net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68 #define XS_TCP_LINGER_TO        (15U * HZ)
  69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71 /*
  72  * We can register our own files under /proc/sys/sunrpc by
  73  * calling register_sysctl_table() again.  The files in that
  74  * directory become the union of all files registered there.
  75  *
  76  * We simply need to make sure that we don't collide with
  77  * someone else's file names!
  78  */
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180
 181 /*
 182  * TCP idle timeout; client drops the transport socket if it is idle
 183  * for this long.  Note that we also timeout UDP sockets to prevent
 184  * holding port numbers when there is no RPC traffic.
 185  */
 186 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 187
 188 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 189 # undef  RPC_DEBUG_DATA
 190 # define RPCDBG_FACILITY        RPCDBG_TRANS
 191 #endif
 192
 193 #ifdef RPC_DEBUG_DATA
 194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 195 {
 196         u8 *buf = (u8 *) packet;
 197         int j;
 198
 199         dprintk("RPC:       %s\n", msg);
 200         for (j = 0; j < count && j < 128; j += 4) {
 201                 if (!(j & 31)) {
 202                         if (j)
 203                                 dprintk("\n");
 204                         dprintk("0x%04x ", j);
 205                 }
 206                 dprintk("%02x%02x%02x%02x ",
 207                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 208         }
 209         dprintk("\n");
 210 }
 211 #else
 212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 213 {
 214         /* NOP */
 215 }
 216 #endif
 217
 218 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 219 {
 220         return (struct rpc_xprt *) sk->sk_user_data;
 221 }
 222
 223 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 224 {
 225         return (struct sockaddr *) &xprt->addr;
 226 }
 227
 228 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 229 {
 230         return (struct sockaddr_un *) &xprt->addr;
 231 }
 232
 233 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 234 {
 235         return (struct sockaddr_in *) &xprt->addr;
 236 }
 237
 238 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 239 {
 240         return (struct sockaddr_in6 *) &xprt->addr;
 241 }
 242
 243 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 244 {
 245         struct sockaddr *sap = xs_addr(xprt);
 246         struct sockaddr_in6 *sin6;
 247         struct sockaddr_in *sin;
 248         struct sockaddr_un *sun;
 249         char buf[128];
 250
 251         switch (sap->sa_family) {
 252         case AF_LOCAL:
 253                 sun = xs_addr_un(xprt);
 254                 strlcpy(buf, sun->sun_path, sizeof(buf));
 255                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 256                                                 kstrdup(buf, GFP_KERNEL);
 257                 break;
 258         case AF_INET:
 259                 (void)rpc_ntop(sap, buf, sizeof(buf));
 260                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 261                                                 kstrdup(buf, GFP_KERNEL);
 262                 sin = xs_addr_in(xprt);
 263                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 264                 break;
 265         case AF_INET6:
 266                 (void)rpc_ntop(sap, buf, sizeof(buf));
 267                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 268                                                 kstrdup(buf, GFP_KERNEL);
 269                 sin6 = xs_addr_in6(xprt);
 270                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 271                 break;
 272         default:
 273                 BUG();
 274         }
 275
 276         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 277 }
 278
 279 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 280 {
 281         struct sockaddr *sap = xs_addr(xprt);
 282         char buf[128];
 283
 284         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 285         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 286
 287         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 288         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 289 }
 290
 291 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 292                                      const char *protocol,
 293                                      const char *netid)
 294 {
 295         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 296         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 297         xs_format_common_peer_addresses(xprt);
 298         xs_format_common_peer_ports(xprt);
 299 }
 300
 301 static void xs_update_peer_port(struct rpc_xprt *xprt)
 302 {
 303         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 304         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 305
 306         xs_format_common_peer_ports(xprt);
 307 }
 308
 309 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 310 {
 311         unsigned int i;
 312
 313         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 314                 switch (i) {
 315                 case RPC_DISPLAY_PROTO:
 316                 case RPC_DISPLAY_NETID:
 317                         continue;
 318                 default:
 319                         kfree(xprt->address_strings[i]);
 320                 }
 321 }
 322
 323 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 324
 325 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 326 {
 327         struct msghdr msg = {
 328                 .msg_name       = addr,
 329                 .msg_namelen    = addrlen,
 330                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 331         };
 332         struct kvec iov = {
 333                 .iov_base       = vec->iov_base + base,
 334                 .iov_len        = vec->iov_len - base,
 335         };
 336
 337         if (iov.iov_len != 0)
 338                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 339         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 340 }
 341
 342 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 343 {
 344         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 345                         int offset, size_t size, int flags);
 346         struct page **ppage;
 347         unsigned int remainder;
 348         int err;
 349
 350         remainder = xdr->page_len - base;
 351         base += xdr->page_base;
 352         ppage = xdr->pages + (base >> PAGE_SHIFT);
 353         base &= ~PAGE_MASK;
 354         do_sendpage = sock->ops->sendpage;
 355         if (!zerocopy)
 356                 do_sendpage = sock_no_sendpage;
 357         for(;;) {
 358                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 359                 int flags = XS_SENDMSG_FLAGS;
 360
 361                 remainder -= len;
 362                 if (more)
 363                         flags |= MSG_MORE;
 364                 if (remainder != 0)
 365                         flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
 366                 err = do_sendpage(sock, *ppage, base, len, flags);
 367                 if (remainder == 0 || err != len)
 368                         break;
 369                 *sent_p += err;
 370                 ppage++;
 371                 base = 0;
 372         }
 373         if (err > 0) {
 374                 *sent_p += err;
 375                 err = 0;
 376         }
 377         return err;
 378 }
 379
 380 /**
 381  * xs_sendpages - write pages directly to a socket
 382  * @sock: socket to send on
 383  * @addr: UDP only -- address of destination
 384  * @addrlen: UDP only -- length of destination address
 385  * @xdr: buffer containing this request
 386  * @base: starting position in the buffer
 387  * @zerocopy: true if it is safe to use sendpage()
 388  * @sent_p: return the total number of bytes successfully queued for sending
 389  *
 390  */
 391 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 392 {
 393         unsigned int remainder = xdr->len - base;
 394         int err = 0;
 395         int sent = 0;
 396
 397         if (unlikely(!sock))
 398                 return -ENOTSOCK;
 399
 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 }
 444
 445 /**
 446  * xs_nospace - place task on wait queue if transmit was incomplete
 447  * @task: task to put to sleep
 448  *
 449  */
 450 static int xs_nospace(struct rpc_task *task)
 451 {
 452         struct rpc_rqst *req = task->tk_rqstp;
 453         struct rpc_xprt *xprt = req->rq_xprt;
 454         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 455         struct sock *sk = transport->inet;
 456         int ret = -EAGAIN;
 457
 458         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 459                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 460                         req->rq_slen);
 461
 462         /* Protect against races with write_space */
 463         spin_lock_bh(&xprt->transport_lock);
 464
 465         /* Don't race with disconnect */
 466         if (xprt_connected(xprt)) {
 467                 /* wait for more buffer space */
 468                 sk->sk_write_pending++;
 469                 xprt_wait_for_buffer_space(task, xs_nospace_callback);
 470         } else
 471                 ret = -ENOTCONN;
 472
 473         spin_unlock_bh(&xprt->transport_lock);
 474
 475         /* Race breaker in case memory is freed before above code is called */
 476         if (ret == -EAGAIN) {
 477                 struct socket_wq *wq;
 478
 479                 rcu_read_lock();
 480                 wq = rcu_dereference(sk->sk_wq);
 481                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 482                 rcu_read_unlock();
 483
 484                 sk->sk_write_space(sk);
 485         }
 486         return ret;
 487 }
 488
 489 /*
 490  * Construct a stream transport record marker in @buf.
 491  */
 492 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 493 {
 494         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 495         rpc_fraghdr *base = buf->head[0].iov_base;
 496         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 497 }
 498
 499 /**
 500  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 501  * @task: RPC task that manages the state of an RPC request
 502  *
 503  * Return values:
 504  *        0:    The request has been sent
 505  *   EAGAIN:    The socket was blocked, please call again later to
 506  *              complete the request
 507  * ENOTCONN:    Caller needs to invoke connect logic then call again
 508  *    other:    Some other error occured, the request was not sent
 509  */
 510 static int xs_local_send_request(struct rpc_task *task)
 511 {
 512         struct rpc_rqst *req = task->tk_rqstp;
 513         struct rpc_xprt *xprt = req->rq_xprt;
 514         struct sock_xprt *transport =
 515                                 container_of(xprt, struct sock_xprt, xprt);
 516         struct xdr_buf *xdr = &req->rq_snd_buf;
 517         int status;
 518         int sent = 0;
 519
 520         xs_encode_stream_record_marker(&req->rq_snd_buf);
 521
 522         xs_pktdump("packet data:",
 523                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 524
 525         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 526                               true, &sent);
 527         dprintk("RPC:       %s(%u) = %d\n",
 528                         __func__, xdr->len - req->rq_bytes_sent, status);
 529
 530         if (status == -EAGAIN && sock_writeable(transport->inet))
 531                 status = -ENOBUFS;
 532
 533         if (likely(sent > 0) || status == 0) {
 534                 req->rq_bytes_sent += sent;
 535                 req->rq_xmit_bytes_sent += sent;
 536                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 537                         req->rq_bytes_sent = 0;
 538                         return 0;
 539                 }
 540                 status = -EAGAIN;
 541         }
 542
 543         switch (status) {
 544         case -ENOBUFS:
 545                 break;
 546         case -EAGAIN:
 547                 status = xs_nospace(task);
 548                 break;
 549         default:
 550                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 551                         -status);
 552         case -EPIPE:
 553                 xs_close(xprt);
 554                 status = -ENOTCONN;
 555         }
 556
 557         return status;
 558 }
 559
 560 /**
 561  * xs_udp_send_request - write an RPC request to a UDP socket
 562  * @task: address of RPC task that manages the state of an RPC request
 563  *
 564  * Return values:
 565  *        0:    The request has been sent
 566  *   EAGAIN:    The socket was blocked, please call again later to
 567  *              complete the request
 568  * ENOTCONN:    Caller needs to invoke connect logic then call again
 569  *    other:    Some other error occurred, the request was not sent
 570  */
 571 static int xs_udp_send_request(struct rpc_task *task)
 572 {
 573         struct rpc_rqst *req = task->tk_rqstp;
 574         struct rpc_xprt *xprt = req->rq_xprt;
 575         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 576         struct xdr_buf *xdr = &req->rq_snd_buf;
 577         int sent = 0;
 578         int status;
 579
 580         xs_pktdump("packet data:",
 581                                 req->rq_svec->iov_base,
 582                                 req->rq_svec->iov_len);
 583
 584         if (!xprt_bound(xprt))
 585                 return -ENOTCONN;
 586         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 587                               xdr, req->rq_bytes_sent, true, &sent);
 588
 589         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 590                         xdr->len - req->rq_bytes_sent, status);
 591
 592         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 593         if (status == -EPERM)
 594                 goto process_status;
 595
 596         if (status == -EAGAIN && sock_writeable(transport->inet))
 597                 status = -ENOBUFS;
 598
 599         if (sent > 0 || status == 0) {
 600                 req->rq_xmit_bytes_sent += sent;
 601                 if (sent >= req->rq_slen)
 602                         return 0;
 603                 /* Still some bytes left; set up for a retry later. */
 604                 status = -EAGAIN;
 605         }
 606
 607 process_status:
 608         switch (status) {
 609         case -ENOTSOCK:
 610                 status = -ENOTCONN;
 611                 /* Should we call xs_close() here? */
 612                 break;
 613         case -EAGAIN:
 614                 status = xs_nospace(task);
 615                 break;
 616         case -ENETUNREACH:
 617         case -ENOBUFS:
 618         case -EPIPE:
 619         case -ECONNREFUSED:
 620         case -EPERM:
 621                 /* When the server has died, an ICMP port unreachable message
 622                  * prompts ECONNREFUSED. */
 623                 break;
 624         default:
 625                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 626                         -status);
 627         }
 628
 629         return status;
 630 }
 631
 632 /**
 633  * xs_tcp_send_request - write an RPC request to a TCP socket
 634  * @task: address of RPC task that manages the state of an RPC request
 635  *
 636  * Return values:
 637  *        0:    The request has been sent
 638  *   EAGAIN:    The socket was blocked, please call again later to
 639  *              complete the request
 640  * ENOTCONN:    Caller needs to invoke connect logic then call again
 641  *    other:    Some other error occurred, the request was not sent
 642  *
 643  * XXX: In the case of soft timeouts, should we eventually give up
 644  *      if sendmsg is not able to make progress?
 645  */
 646 static int xs_tcp_send_request(struct rpc_task *task)
 647 {
 648         struct rpc_rqst *req = task->tk_rqstp;
 649         struct rpc_xprt *xprt = req->rq_xprt;
 650         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 651         struct xdr_buf *xdr = &req->rq_snd_buf;
 652         bool zerocopy = true;
 653         bool vm_wait = false;
 654         int status;
 655         int sent;
 656
 657         xs_encode_stream_record_marker(&req->rq_snd_buf);
 658
 659         xs_pktdump("packet data:",
 660                                 req->rq_svec->iov_base,
 661                                 req->rq_svec->iov_len);
 662         /* Don't use zero copy if this is a resend. If the RPC call
 663          * completes while the socket holds a reference to the pages,
 664          * then we may end up resending corrupted data.
 665          */
 666         if (task->tk_flags & RPC_TASK_SENT)
 667                 zerocopy = false;
 668
 669         /* Continue transmitting the packet/record. We must be careful
 670          * to cope with writespace callbacks arriving _after_ we have
 671          * called sendmsg(). */
 672         while (1) {
 673                 sent = 0;
 674                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 675                                       req->rq_bytes_sent, zerocopy, &sent);
 676
 677                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 678                                 xdr->len - req->rq_bytes_sent, status);
 679
 680                 /* If we've sent the entire packet, immediately
 681                  * reset the count of bytes sent. */
 682                 req->rq_bytes_sent += sent;
 683                 req->rq_xmit_bytes_sent += sent;
 684                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 685                         req->rq_bytes_sent = 0;
 686                         return 0;
 687                 }
 688
 689                 WARN_ON_ONCE(sent == 0 && status == 0);
 690
 691                 if (status == -EAGAIN ) {
 692                         /*
 693                          * Return EAGAIN if we're sure we're hitting the
 694                          * socket send buffer limits.
 695                          */
 696                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 697                                 break;
 698                         /*
 699                          * Did we hit a memory allocation failure?
 700                          */
 701                         if (sent == 0) {
 702                                 status = -ENOBUFS;
 703                                 if (vm_wait)
 704                                         break;
 705                                 /* Retry, knowing now that we're below the
 706                                  * socket send buffer limit
 707                                  */
 708                                 vm_wait = true;
 709                         }
 710                         continue;
 711                 }
 712                 if (status < 0)
 713                         break;
 714                 vm_wait = false;
 715         }
 716
 717         switch (status) {
 718         case -ENOTSOCK:
 719                 status = -ENOTCONN;
 720                 /* Should we call xs_close() here? */
 721                 break;
 722         case -EAGAIN:
 723                 status = xs_nospace(task);
 724                 break;
 725         case -ECONNRESET:
 726         case -ECONNREFUSED:
 727         case -ENOTCONN:
 728         case -EADDRINUSE:
 729         case -ENOBUFS:
 730         case -EPIPE:
 731                 break;
 732         default:
 733                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 734                         -status);
 735         }
 736
 737         return status;
 738 }
 739
 740 /**
 741  * xs_tcp_release_xprt - clean up after a tcp transmission
 742  * @xprt: transport
 743  * @task: rpc task
 744  *
 745  * This cleans up if an error causes us to abort the transmission of a request.
 746  * In this case, the socket may need to be reset in order to avoid confusing
 747  * the server.
 748  */
 749 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 750 {
 751         struct rpc_rqst *req;
 752
 753         if (task != xprt->snd_task)
 754                 return;
 755         if (task == NULL)
 756                 goto out_release;
 757         req = task->tk_rqstp;
 758         if (req == NULL)
 759                 goto out_release;
 760         if (req->rq_bytes_sent == 0)
 761                 goto out_release;
 762         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 763                 goto out_release;
 764         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 765 out_release:
 766         xprt_release_xprt(xprt, task);
 767 }
 768
 769 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 770 {
 771         transport->old_data_ready = sk->sk_data_ready;
 772         transport->old_state_change = sk->sk_state_change;
 773         transport->old_write_space = sk->sk_write_space;
 774         transport->old_error_report = sk->sk_error_report;
 775 }
 776
 777 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 778 {
 779         sk->sk_data_ready = transport->old_data_ready;
 780         sk->sk_state_change = transport->old_state_change;
 781         sk->sk_write_space = transport->old_write_space;
 782         sk->sk_error_report = transport->old_error_report;
 783 }
 784
 785 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
 786 {
 787         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 788
 789         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 790 }
 791
 792 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 793 {
 794         smp_mb__before_atomic();
 795         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 796         clear_bit(XPRT_CLOSING, &xprt->state);
 797         xs_sock_reset_state_flags(xprt);
 798         smp_mb__after_atomic();
 799 }
 800
 801 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 802 {
 803         xs_sock_reset_connection_flags(xprt);
 804         /* Mark transport as closed and wake up all pending tasks */
 805         xprt_disconnect_done(xprt);
 806 }
 807
 808 /**
 809  * xs_error_report - callback to handle TCP socket state errors
 810  * @sk: socket
 811  *
 812  * Note: we don't call sock_error() since there may be a rpc_task
 813  * using the socket, and so we don't want to clear sk->sk_err.
 814  */
 815 static void xs_error_report(struct sock *sk)
 816 {
 817         struct rpc_xprt *xprt;
 818         int err;
 819
 820         read_lock_bh(&sk->sk_callback_lock);
 821         if (!(xprt = xprt_from_sock(sk)))
 822                 goto out;
 823
 824         err = -sk->sk_err;
 825         if (err == 0)
 826                 goto out;
 827         /* Is this a reset event? */
 828         if (sk->sk_state == TCP_CLOSE)
 829                 xs_sock_mark_closed(xprt);
 830         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 831                         xprt, -err);
 832         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 833         xprt_wake_pending_tasks(xprt, err);
 834  out:
 835         read_unlock_bh(&sk->sk_callback_lock);
 836 }
 837
 838 static void xs_reset_transport(struct sock_xprt *transport)
 839 {
 840         struct socket *sock = transport->sock;
 841         struct sock *sk = transport->inet;
 842         struct rpc_xprt *xprt = &transport->xprt;
 843
 844         if (sk == NULL)
 845                 return;
 846
 847         if (atomic_read(&transport->xprt.swapper))
 848                 sk_clear_memalloc(sk);
 849
 850         kernel_sock_shutdown(sock, SHUT_RDWR);
 851
 852         mutex_lock(&transport->recv_mutex);
 853         write_lock_bh(&sk->sk_callback_lock);
 854         transport->inet = NULL;
 855         transport->sock = NULL;
 856
 857         sk->sk_user_data = NULL;
 858
 859         xs_restore_old_callbacks(transport, sk);
 860         xprt_clear_connected(xprt);
 861         write_unlock_bh(&sk->sk_callback_lock);
 862         xs_sock_reset_connection_flags(xprt);
 863         mutex_unlock(&transport->recv_mutex);
 864
 865         trace_rpc_socket_close(xprt, sock);
 866         sock_release(sock);
 867 }
 868
 869 /**
 870  * xs_close - close a socket
 871  * @xprt: transport
 872  *
 873  * This is used when all requests are complete; ie, no DRC state remains
 874  * on the server we want to save.
 875  *
 876  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 877  * xs_reset_transport() zeroing the socket from underneath a writer.
 878  */
 879 static void xs_close(struct rpc_xprt *xprt)
 880 {
 881         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 882
 883         dprintk("RPC:       xs_close xprt %p\n", xprt);
 884
 885         xs_reset_transport(transport);
 886         xprt->reestablish_timeout = 0;
 887
 888         xprt_disconnect_done(xprt);
 889 }
 890
 891 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 892 {
 893         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 894                 xprt);
 895         xprt_disconnect_done(xprt);
 896 }
 897
 898 static void xs_xprt_free(struct rpc_xprt *xprt)
 899 {
 900         xs_free_peer_addresses(xprt);
 901         xprt_free(xprt);
 902 }
 903
 904 /**
 905  * xs_destroy - prepare to shutdown a transport
 906  * @xprt: doomed transport
 907  *
 908  */
 909 static void xs_destroy(struct rpc_xprt *xprt)
 910 {
 911         struct sock_xprt *transport = container_of(xprt,
 912                         struct sock_xprt, xprt);
 913         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 914
 915         cancel_delayed_work_sync(&transport->connect_worker);
 916         xs_close(xprt);
 917         cancel_work_sync(&transport->recv_worker);
 918         xs_xprt_free(xprt);
 919         module_put(THIS_MODULE);
 920 }
 921
 922 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 923 {
 924         struct xdr_skb_reader desc = {
 925                 .skb            = skb,
 926                 .offset         = sizeof(rpc_fraghdr),
 927                 .count          = skb->len - sizeof(rpc_fraghdr),
 928         };
 929
 930         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 931                 return -1;
 932         if (desc.count)
 933                 return -1;
 934         return 0;
 935 }
 936
 937 /**
 938  * xs_local_data_read_skb
 939  * @xprt: transport
 940  * @sk: socket
 941  * @skb: skbuff
 942  *
 943  * Currently this assumes we can read the whole reply in a single gulp.
 944  */
 945 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 946                 struct sock *sk,
 947                 struct sk_buff *skb)
 948 {
 949         struct rpc_task *task;
 950         struct rpc_rqst *rovr;
 951         int repsize, copied;
 952         u32 _xid;
 953         __be32 *xp;
 954
 955         repsize = skb->len - sizeof(rpc_fraghdr);
 956         if (repsize < 4) {
 957                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 958                 return;
 959         }
 960
 961         /* Copy the XID from the skb... */
 962         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 963         if (xp == NULL)
 964                 return;
 965
 966         /* Look up and lock the request corresponding to the given XID */
 967         spin_lock_bh(&xprt->transport_lock);
 968         rovr = xprt_lookup_rqst(xprt, *xp);
 969         if (!rovr)
 970                 goto out_unlock;
 971         task = rovr->rq_task;
 972
 973         copied = rovr->rq_private_buf.buflen;
 974         if (copied > repsize)
 975                 copied = repsize;
 976
 977         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 978                 dprintk("RPC:       sk_buff copy failed\n");
 979                 goto out_unlock;
 980         }
 981
 982         xprt_complete_rqst(task, copied);
 983
 984  out_unlock:
 985         spin_unlock_bh(&xprt->transport_lock);
 986 }
 987
 988 static void xs_local_data_receive(struct sock_xprt *transport)
 989 {
 990         struct sk_buff *skb;
 991         struct sock *sk;
 992         int err;
 993
 994         mutex_lock(&transport->recv_mutex);
 995         sk = transport->inet;
 996         if (sk == NULL)
 997                 goto out;
 998         for (;;) {
 999                 skb = skb_recv_datagram(sk, 0, 1, &err);
1000                 if (skb != NULL) {
1001                         xs_local_data_read_skb(&transport->xprt, sk, skb);
1002                         skb_free_datagram(sk, skb);
1003                         continue;
1004                 }
1005                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1006                         break;
1007         }
1008 out:
1009         mutex_unlock(&transport->recv_mutex);
1010 }
1011
1012 static void xs_local_data_receive_workfn(struct work_struct *work)
1013 {
1014         struct sock_xprt *transport =
1015                 container_of(work, struct sock_xprt, recv_worker);
1016         xs_local_data_receive(transport);
1017 }
1018
1019 /**
1020  * xs_udp_data_read_skb - receive callback for UDP sockets
1021  * @xprt: transport
1022  * @sk: socket
1023  * @skb: skbuff
1024  *
1025  */
1026 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1027                 struct sock *sk,
1028                 struct sk_buff *skb)
1029 {
1030         struct rpc_task *task;
1031         struct rpc_rqst *rovr;
1032         int repsize, copied;
1033         u32 _xid;
1034         __be32 *xp;
1035
1036         repsize = skb->len;
1037         if (repsize < 4) {
1038                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1039                 return;
1040         }
1041
1042         /* Copy the XID from the skb... */
1043         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1044         if (xp == NULL)
1045                 return;
1046
1047         /* Look up and lock the request corresponding to the given XID */
1048         spin_lock_bh(&xprt->transport_lock);
1049         rovr = xprt_lookup_rqst(xprt, *xp);
1050         if (!rovr)
1051                 goto out_unlock;
1052         task = rovr->rq_task;
1053
1054         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1055                 copied = repsize;
1056
1057         /* Suck it into the iovec, verify checksum if not done by hw. */
1058         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1059                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1060                 goto out_unlock;
1061         }
1062
1063         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1064
1065         xprt_adjust_cwnd(xprt, task, copied);
1066         xprt_complete_rqst(task, copied);
1067
1068  out_unlock:
1069         spin_unlock_bh(&xprt->transport_lock);
1070 }
1071
1072 static void xs_udp_data_receive(struct sock_xprt *transport)
1073 {
1074         struct sk_buff *skb;
1075         struct sock *sk;
1076         int err;
1077
1078         mutex_lock(&transport->recv_mutex);
1079         sk = transport->inet;
1080         if (sk == NULL)
1081                 goto out;
1082         for (;;) {
1083                 skb = skb_recv_datagram(sk, 0, 1, &err);
1084                 if (skb != NULL) {
1085                         xs_udp_data_read_skb(&transport->xprt, sk, skb);
1086                         skb_free_datagram_locked(sk, skb);
1087                         continue;
1088                 }
1089                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1090                         break;
1091         }
1092 out:
1093         mutex_unlock(&transport->recv_mutex);
1094 }
1095
1096 static void xs_udp_data_receive_workfn(struct work_struct *work)
1097 {
1098         struct sock_xprt *transport =
1099                 container_of(work, struct sock_xprt, recv_worker);
1100         xs_udp_data_receive(transport);
1101 }
1102
1103 /**
1104  * xs_data_ready - "data ready" callback for UDP sockets
1105  * @sk: socket with data to read
1106  *
1107  */
1108 static void xs_data_ready(struct sock *sk)
1109 {
1110         struct rpc_xprt *xprt;
1111
1112         read_lock_bh(&sk->sk_callback_lock);
1113         dprintk("RPC:       xs_data_ready...\n");
1114         xprt = xprt_from_sock(sk);
1115         if (xprt != NULL) {
1116                 struct sock_xprt *transport = container_of(xprt,
1117                                 struct sock_xprt, xprt);
1118                 transport->old_data_ready(sk);
1119                 /* Any data means we had a useful conversation, so
1120                  * then we don't need to delay the next reconnect
1121                  */
1122                 if (xprt->reestablish_timeout)
1123                         xprt->reestablish_timeout = 0;
1124                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1125                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1126         }
1127         read_unlock_bh(&sk->sk_callback_lock);
1128 }
1129
1130 /*
1131  * Helper function to force a TCP close if the server is sending
1132  * junk and/or it has put us in CLOSE_WAIT
1133  */
1134 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1135 {
1136         xprt_force_disconnect(xprt);
1137 }
1138
1139 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1140 {
1141         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1142         size_t len, used;
1143         char *p;
1144
1145         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1146         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1147         used = xdr_skb_read_bits(desc, p, len);
1148         transport->tcp_offset += used;
1149         if (used != len)
1150                 return;
1151
1152         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1153         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1154                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1155         else
1156                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1157         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1158
1159         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1160         transport->tcp_offset = 0;
1161
1162         /* Sanity check of the record length */
1163         if (unlikely(transport->tcp_reclen < 8)) {
1164                 dprintk("RPC:       invalid TCP record fragment length\n");
1165                 xs_tcp_force_close(xprt);
1166                 return;
1167         }
1168         dprintk("RPC:       reading TCP record fragment of length %d\n",
1169                         transport->tcp_reclen);
1170 }
1171
1172 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1173 {
1174         if (transport->tcp_offset == transport->tcp_reclen) {
1175                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1176                 transport->tcp_offset = 0;
1177                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1178                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1179                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1180                         transport->tcp_copied = 0;
1181                 }
1182         }
1183 }
1184
1185 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1186 {
1187         size_t len, used;
1188         char *p;
1189
1190         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1191         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1192         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1193         used = xdr_skb_read_bits(desc, p, len);
1194         transport->tcp_offset += used;
1195         if (used != len)
1196                 return;
1197         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1198         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1199         transport->tcp_copied = 4;
1200         dprintk("RPC:       reading %s XID %08x\n",
1201                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1202                                                               : "request with",
1203                         ntohl(transport->tcp_xid));
1204         xs_tcp_check_fraghdr(transport);
1205 }
1206
1207 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1208                                        struct xdr_skb_reader *desc)
1209 {
1210         size_t len, used;
1211         u32 offset;
1212         char *p;
1213
1214         /*
1215          * We want transport->tcp_offset to be 8 at the end of this routine
1216          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1217          * When this function is called for the first time,
1218          * transport->tcp_offset is 4 (after having already read the xid).
1219          */
1220         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1221         len = sizeof(transport->tcp_calldir) - offset;
1222         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1223         p = ((char *) &transport->tcp_calldir) + offset;
1224         used = xdr_skb_read_bits(desc, p, len);
1225         transport->tcp_offset += used;
1226         if (used != len)
1227                 return;
1228         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1229         /*
1230          * We don't yet have the XDR buffer, so we will write the calldir
1231          * out after we get the buffer from the 'struct rpc_rqst'
1232          */
1233         switch (ntohl(transport->tcp_calldir)) {
1234         case RPC_REPLY:
1235                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1236                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1237                 transport->tcp_flags |= TCP_RPC_REPLY;
1238                 break;
1239         case RPC_CALL:
1240                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1241                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1242                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1243                 break;
1244         default:
1245                 dprintk("RPC:       invalid request message type\n");
1246                 xs_tcp_force_close(&transport->xprt);
1247         }
1248         xs_tcp_check_fraghdr(transport);
1249 }
1250
1251 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1252                                      struct xdr_skb_reader *desc,
1253                                      struct rpc_rqst *req)
1254 {
1255         struct sock_xprt *transport =
1256                                 container_of(xprt, struct sock_xprt, xprt);
1257         struct xdr_buf *rcvbuf;
1258         size_t len;
1259         ssize_t r;
1260
1261         rcvbuf = &req->rq_private_buf;
1262
1263         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1264                 /*
1265                  * Save the RPC direction in the XDR buffer
1266                  */
1267                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1268                         &transport->tcp_calldir,
1269                         sizeof(transport->tcp_calldir));
1270                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1271                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1272         }
1273
1274         len = desc->count;
1275         if (len > transport->tcp_reclen - transport->tcp_offset) {
1276                 struct xdr_skb_reader my_desc;
1277
1278                 len = transport->tcp_reclen - transport->tcp_offset;
1279                 memcpy(&my_desc, desc, sizeof(my_desc));
1280                 my_desc.count = len;
1281                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1282                                           &my_desc, xdr_skb_read_bits);
1283                 desc->count -= r;
1284                 desc->offset += r;
1285         } else
1286                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1287                                           desc, xdr_skb_read_bits);
1288
1289         if (r > 0) {
1290                 transport->tcp_copied += r;
1291                 transport->tcp_offset += r;
1292         }
1293         if (r != len) {
1294                 /* Error when copying to the receive buffer,
1295                  * usually because we weren't able to allocate
1296                  * additional buffer pages. All we can do now
1297                  * is turn off TCP_RCV_COPY_DATA, so the request
1298                  * will not receive any additional updates,
1299                  * and time out.
1300                  * Any remaining data from this record will
1301                  * be discarded.
1302                  */
1303                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1304                 dprintk("RPC:       XID %08x truncated request\n",
1305                                 ntohl(transport->tcp_xid));
1306                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1307                                 "tcp_offset = %u, tcp_reclen = %u\n",
1308                                 xprt, transport->tcp_copied,
1309                                 transport->tcp_offset, transport->tcp_reclen);
1310                 return;
1311         }
1312
1313         dprintk("RPC:       XID %08x read %Zd bytes\n",
1314                         ntohl(transport->tcp_xid), r);
1315         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1316                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1317                         transport->tcp_offset, transport->tcp_reclen);
1318
1319         if (transport->tcp_copied == req->rq_private_buf.buflen)
1320                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1321         else if (transport->tcp_offset == transport->tcp_reclen) {
1322                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1323                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1324         }
1325 }
1326
1327 /*
1328  * Finds the request corresponding to the RPC xid and invokes the common
1329  * tcp read code to read the data.
1330  */
1331 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1332                                     struct xdr_skb_reader *desc)
1333 {
1334         struct sock_xprt *transport =
1335                                 container_of(xprt, struct sock_xprt, xprt);
1336         struct rpc_rqst *req;
1337
1338         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1339
1340         /* Find and lock the request corresponding to this xid */
1341         spin_lock_bh(&xprt->transport_lock);
1342         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1343         if (!req) {
1344                 dprintk("RPC:       XID %08x request not found!\n",
1345                                 ntohl(transport->tcp_xid));
1346                 spin_unlock_bh(&xprt->transport_lock);
1347                 return -1;
1348         }
1349
1350         xs_tcp_read_common(xprt, desc, req);
1351
1352         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1353                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1354
1355         spin_unlock_bh(&xprt->transport_lock);
1356         return 0;
1357 }
1358
1359 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1360 /*
1361  * Obtains an rpc_rqst previously allocated and invokes the common
1362  * tcp read code to read the data.  The result is placed in the callback
1363  * queue.
1364  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1365  * connection and return -1.
1366  */
1367 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1368                                        struct xdr_skb_reader *desc)
1369 {
1370         struct sock_xprt *transport =
1371                                 container_of(xprt, struct sock_xprt, xprt);
1372         struct rpc_rqst *req;
1373
1374         /* Look up and lock the request corresponding to the given XID */
1375         spin_lock_bh(&xprt->transport_lock);
1376         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1377         if (req == NULL) {
1378                 spin_unlock_bh(&xprt->transport_lock);
1379                 printk(KERN_WARNING "Callback slot table overflowed\n");
1380                 xprt_force_disconnect(xprt);
1381                 return -1;
1382         }
1383
1384         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1385         xs_tcp_read_common(xprt, desc, req);
1386
1387         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1388                 xprt_complete_bc_request(req, transport->tcp_copied);
1389         spin_unlock_bh(&xprt->transport_lock);
1390
1391         return 0;
1392 }
1393
1394 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1395                                         struct xdr_skb_reader *desc)
1396 {
1397         struct sock_xprt *transport =
1398                                 container_of(xprt, struct sock_xprt, xprt);
1399
1400         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1401                 xs_tcp_read_reply(xprt, desc) :
1402                 xs_tcp_read_callback(xprt, desc);
1403 }
1404
1405 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1406 {
1407         int ret;
1408
1409         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1410                               SVC_SOCK_ANONYMOUS);
1411         if (ret < 0)
1412                 return ret;
1413         return 0;
1414 }
1415
1416 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1417 {
1418         return PAGE_SIZE;
1419 }
1420 #else
1421 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1422                                         struct xdr_skb_reader *desc)
1423 {
1424         return xs_tcp_read_reply(xprt, desc);
1425 }
1426 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1427
1428 /*
1429  * Read data off the transport.  This can be either an RPC_CALL or an
1430  * RPC_REPLY.  Relay the processing to helper functions.
1431  */
1432 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1433                                     struct xdr_skb_reader *desc)
1434 {
1435         struct sock_xprt *transport =
1436                                 container_of(xprt, struct sock_xprt, xprt);
1437
1438         if (_xs_tcp_read_data(xprt, desc) == 0)
1439                 xs_tcp_check_fraghdr(transport);
1440         else {
1441                 /*
1442                  * The transport_lock protects the request handling.
1443                  * There's no need to hold it to update the tcp_flags.
1444                  */
1445                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1446         }
1447 }
1448
1449 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1450 {
1451         size_t len;
1452
1453         len = transport->tcp_reclen - transport->tcp_offset;
1454         if (len > desc->count)
1455                 len = desc->count;
1456         desc->count -= len;
1457         desc->offset += len;
1458         transport->tcp_offset += len;
1459         dprintk("RPC:       discarded %Zu bytes\n", len);
1460         xs_tcp_check_fraghdr(transport);
1461 }
1462
1463 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1464 {
1465         struct rpc_xprt *xprt = rd_desc->arg.data;
1466         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1467         struct xdr_skb_reader desc = {
1468                 .skb    = skb,
1469                 .offset = offset,
1470                 .count  = len,
1471         };
1472
1473         dprintk("RPC:       xs_tcp_data_recv started\n");
1474         do {
1475                 trace_xs_tcp_data_recv(transport);
1476                 /* Read in a new fragment marker if necessary */
1477                 /* Can we ever really expect to get completely empty fragments? */
1478                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1479                         xs_tcp_read_fraghdr(xprt, &desc);
1480                         continue;
1481                 }
1482                 /* Read in the xid if necessary */
1483                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1484                         xs_tcp_read_xid(transport, &desc);
1485                         continue;
1486                 }
1487                 /* Read in the call/reply flag */
1488                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1489                         xs_tcp_read_calldir(transport, &desc);
1490                         continue;
1491                 }
1492                 /* Read in the request data */
1493                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1494                         xs_tcp_read_data(xprt, &desc);
1495                         continue;
1496                 }
1497                 /* Skip over any trailing bytes on short reads */
1498                 xs_tcp_read_discard(transport, &desc);
1499         } while (desc.count);
1500         trace_xs_tcp_data_recv(transport);
1501         dprintk("RPC:       xs_tcp_data_recv done\n");
1502         return len - desc.count;
1503 }
1504
1505 static void xs_tcp_data_receive(struct sock_xprt *transport)
1506 {
1507         struct rpc_xprt *xprt = &transport->xprt;
1508         struct sock *sk;
1509         read_descriptor_t rd_desc = {
1510                 .count = 2*1024*1024,
1511                 .arg.data = xprt,
1512         };
1513         unsigned long total = 0;
1514         int read = 0;
1515
1516         mutex_lock(&transport->recv_mutex);
1517         sk = transport->inet;
1518         if (sk == NULL)
1519                 goto out;
1520
1521         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1522         for (;;) {
1523                 lock_sock(sk);
1524                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1525                 if (read <= 0) {
1526                         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1527                         release_sock(sk);
1528                         if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1529                                 break;
1530                 } else {
1531                         release_sock(sk);
1532                         total += read;
1533                 }
1534                 rd_desc.count = 65536;
1535         }
1536 out:
1537         mutex_unlock(&transport->recv_mutex);
1538         trace_xs_tcp_data_ready(xprt, read, total);
1539 }
1540
1541 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1542 {
1543         struct sock_xprt *transport =
1544                 container_of(work, struct sock_xprt, recv_worker);
1545         xs_tcp_data_receive(transport);
1546 }
1547
1548 /**
1549  * xs_tcp_state_change - callback to handle TCP socket state changes
1550  * @sk: socket whose state has changed
1551  *
1552  */
1553 static void xs_tcp_state_change(struct sock *sk)
1554 {
1555         struct rpc_xprt *xprt;
1556         struct sock_xprt *transport;
1557
1558         read_lock_bh(&sk->sk_callback_lock);
1559         if (!(xprt = xprt_from_sock(sk)))
1560                 goto out;
1561         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1562         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1563                         sk->sk_state, xprt_connected(xprt),
1564                         sock_flag(sk, SOCK_DEAD),
1565                         sock_flag(sk, SOCK_ZAPPED),
1566                         sk->sk_shutdown);
1567
1568         transport = container_of(xprt, struct sock_xprt, xprt);
1569         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1570         switch (sk->sk_state) {
1571         case TCP_ESTABLISHED:
1572                 spin_lock(&xprt->transport_lock);
1573                 if (!xprt_test_and_set_connected(xprt)) {
1574
1575                         /* Reset TCP record info */
1576                         transport->tcp_offset = 0;
1577                         transport->tcp_reclen = 0;
1578                         transport->tcp_copied = 0;
1579                         transport->tcp_flags =
1580                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1581                         xprt->connect_cookie++;
1582                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1583                         xprt_clear_connecting(xprt);
1584
1585                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1586                 }
1587                 spin_unlock(&xprt->transport_lock);
1588                 break;
1589         case TCP_FIN_WAIT1:
1590                 /* The client initiated a shutdown of the socket */
1591                 xprt->connect_cookie++;
1592                 xprt->reestablish_timeout = 0;
1593                 set_bit(XPRT_CLOSING, &xprt->state);
1594                 smp_mb__before_atomic();
1595                 clear_bit(XPRT_CONNECTED, &xprt->state);
1596                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1597                 smp_mb__after_atomic();
1598                 break;
1599         case TCP_CLOSE_WAIT:
1600                 /* The server initiated a shutdown of the socket */
1601                 xprt->connect_cookie++;
1602                 clear_bit(XPRT_CONNECTED, &xprt->state);
1603                 xs_tcp_force_close(xprt);
1604         case TCP_CLOSING:
1605                 /*
1606                  * If the server closed down the connection, make sure that
1607                  * we back off before reconnecting
1608                  */
1609                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1610                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1611                 break;
1612         case TCP_LAST_ACK:
1613                 set_bit(XPRT_CLOSING, &xprt->state);
1614                 smp_mb__before_atomic();
1615                 clear_bit(XPRT_CONNECTED, &xprt->state);
1616                 smp_mb__after_atomic();
1617                 break;
1618         case TCP_CLOSE:
1619                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1620                                         &transport->sock_state))
1621                         xprt_clear_connecting(xprt);
1622                 xs_sock_mark_closed(xprt);
1623         }
1624  out:
1625         read_unlock_bh(&sk->sk_callback_lock);
1626 }
1627
1628 static void xs_write_space(struct sock *sk)
1629 {
1630         struct socket_wq *wq;
1631         struct rpc_xprt *xprt;
1632
1633         if (!sk->sk_socket)
1634                 return;
1635         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1636
1637         if (unlikely(!(xprt = xprt_from_sock(sk))))
1638                 return;
1639         rcu_read_lock();
1640         wq = rcu_dereference(sk->sk_wq);
1641         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1642                 goto out;
1643
1644         xprt_write_space(xprt);
1645 out:
1646         rcu_read_unlock();
1647 }
1648
1649 /**
1650  * xs_udp_write_space - callback invoked when socket buffer space
1651  *                             becomes available
1652  * @sk: socket whose state has changed
1653  *
1654  * Called when more output buffer space is available for this socket.
1655  * We try not to wake our writers until they can make "significant"
1656  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1657  * with a bunch of small requests.
1658  */
1659 static void xs_udp_write_space(struct sock *sk)
1660 {
1661         read_lock_bh(&sk->sk_callback_lock);
1662
1663         /* from net/core/sock.c:sock_def_write_space */
1664         if (sock_writeable(sk))
1665                 xs_write_space(sk);
1666
1667         read_unlock_bh(&sk->sk_callback_lock);
1668 }
1669
1670 /**
1671  * xs_tcp_write_space - callback invoked when socket buffer space
1672  *                             becomes available
1673  * @sk: socket whose state has changed
1674  *
1675  * Called when more output buffer space is available for this socket.
1676  * We try not to wake our writers until they can make "significant"
1677  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1678  * with a bunch of small requests.
1679  */
1680 static void xs_tcp_write_space(struct sock *sk)
1681 {
1682         read_lock_bh(&sk->sk_callback_lock);
1683
1684         /* from net/core/stream.c:sk_stream_write_space */
1685         if (sk_stream_is_writeable(sk))
1686                 xs_write_space(sk);
1687
1688         read_unlock_bh(&sk->sk_callback_lock);
1689 }
1690
1691 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1692 {
1693         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1694         struct sock *sk = transport->inet;
1695
1696         if (transport->rcvsize) {
1697                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1698                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1699         }
1700         if (transport->sndsize) {
1701                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1702                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1703                 sk->sk_write_space(sk);
1704         }
1705 }
1706
1707 /**
1708  * xs_udp_set_buffer_size - set send and receive limits
1709  * @xprt: generic transport
1710  * @sndsize: requested size of send buffer, in bytes
1711  * @rcvsize: requested size of receive buffer, in bytes
1712  *
1713  * Set socket send and receive buffer size limits.
1714  */
1715 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1716 {
1717         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1718
1719         transport->sndsize = 0;
1720         if (sndsize)
1721                 transport->sndsize = sndsize + 1024;
1722         transport->rcvsize = 0;
1723         if (rcvsize)
1724                 transport->rcvsize = rcvsize + 1024;
1725
1726         xs_udp_do_set_buffer_size(xprt);
1727 }
1728
1729 /**
1730  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1731  * @task: task that timed out
1732  *
1733  * Adjust the congestion window after a retransmit timeout has occurred.
1734  */
1735 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1736 {
1737         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1738 }
1739
1740 static int xs_get_random_port(void)
1741 {
1742         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1743         unsigned short range;
1744         unsigned short rand;
1745
1746         if (max < min)
1747                 return -EADDRINUSE;
1748         range = max - min + 1;
1749         rand = (unsigned short) prandom_u32() % range;
1750         return rand + min;
1751 }
1752
1753 /**
1754  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1755  * @sock: socket
1756  *
1757  * Note that this function has to be called on all sockets that share the
1758  * same port, and it must be called before binding.
1759  */
1760 static void xs_sock_set_reuseport(struct socket *sock)
1761 {
1762         int opt = 1;
1763
1764         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1765                         (char *)&opt, sizeof(opt));
1766 }
1767
1768 static unsigned short xs_sock_getport(struct socket *sock)
1769 {
1770         struct sockaddr_storage buf;
1771         int buflen;
1772         unsigned short port = 0;
1773
1774         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1775                 goto out;
1776         switch (buf.ss_family) {
1777         case AF_INET6:
1778                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1779                 break;
1780         case AF_INET:
1781                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1782         }
1783 out:
1784         return port;
1785 }
1786
1787 /**
1788  * xs_set_port - reset the port number in the remote endpoint address
1789  * @xprt: generic transport
1790  * @port: new port number
1791  *
1792  */
1793 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1794 {
1795         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1796
1797         rpc_set_port(xs_addr(xprt), port);
1798         xs_update_peer_port(xprt);
1799 }
1800
1801 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1802 {
1803         if (transport->srcport == 0)
1804                 transport->srcport = xs_sock_getport(sock);
1805 }
1806
1807 static int xs_get_srcport(struct sock_xprt *transport)
1808 {
1809         int port = transport->srcport;
1810
1811         if (port == 0 && transport->xprt.resvport)
1812                 port = xs_get_random_port();
1813         return port;
1814 }
1815
1816 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1817 {
1818         if (transport->srcport != 0)
1819                 transport->srcport = 0;
1820         if (!transport->xprt.resvport)
1821                 return 0;
1822         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1823                 return xprt_max_resvport;
1824         return --port;
1825 }
1826 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1827 {
1828         struct sockaddr_storage myaddr;
1829         int err, nloop = 0;
1830         int port = xs_get_srcport(transport);
1831         unsigned short last;
1832
1833         /*
1834          * If we are asking for any ephemeral port (i.e. port == 0 &&
1835          * transport->xprt.resvport == 0), don't bind.  Let the local
1836          * port selection happen implicitly when the socket is used
1837          * (for example at connect time).
1838          *
1839          * This ensures that we can continue to establish TCP
1840          * connections even when all local ephemeral ports are already
1841          * a part of some TCP connection.  This makes no difference
1842          * for UDP sockets, but also doens't harm them.
1843          *
1844          * If we're asking for any reserved port (i.e. port == 0 &&
1845          * transport->xprt.resvport == 1) xs_get_srcport above will
1846          * ensure that port is non-zero and we will bind as needed.
1847          */
1848         if (port <= 0)
1849                 return port;
1850
1851         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1852         do {
1853                 rpc_set_port((struct sockaddr *)&myaddr, port);
1854                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1855                                 transport->xprt.addrlen);
1856                 if (err == 0) {
1857                         transport->srcport = port;
1858                         break;
1859                 }
1860                 last = port;
1861                 port = xs_next_srcport(transport, port);
1862                 if (port > last)
1863                         nloop++;
1864         } while (err == -EADDRINUSE && nloop != 2);
1865
1866         if (myaddr.ss_family == AF_INET)
1867                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1868                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1869                                 port, err ? "failed" : "ok", err);
1870         else
1871                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1872                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1873                                 port, err ? "failed" : "ok", err);
1874         return err;
1875 }
1876
1877 /*
1878  * We don't support autobind on AF_LOCAL sockets
1879  */
1880 static void xs_local_rpcbind(struct rpc_task *task)
1881 {
1882         xprt_set_bound(task->tk_xprt);
1883 }
1884
1885 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1886 {
1887 }
1888
1889 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1890 static struct lock_class_key xs_key[2];
1891 static struct lock_class_key xs_slock_key[2];
1892
1893 static inline void xs_reclassify_socketu(struct socket *sock)
1894 {
1895         struct sock *sk = sock->sk;
1896
1897         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1898                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1899 }
1900
1901 static inline void xs_reclassify_socket4(struct socket *sock)
1902 {
1903         struct sock *sk = sock->sk;
1904
1905         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1906                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1907 }
1908
1909 static inline void xs_reclassify_socket6(struct socket *sock)
1910 {
1911         struct sock *sk = sock->sk;
1912
1913         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1914                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1915 }
1916
1917 static inline void xs_reclassify_socket(int family, struct socket *sock)
1918 {
1919         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1920                 return;
1921
1922         switch (family) {
1923         case AF_LOCAL:
1924                 xs_reclassify_socketu(sock);
1925                 break;
1926         case AF_INET:
1927                 xs_reclassify_socket4(sock);
1928                 break;
1929         case AF_INET6:
1930                 xs_reclassify_socket6(sock);
1931                 break;
1932         }
1933 }
1934 #else
1935 static inline void xs_reclassify_socket(int family, struct socket *sock)
1936 {
1937 }
1938 #endif
1939
1940 static void xs_dummy_setup_socket(struct work_struct *work)
1941 {
1942 }
1943
1944 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1945                 struct sock_xprt *transport, int family, int type,
1946                 int protocol, bool reuseport)
1947 {
1948         struct socket *sock;
1949         int err;
1950
1951         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1952         if (err < 0) {
1953                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1954                                 protocol, -err);
1955                 goto out;
1956         }
1957         xs_reclassify_socket(family, sock);
1958
1959         if (reuseport)
1960                 xs_sock_set_reuseport(sock);
1961
1962         err = xs_bind(transport, sock);
1963         if (err) {
1964                 sock_release(sock);
1965                 goto out;
1966         }
1967
1968         return sock;
1969 out:
1970         return ERR_PTR(err);
1971 }
1972
1973 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1974                                       struct socket *sock)
1975 {
1976         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1977                                                                         xprt);
1978
1979         if (!transport->inet) {
1980                 struct sock *sk = sock->sk;
1981
1982                 write_lock_bh(&sk->sk_callback_lock);
1983
1984                 xs_save_old_callbacks(transport, sk);
1985
1986                 sk->sk_user_data = xprt;
1987                 sk->sk_data_ready = xs_data_ready;
1988                 sk->sk_write_space = xs_udp_write_space;
1989                 sock_set_flag(sk, SOCK_FASYNC);
1990                 sk->sk_error_report = xs_error_report;
1991                 sk->sk_allocation = GFP_NOIO;
1992
1993                 xprt_clear_connected(xprt);
1994
1995                 /* Reset to new socket */
1996                 transport->sock = sock;
1997                 transport->inet = sk;
1998
1999                 write_unlock_bh(&sk->sk_callback_lock);
2000         }
2001
2002         /* Tell the socket layer to start connecting... */
2003         xprt->stat.connect_count++;
2004         xprt->stat.connect_start = jiffies;
2005         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2006 }
2007
2008 /**
2009  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2010  * @transport: socket transport to connect
2011  */
2012 static int xs_local_setup_socket(struct sock_xprt *transport)
2013 {
2014         struct rpc_xprt *xprt = &transport->xprt;
2015         struct socket *sock;
2016         int status = -EIO;
2017
2018         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2019                                         SOCK_STREAM, 0, &sock, 1);
2020         if (status < 0) {
2021                 dprintk("RPC:       can't create AF_LOCAL "
2022                         "transport socket (%d).\n", -status);
2023                 goto out;
2024         }
2025         xs_reclassify_socket(AF_LOCAL, sock);
2026
2027         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2028                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2029
2030         status = xs_local_finish_connecting(xprt, sock);
2031         trace_rpc_socket_connect(xprt, sock, status);
2032         switch (status) {
2033         case 0:
2034                 dprintk("RPC:       xprt %p connected to %s\n",
2035                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2036                 xprt_set_connected(xprt);
2037         case -ENOBUFS:
2038                 break;
2039         case -ENOENT:
2040                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2041                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2042                 break;
2043         case -ECONNREFUSED:
2044                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2045                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2046                 break;
2047         default:
2048                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2049                                 __func__, -status,
2050                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2051         }
2052
2053 out:
2054         xprt_clear_connecting(xprt);
2055         xprt_wake_pending_tasks(xprt, status);
2056         return status;
2057 }
2058
2059 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2060 {
2061         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2062         int ret;
2063
2064          if (RPC_IS_ASYNC(task)) {
2065                 /*
2066                  * We want the AF_LOCAL connect to be resolved in the
2067                  * filesystem namespace of the process making the rpc
2068                  * call.  Thus we connect synchronously.
2069                  *
2070                  * If we want to support asynchronous AF_LOCAL calls,
2071                  * we'll need to figure out how to pass a namespace to
2072                  * connect.
2073                  */
2074                 rpc_exit(task, -ENOTCONN);
2075                 return;
2076         }
2077         ret = xs_local_setup_socket(transport);
2078         if (ret && !RPC_IS_SOFTCONN(task))
2079                 msleep_interruptible(15000);
2080 }
2081
2082 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2083 /*
2084  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2085  * know that we have exclusive access to the socket), to guard against
2086  * races with xs_reset_transport.
2087  */
2088 static void xs_set_memalloc(struct rpc_xprt *xprt)
2089 {
2090         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2091                         xprt);
2092
2093         /*
2094          * If there's no sock, then we have nothing to set. The
2095          * reconnecting process will get it for us.
2096          */
2097         if (!transport->inet)
2098                 return;
2099         if (atomic_read(&xprt->swapper))
2100                 sk_set_memalloc(transport->inet);
2101 }
2102
2103 /**
2104  * xs_enable_swap - Tag this transport as being used for swap.
2105  * @xprt: transport to tag
2106  *
2107  * Take a reference to this transport on behalf of the rpc_clnt, and
2108  * optionally mark it for swapping if it wasn't already.
2109  */
2110 static int
2111 xs_enable_swap(struct rpc_xprt *xprt)
2112 {
2113         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2114
2115         if (atomic_inc_return(&xprt->swapper) != 1)
2116                 return 0;
2117         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2118                 return -ERESTARTSYS;
2119         if (xs->inet)
2120                 sk_set_memalloc(xs->inet);
2121         xprt_release_xprt(xprt, NULL);
2122         return 0;
2123 }
2124
2125 /**
2126  * xs_disable_swap - Untag this transport as being used for swap.
2127  * @xprt: transport to tag
2128  *
2129  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2130  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2131  */
2132 static void
2133 xs_disable_swap(struct rpc_xprt *xprt)
2134 {
2135         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2136
2137         if (!atomic_dec_and_test(&xprt->swapper))
2138                 return;
2139         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2140                 return;
2141         if (xs->inet)
2142                 sk_clear_memalloc(xs->inet);
2143         xprt_release_xprt(xprt, NULL);
2144 }
2145 #else
2146 static void xs_set_memalloc(struct rpc_xprt *xprt)
2147 {
2148 }
2149
2150 static int
2151 xs_enable_swap(struct rpc_xprt *xprt)
2152 {
2153         return -EINVAL;
2154 }
2155
2156 static void
2157 xs_disable_swap(struct rpc_xprt *xprt)
2158 {
2159 }
2160 #endif
2161
2162 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2163 {
2164         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2165
2166         if (!transport->inet) {
2167                 struct sock *sk = sock->sk;
2168
2169                 write_lock_bh(&sk->sk_callback_lock);
2170
2171                 xs_save_old_callbacks(transport, sk);
2172
2173                 sk->sk_user_data = xprt;
2174                 sk->sk_data_ready = xs_data_ready;
2175                 sk->sk_write_space = xs_udp_write_space;
2176                 sock_set_flag(sk, SOCK_FASYNC);
2177                 sk->sk_allocation = GFP_NOIO;
2178
2179                 xprt_set_connected(xprt);
2180
2181                 /* Reset to new socket */
2182                 transport->sock = sock;
2183                 transport->inet = sk;
2184
2185                 xs_set_memalloc(xprt);
2186
2187                 write_unlock_bh(&sk->sk_callback_lock);
2188         }
2189         xs_udp_do_set_buffer_size(xprt);
2190
2191         xprt->stat.connect_start = jiffies;
2192 }
2193
2194 static void xs_udp_setup_socket(struct work_struct *work)
2195 {
2196         struct sock_xprt *transport =
2197                 container_of(work, struct sock_xprt, connect_worker.work);
2198         struct rpc_xprt *xprt = &transport->xprt;
2199         struct socket *sock = transport->sock;
2200         int status = -EIO;
2201
2202         sock = xs_create_sock(xprt, transport,
2203                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2204                         IPPROTO_UDP, false);
2205         if (IS_ERR(sock))
2206                 goto out;
2207
2208         dprintk("RPC:       worker connecting xprt %p via %s to "
2209                                 "%s (port %s)\n", xprt,
2210                         xprt->address_strings[RPC_DISPLAY_PROTO],
2211                         xprt->address_strings[RPC_DISPLAY_ADDR],
2212                         xprt->address_strings[RPC_DISPLAY_PORT]);
2213
2214         xs_udp_finish_connecting(xprt, sock);
2215         trace_rpc_socket_connect(xprt, sock, 0);
2216         status = 0;
2217 out:
2218         xprt_clear_connecting(xprt);
2219         xprt_unlock_connect(xprt, transport);
2220         xprt_wake_pending_tasks(xprt, status);
2221 }
2222
2223 /**
2224  * xs_tcp_shutdown - gracefully shut down a TCP socket
2225  * @xprt: transport
2226  *
2227  * Initiates a graceful shutdown of the TCP socket by calling the
2228  * equivalent of shutdown(SHUT_RDWR);
2229  */
2230 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2231 {
2232         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2233         struct socket *sock = transport->sock;
2234
2235         if (sock == NULL)
2236                 return;
2237         if (xprt_connected(xprt)) {
2238                 kernel_sock_shutdown(sock, SHUT_RDWR);
2239                 trace_rpc_socket_shutdown(xprt, sock);
2240         } else
2241                 xs_reset_transport(transport);
2242 }
2243
2244 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2245 {
2246         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2247         int ret = -ENOTCONN;
2248
2249         if (!transport->inet) {
2250                 struct sock *sk = sock->sk;
2251                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2252                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2253                 unsigned int opt_on = 1;
2254                 unsigned int timeo;
2255                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2256
2257                 /* TCP Keepalive options */
2258                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2259                                 (char *)&opt_on, sizeof(opt_on));
2260                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2261                                 (char *)&keepidle, sizeof(keepidle));
2262                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2263                                 (char *)&keepidle, sizeof(keepidle));
2264                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2265                                 (char *)&keepcnt, sizeof(keepcnt));
2266
2267                 /* Avoid temporary address, they are bad for long-lived
2268                  * connections such as NFS mounts.
2269                  * RFC4941, section 3.6 suggests that:
2270                  *    Individual applications, which have specific
2271                  *    knowledge about the normal duration of connections,
2272                  *    MAY override this as appropriate.
2273                  */
2274                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2275                                 (char *)&addr_pref, sizeof(addr_pref));
2276
2277                 /* TCP user timeout (see RFC5482) */
2278                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2279                         (xprt->timeout->to_retries + 1);
2280                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2281                                 (char *)&timeo, sizeof(timeo));
2282
2283                 write_lock_bh(&sk->sk_callback_lock);
2284
2285                 xs_save_old_callbacks(transport, sk);
2286
2287                 sk->sk_user_data = xprt;
2288                 sk->sk_data_ready = xs_data_ready;
2289                 sk->sk_state_change = xs_tcp_state_change;
2290                 sk->sk_write_space = xs_tcp_write_space;
2291                 sock_set_flag(sk, SOCK_FASYNC);
2292                 sk->sk_error_report = xs_error_report;
2293                 sk->sk_allocation = GFP_NOIO;
2294
2295                 /* socket options */
2296                 sock_reset_flag(sk, SOCK_LINGER);
2297                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2298
2299                 xprt_clear_connected(xprt);
2300
2301                 /* Reset to new socket */
2302                 transport->sock = sock;
2303                 transport->inet = sk;
2304
2305                 write_unlock_bh(&sk->sk_callback_lock);
2306         }
2307
2308         if (!xprt_bound(xprt))
2309                 goto out;
2310
2311         xs_set_memalloc(xprt);
2312
2313         /* Tell the socket layer to start connecting... */
2314         xprt->stat.connect_count++;
2315         xprt->stat.connect_start = jiffies;
2316         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2317         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2318         switch (ret) {
2319         case 0:
2320                 xs_set_srcport(transport, sock);
2321         case -EINPROGRESS:
2322                 /* SYN_SENT! */
2323                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2324                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2325                 break;
2326         case -EADDRNOTAVAIL:
2327                 /* Source port number is unavailable. Try a new one! */
2328                 transport->srcport = 0;
2329         }
2330 out:
2331         return ret;
2332 }
2333
2334 /**
2335  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2336  *
2337  * Invoked by a work queue tasklet.
2338  */
2339 static void xs_tcp_setup_socket(struct work_struct *work)
2340 {
2341         struct sock_xprt *transport =
2342                 container_of(work, struct sock_xprt, connect_worker.work);
2343         struct socket *sock = transport->sock;
2344         struct rpc_xprt *xprt = &transport->xprt;
2345         int status = -EIO;
2346
2347         if (!sock) {
2348                 sock = xs_create_sock(xprt, transport,
2349                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2350                                 IPPROTO_TCP, true);
2351                 if (IS_ERR(sock)) {
2352                         status = PTR_ERR(sock);
2353                         goto out;
2354                 }
2355         }
2356
2357         dprintk("RPC:       worker connecting xprt %p via %s to "
2358                                 "%s (port %s)\n", xprt,
2359                         xprt->address_strings[RPC_DISPLAY_PROTO],
2360                         xprt->address_strings[RPC_DISPLAY_ADDR],
2361                         xprt->address_strings[RPC_DISPLAY_PORT]);
2362
2363         status = xs_tcp_finish_connecting(xprt, sock);
2364         trace_rpc_socket_connect(xprt, sock, status);
2365         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2366                         xprt, -status, xprt_connected(xprt),
2367                         sock->sk->sk_state);
2368         switch (status) {
2369         default:
2370                 printk("%s: connect returned unhandled error %d\n",
2371                         __func__, status);
2372         case -EADDRNOTAVAIL:
2373                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2374                  * and retry
2375                  */
2376                 xs_tcp_force_close(xprt);
2377                 break;
2378         case 0:
2379         case -EINPROGRESS:
2380         case -EALREADY:
2381                 xprt_unlock_connect(xprt, transport);
2382                 return;
2383         case -EINVAL:
2384                 /* Happens, for instance, if the user specified a link
2385                  * local IPv6 address without a scope-id.
2386                  */
2387         case -ECONNREFUSED:
2388         case -ECONNRESET:
2389         case -ENETUNREACH:
2390         case -EHOSTUNREACH:
2391         case -EADDRINUSE:
2392         case -ENOBUFS:
2393                 /*
2394                  * xs_tcp_force_close() wakes tasks with -EIO.
2395                  * We need to wake them first to ensure the
2396                  * correct error code.
2397                  */
2398                 xprt_wake_pending_tasks(xprt, status);
2399                 xs_tcp_force_close(xprt);
2400                 goto out;
2401         }
2402         status = -EAGAIN;
2403 out:
2404         xprt_clear_connecting(xprt);
2405         xprt_unlock_connect(xprt, transport);
2406         xprt_wake_pending_tasks(xprt, status);
2407 }
2408
2409 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2410 {
2411         unsigned long start, now = jiffies;
2412
2413         start = xprt->stat.connect_start + xprt->reestablish_timeout;
2414         if (time_after(start, now))
2415                 return start - now;
2416         return 0;
2417 }
2418
2419 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2420 {
2421         xprt->reestablish_timeout <<= 1;
2422         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2423                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2424         if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2425                 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2426 }
2427
2428 /**
2429  * xs_connect - connect a socket to a remote endpoint
2430  * @xprt: pointer to transport structure
2431  * @task: address of RPC task that manages state of connect request
2432  *
2433  * TCP: If the remote end dropped the connection, delay reconnecting.
2434  *
2435  * UDP socket connects are synchronous, but we use a work queue anyway
2436  * to guarantee that even unprivileged user processes can set up a
2437  * socket on a privileged port.
2438  *
2439  * If a UDP socket connect fails, the delay behavior here prevents
2440  * retry floods (hard mounts).
2441  */
2442 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2443 {
2444         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2445         unsigned long delay = 0;
2446
2447         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2448
2449         if (transport->sock != NULL) {
2450                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2451                                 "seconds\n",
2452                                 xprt, xprt->reestablish_timeout / HZ);
2453
2454                 /* Start by resetting any existing state */
2455                 xs_reset_transport(transport);
2456
2457                 delay = xs_reconnect_delay(xprt);
2458                 xs_reconnect_backoff(xprt);
2459
2460         } else
2461                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2462
2463         queue_delayed_work(xprtiod_workqueue,
2464                         &transport->connect_worker,
2465                         delay);
2466 }
2467
2468 /**
2469  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2470  * @xprt: rpc_xprt struct containing statistics
2471  * @seq: output file
2472  *
2473  */
2474 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2475 {
2476         long idle_time = 0;
2477
2478         if (xprt_connected(xprt))
2479                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2480
2481         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2482                         "%llu %llu %lu %llu %llu\n",
2483                         xprt->stat.bind_count,
2484                         xprt->stat.connect_count,
2485                         xprt->stat.connect_time,
2486                         idle_time,
2487                         xprt->stat.sends,
2488                         xprt->stat.recvs,
2489                         xprt->stat.bad_xids,
2490                         xprt->stat.req_u,
2491                         xprt->stat.bklog_u,
2492                         xprt->stat.max_slots,
2493                         xprt->stat.sending_u,
2494                         xprt->stat.pending_u);
2495 }
2496
2497 /**
2498  * xs_udp_print_stats - display UDP socket-specifc stats
2499  * @xprt: rpc_xprt struct containing statistics
2500  * @seq: output file
2501  *
2502  */
2503 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2504 {
2505         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2506
2507         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2508                         "%lu %llu %llu\n",
2509                         transport->srcport,
2510                         xprt->stat.bind_count,
2511                         xprt->stat.sends,
2512                         xprt->stat.recvs,
2513                         xprt->stat.bad_xids,
2514                         xprt->stat.req_u,
2515                         xprt->stat.bklog_u,
2516                         xprt->stat.max_slots,
2517                         xprt->stat.sending_u,
2518                         xprt->stat.pending_u);
2519 }
2520
2521 /**
2522  * xs_tcp_print_stats - display TCP socket-specifc stats
2523  * @xprt: rpc_xprt struct containing statistics
2524  * @seq: output file
2525  *
2526  */
2527 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2528 {
2529         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2530         long idle_time = 0;
2531
2532         if (xprt_connected(xprt))
2533                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2534
2535         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2536                         "%llu %llu %lu %llu %llu\n",
2537                         transport->srcport,
2538                         xprt->stat.bind_count,
2539                         xprt->stat.connect_count,
2540                         xprt->stat.connect_time,
2541                         idle_time,
2542                         xprt->stat.sends,
2543                         xprt->stat.recvs,
2544                         xprt->stat.bad_xids,
2545                         xprt->stat.req_u,
2546                         xprt->stat.bklog_u,
2547                         xprt->stat.max_slots,
2548                         xprt->stat.sending_u,
2549                         xprt->stat.pending_u);
2550 }
2551
2552 /*
2553  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2554  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2555  * to use the server side send routines.
2556  */
2557 static int bc_malloc(struct rpc_task *task)
2558 {
2559         struct rpc_rqst *rqst = task->tk_rqstp;
2560         size_t size = rqst->rq_callsize;
2561         struct page *page;
2562         struct rpc_buffer *buf;
2563
2564         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2565                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2566                           size);
2567                 return -EINVAL;
2568         }
2569
2570         page = alloc_page(GFP_KERNEL);
2571         if (!page)
2572                 return -ENOMEM;
2573
2574         buf = page_address(page);
2575         buf->len = PAGE_SIZE;
2576
2577         rqst->rq_buffer = buf->data;
2578         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2579         return 0;
2580 }
2581
2582 /*
2583  * Free the space allocated in the bc_alloc routine
2584  */
2585 static void bc_free(struct rpc_task *task)
2586 {
2587         void *buffer = task->tk_rqstp->rq_buffer;
2588         struct rpc_buffer *buf;
2589
2590         buf = container_of(buffer, struct rpc_buffer, data);
2591         free_page((unsigned long)buf);
2592 }
2593
2594 /*
2595  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2596  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2597  */
2598 static int bc_sendto(struct rpc_rqst *req)
2599 {
2600         int len;
2601         struct xdr_buf *xbufp = &req->rq_snd_buf;
2602         struct rpc_xprt *xprt = req->rq_xprt;
2603         struct sock_xprt *transport =
2604                                 container_of(xprt, struct sock_xprt, xprt);
2605         struct socket *sock = transport->sock;
2606         unsigned long headoff;
2607         unsigned long tailoff;
2608
2609         xs_encode_stream_record_marker(xbufp);
2610
2611         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2612         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2613         len = svc_send_common(sock, xbufp,
2614                               virt_to_page(xbufp->head[0].iov_base), headoff,
2615                               xbufp->tail[0].iov_base, tailoff);
2616
2617         if (len != xbufp->len) {
2618                 printk(KERN_NOTICE "Error sending entire callback!\n");
2619                 len = -EAGAIN;
2620         }
2621
2622         return len;
2623 }
2624
2625 /*
2626  * The send routine. Borrows from svc_send
2627  */
2628 static int bc_send_request(struct rpc_task *task)
2629 {
2630         struct rpc_rqst *req = task->tk_rqstp;
2631         struct svc_xprt *xprt;
2632         int len;
2633
2634         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2635         /*
2636          * Get the server socket associated with this callback xprt
2637          */
2638         xprt = req->rq_xprt->bc_xprt;
2639
2640         /*
2641          * Grab the mutex to serialize data as the connection is shared
2642          * with the fore channel
2643          */
2644         if (!mutex_trylock(&xprt->xpt_mutex)) {
2645                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2646                 if (!mutex_trylock(&xprt->xpt_mutex))
2647                         return -EAGAIN;
2648                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2649         }
2650         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2651                 len = -ENOTCONN;
2652         else
2653                 len = bc_sendto(req);
2654         mutex_unlock(&xprt->xpt_mutex);
2655
2656         if (len > 0)
2657                 len = 0;
2658
2659         return len;
2660 }
2661
2662 /*
2663  * The close routine. Since this is client initiated, we do nothing
2664  */
2665
2666 static void bc_close(struct rpc_xprt *xprt)
2667 {
2668 }
2669
2670 /*
2671  * The xprt destroy routine. Again, because this connection is client
2672  * initiated, we do nothing
2673  */
2674
2675 static void bc_destroy(struct rpc_xprt *xprt)
2676 {
2677         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2678
2679         xs_xprt_free(xprt);
2680         module_put(THIS_MODULE);
2681 }
2682
2683 static struct rpc_xprt_ops xs_local_ops = {
2684         .reserve_xprt           = xprt_reserve_xprt,
2685         .release_xprt           = xs_tcp_release_xprt,
2686         .alloc_slot             = xprt_alloc_slot,
2687         .rpcbind                = xs_local_rpcbind,
2688         .set_port               = xs_local_set_port,
2689         .connect                = xs_local_connect,
2690         .buf_alloc              = rpc_malloc,
2691         .buf_free               = rpc_free,
2692         .send_request           = xs_local_send_request,
2693         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2694         .close                  = xs_close,
2695         .destroy                = xs_destroy,
2696         .print_stats            = xs_local_print_stats,
2697         .enable_swap            = xs_enable_swap,
2698         .disable_swap           = xs_disable_swap,
2699 };
2700
2701 static struct rpc_xprt_ops xs_udp_ops = {
2702         .set_buffer_size        = xs_udp_set_buffer_size,
2703         .reserve_xprt           = xprt_reserve_xprt_cong,
2704         .release_xprt           = xprt_release_xprt_cong,
2705         .alloc_slot             = xprt_alloc_slot,
2706         .rpcbind                = rpcb_getport_async,
2707         .set_port               = xs_set_port,
2708         .connect                = xs_connect,
2709         .buf_alloc              = rpc_malloc,
2710         .buf_free               = rpc_free,
2711         .send_request           = xs_udp_send_request,
2712         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2713         .timer                  = xs_udp_timer,
2714         .release_request        = xprt_release_rqst_cong,
2715         .close                  = xs_close,
2716         .destroy                = xs_destroy,
2717         .print_stats            = xs_udp_print_stats,
2718         .enable_swap            = xs_enable_swap,
2719         .disable_swap           = xs_disable_swap,
2720         .inject_disconnect      = xs_inject_disconnect,
2721 };
2722
2723 static struct rpc_xprt_ops xs_tcp_ops = {
2724         .reserve_xprt           = xprt_reserve_xprt,
2725         .release_xprt           = xs_tcp_release_xprt,
2726         .alloc_slot             = xprt_lock_and_alloc_slot,
2727         .rpcbind                = rpcb_getport_async,
2728         .set_port               = xs_set_port,
2729         .connect                = xs_connect,
2730         .buf_alloc              = rpc_malloc,
2731         .buf_free               = rpc_free,
2732         .send_request           = xs_tcp_send_request,
2733         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2734         .close                  = xs_tcp_shutdown,
2735         .destroy                = xs_destroy,
2736         .print_stats            = xs_tcp_print_stats,
2737         .enable_swap            = xs_enable_swap,
2738         .disable_swap           = xs_disable_swap,
2739         .inject_disconnect      = xs_inject_disconnect,
2740 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2741         .bc_setup               = xprt_setup_bc,
2742         .bc_up                  = xs_tcp_bc_up,
2743         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2744         .bc_free_rqst           = xprt_free_bc_rqst,
2745         .bc_destroy             = xprt_destroy_bc,
2746 #endif
2747 };
2748
2749 /*
2750  * The rpc_xprt_ops for the server backchannel
2751  */
2752
2753 static struct rpc_xprt_ops bc_tcp_ops = {
2754         .reserve_xprt           = xprt_reserve_xprt,
2755         .release_xprt           = xprt_release_xprt,
2756         .alloc_slot             = xprt_alloc_slot,
2757         .buf_alloc              = bc_malloc,
2758         .buf_free               = bc_free,
2759         .send_request           = bc_send_request,
2760         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2761         .close                  = bc_close,
2762         .destroy                = bc_destroy,
2763         .print_stats            = xs_tcp_print_stats,
2764         .enable_swap            = xs_enable_swap,
2765         .disable_swap           = xs_disable_swap,
2766         .inject_disconnect      = xs_inject_disconnect,
2767 };
2768
2769 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2770 {
2771         static const struct sockaddr_in sin = {
2772                 .sin_family             = AF_INET,
2773                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2774         };
2775         static const struct sockaddr_in6 sin6 = {
2776                 .sin6_family            = AF_INET6,
2777                 .sin6_addr              = IN6ADDR_ANY_INIT,
2778         };
2779
2780         switch (family) {
2781         case AF_LOCAL:
2782                 break;
2783         case AF_INET:
2784                 memcpy(sap, &sin, sizeof(sin));
2785                 break;
2786         case AF_INET6:
2787                 memcpy(sap, &sin6, sizeof(sin6));
2788                 break;
2789         default:
2790                 dprintk("RPC:       %s: Bad address family\n", __func__);
2791                 return -EAFNOSUPPORT;
2792         }
2793         return 0;
2794 }
2795
2796 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2797                                       unsigned int slot_table_size,
2798                                       unsigned int max_slot_table_size)
2799 {
2800         struct rpc_xprt *xprt;
2801         struct sock_xprt *new;
2802
2803         if (args->addrlen > sizeof(xprt->addr)) {
2804                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2805                 return ERR_PTR(-EBADF);
2806         }
2807
2808         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2809                         max_slot_table_size);
2810         if (xprt == NULL) {
2811                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2812                                 "rpc_xprt\n");
2813                 return ERR_PTR(-ENOMEM);
2814         }
2815
2816         new = container_of(xprt, struct sock_xprt, xprt);
2817         mutex_init(&new->recv_mutex);
2818         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2819         xprt->addrlen = args->addrlen;
2820         if (args->srcaddr)
2821                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2822         else {
2823                 int err;
2824                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2825                                         (struct sockaddr *)&new->srcaddr);
2826                 if (err != 0) {
2827                         xprt_free(xprt);
2828                         return ERR_PTR(err);
2829                 }
2830         }
2831
2832         return xprt;
2833 }
2834
2835 static const struct rpc_timeout xs_local_default_timeout = {
2836         .to_initval = 10 * HZ,
2837         .to_maxval = 10 * HZ,
2838         .to_retries = 2,
2839 };
2840
2841 /**
2842  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2843  * @args: rpc transport creation arguments
2844  *
2845  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2846  */
2847 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2848 {
2849         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2850         struct sock_xprt *transport;
2851         struct rpc_xprt *xprt;
2852         struct rpc_xprt *ret;
2853
2854         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2855                         xprt_max_tcp_slot_table_entries);
2856         if (IS_ERR(xprt))
2857                 return xprt;
2858         transport = container_of(xprt, struct sock_xprt, xprt);
2859
2860         xprt->prot = 0;
2861         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2862         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2863
2864         xprt->bind_timeout = XS_BIND_TO;
2865         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2866         xprt->idle_timeout = XS_IDLE_DISC_TO;
2867
2868         xprt->ops = &xs_local_ops;
2869         xprt->timeout = &xs_local_default_timeout;
2870
2871         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2872         INIT_DELAYED_WORK(&transport->connect_worker,
2873                         xs_dummy_setup_socket);
2874
2875         switch (sun->sun_family) {
2876         case AF_LOCAL:
2877                 if (sun->sun_path[0] != '/') {
2878                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2879                                         sun->sun_path);
2880                         ret = ERR_PTR(-EINVAL);
2881                         goto out_err;
2882                 }
2883                 xprt_set_bound(xprt);
2884                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2885                 break;
2886         default:
2887                 ret = ERR_PTR(-EAFNOSUPPORT);
2888                 goto out_err;
2889         }
2890
2891         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2892                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2893
2894         if (try_module_get(THIS_MODULE))
2895                 return xprt;
2896         ret = ERR_PTR(-EINVAL);
2897 out_err:
2898         xs_xprt_free(xprt);
2899         return ret;
2900 }
2901
2902 static const struct rpc_timeout xs_udp_default_timeout = {
2903         .to_initval = 5 * HZ,
2904         .to_maxval = 30 * HZ,
2905         .to_increment = 5 * HZ,
2906         .to_retries = 5,
2907 };
2908
2909 /**
2910  * xs_setup_udp - Set up transport to use a UDP socket
2911  * @args: rpc transport creation arguments
2912  *
2913  */
2914 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2915 {
2916         struct sockaddr *addr = args->dstaddr;
2917         struct rpc_xprt *xprt;
2918         struct sock_xprt *transport;
2919         struct rpc_xprt *ret;
2920
2921         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2922                         xprt_udp_slot_table_entries);
2923         if (IS_ERR(xprt))
2924                 return xprt;
2925         transport = container_of(xprt, struct sock_xprt, xprt);
2926
2927         xprt->prot = IPPROTO_UDP;
2928         xprt->tsh_size = 0;
2929         /* XXX: header size can vary due to auth type, IPv6, etc. */
2930         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2931
2932         xprt->bind_timeout = XS_BIND_TO;
2933         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2934         xprt->idle_timeout = XS_IDLE_DISC_TO;
2935
2936         xprt->ops = &xs_udp_ops;
2937
2938         xprt->timeout = &xs_udp_default_timeout;
2939
2940         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2941         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2942
2943         switch (addr->sa_family) {
2944         case AF_INET:
2945                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2946                         xprt_set_bound(xprt);
2947
2948                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2949                 break;
2950         case AF_INET6:
2951                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2952                         xprt_set_bound(xprt);
2953
2954                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2955                 break;
2956         default:
2957                 ret = ERR_PTR(-EAFNOSUPPORT);
2958                 goto out_err;
2959         }
2960
2961         if (xprt_bound(xprt))
2962                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2963                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2964                                 xprt->address_strings[RPC_DISPLAY_PORT],
2965                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2966         else
2967                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2968                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2969                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2970
2971         if (try_module_get(THIS_MODULE))
2972                 return xprt;
2973         ret = ERR_PTR(-EINVAL);
2974 out_err:
2975         xs_xprt_free(xprt);
2976         return ret;
2977 }
2978
2979 static const struct rpc_timeout xs_tcp_default_timeout = {
2980         .to_initval = 60 * HZ,
2981         .to_maxval = 60 * HZ,
2982         .to_retries = 2,
2983 };
2984
2985 /**
2986  * xs_setup_tcp - Set up transport to use a TCP socket
2987  * @args: rpc transport creation arguments
2988  *
2989  */
2990 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2991 {
2992         struct sockaddr *addr = args->dstaddr;
2993         struct rpc_xprt *xprt;
2994         struct sock_xprt *transport;
2995         struct rpc_xprt *ret;
2996         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2997
2998         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2999                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3000
3001         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3002                         max_slot_table_size);
3003         if (IS_ERR(xprt))
3004                 return xprt;
3005         transport = container_of(xprt, struct sock_xprt, xprt);
3006
3007         xprt->prot = IPPROTO_TCP;
3008         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3009         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3010
3011         xprt->bind_timeout = XS_BIND_TO;
3012         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3013         xprt->idle_timeout = XS_IDLE_DISC_TO;
3014
3015         xprt->ops = &xs_tcp_ops;
3016         xprt->timeout = &xs_tcp_default_timeout;
3017
3018         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3019
3020         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3021         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3022
3023         switch (addr->sa_family) {
3024         case AF_INET:
3025                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3026                         xprt_set_bound(xprt);
3027
3028                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3029                 break;
3030         case AF_INET6:
3031                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3032                         xprt_set_bound(xprt);
3033
3034                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3035                 break;
3036         default:
3037                 ret = ERR_PTR(-EAFNOSUPPORT);
3038                 goto out_err;
3039         }
3040
3041         if (xprt_bound(xprt))
3042                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3043                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3044                                 xprt->address_strings[RPC_DISPLAY_PORT],
3045                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3046         else
3047                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3048                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3049                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3050
3051         if (try_module_get(THIS_MODULE))
3052                 return xprt;
3053         ret = ERR_PTR(-EINVAL);
3054 out_err:
3055         xs_xprt_free(xprt);
3056         return ret;
3057 }
3058
3059 /**
3060  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3061  * @args: rpc transport creation arguments
3062  *
3063  */
3064 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3065 {
3066         struct sockaddr *addr = args->dstaddr;
3067         struct rpc_xprt *xprt;
3068         struct sock_xprt *transport;
3069         struct svc_sock *bc_sock;
3070         struct rpc_xprt *ret;
3071
3072         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3073                         xprt_tcp_slot_table_entries);
3074         if (IS_ERR(xprt))
3075                 return xprt;
3076         transport = container_of(xprt, struct sock_xprt, xprt);
3077
3078         xprt->prot = IPPROTO_TCP;
3079         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3080         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3081         xprt->timeout = &xs_tcp_default_timeout;
3082
3083         /* backchannel */
3084         xprt_set_bound(xprt);
3085         xprt->bind_timeout = 0;
3086         xprt->reestablish_timeout = 0;
3087         xprt->idle_timeout = 0;
3088
3089         xprt->ops = &bc_tcp_ops;
3090
3091         switch (addr->sa_family) {
3092         case AF_INET:
3093                 xs_format_peer_addresses(xprt, "tcp",
3094                                          RPCBIND_NETID_TCP);
3095                 break;
3096         case AF_INET6:
3097                 xs_format_peer_addresses(xprt, "tcp",
3098                                    RPCBIND_NETID_TCP6);
3099                 break;
3100         default:
3101                 ret = ERR_PTR(-EAFNOSUPPORT);
3102                 goto out_err;
3103         }
3104
3105         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3106                         xprt->address_strings[RPC_DISPLAY_ADDR],
3107                         xprt->address_strings[RPC_DISPLAY_PORT],
3108                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3109
3110         /*
3111          * Once we've associated a backchannel xprt with a connection,
3112          * we want to keep it around as long as the connection lasts,
3113          * in case we need to start using it for a backchannel again;
3114          * this reference won't be dropped until bc_xprt is destroyed.
3115          */
3116         xprt_get(xprt);
3117         args->bc_xprt->xpt_bc_xprt = xprt;
3118         xprt->bc_xprt = args->bc_xprt;
3119         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3120         transport->sock = bc_sock->sk_sock;
3121         transport->inet = bc_sock->sk_sk;
3122
3123         /*
3124          * Since we don't want connections for the backchannel, we set
3125          * the xprt status to connected
3126          */
3127         xprt_set_connected(xprt);
3128
3129         if (try_module_get(THIS_MODULE))
3130                 return xprt;
3131
3132         args->bc_xprt->xpt_bc_xprt = NULL;
3133         args->bc_xprt->xpt_bc_xps = NULL;
3134         xprt_put(xprt);
3135         ret = ERR_PTR(-EINVAL);
3136 out_err:
3137         xs_xprt_free(xprt);
3138         return ret;
3139 }
3140
3141 static struct xprt_class        xs_local_transport = {
3142         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3143         .name           = "named UNIX socket",
3144         .owner          = THIS_MODULE,
3145         .ident          = XPRT_TRANSPORT_LOCAL,
3146         .setup          = xs_setup_local,
3147         .netid          = { "" },
3148 };
3149
3150 static struct xprt_class        xs_udp_transport = {
3151         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3152         .name           = "udp",
3153         .owner          = THIS_MODULE,
3154         .ident          = XPRT_TRANSPORT_UDP,
3155         .setup          = xs_setup_udp,
3156         .netid          = { "udp", "udp6", "" },
3157 };
3158
3159 static struct xprt_class        xs_tcp_transport = {
3160         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3161         .name           = "tcp",
3162         .owner          = THIS_MODULE,
3163         .ident          = XPRT_TRANSPORT_TCP,
3164         .setup          = xs_setup_tcp,
3165         .netid          = { "tcp", "tcp6", "" },
3166 };
3167
3168 static struct xprt_class        xs_bc_tcp_transport = {
3169         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3170         .name           = "tcp NFSv4.1 backchannel",
3171         .owner          = THIS_MODULE,
3172         .ident          = XPRT_TRANSPORT_BC_TCP,
3173         .setup          = xs_setup_bc_tcp,
3174         .netid          = { "" },
3175 };
3176
3177 /**
3178  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3179  *
3180  */
3181 int init_socket_xprt(void)
3182 {
3183 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3184         if (!sunrpc_table_header)
3185                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3186 #endif
3187
3188         xprt_register_transport(&xs_local_transport);
3189         xprt_register_transport(&xs_udp_transport);
3190         xprt_register_transport(&xs_tcp_transport);
3191         xprt_register_transport(&xs_bc_tcp_transport);
3192
3193         return 0;
3194 }
3195
3196 /**
3197  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3198  *
3199  */
3200 void cleanup_socket_xprt(void)
3201 {
3202 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3203         if (sunrpc_table_header) {
3204                 unregister_sysctl_table(sunrpc_table_header);
3205                 sunrpc_table_header = NULL;
3206         }
3207 #endif
3208
3209         xprt_unregister_transport(&xs_local_transport);
3210         xprt_unregister_transport(&xs_udp_transport);
3211         xprt_unregister_transport(&xs_tcp_transport);
3212         xprt_unregister_transport(&xs_bc_tcp_transport);
3213 }
3214
3215 static int param_set_uint_minmax(const char *val,
3216                 const struct kernel_param *kp,
3217                 unsigned int min, unsigned int max)
3218 {
3219         unsigned int num;
3220         int ret;
3221
3222         if (!val)
3223                 return -EINVAL;
3224         ret = kstrtouint(val, 0, &num);
3225         if (ret == -EINVAL || num < min || num > max)
3226                 return -EINVAL;
3227         *((unsigned int *)kp->arg) = num;
3228         return 0;
3229 }
3230
3231 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3232 {
3233         return param_set_uint_minmax(val, kp,
3234                         RPC_MIN_RESVPORT,
3235                         RPC_MAX_RESVPORT);
3236 }
3237
3238 static const struct kernel_param_ops param_ops_portnr = {
3239         .set = param_set_portnr,
3240         .get = param_get_uint,
3241 };
3242
3243 #define param_check_portnr(name, p) \
3244         __param_check(name, p, unsigned int);
3245
3246 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3247 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3248
3249 static int param_set_slot_table_size(const char *val,
3250                                      const struct kernel_param *kp)
3251 {
3252         return param_set_uint_minmax(val, kp,
3253                         RPC_MIN_SLOT_TABLE,
3254                         RPC_MAX_SLOT_TABLE);
3255 }
3256
3257 static const struct kernel_param_ops param_ops_slot_table_size = {
3258         .set = param_set_slot_table_size,
3259         .get = param_get_uint,
3260 };
3261
3262 #define param_check_slot_table_size(name, p) \
3263         __param_check(name, p, unsigned int);
3264
3265 static int param_set_max_slot_table_size(const char *val,
3266                                      const struct kernel_param *kp)
3267 {
3268         return param_set_uint_minmax(val, kp,
3269                         RPC_MIN_SLOT_TABLE,
3270                         RPC_MAX_SLOT_TABLE_LIMIT);
3271 }
3272
3273 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3274         .set = param_set_max_slot_table_size,
3275         .get = param_get_uint,
3276 };
3277
3278 #define param_check_max_slot_table_size(name, p) \
3279         __param_check(name, p, unsigned int);
3280
3281 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3282                    slot_table_size, 0644);
3283 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3284                    max_slot_table_size, 0644);
3285 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3286                    slot_table_size, 0644);
3287