "tcp",
2916         .owner          = THIS_MODULE,
2917         .ident          = XPRT_TRANSPORT_TCP,
2918         .setup          = xs_setup_tcp,
2919 };
2920
2921 static struct xprt_class        xs_bc_tcp_transport = {
2922         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2923         .name           = "tcp NFSv4.1 backchannel",
2924         .owner          = THIS_MODULE,
2925         .ident          = XPRT_TRANSPORT_BC_TCP,
2926         .setup          = xs_setup_bc_tcp,
2927 };
2928
2929 /**
2930  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2931  *
2932  */
2933 int init_socket_xprt(void)
2934 {
2935 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2936         if (!sunrpc_table_header)
2937                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2938 #endif
2939
2940         xprt_register_transport(&xs_local_transport);
2941         xprt_register_transport(&xs_udp_transport);
2942         xprt_register_transport(&xs_tcp_transport);
2943         xprt_register_transport(&xs_bc_tcp_transport);
2944
2945         return 0;
2946 }
2947
2948 /**
2949  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2950  *
2951  */
2952 void cleanup_socket_xprt(void)
2953 {
2954 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2955         if (sunrpc_table_header) {
2956                 unregister_sysctl_table(sunrpc_table_header);
2957                 sunrpc_table_header = NULL;
2958         }
2959 #endif
2960
2961         xprt_unregister_transport(&xs_local_transport);
2962         xprt_unregister_transport(&xs_udp_transport);
2963         xprt_unregister_transport(&xs_tcp_transport);
2964         xprt_unregister_transport(&xs_bc_tcp_transport);
2965 }
2966
2967 static int param_set_uint_minmax(const char *val,
2968                 const struct kernel_param *kp,
2969                 unsigned int min, unsigned int max)
2970 {
2971         unsigned int num;
2972         int ret;
2973
2974         if (!val)
2975                 return -EINVAL;
2976         ret = kstrtouint(val, 0, &num);
2977         if (ret == -EINVAL || num < min || num > max)
2978                 return -EINVAL;
2979         *((unsigned int *)kp->arg) = num;
2980         return 0;
2981 }
2982
2983 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2984 {
2985         return param_set_uint_minmax(val, kp,
2986                         RPC_MIN_RESVPORT,
2987                         RPC_MAX_RESVPORT);
2988 }
2989
2990 static struct kernel_param_ops param_ops_portnr = {
2991         .set = param_set_portnr,
2992         .get = param_get_uint,
2993 };
2994
2995 #define param_check_portnr(name, p) \
2996         __param_check(name, p, unsigned int);
2997
2998 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2999 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3000
3001 static int param_set_slot_table_size(const char *val,
3002                                      const struct kernel_param *kp)
3003 {
3004         return param_set_uint_minmax(val, kp,
3005                         RPC_MIN_SLOT_TABLE,
3006                         RPC_MAX_SLOT_TABLE);
3007 }
3008
3009 static struct kernel_param_ops param_ops_slot_table_size = {
3010         .set = param_set_slot_table_size,
3011         .get = param_get_uint,
3012 };
3013
3014 #define param_check_slot_table_size(name, p) \
3015         __param_check(name, p, unsigned int);
3016
3017 static int param_set_max_slot_table_size(const char *val,
3018                                      const struct kernel_param *kp)
3019 {
3020         return param_set_uint_minmax(val, kp,
3021                         RPC_MIN_SLOT_TABLE,
3022                         RPC_MAX_SLOT_TABLE_LIMIT);
3023 }
3024
3025 static struct kernel_param_ops param_ops_max_slot_table_size = {
3026         .set = param_set_max_slot_table_size,
3027         .get = param_get_uint,
3028 };
3029
3030 #define param_check_max_slot_table_size(name, p) \
3031         __param_check(name, p, unsigned int);
3032
3033 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3034                    slot_table_size, 0644);
3035 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3036                    max_slot_table_size, 0644);
3037 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3038                    slot_table_size, 0644);
3039