net/sunrpc/xprtsock.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/net/sunrpc/xprtsock.c
   4  *
   5  * Client-side transport implementation for sockets.
   6  *
   7  * TCP callback races fixes (C) 1998 Red Hat
   8  * TCP send fixes (C) 1998 Red Hat
   9  * TCP NFS related read + write fixes
  10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  11  *
  12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  13  * Fix behaviour when socket buffer is full.
  14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  15  *
  16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  17  *
  18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  19  *   <gilles.quillard@bull.net>
  20  */
  21
  22 #include <linux/types.h>
  23 #include <linux/string.h>
  24 #include <linux/slab.h>
  25 #include <linux/module.h>
  26 #include <linux/capability.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/errno.h>
  29 #include <linux/socket.h>
  30 #include <linux/in.h>
  31 #include <linux/net.h>
  32 #include <linux/mm.h>
  33 #include <linux/un.h>
  34 #include <linux/udp.h>
  35 #include <linux/tcp.h>
  36 #include <linux/sunrpc/clnt.h>
  37 #include <linux/sunrpc/addr.h>
  38 #include <linux/sunrpc/sched.h>
  39 #include <linux/sunrpc/svcsock.h>
  40 #include <linux/sunrpc/xprtsock.h>
  41 #include <linux/file.h>
  42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  43 #include <linux/sunrpc/bc_xprt.h>
  44 #endif
  45
  46 #include <net/sock.h>
  47 #include <net/checksum.h>
  48 #include <net/udp.h>
  49 #include <net/tcp.h>
  50 #include <linux/bvec.h>
  51 #include <linux/highmem.h>
  52 #include <linux/uio.h>
  53 #include <linux/sched/mm.h>
  54
  55 #include <trace/events/sunrpc.h>
  56
  57 #include "sunrpc.h"
  58
  59 static void xs_close(struct rpc_xprt *xprt);
  60 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  61                 struct socket *sock);
  62
  63 /*
  64  * xprtsock tunables
  65  */
  66 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  67 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  68 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  69
  70 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  71 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  72
  73 #define XS_TCP_LINGER_TO        (15U * HZ)
  74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  75
  76 /*
  77  * We can register our own files under /proc/sys/sunrpc by
  78  * calling register_sysctl_table() again.  The files in that
  79  * directory become the union of all files registered there.
  80  *
  81  * We simply need to make sure that we don't collide with
  82  * someone else's file names!
  83  */
  84
  85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  90
  91 static struct ctl_table_header *sunrpc_table_header;
  92
  93 /*
  94  * FIXME: changing the UDP slot table size should also resize the UDP
  95  *        socket buffers for existing UDP transports
  96  */
  97 static struct ctl_table xs_tunables_table[] = {
  98         {
  99                 .procname       = "udp_slot_table_entries",
 100                 .data           = &xprt_udp_slot_table_entries,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = proc_dointvec_minmax,
 104                 .extra1         = &min_slot_table_size,
 105                 .extra2         = &max_slot_table_size
 106         },
 107         {
 108                 .procname       = "tcp_slot_table_entries",
 109                 .data           = &xprt_tcp_slot_table_entries,
 110                 .maxlen         = sizeof(unsigned int),
 111                 .mode           = 0644,
 112                 .proc_handler   = proc_dointvec_minmax,
 113                 .extra1         = &min_slot_table_size,
 114                 .extra2         = &max_slot_table_size
 115         },
 116         {
 117                 .procname       = "tcp_max_slot_table_entries",
 118                 .data           = &xprt_max_tcp_slot_table_entries,
 119                 .maxlen         = sizeof(unsigned int),
 120                 .mode           = 0644,
 121                 .proc_handler   = proc_dointvec_minmax,
 122                 .extra1         = &min_slot_table_size,
 123                 .extra2         = &max_tcp_slot_table_limit
 124         },
 125         {
 126                 .procname       = "min_resvport",
 127                 .data           = &xprt_min_resvport,
 128                 .maxlen         = sizeof(unsigned int),
 129                 .mode           = 0644,
 130                 .proc_handler   = proc_dointvec_minmax,
 131                 .extra1         = &xprt_min_resvport_limit,
 132                 .extra2         = &xprt_max_resvport_limit
 133         },
 134         {
 135                 .procname       = "max_resvport",
 136                 .data           = &xprt_max_resvport,
 137                 .maxlen         = sizeof(unsigned int),
 138                 .mode           = 0644,
 139                 .proc_handler   = proc_dointvec_minmax,
 140                 .extra1         = &xprt_min_resvport_limit,
 141                 .extra2         = &xprt_max_resvport_limit
 142         },
 143         {
 144                 .procname       = "tcp_fin_timeout",
 145                 .data           = &xs_tcp_fin_timeout,
 146                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 147                 .mode           = 0644,
 148                 .proc_handler   = proc_dointvec_jiffies,
 149         },
 150         { },
 151 };
 152
 153 static struct ctl_table sunrpc_table[] = {
 154         {
 155                 .procname       = "sunrpc",
 156                 .mode           = 0555,
 157                 .child          = xs_tunables_table
 158         },
 159         { },
 160 };
 161
 162 /*
 163  * Wait duration for a reply from the RPC portmapper.
 164  */
 165 #define XS_BIND_TO              (60U * HZ)
 166
 167 /*
 168  * Delay if a UDP socket connect error occurs.  This is most likely some
 169  * kind of resource problem on the local host.
 170  */
 171 #define XS_UDP_REEST_TO         (2U * HZ)
 172
 173 /*
 174  * The reestablish timeout allows clients to delay for a bit before attempting
 175  * to reconnect to a server that just dropped our connection.
 176  *
 177  * We implement an exponential backoff when trying to reestablish a TCP
 178  * transport connection with the server.  Some servers like to drop a TCP
 179  * connection when they are overworked, so we start with a short timeout and
 180  * increase over time if the server is down or not responding.
 181  */
 182 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 183
 184 /*
 185  * TCP idle timeout; client drops the transport socket if it is idle
 186  * for this long.  Note that we also timeout UDP sockets to prevent
 187  * holding port numbers when there is no RPC traffic.
 188  */
 189 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 190
 191 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 192 # undef  RPC_DEBUG_DATA
 193 # define RPCDBG_FACILITY        RPCDBG_TRANS
 194 #endif
 195
 196 #ifdef RPC_DEBUG_DATA
 197 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 198 {
 199         u8 *buf = (u8 *) packet;
 200         int j;
 201
 202         dprintk("RPC:       %s\n", msg);
 203         for (j = 0; j < count && j < 128; j += 4) {
 204                 if (!(j & 31)) {
 205                         if (j)
 206                                 dprintk("\n");
 207                         dprintk("0x%04x ", j);
 208                 }
 209                 dprintk("%02x%02x%02x%02x ",
 210                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 211         }
 212         dprintk("\n");
 213 }
 214 #else
 215 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 216 {
 217         /* NOP */
 218 }
 219 #endif
 220
 221 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 222 {
 223         return (struct rpc_xprt *) sk->sk_user_data;
 224 }
 225
 226 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 227 {
 228         return (struct sockaddr *) &xprt->addr;
 229 }
 230
 231 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 232 {
 233         return (struct sockaddr_un *) &xprt->addr;
 234 }
 235
 236 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 237 {
 238         return (struct sockaddr_in *) &xprt->addr;
 239 }
 240
 241 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 242 {
 243         return (struct sockaddr_in6 *) &xprt->addr;
 244 }
 245
 246 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 247 {
 248         struct sockaddr *sap = xs_addr(xprt);
 249         struct sockaddr_in6 *sin6;
 250         struct sockaddr_in *sin;
 251         struct sockaddr_un *sun;
 252         char buf[128];
 253
 254         switch (sap->sa_family) {
 255         case AF_LOCAL:
 256                 sun = xs_addr_un(xprt);
 257                 strlcpy(buf, sun->sun_path, sizeof(buf));
 258                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 259                                                 kstrdup(buf, GFP_KERNEL);
 260                 break;
 261         case AF_INET:
 262                 (void)rpc_ntop(sap, buf, sizeof(buf));
 263                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 264                                                 kstrdup(buf, GFP_KERNEL);
 265                 sin = xs_addr_in(xprt);
 266                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 267                 break;
 268         case AF_INET6:
 269                 (void)rpc_ntop(sap, buf, sizeof(buf));
 270                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 271                                                 kstrdup(buf, GFP_KERNEL);
 272                 sin6 = xs_addr_in6(xprt);
 273                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 274                 break;
 275         default:
 276                 BUG();
 277         }
 278
 279         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 280 }
 281
 282 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 283 {
 284         struct sockaddr *sap = xs_addr(xprt);
 285         char buf[128];
 286
 287         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 288         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 289
 290         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 291         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 292 }
 293
 294 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 295                                      const char *protocol,
 296                                      const char *netid)
 297 {
 298         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 299         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 300         xs_format_common_peer_addresses(xprt);
 301         xs_format_common_peer_ports(xprt);
 302 }
 303
 304 static void xs_update_peer_port(struct rpc_xprt *xprt)
 305 {
 306         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 307         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 308
 309         xs_format_common_peer_ports(xprt);
 310 }
 311
 312 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 313 {
 314         unsigned int i;
 315
 316         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 317                 switch (i) {
 318                 case RPC_DISPLAY_PROTO:
 319                 case RPC_DISPLAY_NETID:
 320                         continue;
 321                 default:
 322                         kfree(xprt->address_strings[i]);
 323                 }
 324 }
 325
 326 static size_t
 327 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 328 {
 329         size_t i,n;
 330
 331         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 332                 return want;
 333         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 334         for (i = 0; i < n; i++) {
 335                 if (buf->pages[i])
 336                         continue;
 337                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 338                 if (!buf->pages[i]) {
 339                         i *= PAGE_SIZE;
 340                         return i > buf->page_base ? i - buf->page_base : 0;
 341                 }
 342         }
 343         return want;
 344 }
 345
 346 static ssize_t
 347 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 348 {
 349         ssize_t ret;
 350         if (seek != 0)
 351                 iov_iter_advance(&msg->msg_iter, seek);
 352         ret = sock_recvmsg(sock, msg, flags);
 353         return ret > 0 ? ret + seek : ret;
 354 }
 355
 356 static ssize_t
 357 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 358                 struct kvec *kvec, size_t count, size_t seek)
 359 {
 360         iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
 361         return xs_sock_recvmsg(sock, msg, flags, seek);
 362 }
 363
 364 static ssize_t
 365 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 366                 struct bio_vec *bvec, unsigned long nr, size_t count,
 367                 size_t seek)
 368 {
 369         iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
 370         return xs_sock_recvmsg(sock, msg, flags, seek);
 371 }
 372
 373 static ssize_t
 374 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 375                 size_t count)
 376 {
 377         iov_iter_discard(&msg->msg_iter, READ, count);
 378         return sock_recvmsg(sock, msg, flags);
 379 }
 380
 381 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 382 static void
 383 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 384 {
 385         struct bvec_iter bi = {
 386                 .bi_size = count,
 387         };
 388         struct bio_vec bv;
 389
 390         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 391         for_each_bvec(bv, bvec, bi, bi)
 392                 flush_dcache_page(bv.bv_page);
 393 }
 394 #else
 395 static inline void
 396 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 397 {
 398 }
 399 #endif
 400
 401 static ssize_t
 402 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 403                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 404 {
 405         size_t want, seek_init = seek, offset = 0;
 406         ssize_t ret;
 407
 408         want = min_t(size_t, count, buf->head[0].iov_len);
 409         if (seek < want) {
 410                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 411                 if (ret <= 0)
 412                         goto sock_err;
 413                 offset += ret;
 414                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 415                         goto out;
 416                 if (ret != want)
 417                         goto out;
 418                 seek = 0;
 419         } else {
 420                 seek -= want;
 421                 offset += want;
 422         }
 423
 424         want = xs_alloc_sparse_pages(buf,
 425                         min_t(size_t, count - offset, buf->page_len),
 426                         GFP_KERNEL);
 427         if (seek < want) {
 428                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 429                                 xdr_buf_pagecount(buf),
 430                                 want + buf->page_base,
 431                                 seek + buf->page_base);
 432                 if (ret <= 0)
 433                         goto sock_err;
 434                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 435                 offset += ret - buf->page_base;
 436                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 437                         goto out;
 438                 if (ret != want)
 439                         goto out;
 440                 seek = 0;
 441         } else {
 442                 seek -= want;
 443                 offset += want;
 444         }
 445
 446         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 447         if (seek < want) {
 448                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 449                 if (ret <= 0)
 450                         goto sock_err;
 451                 offset += ret;
 452                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 453                         goto out;
 454                 if (ret != want)
 455                         goto out;
 456         } else if (offset < seek_init)
 457                 offset = seek_init;
 458         ret = -EMSGSIZE;
 459 out:
 460         *read = offset - seek_init;
 461         return ret;
 462 sock_err:
 463         offset += seek;
 464         goto out;
 465 }
 466
 467 static void
 468 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 469 {
 470         if (!transport->recv.copied) {
 471                 if (buf->head[0].iov_len >= transport->recv.offset)
 472                         memcpy(buf->head[0].iov_base,
 473                                         &transport->recv.xid,
 474                                         transport->recv.offset);
 475                 transport->recv.copied = transport->recv.offset;
 476         }
 477 }
 478
 479 static bool
 480 xs_read_stream_request_done(struct sock_xprt *transport)
 481 {
 482         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 483 }
 484
 485 static void
 486 xs_read_stream_check_eor(struct sock_xprt *transport,
 487                 struct msghdr *msg)
 488 {
 489         if (xs_read_stream_request_done(transport))
 490                 msg->msg_flags |= MSG_EOR;
 491 }
 492
 493 static ssize_t
 494 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 495                 int flags, struct rpc_rqst *req)
 496 {
 497         struct xdr_buf *buf = &req->rq_private_buf;
 498         size_t want, uninitialized_var(read);
 499         ssize_t uninitialized_var(ret);
 500
 501         xs_read_header(transport, buf);
 502
 503         want = transport->recv.len - transport->recv.offset;
 504         if (want != 0) {
 505                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 506                                 transport->recv.copied + want,
 507                                 transport->recv.copied,
 508                                 &read);
 509                 transport->recv.offset += read;
 510                 transport->recv.copied += read;
 511         }
 512
 513         if (transport->recv.offset == transport->recv.len)
 514                 xs_read_stream_check_eor(transport, msg);
 515
 516         if (want == 0)
 517                 return 0;
 518
 519         switch (ret) {
 520         default:
 521                 break;
 522         case -EFAULT:
 523         case -EMSGSIZE:
 524                 msg->msg_flags |= MSG_TRUNC;
 525                 return read;
 526         case 0:
 527                 return -ESHUTDOWN;
 528         }
 529         return ret < 0 ? ret : read;
 530 }
 531
 532 static size_t
 533 xs_read_stream_headersize(bool isfrag)
 534 {
 535         if (isfrag)
 536                 return sizeof(__be32);
 537         return 3 * sizeof(__be32);
 538 }
 539
 540 static ssize_t
 541 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 542                 int flags, size_t want, size_t seek)
 543 {
 544         struct kvec kvec = {
 545                 .iov_base = &transport->recv.fraghdr,
 546                 .iov_len = want,
 547         };
 548         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 549 }
 550
 551 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 552 static ssize_t
 553 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 554 {
 555         struct rpc_xprt *xprt = &transport->xprt;
 556         struct rpc_rqst *req;
 557         ssize_t ret;
 558
 559         /* Look up and lock the request corresponding to the given XID */
 560         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 561         if (!req) {
 562                 printk(KERN_WARNING "Callback slot table overflowed\n");
 563                 return -ESHUTDOWN;
 564         }
 565
 566         ret = xs_read_stream_request(transport, msg, flags, req);
 567         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 568                 xprt_complete_bc_request(req, transport->recv.copied);
 569
 570         return ret;
 571 }
 572 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 573 static ssize_t
 574 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 575 {
 576         return -ESHUTDOWN;
 577 }
 578 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 579
 580 static ssize_t
 581 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 582 {
 583         struct rpc_xprt *xprt = &transport->xprt;
 584         struct rpc_rqst *req;
 585         ssize_t ret = 0;
 586
 587         /* Look up and lock the request corresponding to the given XID */
 588         spin_lock(&xprt->queue_lock);
 589         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 590         if (!req) {
 591                 msg->msg_flags |= MSG_TRUNC;
 592                 goto out;
 593         }
 594         xprt_pin_rqst(req);
 595         spin_unlock(&xprt->queue_lock);
 596
 597         ret = xs_read_stream_request(transport, msg, flags, req);
 598
 599         spin_lock(&xprt->queue_lock);
 600         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 601                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 602         xprt_unpin_rqst(req);
 603 out:
 604         spin_unlock(&xprt->queue_lock);
 605         return ret;
 606 }
 607
 608 static ssize_t
 609 xs_read_stream(struct sock_xprt *transport, int flags)
 610 {
 611         struct msghdr msg = { 0 };
 612         size_t want, read = 0;
 613         ssize_t ret = 0;
 614
 615         if (transport->recv.len == 0) {
 616                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 617                 ret = xs_read_stream_header(transport, &msg, flags, want,
 618                                 transport->recv.offset);
 619                 if (ret <= 0)
 620                         goto out_err;
 621                 transport->recv.offset = ret;
 622                 if (transport->recv.offset != want)
 623                         return transport->recv.offset;
 624                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 625                         RPC_FRAGMENT_SIZE_MASK;
 626                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 627                 read = ret;
 628         }
 629
 630         switch (be32_to_cpu(transport->recv.calldir)) {
 631         default:
 632                 msg.msg_flags |= MSG_TRUNC;
 633                 break;
 634         case RPC_CALL:
 635                 ret = xs_read_stream_call(transport, &msg, flags);
 636                 break;
 637         case RPC_REPLY:
 638                 ret = xs_read_stream_reply(transport, &msg, flags);
 639         }
 640         if (msg.msg_flags & MSG_TRUNC) {
 641                 transport->recv.calldir = cpu_to_be32(-1);
 642                 transport->recv.copied = -1;
 643         }
 644         if (ret < 0)
 645                 goto out_err;
 646         read += ret;
 647         if (transport->recv.offset < transport->recv.len) {
 648                 if (!(msg.msg_flags & MSG_TRUNC))
 649                         return read;
 650                 msg.msg_flags = 0;
 651                 ret = xs_read_discard(transport->sock, &msg, flags,
 652                                 transport->recv.len - transport->recv.offset);
 653                 if (ret <= 0)
 654                         goto out_err;
 655                 transport->recv.offset += ret;
 656                 read += ret;
 657                 if (transport->recv.offset != transport->recv.len)
 658                         return read;
 659         }
 660         if (xs_read_stream_request_done(transport)) {
 661                 trace_xs_stream_read_request(transport);
 662                 transport->recv.copied = 0;
 663         }
 664         transport->recv.offset = 0;
 665         transport->recv.len = 0;
 666         return read;
 667 out_err:
 668         return ret != 0 ? ret : -ESHUTDOWN;
 669 }
 670
 671 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 672 {
 673         return transport->sock->ops->poll(transport->file, transport->sock,
 674                         NULL);
 675 }
 676
 677 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 678 {
 679         __poll_t events = xs_poll_socket(transport);
 680
 681         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 682 }
 683
 684 static void xs_poll_check_readable(struct sock_xprt *transport)
 685 {
 686
 687         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 688         if (!xs_poll_socket_readable(transport))
 689                 return;
 690         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 691                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 692 }
 693
 694 static void xs_stream_data_receive(struct sock_xprt *transport)
 695 {
 696         size_t read = 0;
 697         ssize_t ret = 0;
 698
 699         mutex_lock(&transport->recv_mutex);
 700         if (transport->sock == NULL)
 701                 goto out;
 702         for (;;) {
 703                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 704                 if (ret < 0)
 705                         break;
 706                 read += ret;
 707                 cond_resched();
 708         }
 709         if (ret == -ESHUTDOWN)
 710                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 711         else
 712                 xs_poll_check_readable(transport);
 713 out:
 714         mutex_unlock(&transport->recv_mutex);
 715         trace_xs_stream_read_data(&transport->xprt, ret, read);
 716 }
 717
 718 static void xs_stream_data_receive_workfn(struct work_struct *work)
 719 {
 720         struct sock_xprt *transport =
 721                 container_of(work, struct sock_xprt, recv_worker);
 722         unsigned int pflags = memalloc_nofs_save();
 723
 724         xs_stream_data_receive(transport);
 725         memalloc_nofs_restore(pflags);
 726 }
 727
 728 static void
 729 xs_stream_reset_connect(struct sock_xprt *transport)
 730 {
 731         transport->recv.offset = 0;
 732         transport->recv.len = 0;
 733         transport->recv.copied = 0;
 734         transport->xmit.offset = 0;
 735 }
 736
 737 static void
 738 xs_stream_start_connect(struct sock_xprt *transport)
 739 {
 740         transport->xprt.stat.connect_count++;
 741         transport->xprt.stat.connect_start = jiffies;
 742 }
 743
 744 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 745
 746 static int xs_sendmsg(struct socket *sock, struct msghdr *msg, size_t seek)
 747 {
 748         if (seek)
 749                 iov_iter_advance(&msg->msg_iter, seek);
 750         return sock_sendmsg(sock, msg);
 751 }
 752
 753 static int xs_send_kvec(struct socket *sock, struct msghdr *msg, struct kvec *vec, size_t seek)
 754 {
 755         iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
 756         return xs_sendmsg(sock, msg, seek);
 757 }
 758
 759 static int xs_send_pagedata(struct socket *sock, struct msghdr *msg, struct xdr_buf *xdr, size_t base)
 760 {
 761         int err;
 762
 763         err = xdr_alloc_bvec(xdr, GFP_KERNEL);
 764         if (err < 0)
 765                 return err;
 766
 767         iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec,
 768                         xdr_buf_pagecount(xdr),
 769                         xdr->page_len + xdr->page_base);
 770         return xs_sendmsg(sock, msg, base + xdr->page_base);
 771 }
 772
 773 #define xs_record_marker_len() sizeof(rpc_fraghdr)
 774
 775 /* Common case:
 776  *  - stream transport
 777  *  - sending from byte 0 of the message
 778  *  - the message is wholly contained in @xdr's head iovec
 779  */
 780 static int xs_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
 781                 rpc_fraghdr marker, struct kvec *vec, size_t base)
 782 {
 783         struct kvec iov[2] = {
 784                 [0] = {
 785                         .iov_base       = &marker,
 786                         .iov_len        = sizeof(marker)
 787                 },
 788                 [1] = *vec,
 789         };
 790         size_t len = iov[0].iov_len + iov[1].iov_len;
 791
 792         iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
 793         return xs_sendmsg(sock, msg, base);
 794 }
 795
 796 /**
 797  * xs_sendpages - write pages directly to a socket
 798  * @sock: socket to send on
 799  * @addr: UDP only -- address of destination
 800  * @addrlen: UDP only -- length of destination address
 801  * @xdr: buffer containing this request
 802  * @base: starting position in the buffer
 803  * @rm: stream record marker field
 804  * @sent_p: return the total number of bytes successfully queued for sending
 805  *
 806  */
 807 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, rpc_fraghdr rm, int *sent_p)
 808 {
 809         struct msghdr msg = {
 810                 .msg_name = addr,
 811                 .msg_namelen = addrlen,
 812                 .msg_flags = XS_SENDMSG_FLAGS | MSG_MORE,
 813         };
 814         unsigned int rmsize = rm ? sizeof(rm) : 0;
 815         unsigned int remainder = rmsize + xdr->len - base;
 816         unsigned int want;
 817         int err = 0;
 818
 819         if (unlikely(!sock))
 820                 return -ENOTSOCK;
 821
 822         want = xdr->head[0].iov_len + rmsize;
 823         if (base < want) {
 824                 unsigned int len = want - base;
 825                 remainder -= len;
 826                 if (remainder == 0)
 827                         msg.msg_flags &= ~MSG_MORE;
 828                 if (rmsize)
 829                         err = xs_send_rm_and_kvec(sock, &msg, rm,
 830                                         &xdr->head[0], base);
 831                 else
 832                         err = xs_send_kvec(sock, &msg, &xdr->head[0], base);
 833                 if (remainder == 0 || err != len)
 834                         goto out;
 835                 *sent_p += err;
 836                 base = 0;
 837         } else
 838                 base -= want;
 839
 840         if (base < xdr->page_len) {
 841                 unsigned int len = xdr->page_len - base;
 842                 remainder -= len;
 843                 if (remainder == 0)
 844                         msg.msg_flags &= ~MSG_MORE;
 845                 err = xs_send_pagedata(sock, &msg, xdr, base);
 846                 if (remainder == 0 || err != len)
 847                         goto out;
 848                 *sent_p += err;
 849                 base = 0;
 850         } else
 851                 base -= xdr->page_len;
 852
 853         if (base >= xdr->tail[0].iov_len)
 854                 return 0;
 855         msg.msg_flags &= ~MSG_MORE;
 856         err = xs_send_kvec(sock, &msg, &xdr->tail[0], base);
 857 out:
 858         if (err > 0) {
 859                 *sent_p += err;
 860                 err = 0;
 861         }
 862         return err;
 863 }
 864
 865 /**
 866  * xs_nospace - handle transmit was incomplete
 867  * @req: pointer to RPC request
 868  *
 869  */
 870 static int xs_nospace(struct rpc_rqst *req)
 871 {
 872         struct rpc_xprt *xprt = req->rq_xprt;
 873         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 874         struct sock *sk = transport->inet;
 875         int ret = -EAGAIN;
 876
 877         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 878                         req->rq_task->tk_pid,
 879                         req->rq_slen - transport->xmit.offset,
 880                         req->rq_slen);
 881
 882         /* Protect against races with write_space */
 883         spin_lock(&xprt->transport_lock);
 884
 885         /* Don't race with disconnect */
 886         if (xprt_connected(xprt)) {
 887                 /* wait for more buffer space */
 888                 sk->sk_write_pending++;
 889                 xprt_wait_for_buffer_space(xprt);
 890         } else
 891                 ret = -ENOTCONN;
 892
 893         spin_unlock(&xprt->transport_lock);
 894
 895         /* Race breaker in case memory is freed before above code is called */
 896         if (ret == -EAGAIN) {
 897                 struct socket_wq *wq;
 898
 899                 rcu_read_lock();
 900                 wq = rcu_dereference(sk->sk_wq);
 901                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 902                 rcu_read_unlock();
 903
 904                 sk->sk_write_space(sk);
 905         }
 906         return ret;
 907 }
 908
 909 static void
 910 xs_stream_prepare_request(struct rpc_rqst *req)
 911 {
 912         req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
 913 }
 914
 915 /*
 916  * Determine if the previous message in the stream was aborted before it
 917  * could complete transmission.
 918  */
 919 static bool
 920 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 921 {
 922         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 923 }
 924
 925 /*
 926  * Return the stream record marker field for a record of length < 2^31-1
 927  */
 928 static rpc_fraghdr
 929 xs_stream_record_marker(struct xdr_buf *xdr)
 930 {
 931         if (!xdr->len)
 932                 return 0;
 933         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 934 }
 935
 936 /**
 937  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 938  * @req: pointer to RPC request
 939  *
 940  * Return values:
 941  *        0:    The request has been sent
 942  *   EAGAIN:    The socket was blocked, please call again later to
 943  *              complete the request
 944  * ENOTCONN:    Caller needs to invoke connect logic then call again
 945  *    other:    Some other error occured, the request was not sent
 946  */
 947 static int xs_local_send_request(struct rpc_rqst *req)
 948 {
 949         struct rpc_xprt *xprt = req->rq_xprt;
 950         struct sock_xprt *transport =
 951                                 container_of(xprt, struct sock_xprt, xprt);
 952         struct xdr_buf *xdr = &req->rq_snd_buf;
 953         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 954         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 955         int status;
 956         int sent = 0;
 957
 958         /* Close the stream if the previous transmission was incomplete */
 959         if (xs_send_request_was_aborted(transport, req)) {
 960                 xs_close(xprt);
 961                 return -ENOTCONN;
 962         }
 963
 964         xs_pktdump("packet data:",
 965                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 966
 967         req->rq_xtime = ktime_get();
 968         status = xs_sendpages(transport->sock, NULL, 0, xdr,
 969                               transport->xmit.offset, rm, &sent);
 970         dprintk("RPC:       %s(%u) = %d\n",
 971                         __func__, xdr->len - transport->xmit.offset, status);
 972
 973         if (status == -EAGAIN && sock_writeable(transport->inet))
 974                 status = -ENOBUFS;
 975
 976         if (likely(sent > 0) || status == 0) {
 977                 transport->xmit.offset += sent;
 978                 req->rq_bytes_sent = transport->xmit.offset;
 979                 if (likely(req->rq_bytes_sent >= msglen)) {
 980                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 981                         transport->xmit.offset = 0;
 982                         return 0;
 983                 }
 984                 status = -EAGAIN;
 985         }
 986
 987         switch (status) {
 988         case -ENOBUFS:
 989                 break;
 990         case -EAGAIN:
 991                 status = xs_nospace(req);
 992                 break;
 993         default:
 994                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 995                         -status);
 996                 /* fall through */
 997         case -EPIPE:
 998                 xs_close(xprt);
 999                 status = -ENOTCONN;
1000         }
1001
1002         return status;
1003 }
1004
1005 /**
1006  * xs_udp_send_request - write an RPC request to a UDP socket
1007  * @req: pointer to RPC request
1008  *
1009  * Return values:
1010  *        0:    The request has been sent
1011  *   EAGAIN:    The socket was blocked, please call again later to
1012  *              complete the request
1013  * ENOTCONN:    Caller needs to invoke connect logic then call again
1014  *    other:    Some other error occurred, the request was not sent
1015  */
1016 static int xs_udp_send_request(struct rpc_rqst *req)
1017 {
1018         struct rpc_xprt *xprt = req->rq_xprt;
1019         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1020         struct xdr_buf *xdr = &req->rq_snd_buf;
1021         int sent = 0;
1022         int status;
1023
1024         xs_pktdump("packet data:",
1025                                 req->rq_svec->iov_base,
1026                                 req->rq_svec->iov_len);
1027
1028         if (!xprt_bound(xprt))
1029                 return -ENOTCONN;
1030
1031         if (!xprt_request_get_cong(xprt, req))
1032                 return -EBADSLT;
1033
1034         req->rq_xtime = ktime_get();
1035         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
1036                               xdr, 0, 0, &sent);
1037
1038         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
1039                         xdr->len, status);
1040
1041         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1042         if (status == -EPERM)
1043                 goto process_status;
1044
1045         if (status == -EAGAIN && sock_writeable(transport->inet))
1046                 status = -ENOBUFS;
1047
1048         if (sent > 0 || status == 0) {
1049                 req->rq_xmit_bytes_sent += sent;
1050                 if (sent >= req->rq_slen)
1051                         return 0;
1052                 /* Still some bytes left; set up for a retry later. */
1053                 status = -EAGAIN;
1054         }
1055
1056 process_status:
1057         switch (status) {
1058         case -ENOTSOCK:
1059                 status = -ENOTCONN;
1060                 /* Should we call xs_close() here? */
1061                 break;
1062         case -EAGAIN:
1063                 status = xs_nospace(req);
1064                 break;
1065         case -ENETUNREACH:
1066         case -ENOBUFS:
1067         case -EPIPE:
1068         case -ECONNREFUSED:
1069         case -EPERM:
1070                 /* When the server has died, an ICMP port unreachable message
1071                  * prompts ECONNREFUSED. */
1072                 break;
1073         default:
1074                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1075                         -status);
1076         }
1077
1078         return status;
1079 }
1080
1081 /**
1082  * xs_tcp_send_request - write an RPC request to a TCP socket
1083  * @req: pointer to RPC request
1084  *
1085  * Return values:
1086  *        0:    The request has been sent
1087  *   EAGAIN:    The socket was blocked, please call again later to
1088  *              complete the request
1089  * ENOTCONN:    Caller needs to invoke connect logic then call again
1090  *    other:    Some other error occurred, the request was not sent
1091  *
1092  * XXX: In the case of soft timeouts, should we eventually give up
1093  *      if sendmsg is not able to make progress?
1094  */
1095 static int xs_tcp_send_request(struct rpc_rqst *req)
1096 {
1097         struct rpc_xprt *xprt = req->rq_xprt;
1098         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1099         struct xdr_buf *xdr = &req->rq_snd_buf;
1100         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1101         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1102         bool vm_wait = false;
1103         int status;
1104         int sent;
1105
1106         /* Close the stream if the previous transmission was incomplete */
1107         if (xs_send_request_was_aborted(transport, req)) {
1108                 if (transport->sock != NULL)
1109                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1110                 return -ENOTCONN;
1111         }
1112
1113         xs_pktdump("packet data:",
1114                                 req->rq_svec->iov_base,
1115                                 req->rq_svec->iov_len);
1116
1117         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1118                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1119
1120         /* Continue transmitting the packet/record. We must be careful
1121          * to cope with writespace callbacks arriving _after_ we have
1122          * called sendmsg(). */
1123         req->rq_xtime = ktime_get();
1124         while (1) {
1125                 sent = 0;
1126                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1127                                       transport->xmit.offset, rm, &sent);
1128
1129                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1130                                 xdr->len - transport->xmit.offset, status);
1131
1132                 /* If we've sent the entire packet, immediately
1133                  * reset the count of bytes sent. */
1134                 transport->xmit.offset += sent;
1135                 req->rq_bytes_sent = transport->xmit.offset;
1136                 if (likely(req->rq_bytes_sent >= msglen)) {
1137                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1138                         transport->xmit.offset = 0;
1139                         return 0;
1140                 }
1141
1142                 WARN_ON_ONCE(sent == 0 && status == 0);
1143
1144                 if (status == -EAGAIN ) {
1145                         /*
1146                          * Return EAGAIN if we're sure we're hitting the
1147                          * socket send buffer limits.
1148                          */
1149                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1150                                 break;
1151                         /*
1152                          * Did we hit a memory allocation failure?
1153                          */
1154                         if (sent == 0) {
1155                                 status = -ENOBUFS;
1156                                 if (vm_wait)
1157                                         break;
1158                                 /* Retry, knowing now that we're below the
1159                                  * socket send buffer limit
1160                                  */
1161                                 vm_wait = true;
1162                         }
1163                         continue;
1164                 }
1165                 if (status < 0)
1166                         break;
1167                 vm_wait = false;
1168         }
1169
1170         switch (status) {
1171         case -ENOTSOCK:
1172                 status = -ENOTCONN;
1173                 /* Should we call xs_close() here? */
1174                 break;
1175         case -EAGAIN:
1176                 status = xs_nospace(req);
1177                 break;
1178         case -ECONNRESET:
1179         case -ECONNREFUSED:
1180         case -ENOTCONN:
1181         case -EADDRINUSE:
1182         case -ENOBUFS:
1183         case -EPIPE:
1184                 break;
1185         default:
1186                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1187                         -status);
1188         }
1189
1190         return status;
1191 }
1192
1193 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1194 {
1195         transport->old_data_ready = sk->sk_data_ready;
1196         transport->old_state_change = sk->sk_state_change;
1197         transport->old_write_space = sk->sk_write_space;
1198         transport->old_error_report = sk->sk_error_report;
1199 }
1200
1201 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1202 {
1203         sk->sk_data_ready = transport->old_data_ready;
1204         sk->sk_state_change = transport->old_state_change;
1205         sk->sk_write_space = transport->old_write_space;
1206         sk->sk_error_report = transport->old_error_report;
1207 }
1208
1209 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1210 {
1211         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1212
1213         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1214         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1215         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1216         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1217 }
1218
1219 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1220 {
1221         set_bit(nr, &transport->sock_state);
1222         queue_work(xprtiod_workqueue, &transport->error_worker);
1223 }
1224
1225 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1226 {
1227         smp_mb__before_atomic();
1228         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1229         clear_bit(XPRT_CLOSING, &xprt->state);
1230         xs_sock_reset_state_flags(xprt);
1231         smp_mb__after_atomic();
1232 }
1233
1234 /**
1235  * xs_error_report - callback to handle TCP socket state errors
1236  * @sk: socket
1237  *
1238  * Note: we don't call sock_error() since there may be a rpc_task
1239  * using the socket, and so we don't want to clear sk->sk_err.
1240  */
1241 static void xs_error_report(struct sock *sk)
1242 {
1243         struct sock_xprt *transport;
1244         struct rpc_xprt *xprt;
1245         int err;
1246
1247         read_lock_bh(&sk->sk_callback_lock);
1248         if (!(xprt = xprt_from_sock(sk)))
1249                 goto out;
1250
1251         transport = container_of(xprt, struct sock_xprt, xprt);
1252         err = -sk->sk_err;
1253         if (err == 0)
1254                 goto out;
1255         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1256                         xprt, -err);
1257         trace_rpc_socket_error(xprt, sk->sk_socket, err);
1258         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1259  out:
1260         read_unlock_bh(&sk->sk_callback_lock);
1261 }
1262
1263 static void xs_reset_transport(struct sock_xprt *transport)
1264 {
1265         struct socket *sock = transport->sock;
1266         struct sock *sk = transport->inet;
1267         struct rpc_xprt *xprt = &transport->xprt;
1268         struct file *filp = transport->file;
1269
1270         if (sk == NULL)
1271                 return;
1272
1273         if (atomic_read(&transport->xprt.swapper))
1274                 sk_clear_memalloc(sk);
1275
1276         kernel_sock_shutdown(sock, SHUT_RDWR);
1277
1278         mutex_lock(&transport->recv_mutex);
1279         write_lock_bh(&sk->sk_callback_lock);
1280         transport->inet = NULL;
1281         transport->sock = NULL;
1282         transport->file = NULL;
1283
1284         sk->sk_user_data = NULL;
1285
1286         xs_restore_old_callbacks(transport, sk);
1287         xprt_clear_connected(xprt);
1288         write_unlock_bh(&sk->sk_callback_lock);
1289         xs_sock_reset_connection_flags(xprt);
1290         /* Reset stream record info */
1291         xs_stream_reset_connect(transport);
1292         mutex_unlock(&transport->recv_mutex);
1293
1294         trace_rpc_socket_close(xprt, sock);
1295         fput(filp);
1296
1297         xprt_disconnect_done(xprt);
1298 }
1299
1300 /**
1301  * xs_close - close a socket
1302  * @xprt: transport
1303  *
1304  * This is used when all requests are complete; ie, no DRC state remains
1305  * on the server we want to save.
1306  *
1307  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1308  * xs_reset_transport() zeroing the socket from underneath a writer.
1309  */
1310 static void xs_close(struct rpc_xprt *xprt)
1311 {
1312         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1313
1314         dprintk("RPC:       xs_close xprt %p\n", xprt);
1315
1316         xs_reset_transport(transport);
1317         xprt->reestablish_timeout = 0;
1318 }
1319
1320 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1321 {
1322         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1323                 xprt);
1324         xprt_disconnect_done(xprt);
1325 }
1326
1327 static void xs_xprt_free(struct rpc_xprt *xprt)
1328 {
1329         xs_free_peer_addresses(xprt);
1330         xprt_free(xprt);
1331 }
1332
1333 /**
1334  * xs_destroy - prepare to shutdown a transport
1335  * @xprt: doomed transport
1336  *
1337  */
1338 static void xs_destroy(struct rpc_xprt *xprt)
1339 {
1340         struct sock_xprt *transport = container_of(xprt,
1341                         struct sock_xprt, xprt);
1342         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1343
1344         cancel_delayed_work_sync(&transport->connect_worker);
1345         xs_close(xprt);
1346         cancel_work_sync(&transport->recv_worker);
1347         cancel_work_sync(&transport->error_worker);
1348         xs_xprt_free(xprt);
1349         module_put(THIS_MODULE);
1350 }
1351
1352 /**
1353  * xs_udp_data_read_skb - receive callback for UDP sockets
1354  * @xprt: transport
1355  * @sk: socket
1356  * @skb: skbuff
1357  *
1358  */
1359 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1360                 struct sock *sk,
1361                 struct sk_buff *skb)
1362 {
1363         struct rpc_task *task;
1364         struct rpc_rqst *rovr;
1365         int repsize, copied;
1366         u32 _xid;
1367         __be32 *xp;
1368
1369         repsize = skb->len;
1370         if (repsize < 4) {
1371                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1372                 return;
1373         }
1374
1375         /* Copy the XID from the skb... */
1376         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1377         if (xp == NULL)
1378                 return;
1379
1380         /* Look up and lock the request corresponding to the given XID */
1381         spin_lock(&xprt->queue_lock);
1382         rovr = xprt_lookup_rqst(xprt, *xp);
1383         if (!rovr)
1384                 goto out_unlock;
1385         xprt_pin_rqst(rovr);
1386         xprt_update_rtt(rovr->rq_task);
1387         spin_unlock(&xprt->queue_lock);
1388         task = rovr->rq_task;
1389
1390         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1391                 copied = repsize;
1392
1393         /* Suck it into the iovec, verify checksum if not done by hw. */
1394         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1395                 spin_lock(&xprt->queue_lock);
1396                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1397                 goto out_unpin;
1398         }
1399
1400
1401         spin_lock(&xprt->transport_lock);
1402         xprt_adjust_cwnd(xprt, task, copied);
1403         spin_unlock(&xprt->transport_lock);
1404         spin_lock(&xprt->queue_lock);
1405         xprt_complete_rqst(task, copied);
1406         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1407 out_unpin:
1408         xprt_unpin_rqst(rovr);
1409  out_unlock:
1410         spin_unlock(&xprt->queue_lock);
1411 }
1412
1413 static void xs_udp_data_receive(struct sock_xprt *transport)
1414 {
1415         struct sk_buff *skb;
1416         struct sock *sk;
1417         int err;
1418
1419         mutex_lock(&transport->recv_mutex);
1420         sk = transport->inet;
1421         if (sk == NULL)
1422                 goto out;
1423         for (;;) {
1424                 skb = skb_recv_udp(sk, 0, 1, &err);
1425                 if (skb == NULL)
1426                         break;
1427                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1428                 consume_skb(skb);
1429                 cond_resched();
1430         }
1431         xs_poll_check_readable(transport);
1432 out:
1433         mutex_unlock(&transport->recv_mutex);
1434 }
1435
1436 static void xs_udp_data_receive_workfn(struct work_struct *work)
1437 {
1438         struct sock_xprt *transport =
1439                 container_of(work, struct sock_xprt, recv_worker);
1440         unsigned int pflags = memalloc_nofs_save();
1441
1442         xs_udp_data_receive(transport);
1443         memalloc_nofs_restore(pflags);
1444 }
1445
1446 /**
1447  * xs_data_ready - "data ready" callback for UDP sockets
1448  * @sk: socket with data to read
1449  *
1450  */
1451 static void xs_data_ready(struct sock *sk)
1452 {
1453         struct rpc_xprt *xprt;
1454
1455         read_lock_bh(&sk->sk_callback_lock);
1456         dprintk("RPC:       xs_data_ready...\n");
1457         xprt = xprt_from_sock(sk);
1458         if (xprt != NULL) {
1459                 struct sock_xprt *transport = container_of(xprt,
1460                                 struct sock_xprt, xprt);
1461                 transport->old_data_ready(sk);
1462                 /* Any data means we had a useful conversation, so
1463                  * then we don't need to delay the next reconnect
1464                  */
1465                 if (xprt->reestablish_timeout)
1466                         xprt->reestablish_timeout = 0;
1467                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1468                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1469         }
1470         read_unlock_bh(&sk->sk_callback_lock);
1471 }
1472
1473 /*
1474  * Helper function to force a TCP close if the server is sending
1475  * junk and/or it has put us in CLOSE_WAIT
1476  */
1477 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1478 {
1479         xprt_force_disconnect(xprt);
1480 }
1481
1482 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1483 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1484 {
1485         return PAGE_SIZE;
1486 }
1487 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1488
1489 /**
1490  * xs_tcp_state_change - callback to handle TCP socket state changes
1491  * @sk: socket whose state has changed
1492  *
1493  */
1494 static void xs_tcp_state_change(struct sock *sk)
1495 {
1496         struct rpc_xprt *xprt;
1497         struct sock_xprt *transport;
1498
1499         read_lock_bh(&sk->sk_callback_lock);
1500         if (!(xprt = xprt_from_sock(sk)))
1501                 goto out;
1502         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1503         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1504                         sk->sk_state, xprt_connected(xprt),
1505                         sock_flag(sk, SOCK_DEAD),
1506                         sock_flag(sk, SOCK_ZAPPED),
1507                         sk->sk_shutdown);
1508
1509         transport = container_of(xprt, struct sock_xprt, xprt);
1510         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1511         switch (sk->sk_state) {
1512         case TCP_ESTABLISHED:
1513                 if (!xprt_test_and_set_connected(xprt)) {
1514                         xprt->connect_cookie++;
1515                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1516                         xprt_clear_connecting(xprt);
1517
1518                         xprt->stat.connect_count++;
1519                         xprt->stat.connect_time += (long)jiffies -
1520                                                    xprt->stat.connect_start;
1521                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1522                 }
1523                 break;
1524         case TCP_FIN_WAIT1:
1525                 /* The client initiated a shutdown of the socket */
1526                 xprt->connect_cookie++;
1527                 xprt->reestablish_timeout = 0;
1528                 set_bit(XPRT_CLOSING, &xprt->state);
1529                 smp_mb__before_atomic();
1530                 clear_bit(XPRT_CONNECTED, &xprt->state);
1531                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1532                 smp_mb__after_atomic();
1533                 break;
1534         case TCP_CLOSE_WAIT:
1535                 /* The server initiated a shutdown of the socket */
1536                 xprt->connect_cookie++;
1537                 clear_bit(XPRT_CONNECTED, &xprt->state);
1538                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1539                 /* fall through */
1540         case TCP_CLOSING:
1541                 /*
1542                  * If the server closed down the connection, make sure that
1543                  * we back off before reconnecting
1544                  */
1545                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1546                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1547                 break;
1548         case TCP_LAST_ACK:
1549                 set_bit(XPRT_CLOSING, &xprt->state);
1550                 smp_mb__before_atomic();
1551                 clear_bit(XPRT_CONNECTED, &xprt->state);
1552                 smp_mb__after_atomic();
1553                 break;
1554         case TCP_CLOSE:
1555                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1556                                         &transport->sock_state))
1557                         xprt_clear_connecting(xprt);
1558                 clear_bit(XPRT_CLOSING, &xprt->state);
1559                 /* Trigger the socket release */
1560                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1561         }
1562  out:
1563         read_unlock_bh(&sk->sk_callback_lock);
1564 }
1565
1566 static void xs_write_space(struct sock *sk)
1567 {
1568         struct socket_wq *wq;
1569         struct sock_xprt *transport;
1570         struct rpc_xprt *xprt;
1571
1572         if (!sk->sk_socket)
1573                 return;
1574         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1575
1576         if (unlikely(!(xprt = xprt_from_sock(sk))))
1577                 return;
1578         transport = container_of(xprt, struct sock_xprt, xprt);
1579         rcu_read_lock();
1580         wq = rcu_dereference(sk->sk_wq);
1581         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1582                 goto out;
1583
1584         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1585         sk->sk_write_pending--;
1586 out:
1587         rcu_read_unlock();
1588 }
1589
1590 /**
1591  * xs_udp_write_space - callback invoked when socket buffer space
1592  *                             becomes available
1593  * @sk: socket whose state has changed
1594  *
1595  * Called when more output buffer space is available for this socket.
1596  * We try not to wake our writers until they can make "significant"
1597  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1598  * with a bunch of small requests.
1599  */
1600 static void xs_udp_write_space(struct sock *sk)
1601 {
1602         read_lock_bh(&sk->sk_callback_lock);
1603
1604         /* from net/core/sock.c:sock_def_write_space */
1605         if (sock_writeable(sk))
1606                 xs_write_space(sk);
1607
1608         read_unlock_bh(&sk->sk_callback_lock);
1609 }
1610
1611 /**
1612  * xs_tcp_write_space - callback invoked when socket buffer space
1613  *                             becomes available
1614  * @sk: socket whose state has changed
1615  *
1616  * Called when more output buffer space is available for this socket.
1617  * We try not to wake our writers until they can make "significant"
1618  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1619  * with a bunch of small requests.
1620  */
1621 static void xs_tcp_write_space(struct sock *sk)
1622 {
1623         read_lock_bh(&sk->sk_callback_lock);
1624
1625         /* from net/core/stream.c:sk_stream_write_space */
1626         if (sk_stream_is_writeable(sk))
1627                 xs_write_space(sk);
1628
1629         read_unlock_bh(&sk->sk_callback_lock);
1630 }
1631
1632 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1633 {
1634         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1635         struct sock *sk = transport->inet;
1636
1637         if (transport->rcvsize) {
1638                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1639                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1640         }
1641         if (transport->sndsize) {
1642                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1643                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1644                 sk->sk_write_space(sk);
1645         }
1646 }
1647
1648 /**
1649  * xs_udp_set_buffer_size - set send and receive limits
1650  * @xprt: generic transport
1651  * @sndsize: requested size of send buffer, in bytes
1652  * @rcvsize: requested size of receive buffer, in bytes
1653  *
1654  * Set socket send and receive buffer size limits.
1655  */
1656 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1657 {
1658         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1659
1660         transport->sndsize = 0;
1661         if (sndsize)
1662                 transport->sndsize = sndsize + 1024;
1663         transport->rcvsize = 0;
1664         if (rcvsize)
1665                 transport->rcvsize = rcvsize + 1024;
1666
1667         xs_udp_do_set_buffer_size(xprt);
1668 }
1669
1670 /**
1671  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1672  * @xprt: controlling transport
1673  * @task: task that timed out
1674  *
1675  * Adjust the congestion window after a retransmit timeout has occurred.
1676  */
1677 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1678 {
1679         spin_lock(&xprt->transport_lock);
1680         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1681         spin_unlock(&xprt->transport_lock);
1682 }
1683
1684 static int xs_get_random_port(void)
1685 {
1686         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1687         unsigned short range;
1688         unsigned short rand;
1689
1690         if (max < min)
1691                 return -EADDRINUSE;
1692         range = max - min + 1;
1693         rand = (unsigned short) prandom_u32() % range;
1694         return rand + min;
1695 }
1696
1697 /**
1698  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1699  * @sock: socket
1700  *
1701  * Note that this function has to be called on all sockets that share the
1702  * same port, and it must be called before binding.
1703  */
1704 static void xs_sock_set_reuseport(struct socket *sock)
1705 {
1706         int opt = 1;
1707
1708         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1709                         (char *)&opt, sizeof(opt));
1710 }
1711
1712 static unsigned short xs_sock_getport(struct socket *sock)
1713 {
1714         struct sockaddr_storage buf;
1715         unsigned short port = 0;
1716
1717         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1718                 goto out;
1719         switch (buf.ss_family) {
1720         case AF_INET6:
1721                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1722                 break;
1723         case AF_INET:
1724                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1725         }
1726 out:
1727         return port;
1728 }
1729
1730 /**
1731  * xs_set_port - reset the port number in the remote endpoint address
1732  * @xprt: generic transport
1733  * @port: new port number
1734  *
1735  */
1736 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1737 {
1738         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1739
1740         rpc_set_port(xs_addr(xprt), port);
1741         xs_update_peer_port(xprt);
1742 }
1743
1744 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1745 {
1746         if (transport->srcport == 0)
1747                 transport->srcport = xs_sock_getport(sock);
1748 }
1749
1750 static int xs_get_srcport(struct sock_xprt *transport)
1751 {
1752         int port = transport->srcport;
1753
1754         if (port == 0 && transport->xprt.resvport)
1755                 port = xs_get_random_port();
1756         return port;
1757 }
1758
1759 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1760 {
1761         if (transport->srcport != 0)
1762                 transport->srcport = 0;
1763         if (!transport->xprt.resvport)
1764                 return 0;
1765         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1766                 return xprt_max_resvport;
1767         return --port;
1768 }
1769 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1770 {
1771         struct sockaddr_storage myaddr;
1772         int err, nloop = 0;
1773         int port = xs_get_srcport(transport);
1774         unsigned short last;
1775
1776         /*
1777          * If we are asking for any ephemeral port (i.e. port == 0 &&
1778          * transport->xprt.resvport == 0), don't bind.  Let the local
1779          * port selection happen implicitly when the socket is used
1780          * (for example at connect time).
1781          *
1782          * This ensures that we can continue to establish TCP
1783          * connections even when all local ephemeral ports are already
1784          * a part of some TCP connection.  This makes no difference
1785          * for UDP sockets, but also doens't harm them.
1786          *
1787          * If we're asking for any reserved port (i.e. port == 0 &&
1788          * transport->xprt.resvport == 1) xs_get_srcport above will
1789          * ensure that port is non-zero and we will bind as needed.
1790          */
1791         if (port <= 0)
1792                 return port;
1793
1794         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1795         do {
1796                 rpc_set_port((struct sockaddr *)&myaddr, port);
1797                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1798                                 transport->xprt.addrlen);
1799                 if (err == 0) {
1800                         transport->srcport = port;
1801                         break;
1802                 }
1803                 last = port;
1804                 port = xs_next_srcport(transport, port);
1805                 if (port > last)
1806                         nloop++;
1807         } while (err == -EADDRINUSE && nloop != 2);
1808
1809         if (myaddr.ss_family == AF_INET)
1810                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1811                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1812                                 port, err ? "failed" : "ok", err);
1813         else
1814                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1815                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1816                                 port, err ? "failed" : "ok", err);
1817         return err;
1818 }
1819
1820 /*
1821  * We don't support autobind on AF_LOCAL sockets
1822  */
1823 static void xs_local_rpcbind(struct rpc_task *task)
1824 {
1825         xprt_set_bound(task->tk_xprt);
1826 }
1827
1828 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1829 {
1830 }
1831
1832 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1833 static struct lock_class_key xs_key[2];
1834 static struct lock_class_key xs_slock_key[2];
1835
1836 static inline void xs_reclassify_socketu(struct socket *sock)
1837 {
1838         struct sock *sk = sock->sk;
1839
1840         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1841                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1842 }
1843
1844 static inline void xs_reclassify_socket4(struct socket *sock)
1845 {
1846         struct sock *sk = sock->sk;
1847
1848         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1849                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1850 }
1851
1852 static inline void xs_reclassify_socket6(struct socket *sock)
1853 {
1854         struct sock *sk = sock->sk;
1855
1856         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1857                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1858 }
1859
1860 static inline void xs_reclassify_socket(int family, struct socket *sock)
1861 {
1862         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1863                 return;
1864
1865         switch (family) {
1866         case AF_LOCAL:
1867                 xs_reclassify_socketu(sock);
1868                 break;
1869         case AF_INET:
1870                 xs_reclassify_socket4(sock);
1871                 break;
1872         case AF_INET6:
1873                 xs_reclassify_socket6(sock);
1874                 break;
1875         }
1876 }
1877 #else
1878 static inline void xs_reclassify_socket(int family, struct socket *sock)
1879 {
1880 }
1881 #endif
1882
1883 static void xs_dummy_setup_socket(struct work_struct *work)
1884 {
1885 }
1886
1887 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1888                 struct sock_xprt *transport, int family, int type,
1889                 int protocol, bool reuseport)
1890 {
1891         struct file *filp;
1892         struct socket *sock;
1893         int err;
1894
1895         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1896         if (err < 0) {
1897                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1898                                 protocol, -err);
1899                 goto out;
1900         }
1901         xs_reclassify_socket(family, sock);
1902
1903         if (reuseport)
1904                 xs_sock_set_reuseport(sock);
1905
1906         err = xs_bind(transport, sock);
1907         if (err) {
1908                 sock_release(sock);
1909                 goto out;
1910         }
1911
1912         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1913         if (IS_ERR(filp))
1914                 return ERR_CAST(filp);
1915         transport->file = filp;
1916
1917         return sock;
1918 out:
1919         return ERR_PTR(err);
1920 }
1921
1922 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1923                                       struct socket *sock)
1924 {
1925         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1926                                                                         xprt);
1927
1928         if (!transport->inet) {
1929                 struct sock *sk = sock->sk;
1930
1931                 write_lock_bh(&sk->sk_callback_lock);
1932
1933                 xs_save_old_callbacks(transport, sk);
1934
1935                 sk->sk_user_data = xprt;
1936                 sk->sk_data_ready = xs_data_ready;
1937                 sk->sk_write_space = xs_udp_write_space;
1938                 sock_set_flag(sk, SOCK_FASYNC);
1939                 sk->sk_error_report = xs_error_report;
1940
1941                 xprt_clear_connected(xprt);
1942
1943                 /* Reset to new socket */
1944                 transport->sock = sock;
1945                 transport->inet = sk;
1946
1947                 write_unlock_bh(&sk->sk_callback_lock);
1948         }
1949
1950         xs_stream_start_connect(transport);
1951
1952         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1953 }
1954
1955 /**
1956  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1957  * @transport: socket transport to connect
1958  */
1959 static int xs_local_setup_socket(struct sock_xprt *transport)
1960 {
1961         struct rpc_xprt *xprt = &transport->xprt;
1962         struct file *filp;
1963         struct socket *sock;
1964         int status = -EIO;
1965
1966         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1967                                         SOCK_STREAM, 0, &sock, 1);
1968         if (status < 0) {
1969                 dprintk("RPC:       can't create AF_LOCAL "
1970                         "transport socket (%d).\n", -status);
1971                 goto out;
1972         }
1973         xs_reclassify_socket(AF_LOCAL, sock);
1974
1975         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1976         if (IS_ERR(filp)) {
1977                 status = PTR_ERR(filp);
1978                 goto out;
1979         }
1980         transport->file = filp;
1981
1982         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1983                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1984
1985         status = xs_local_finish_connecting(xprt, sock);
1986         trace_rpc_socket_connect(xprt, sock, status);
1987         switch (status) {
1988         case 0:
1989                 dprintk("RPC:       xprt %p connected to %s\n",
1990                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1991                 xprt->stat.connect_count++;
1992                 xprt->stat.connect_time += (long)jiffies -
1993                                            xprt->stat.connect_start;
1994                 xprt_set_connected(xprt);
1995         case -ENOBUFS:
1996                 break;
1997         case -ENOENT:
1998                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1999                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2000                 break;
2001         case -ECONNREFUSED:
2002                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2003                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2004                 break;
2005         default:
2006                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2007                                 __func__, -status,
2008                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2009         }
2010
2011 out:
2012         xprt_clear_connecting(xprt);
2013         xprt_wake_pending_tasks(xprt, status);
2014         return status;
2015 }
2016
2017 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2018 {
2019         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2020         int ret;
2021
2022          if (RPC_IS_ASYNC(task)) {
2023                 /*
2024                  * We want the AF_LOCAL connect to be resolved in the
2025                  * filesystem namespace of the process making the rpc
2026                  * call.  Thus we connect synchronously.
2027                  *
2028                  * If we want to support asynchronous AF_LOCAL calls,
2029                  * we'll need to figure out how to pass a namespace to
2030                  * connect.
2031                  */
2032                 task->tk_rpc_status = -ENOTCONN;
2033                 rpc_exit(task, -ENOTCONN);
2034                 return;
2035         }
2036         ret = xs_local_setup_socket(transport);
2037         if (ret && !RPC_IS_SOFTCONN(task))
2038                 msleep_interruptible(15000);
2039 }
2040
2041 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2042 /*
2043  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2044  * know that we have exclusive access to the socket), to guard against
2045  * races with xs_reset_transport.
2046  */
2047 static void xs_set_memalloc(struct rpc_xprt *xprt)
2048 {
2049         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2050                         xprt);
2051
2052         /*
2053          * If there's no sock, then we have nothing to set. The
2054          * reconnecting process will get it for us.
2055          */
2056         if (!transport->inet)
2057                 return;
2058         if (atomic_read(&xprt->swapper))
2059                 sk_set_memalloc(transport->inet);
2060 }
2061
2062 /**
2063  * xs_enable_swap - Tag this transport as being used for swap.
2064  * @xprt: transport to tag
2065  *
2066  * Take a reference to this transport on behalf of the rpc_clnt, and
2067  * optionally mark it for swapping if it wasn't already.
2068  */
2069 static int
2070 xs_enable_swap(struct rpc_xprt *xprt)
2071 {
2072         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2073
2074         if (atomic_inc_return(&xprt->swapper) != 1)
2075                 return 0;
2076         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2077                 return -ERESTARTSYS;
2078         if (xs->inet)
2079                 sk_set_memalloc(xs->inet);
2080         xprt_release_xprt(xprt, NULL);
2081         return 0;
2082 }
2083
2084 /**
2085  * xs_disable_swap - Untag this transport as being used for swap.
2086  * @xprt: transport to tag
2087  *
2088  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2089  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2090  */
2091 static void
2092 xs_disable_swap(struct rpc_xprt *xprt)
2093 {
2094         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2095
2096         if (!atomic_dec_and_test(&xprt->swapper))
2097                 return;
2098         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2099                 return;
2100         if (xs->inet)
2101                 sk_clear_memalloc(xs->inet);
2102         xprt_release_xprt(xprt, NULL);
2103 }
2104 #else
2105 static void xs_set_memalloc(struct rpc_xprt *xprt)
2106 {
2107 }
2108
2109 static int
2110 xs_enable_swap(struct rpc_xprt *xprt)
2111 {
2112         return -EINVAL;
2113 }
2114
2115 static void
2116 xs_disable_swap(struct rpc_xprt *xprt)
2117 {
2118 }
2119 #endif
2120
2121 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2122 {
2123         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2124
2125         if (!transport->inet) {
2126                 struct sock *sk = sock->sk;
2127
2128                 write_lock_bh(&sk->sk_callback_lock);
2129
2130                 xs_save_old_callbacks(transport, sk);
2131
2132                 sk->sk_user_data = xprt;
2133                 sk->sk_data_ready = xs_data_ready;
2134                 sk->sk_write_space = xs_udp_write_space;
2135                 sock_set_flag(sk, SOCK_FASYNC);
2136
2137                 xprt_set_connected(xprt);
2138
2139                 /* Reset to new socket */
2140                 transport->sock = sock;
2141                 transport->inet = sk;
2142
2143                 xs_set_memalloc(xprt);
2144
2145                 write_unlock_bh(&sk->sk_callback_lock);
2146         }
2147         xs_udp_do_set_buffer_size(xprt);
2148
2149         xprt->stat.connect_start = jiffies;
2150 }
2151
2152 static void xs_udp_setup_socket(struct work_struct *work)
2153 {
2154         struct sock_xprt *transport =
2155                 container_of(work, struct sock_xprt, connect_worker.work);
2156         struct rpc_xprt *xprt = &transport->xprt;
2157         struct socket *sock;
2158         int status = -EIO;
2159
2160         sock = xs_create_sock(xprt, transport,
2161                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2162                         IPPROTO_UDP, false);
2163         if (IS_ERR(sock))
2164                 goto out;
2165
2166         dprintk("RPC:       worker connecting xprt %p via %s to "
2167                                 "%s (port %s)\n", xprt,
2168                         xprt->address_strings[RPC_DISPLAY_PROTO],
2169                         xprt->address_strings[RPC_DISPLAY_ADDR],
2170                         xprt->address_strings[RPC_DISPLAY_PORT]);
2171
2172         xs_udp_finish_connecting(xprt, sock);
2173         trace_rpc_socket_connect(xprt, sock, 0);
2174         status = 0;
2175 out:
2176         xprt_clear_connecting(xprt);
2177         xprt_unlock_connect(xprt, transport);
2178         xprt_wake_pending_tasks(xprt, status);
2179 }
2180
2181 /**
2182  * xs_tcp_shutdown - gracefully shut down a TCP socket
2183  * @xprt: transport
2184  *
2185  * Initiates a graceful shutdown of the TCP socket by calling the
2186  * equivalent of shutdown(SHUT_RDWR);
2187  */
2188 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2189 {
2190         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2191         struct socket *sock = transport->sock;
2192         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2193
2194         if (sock == NULL)
2195                 return;
2196         switch (skst) {
2197         default:
2198                 kernel_sock_shutdown(sock, SHUT_RDWR);
2199                 trace_rpc_socket_shutdown(xprt, sock);
2200                 break;
2201         case TCP_CLOSE:
2202         case TCP_TIME_WAIT:
2203                 xs_reset_transport(transport);
2204         }
2205 }
2206
2207 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2208                 struct socket *sock)
2209 {
2210         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2211         unsigned int keepidle;
2212         unsigned int keepcnt;
2213         unsigned int opt_on = 1;
2214         unsigned int timeo;
2215
2216         spin_lock(&xprt->transport_lock);
2217         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2218         keepcnt = xprt->timeout->to_retries + 1;
2219         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2220                 (xprt->timeout->to_retries + 1);
2221         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2222         spin_unlock(&xprt->transport_lock);
2223
2224         /* TCP Keepalive options */
2225         kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2226                         (char *)&opt_on, sizeof(opt_on));
2227         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2228                         (char *)&keepidle, sizeof(keepidle));
2229         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2230                         (char *)&keepidle, sizeof(keepidle));
2231         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2232                         (char *)&keepcnt, sizeof(keepcnt));
2233
2234         /* TCP user timeout (see RFC5482) */
2235         kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2236                         (char *)&timeo, sizeof(timeo));
2237 }
2238
2239 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2240                 unsigned long connect_timeout,
2241                 unsigned long reconnect_timeout)
2242 {
2243         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2244         struct rpc_timeout to;
2245         unsigned long initval;
2246
2247         spin_lock(&xprt->transport_lock);
2248         if (reconnect_timeout < xprt->max_reconnect_timeout)
2249                 xprt->max_reconnect_timeout = reconnect_timeout;
2250         if (connect_timeout < xprt->connect_timeout) {
2251                 memcpy(&to, xprt->timeout, sizeof(to));
2252                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2253                 /* Arbitrary lower limit */
2254                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2255                         initval = XS_TCP_INIT_REEST_TO << 1;
2256                 to.to_initval = initval;
2257                 to.to_maxval = initval;
2258                 memcpy(&transport->tcp_timeout, &to,
2259                                 sizeof(transport->tcp_timeout));
2260                 xprt->timeout = &transport->tcp_timeout;
2261                 xprt->connect_timeout = connect_timeout;
2262         }
2263         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2264         spin_unlock(&xprt->transport_lock);
2265 }
2266
2267 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2268 {
2269         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2270         int ret = -ENOTCONN;
2271
2272         if (!transport->inet) {
2273                 struct sock *sk = sock->sk;
2274                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2275
2276                 /* Avoid temporary address, they are bad for long-lived
2277                  * connections such as NFS mounts.
2278                  * RFC4941, section 3.6 suggests that:
2279                  *    Individual applications, which have specific
2280                  *    knowledge about the normal duration of connections,
2281                  *    MAY override this as appropriate.
2282                  */
2283                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2284                                 (char *)&addr_pref, sizeof(addr_pref));
2285
2286                 xs_tcp_set_socket_timeouts(xprt, sock);
2287
2288                 write_lock_bh(&sk->sk_callback_lock);
2289
2290                 xs_save_old_callbacks(transport, sk);
2291
2292                 sk->sk_user_data = xprt;
2293                 sk->sk_data_ready = xs_data_ready;
2294                 sk->sk_state_change = xs_tcp_state_change;
2295                 sk->sk_write_space = xs_tcp_write_space;
2296                 sock_set_flag(sk, SOCK_FASYNC);
2297                 sk->sk_error_report = xs_error_report;
2298
2299                 /* socket options */
2300                 sock_reset_flag(sk, SOCK_LINGER);
2301                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2302
2303                 xprt_clear_connected(xprt);
2304
2305                 /* Reset to new socket */
2306                 transport->sock = sock;
2307                 transport->inet = sk;
2308
2309                 write_unlock_bh(&sk->sk_callback_lock);
2310         }
2311
2312         if (!xprt_bound(xprt))
2313                 goto out;
2314
2315         xs_set_memalloc(xprt);
2316
2317         xs_stream_start_connect(transport);
2318
2319         /* Tell the socket layer to start connecting... */
2320         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2321         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2322         switch (ret) {
2323         case 0:
2324                 xs_set_srcport(transport, sock);
2325                 /* fall through */
2326         case -EINPROGRESS:
2327                 /* SYN_SENT! */
2328                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2329                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2330                 break;
2331         case -EADDRNOTAVAIL:
2332                 /* Source port number is unavailable. Try a new one! */
2333                 transport->srcport = 0;
2334         }
2335 out:
2336         return ret;
2337 }
2338
2339 /**
2340  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2341  * @work: queued work item
2342  *
2343  * Invoked by a work queue tasklet.
2344  */
2345 static void xs_tcp_setup_socket(struct work_struct *work)
2346 {
2347         struct sock_xprt *transport =
2348                 container_of(work, struct sock_xprt, connect_worker.work);
2349         struct socket *sock = transport->sock;
2350         struct rpc_xprt *xprt = &transport->xprt;
2351         int status = -EIO;
2352
2353         if (!sock) {
2354                 sock = xs_create_sock(xprt, transport,
2355                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2356                                 IPPROTO_TCP, true);
2357                 if (IS_ERR(sock)) {
2358                         status = PTR_ERR(sock);
2359                         goto out;
2360                 }
2361         }
2362
2363         dprintk("RPC:       worker connecting xprt %p via %s to "
2364                                 "%s (port %s)\n", xprt,
2365                         xprt->address_strings[RPC_DISPLAY_PROTO],
2366                         xprt->address_strings[RPC_DISPLAY_ADDR],
2367                         xprt->address_strings[RPC_DISPLAY_PORT]);
2368
2369         status = xs_tcp_finish_connecting(xprt, sock);
2370         trace_rpc_socket_connect(xprt, sock, status);
2371         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2372                         xprt, -status, xprt_connected(xprt),
2373                         sock->sk->sk_state);
2374         switch (status) {
2375         default:
2376                 printk("%s: connect returned unhandled error %d\n",
2377                         __func__, status);
2378                 /* fall through */
2379         case -EADDRNOTAVAIL:
2380                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2381                  * and retry
2382                  */
2383                 xs_tcp_force_close(xprt);
2384                 break;
2385         case 0:
2386         case -EINPROGRESS:
2387         case -EALREADY:
2388                 xprt_unlock_connect(xprt, transport);
2389                 return;
2390         case -EINVAL:
2391                 /* Happens, for instance, if the user specified a link
2392                  * local IPv6 address without a scope-id.
2393                  */
2394         case -ECONNREFUSED:
2395         case -ECONNRESET:
2396         case -ENETDOWN:
2397         case -ENETUNREACH:
2398         case -EHOSTUNREACH:
2399         case -EADDRINUSE:
2400         case -ENOBUFS:
2401                 /*
2402                  * xs_tcp_force_close() wakes tasks with -EIO.
2403                  * We need to wake them first to ensure the
2404                  * correct error code.
2405                  */
2406                 xprt_wake_pending_tasks(xprt, status);
2407                 xs_tcp_force_close(xprt);
2408                 goto out;
2409         }
2410         status = -EAGAIN;
2411 out:
2412         xprt_clear_connecting(xprt);
2413         xprt_unlock_connect(xprt, transport);
2414         xprt_wake_pending_tasks(xprt, status);
2415 }
2416
2417 /**
2418  * xs_connect - connect a socket to a remote endpoint
2419  * @xprt: pointer to transport structure
2420  * @task: address of RPC task that manages state of connect request
2421  *
2422  * TCP: If the remote end dropped the connection, delay reconnecting.
2423  *
2424  * UDP socket connects are synchronous, but we use a work queue anyway
2425  * to guarantee that even unprivileged user processes can set up a
2426  * socket on a privileged port.
2427  *
2428  * If a UDP socket connect fails, the delay behavior here prevents
2429  * retry floods (hard mounts).
2430  */
2431 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2432 {
2433         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2434         unsigned long delay = 0;
2435
2436         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2437
2438         if (transport->sock != NULL) {
2439                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2440                                 "seconds\n",
2441                                 xprt, xprt->reestablish_timeout / HZ);
2442
2443                 /* Start by resetting any existing state */
2444                 xs_reset_transport(transport);
2445
2446                 delay = xprt_reconnect_delay(xprt);
2447                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2448
2449         } else
2450                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2451
2452         queue_delayed_work(xprtiod_workqueue,
2453                         &transport->connect_worker,
2454                         delay);
2455 }
2456
2457 static void xs_wake_disconnect(struct sock_xprt *transport)
2458 {
2459         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2460                 xs_tcp_force_close(&transport->xprt);
2461 }
2462
2463 static void xs_wake_write(struct sock_xprt *transport)
2464 {
2465         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2466                 xprt_write_space(&transport->xprt);
2467 }
2468
2469 static void xs_wake_error(struct sock_xprt *transport)
2470 {
2471         int sockerr;
2472         int sockerr_len = sizeof(sockerr);
2473
2474         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2475                 return;
2476         mutex_lock(&transport->recv_mutex);
2477         if (transport->sock == NULL)
2478                 goto out;
2479         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2480                 goto out;
2481         if (kernel_getsockopt(transport->sock, SOL_SOCKET, SO_ERROR,
2482                                 (char *)&sockerr, &sockerr_len) != 0)
2483                 goto out;
2484         if (sockerr < 0)
2485                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2486 out:
2487         mutex_unlock(&transport->recv_mutex);
2488 }
2489
2490 static void xs_wake_pending(struct sock_xprt *transport)
2491 {
2492         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2493                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2494 }
2495
2496 static void xs_error_handle(struct work_struct *work)
2497 {
2498         struct sock_xprt *transport = container_of(work,
2499                         struct sock_xprt, error_worker);
2500
2501         xs_wake_disconnect(transport);
2502         xs_wake_write(transport);
2503         xs_wake_error(transport);
2504         xs_wake_pending(transport);
2505 }
2506
2507 /**
2508  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2509  * @xprt: rpc_xprt struct containing statistics
2510  * @seq: output file
2511  *
2512  */
2513 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2514 {
2515         long idle_time = 0;
2516
2517         if (xprt_connected(xprt))
2518                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2519
2520         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2521                         "%llu %llu %lu %llu %llu\n",
2522                         xprt->stat.bind_count,
2523                         xprt->stat.connect_count,
2524                         xprt->stat.connect_time / HZ,
2525                         idle_time,
2526                         xprt->stat.sends,
2527                         xprt->stat.recvs,
2528                         xprt->stat.bad_xids,
2529                         xprt->stat.req_u,
2530                         xprt->stat.bklog_u,
2531                         xprt->stat.max_slots,
2532                         xprt->stat.sending_u,
2533                         xprt->stat.pending_u);
2534 }
2535
2536 /**
2537  * xs_udp_print_stats - display UDP socket-specifc stats
2538  * @xprt: rpc_xprt struct containing statistics
2539  * @seq: output file
2540  *
2541  */
2542 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2543 {
2544         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2545
2546         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2547                         "%lu %llu %llu\n",
2548                         transport->srcport,
2549                         xprt->stat.bind_count,
2550                         xprt->stat.sends,
2551                         xprt->stat.recvs,
2552                         xprt->stat.bad_xids,
2553                         xprt->stat.req_u,
2554                         xprt->stat.bklog_u,
2555                         xprt->stat.max_slots,
2556                         xprt->stat.sending_u,
2557                         xprt->stat.pending_u);
2558 }
2559
2560 /**
2561  * xs_tcp_print_stats - display TCP socket-specifc stats
2562  * @xprt: rpc_xprt struct containing statistics
2563  * @seq: output file
2564  *
2565  */
2566 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2567 {
2568         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2569         long idle_time = 0;
2570
2571         if (xprt_connected(xprt))
2572                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2573
2574         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2575                         "%llu %llu %lu %llu %llu\n",
2576                         transport->srcport,
2577                         xprt->stat.bind_count,
2578                         xprt->stat.connect_count,
2579                         xprt->stat.connect_time / HZ,
2580                         idle_time,
2581                         xprt->stat.sends,
2582                         xprt->stat.recvs,
2583                         xprt->stat.bad_xids,
2584                         xprt->stat.req_u,
2585                         xprt->stat.bklog_u,
2586                         xprt->stat.max_slots,
2587                         xprt->stat.sending_u,
2588                         xprt->stat.pending_u);
2589 }
2590
2591 /*
2592  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2593  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2594  * to use the server side send routines.
2595  */
2596 static int bc_malloc(struct rpc_task *task)
2597 {
2598         struct rpc_rqst *rqst = task->tk_rqstp;
2599         size_t size = rqst->rq_callsize;
2600         struct page *page;
2601         struct rpc_buffer *buf;
2602
2603         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2604                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2605                           size);
2606                 return -EINVAL;
2607         }
2608
2609         page = alloc_page(GFP_KERNEL);
2610         if (!page)
2611                 return -ENOMEM;
2612
2613         buf = page_address(page);
2614         buf->len = PAGE_SIZE;
2615
2616         rqst->rq_buffer = buf->data;
2617         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2618         return 0;
2619 }
2620
2621 /*
2622  * Free the space allocated in the bc_alloc routine
2623  */
2624 static void bc_free(struct rpc_task *task)
2625 {
2626         void *buffer = task->tk_rqstp->rq_buffer;
2627         struct rpc_buffer *buf;
2628
2629         buf = container_of(buffer, struct rpc_buffer, data);
2630         free_page((unsigned long)buf);
2631 }
2632
2633 /*
2634  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2635  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2636  */
2637 static int bc_sendto(struct rpc_rqst *req)
2638 {
2639         int len;
2640         struct xdr_buf *xbufp = &req->rq_snd_buf;
2641         struct sock_xprt *transport =
2642                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2643         unsigned long headoff;
2644         unsigned long tailoff;
2645         struct page *tailpage;
2646         struct msghdr msg = {
2647                 .msg_flags      = MSG_MORE
2648         };
2649         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2650                                          (u32)xbufp->len);
2651         struct kvec iov = {
2652                 .iov_base       = &marker,
2653                 .iov_len        = sizeof(marker),
2654         };
2655
2656         len = kernel_sendmsg(transport->sock, &msg, &iov, 1, iov.iov_len);
2657         if (len != iov.iov_len)
2658                 return -EAGAIN;
2659
2660         tailpage = NULL;
2661         if (xbufp->tail[0].iov_len)
2662                 tailpage = virt_to_page(xbufp->tail[0].iov_base);
2663         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2664         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2665         len = svc_send_common(transport->sock, xbufp,
2666                               virt_to_page(xbufp->head[0].iov_base), headoff,
2667                               tailpage, tailoff);
2668         if (len != xbufp->len)
2669                 return -EAGAIN;
2670         return len;
2671 }
2672
2673 /*
2674  * The send routine. Borrows from svc_send
2675  */
2676 static int bc_send_request(struct rpc_rqst *req)
2677 {
2678         struct svc_xprt *xprt;
2679         int len;
2680
2681         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2682         /*
2683          * Get the server socket associated with this callback xprt
2684          */
2685         xprt = req->rq_xprt->bc_xprt;
2686
2687         /*
2688          * Grab the mutex to serialize data as the connection is shared
2689          * with the fore channel
2690          */
2691         mutex_lock(&xprt->xpt_mutex);
2692         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2693                 len = -ENOTCONN;
2694         else
2695                 len = bc_sendto(req);
2696         mutex_unlock(&xprt->xpt_mutex);
2697
2698         if (len > 0)
2699                 len = 0;
2700
2701         return len;
2702 }
2703
2704 /*
2705  * The close routine. Since this is client initiated, we do nothing
2706  */
2707
2708 static void bc_close(struct rpc_xprt *xprt)
2709 {
2710 }
2711
2712 /*
2713  * The xprt destroy routine. Again, because this connection is client
2714  * initiated, we do nothing
2715  */
2716
2717 static void bc_destroy(struct rpc_xprt *xprt)
2718 {
2719         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2720
2721         xs_xprt_free(xprt);
2722         module_put(THIS_MODULE);
2723 }
2724
2725 static const struct rpc_xprt_ops xs_local_ops = {
2726         .reserve_xprt           = xprt_reserve_xprt,
2727         .release_xprt           = xprt_release_xprt,
2728         .alloc_slot             = xprt_alloc_slot,
2729         .free_slot              = xprt_free_slot,
2730         .rpcbind                = xs_local_rpcbind,
2731         .set_port               = xs_local_set_port,
2732         .connect                = xs_local_connect,
2733         .buf_alloc              = rpc_malloc,
2734         .buf_free               = rpc_free,
2735         .prepare_request        = xs_stream_prepare_request,
2736         .send_request           = xs_local_send_request,
2737         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2738         .close                  = xs_close,
2739         .destroy                = xs_destroy,
2740         .print_stats            = xs_local_print_stats,
2741         .enable_swap            = xs_enable_swap,
2742         .disable_swap           = xs_disable_swap,
2743 };
2744
2745 static const struct rpc_xprt_ops xs_udp_ops = {
2746         .set_buffer_size        = xs_udp_set_buffer_size,
2747         .reserve_xprt           = xprt_reserve_xprt_cong,
2748         .release_xprt           = xprt_release_xprt_cong,
2749         .alloc_slot             = xprt_alloc_slot,
2750         .free_slot              = xprt_free_slot,
2751         .rpcbind                = rpcb_getport_async,
2752         .set_port               = xs_set_port,
2753         .connect                = xs_connect,
2754         .buf_alloc              = rpc_malloc,
2755         .buf_free               = rpc_free,
2756         .send_request           = xs_udp_send_request,
2757         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2758         .timer                  = xs_udp_timer,
2759         .release_request        = xprt_release_rqst_cong,
2760         .close                  = xs_close,
2761         .destroy                = xs_destroy,
2762         .print_stats            = xs_udp_print_stats,
2763         .enable_swap            = xs_enable_swap,
2764         .disable_swap           = xs_disable_swap,
2765         .inject_disconnect      = xs_inject_disconnect,
2766 };
2767
2768 static const struct rpc_xprt_ops xs_tcp_ops = {
2769         .reserve_xprt           = xprt_reserve_xprt,
2770         .release_xprt           = xprt_release_xprt,
2771         .alloc_slot             = xprt_alloc_slot,
2772         .free_slot              = xprt_free_slot,
2773         .rpcbind                = rpcb_getport_async,
2774         .set_port               = xs_set_port,
2775         .connect                = xs_connect,
2776         .buf_alloc              = rpc_malloc,
2777         .buf_free               = rpc_free,
2778         .prepare_request        = xs_stream_prepare_request,
2779         .send_request           = xs_tcp_send_request,
2780         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2781         .close                  = xs_tcp_shutdown,
2782         .destroy                = xs_destroy,
2783         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2784         .print_stats            = xs_tcp_print_stats,
2785         .enable_swap            = xs_enable_swap,
2786         .disable_swap           = xs_disable_swap,
2787         .inject_disconnect      = xs_inject_disconnect,
2788 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2789         .bc_setup               = xprt_setup_bc,
2790         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2791         .bc_free_rqst           = xprt_free_bc_rqst,
2792         .bc_destroy             = xprt_destroy_bc,
2793 #endif
2794 };
2795
2796 /*
2797  * The rpc_xprt_ops for the server backchannel
2798  */
2799
2800 static const struct rpc_xprt_ops bc_tcp_ops = {
2801         .reserve_xprt           = xprt_reserve_xprt,
2802         .release_xprt           = xprt_release_xprt,
2803         .alloc_slot             = xprt_alloc_slot,
2804         .free_slot              = xprt_free_slot,
2805         .buf_alloc              = bc_malloc,
2806         .buf_free               = bc_free,
2807         .send_request           = bc_send_request,
2808         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2809         .close                  = bc_close,
2810         .destroy                = bc_destroy,
2811         .print_stats            = xs_tcp_print_stats,
2812         .enable_swap            = xs_enable_swap,
2813         .disable_swap           = xs_disable_swap,
2814         .inject_disconnect      = xs_inject_disconnect,
2815 };
2816
2817 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2818 {
2819         static const struct sockaddr_in sin = {
2820                 .sin_family             = AF_INET,
2821                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2822         };
2823         static const struct sockaddr_in6 sin6 = {
2824                 .sin6_family            = AF_INET6,
2825                 .sin6_addr              = IN6ADDR_ANY_INIT,
2826         };
2827
2828         switch (family) {
2829         case AF_LOCAL:
2830                 break;
2831         case AF_INET:
2832                 memcpy(sap, &sin, sizeof(sin));
2833                 break;
2834         case AF_INET6:
2835                 memcpy(sap, &sin6, sizeof(sin6));
2836                 break;
2837         default:
2838                 dprintk("RPC:       %s: Bad address family\n", __func__);
2839                 return -EAFNOSUPPORT;
2840         }
2841         return 0;
2842 }
2843
2844 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2845                                       unsigned int slot_table_size,
2846                                       unsigned int max_slot_table_size)
2847 {
2848         struct rpc_xprt *xprt;
2849         struct sock_xprt *new;
2850
2851         if (args->addrlen > sizeof(xprt->addr)) {
2852                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2853                 return ERR_PTR(-EBADF);
2854         }
2855
2856         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2857                         max_slot_table_size);
2858         if (xprt == NULL) {
2859                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2860                                 "rpc_xprt\n");
2861                 return ERR_PTR(-ENOMEM);
2862         }
2863
2864         new = container_of(xprt, struct sock_xprt, xprt);
2865         mutex_init(&new->recv_mutex);
2866         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2867         xprt->addrlen = args->addrlen;
2868         if (args->srcaddr)
2869                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2870         else {
2871                 int err;
2872                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2873                                         (struct sockaddr *)&new->srcaddr);
2874                 if (err != 0) {
2875                         xprt_free(xprt);
2876                         return ERR_PTR(err);
2877                 }
2878         }
2879
2880         return xprt;
2881 }
2882
2883 static const struct rpc_timeout xs_local_default_timeout = {
2884         .to_initval = 10 * HZ,
2885         .to_maxval = 10 * HZ,
2886         .to_retries = 2,
2887 };
2888
2889 /**
2890  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2891  * @args: rpc transport creation arguments
2892  *
2893  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2894  */
2895 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2896 {
2897         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2898         struct sock_xprt *transport;
2899         struct rpc_xprt *xprt;
2900         struct rpc_xprt *ret;
2901
2902         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2903                         xprt_max_tcp_slot_table_entries);
2904         if (IS_ERR(xprt))
2905                 return xprt;
2906         transport = container_of(xprt, struct sock_xprt, xprt);
2907
2908         xprt->prot = 0;
2909         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2910
2911         xprt->bind_timeout = XS_BIND_TO;
2912         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2913         xprt->idle_timeout = XS_IDLE_DISC_TO;
2914
2915         xprt->ops = &xs_local_ops;
2916         xprt->timeout = &xs_local_default_timeout;
2917
2918         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2919         INIT_WORK(&transport->error_worker, xs_error_handle);
2920         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2921
2922         switch (sun->sun_family) {
2923         case AF_LOCAL:
2924                 if (sun->sun_path[0] != '/') {
2925                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2926                                         sun->sun_path);
2927                         ret = ERR_PTR(-EINVAL);
2928                         goto out_err;
2929                 }
2930                 xprt_set_bound(xprt);
2931                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2932                 ret = ERR_PTR(xs_local_setup_socket(transport));
2933                 if (ret)
2934                         goto out_err;
2935                 break;
2936         default:
2937                 ret = ERR_PTR(-EAFNOSUPPORT);
2938                 goto out_err;
2939         }
2940
2941         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2942                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2943
2944         if (try_module_get(THIS_MODULE))
2945                 return xprt;
2946         ret = ERR_PTR(-EINVAL);
2947 out_err:
2948         xs_xprt_free(xprt);
2949         return ret;
2950 }
2951
2952 static const struct rpc_timeout xs_udp_default_timeout = {
2953         .to_initval = 5 * HZ,
2954         .to_maxval = 30 * HZ,
2955         .to_increment = 5 * HZ,
2956         .to_retries = 5,
2957 };
2958
2959 /**
2960  * xs_setup_udp - Set up transport to use a UDP socket
2961  * @args: rpc transport creation arguments
2962  *
2963  */
2964 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2965 {
2966         struct sockaddr *addr = args->dstaddr;
2967         struct rpc_xprt *xprt;
2968         struct sock_xprt *transport;
2969         struct rpc_xprt *ret;
2970
2971         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2972                         xprt_udp_slot_table_entries);
2973         if (IS_ERR(xprt))
2974                 return xprt;
2975         transport = container_of(xprt, struct sock_xprt, xprt);
2976
2977         xprt->prot = IPPROTO_UDP;
2978         /* XXX: header size can vary due to auth type, IPv6, etc. */
2979         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2980
2981         xprt->bind_timeout = XS_BIND_TO;
2982         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2983         xprt->idle_timeout = XS_IDLE_DISC_TO;
2984
2985         xprt->ops = &xs_udp_ops;
2986
2987         xprt->timeout = &xs_udp_default_timeout;
2988
2989         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2990         INIT_WORK(&transport->error_worker, xs_error_handle);
2991         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2992
2993         switch (addr->sa_family) {
2994         case AF_INET:
2995                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2996                         xprt_set_bound(xprt);
2997
2998                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2999                 break;
3000         case AF_INET6:
3001                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3002                         xprt_set_bound(xprt);
3003
3004                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3005                 break;
3006         default:
3007                 ret = ERR_PTR(-EAFNOSUPPORT);
3008                 goto out_err;
3009         }
3010
3011         if (xprt_bound(xprt))
3012                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3013                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3014                                 xprt->address_strings[RPC_DISPLAY_PORT],
3015                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3016         else
3017                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3018                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3019                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3020
3021         if (try_module_get(THIS_MODULE))
3022                 return xprt;
3023         ret = ERR_PTR(-EINVAL);
3024 out_err:
3025         xs_xprt_free(xprt);
3026         return ret;
3027 }
3028
3029 static const struct rpc_timeout xs_tcp_default_timeout = {
3030         .to_initval = 60 * HZ,
3031         .to_maxval = 60 * HZ,
3032         .to_retries = 2,
3033 };
3034
3035 /**
3036  * xs_setup_tcp - Set up transport to use a TCP socket
3037  * @args: rpc transport creation arguments
3038  *
3039  */
3040 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3041 {
3042         struct sockaddr *addr = args->dstaddr;
3043         struct rpc_xprt *xprt;
3044         struct sock_xprt *transport;
3045         struct rpc_xprt *ret;
3046         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3047
3048         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3049                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3050
3051         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3052                         max_slot_table_size);
3053         if (IS_ERR(xprt))
3054                 return xprt;
3055         transport = container_of(xprt, struct sock_xprt, xprt);
3056
3057         xprt->prot = IPPROTO_TCP;
3058         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3059
3060         xprt->bind_timeout = XS_BIND_TO;
3061         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3062         xprt->idle_timeout = XS_IDLE_DISC_TO;
3063
3064         xprt->ops = &xs_tcp_ops;
3065         xprt->timeout = &xs_tcp_default_timeout;
3066
3067         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3068         xprt->connect_timeout = xprt->timeout->to_initval *
3069                 (xprt->timeout->to_retries + 1);
3070
3071         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3072         INIT_WORK(&transport->error_worker, xs_error_handle);
3073         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3074
3075         switch (addr->sa_family) {
3076         case AF_INET:
3077                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3078                         xprt_set_bound(xprt);
3079
3080                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3081                 break;
3082         case AF_INET6:
3083                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3084                         xprt_set_bound(xprt);
3085
3086                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3087                 break;
3088         default:
3089                 ret = ERR_PTR(-EAFNOSUPPORT);
3090                 goto out_err;
3091         }
3092
3093         if (xprt_bound(xprt))
3094                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3095                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3096                                 xprt->address_strings[RPC_DISPLAY_PORT],
3097                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3098         else
3099                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3100                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3101                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3102
3103         if (try_module_get(THIS_MODULE))
3104                 return xprt;
3105         ret = ERR_PTR(-EINVAL);
3106 out_err:
3107         xs_xprt_free(xprt);
3108         return ret;
3109 }
3110
3111 /**
3112  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3113  * @args: rpc transport creation arguments
3114  *
3115  */
3116 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3117 {
3118         struct sockaddr *addr = args->dstaddr;
3119         struct rpc_xprt *xprt;
3120         struct sock_xprt *transport;
3121         struct svc_sock *bc_sock;
3122         struct rpc_xprt *ret;
3123
3124         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3125                         xprt_tcp_slot_table_entries);
3126         if (IS_ERR(xprt))
3127                 return xprt;
3128         transport = container_of(xprt, struct sock_xprt, xprt);
3129
3130         xprt->prot = IPPROTO_TCP;
3131         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3132         xprt->timeout = &xs_tcp_default_timeout;
3133
3134         /* backchannel */
3135         xprt_set_bound(xprt);
3136         xprt->bind_timeout = 0;
3137         xprt->reestablish_timeout = 0;
3138         xprt->idle_timeout = 0;
3139
3140         xprt->ops = &bc_tcp_ops;
3141
3142         switch (addr->sa_family) {
3143         case AF_INET:
3144                 xs_format_peer_addresses(xprt, "tcp",
3145                                          RPCBIND_NETID_TCP);
3146                 break;
3147         case AF_INET6:
3148                 xs_format_peer_addresses(xprt, "tcp",
3149                                    RPCBIND_NETID_TCP6);
3150                 break;
3151         default:
3152                 ret = ERR_PTR(-EAFNOSUPPORT);
3153                 goto out_err;
3154         }
3155
3156         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3157                         xprt->address_strings[RPC_DISPLAY_ADDR],
3158                         xprt->address_strings[RPC_DISPLAY_PORT],
3159                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3160
3161         /*
3162          * Once we've associated a backchannel xprt with a connection,
3163          * we want to keep it around as long as the connection lasts,
3164          * in case we need to start using it for a backchannel again;
3165          * this reference won't be dropped until bc_xprt is destroyed.
3166          */
3167         xprt_get(xprt);
3168         args->bc_xprt->xpt_bc_xprt = xprt;
3169         xprt->bc_xprt = args->bc_xprt;
3170         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3171         transport->sock = bc_sock->sk_sock;
3172         transport->inet = bc_sock->sk_sk;
3173
3174         /*
3175          * Since we don't want connections for the backchannel, we set
3176          * the xprt status to connected
3177          */
3178         xprt_set_connected(xprt);
3179
3180         if (try_module_get(THIS_MODULE))
3181                 return xprt;
3182
3183         args->bc_xprt->xpt_bc_xprt = NULL;
3184         args->bc_xprt->xpt_bc_xps = NULL;
3185         xprt_put(xprt);
3186         ret = ERR_PTR(-EINVAL);
3187 out_err:
3188         xs_xprt_free(xprt);
3189         return ret;
3190 }
3191
3192 static struct xprt_class        xs_local_transport = {
3193         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3194         .name           = "named UNIX socket",
3195         .owner          = THIS_MODULE,
3196         .ident          = XPRT_TRANSPORT_LOCAL,
3197         .setup          = xs_setup_local,
3198 };
3199
3200 static struct xprt_class        xs_udp_transport = {
3201         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3202         .name           = "udp",
3203         .owner          = THIS_MODULE,
3204         .ident          = XPRT_TRANSPORT_UDP,
3205         .setup          = xs_setup_udp,
3206 };
3207
3208 static struct xprt_class        xs_tcp_transport = {
3209         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3210         .name           = "tcp",
3211         .owner          = THIS_MODULE,
3212         .ident          = XPRT_TRANSPORT_TCP,
3213         .setup          = xs_setup_tcp,
3214 };
3215
3216 static struct xprt_class        xs_bc_tcp_transport = {
3217         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3218         .name           = "tcp NFSv4.1 backchannel",
3219         .owner          = THIS_MODULE,
3220         .ident          = XPRT_TRANSPORT_BC_TCP,
3221         .setup          = xs_setup_bc_tcp,
3222 };
3223
3224 /**
3225  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3226  *
3227  */
3228 int init_socket_xprt(void)
3229 {
3230         if (!sunrpc_table_header)
3231                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3232
3233         xprt_register_transport(&xs_local_transport);
3234         xprt_register_transport(&xs_udp_transport);
3235         xprt_register_transport(&xs_tcp_transport);
3236         xprt_register_transport(&xs_bc_tcp_transport);
3237
3238         return 0;
3239 }
3240
3241 /**
3242  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3243  *
3244  */
3245 void cleanup_socket_xprt(void)
3246 {
3247         if (sunrpc_table_header) {
3248                 unregister_sysctl_table(sunrpc_table_header);
3249                 sunrpc_table_header = NULL;
3250         }
3251
3252         xprt_unregister_transport(&xs_local_transport);
3253         xprt_unregister_transport(&xs_udp_transport);
3254         xprt_unregister_transport(&xs_tcp_transport);
3255         xprt_unregister_transport(&xs_bc_tcp_transport);
3256 }
3257
3258 static int param_set_uint_minmax(const char *val,
3259                 const struct kernel_param *kp,
3260                 unsigned int min, unsigned int max)
3261 {
3262         unsigned int num;
3263         int ret;
3264
3265         if (!val)
3266                 return -EINVAL;
3267         ret = kstrtouint(val, 0, &num);
3268         if (ret)
3269                 return ret;
3270         if (num < min || num > max)
3271                 return -EINVAL;
3272         *((unsigned int *)kp->arg) = num;
3273         return 0;
3274 }
3275
3276 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3277 {
3278         return param_set_uint_minmax(val, kp,
3279                         RPC_MIN_RESVPORT,
3280                         RPC_MAX_RESVPORT);
3281 }
3282
3283 static const struct kernel_param_ops param_ops_portnr = {
3284         .set = param_set_portnr,
3285         .get = param_get_uint,
3286 };
3287
3288 #define param_check_portnr(name, p) \
3289         __param_check(name, p, unsigned int);
3290
3291 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3292 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3293
3294 static int param_set_slot_table_size(const char *val,
3295                                      const struct kernel_param *kp)
3296 {
3297         return param_set_uint_minmax(val, kp,
3298                         RPC_MIN_SLOT_TABLE,
3299                         RPC_MAX_SLOT_TABLE);
3300 }
3301
3302 static const struct kernel_param_ops param_ops_slot_table_size = {
3303         .set = param_set_slot_table_size,
3304         .get = param_get_uint,
3305 };
3306
3307 #define param_check_slot_table_size(name, p) \
3308         __param_check(name, p, unsigned int);
3309
3310 static int param_set_max_slot_table_size(const char *val,
3311                                      const struct kernel_param *kp)
3312 {
3313         return param_set_uint_minmax(val, kp,
3314                         RPC_MIN_SLOT_TABLE,
3315                         RPC_MAX_SLOT_TABLE_LIMIT);
3316 }
3317
3318 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3319         .set = param_set_max_slot_table_size,
3320         .get = param_get_uint,
3321 };
3322
3323 #define param_check_max_slot_table_size(name, p) \
3324         __param_check(name, p, unsigned int);
3325
3326 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3327                    slot_table_size, 0644);
3328 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3329                    max_slot_table_size, 0644);
3330 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3331                    slot_table_size, 0644);