net/sunrpc/xprtsock.c

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