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