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 <net/tls.h>
  51 #include <net/tls_prot.h>
  52 #include <net/handshake.h>
  53
  54 #include <linux/bvec.h>
  55 #include <linux/highmem.h>
  56 #include <linux/uio.h>
  57 #include <linux/sched/mm.h>
  58
  59 #include <trace/events/sock.h>
  60 #include <trace/events/sunrpc.h>
  61
  62 #include "socklib.h"
  63 #include "sunrpc.h"
  64
  65 static void xs_close(struct rpc_xprt *xprt);
  66 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
  67 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  68                 struct socket *sock);
  69
  70 /*
  71  * xprtsock tunables
  72  */
  73 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  74 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  75 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  76
  77 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  78 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  79
  80 #define XS_TCP_LINGER_TO        (15U * HZ)
  81 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  82
  83 /*
  84  * We can register our own files under /proc/sys/sunrpc by
  85  * calling register_sysctl() again.  The files in that
  86  * directory become the union of all files registered there.
  87  *
  88  * We simply need to make sure that we don't collide with
  89  * someone else's file names!
  90  */
  91
  92 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  93 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  94 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  95 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  96 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  97
  98 static struct ctl_table_header *sunrpc_table_header;
  99
 100 static struct xprt_class xs_local_transport;
 101 static struct xprt_class xs_udp_transport;
 102 static struct xprt_class xs_tcp_transport;
 103 static struct xprt_class xs_tcp_tls_transport;
 104 static struct xprt_class xs_bc_tcp_transport;
 105
 106 /*
 107  * FIXME: changing the UDP slot table size should also resize the UDP
 108  *        socket buffers for existing UDP transports
 109  */
 110 static struct ctl_table xs_tunables_table[] = {
 111         {
 112                 .procname       = "udp_slot_table_entries",
 113                 .data           = &xprt_udp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_slot_table_size
 119         },
 120         {
 121                 .procname       = "tcp_slot_table_entries",
 122                 .data           = &xprt_tcp_slot_table_entries,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &min_slot_table_size,
 127                 .extra2         = &max_slot_table_size
 128         },
 129         {
 130                 .procname       = "tcp_max_slot_table_entries",
 131                 .data           = &xprt_max_tcp_slot_table_entries,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &min_slot_table_size,
 136                 .extra2         = &max_tcp_slot_table_limit
 137         },
 138         {
 139                 .procname       = "min_resvport",
 140                 .data           = &xprt_min_resvport,
 141                 .maxlen         = sizeof(unsigned int),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_minmax,
 144                 .extra1         = &xprt_min_resvport_limit,
 145                 .extra2         = &xprt_max_resvport_limit
 146         },
 147         {
 148                 .procname       = "max_resvport",
 149                 .data           = &xprt_max_resvport,
 150                 .maxlen         = sizeof(unsigned int),
 151                 .mode           = 0644,
 152                 .proc_handler   = proc_dointvec_minmax,
 153                 .extra1         = &xprt_min_resvport_limit,
 154                 .extra2         = &xprt_max_resvport_limit
 155         },
 156         {
 157                 .procname       = "tcp_fin_timeout",
 158                 .data           = &xs_tcp_fin_timeout,
 159                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 160                 .mode           = 0644,
 161                 .proc_handler   = proc_dointvec_jiffies,
 162         },
 163         { },
 164 };
 165
 166 /*
 167  * Wait duration for a reply from the RPC portmapper.
 168  */
 169 #define XS_BIND_TO              (60U * HZ)
 170
 171 /*
 172  * Delay if a UDP socket connect error occurs.  This is most likely some
 173  * kind of resource problem on the local host.
 174  */
 175 #define XS_UDP_REEST_TO         (2U * HZ)
 176
 177 /*
 178  * The reestablish timeout allows clients to delay for a bit before attempting
 179  * to reconnect to a server that just dropped our connection.
 180  *
 181  * We implement an exponential backoff when trying to reestablish a TCP
 182  * transport connection with the server.  Some servers like to drop a TCP
 183  * connection when they are overworked, so we start with a short timeout and
 184  * increase over time if the server is down or not responding.
 185  */
 186 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 187
 188 /*
 189  * TCP idle timeout; client drops the transport socket if it is idle
 190  * for this long.  Note that we also timeout UDP sockets to prevent
 191  * holding port numbers when there is no RPC traffic.
 192  */
 193 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 194
 195 /*
 196  * TLS handshake timeout.
 197  */
 198 #define XS_TLS_HANDSHAKE_TO     (10U * HZ)
 199
 200 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 201 # undef  RPC_DEBUG_DATA
 202 # define RPCDBG_FACILITY        RPCDBG_TRANS
 203 #endif
 204
 205 #ifdef RPC_DEBUG_DATA
 206 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 207 {
 208         u8 *buf = (u8 *) packet;
 209         int j;
 210
 211         dprintk("RPC:       %s\n", msg);
 212         for (j = 0; j < count && j < 128; j += 4) {
 213                 if (!(j & 31)) {
 214                         if (j)
 215                                 dprintk("\n");
 216                         dprintk("0x%04x ", j);
 217                 }
 218                 dprintk("%02x%02x%02x%02x ",
 219                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 220         }
 221         dprintk("\n");
 222 }
 223 #else
 224 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 225 {
 226         /* NOP */
 227 }
 228 #endif
 229
 230 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 231 {
 232         return (struct rpc_xprt *) sk->sk_user_data;
 233 }
 234
 235 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 236 {
 237         return (struct sockaddr *) &xprt->addr;
 238 }
 239
 240 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 241 {
 242         return (struct sockaddr_un *) &xprt->addr;
 243 }
 244
 245 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 246 {
 247         return (struct sockaddr_in *) &xprt->addr;
 248 }
 249
 250 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 251 {
 252         return (struct sockaddr_in6 *) &xprt->addr;
 253 }
 254
 255 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 256 {
 257         struct sockaddr *sap = xs_addr(xprt);
 258         struct sockaddr_in6 *sin6;
 259         struct sockaddr_in *sin;
 260         struct sockaddr_un *sun;
 261         char buf[128];
 262
 263         switch (sap->sa_family) {
 264         case AF_LOCAL:
 265                 sun = xs_addr_un(xprt);
 266                 if (sun->sun_path[0]) {
 267                         strscpy(buf, sun->sun_path, sizeof(buf));
 268                 } else {
 269                         buf[0] = '@';
 270                         strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1);
 271                 }
 272                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 273                                                 kstrdup(buf, GFP_KERNEL);
 274                 break;
 275         case AF_INET:
 276                 (void)rpc_ntop(sap, buf, sizeof(buf));
 277                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 278                                                 kstrdup(buf, GFP_KERNEL);
 279                 sin = xs_addr_in(xprt);
 280                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 281                 break;
 282         case AF_INET6:
 283                 (void)rpc_ntop(sap, buf, sizeof(buf));
 284                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 285                                                 kstrdup(buf, GFP_KERNEL);
 286                 sin6 = xs_addr_in6(xprt);
 287                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 288                 break;
 289         default:
 290                 BUG();
 291         }
 292
 293         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 294 }
 295
 296 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 297 {
 298         struct sockaddr *sap = xs_addr(xprt);
 299         char buf[128];
 300
 301         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 302         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 303
 304         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 305         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 306 }
 307
 308 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 309                                      const char *protocol,
 310                                      const char *netid)
 311 {
 312         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 313         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 314         xs_format_common_peer_addresses(xprt);
 315         xs_format_common_peer_ports(xprt);
 316 }
 317
 318 static void xs_update_peer_port(struct rpc_xprt *xprt)
 319 {
 320         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 321         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 322
 323         xs_format_common_peer_ports(xprt);
 324 }
 325
 326 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 327 {
 328         unsigned int i;
 329
 330         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 331                 switch (i) {
 332                 case RPC_DISPLAY_PROTO:
 333                 case RPC_DISPLAY_NETID:
 334                         continue;
 335                 default:
 336                         kfree(xprt->address_strings[i]);
 337                 }
 338 }
 339
 340 static size_t
 341 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 342 {
 343         size_t i,n;
 344
 345         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 346                 return want;
 347         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 348         for (i = 0; i < n; i++) {
 349                 if (buf->pages[i])
 350                         continue;
 351                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 352                 if (!buf->pages[i]) {
 353                         i *= PAGE_SIZE;
 354                         return i > buf->page_base ? i - buf->page_base : 0;
 355                 }
 356         }
 357         return want;
 358 }
 359
 360 static int
 361 xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
 362                      struct cmsghdr *cmsg, int ret)
 363 {
 364         if (cmsg->cmsg_level == SOL_TLS &&
 365             cmsg->cmsg_type == TLS_GET_RECORD_TYPE) {
 366                 u8 content_type = *((u8 *)CMSG_DATA(cmsg));
 367
 368                 switch (content_type) {
 369                 case TLS_RECORD_TYPE_DATA:
 370                         /* TLS sets EOR at the end of each application data
 371                          * record, even though there might be more frames
 372                          * waiting to be decrypted.
 373                          */
 374                         msg->msg_flags &= ~MSG_EOR;
 375                         break;
 376                 case TLS_RECORD_TYPE_ALERT:
 377                         ret = -ENOTCONN;
 378                         break;
 379                 default:
 380                         ret = -EAGAIN;
 381                 }
 382         }
 383         return ret;
 384 }
 385
 386 static int
 387 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
 388 {
 389         union {
 390                 struct cmsghdr  cmsg;
 391                 u8              buf[CMSG_SPACE(sizeof(u8))];
 392         } u;
 393         int ret;
 394
 395         msg->msg_control = &u;
 396         msg->msg_controllen = sizeof(u);
 397         ret = sock_recvmsg(sock, msg, flags);
 398         if (msg->msg_controllen != sizeof(u))
 399                 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
 400         return ret;
 401 }
 402
 403 static ssize_t
 404 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 405 {
 406         ssize_t ret;
 407         if (seek != 0)
 408                 iov_iter_advance(&msg->msg_iter, seek);
 409         ret = xs_sock_recv_cmsg(sock, msg, flags);
 410         return ret > 0 ? ret + seek : ret;
 411 }
 412
 413 static ssize_t
 414 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 415                 struct kvec *kvec, size_t count, size_t seek)
 416 {
 417         iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
 418         return xs_sock_recvmsg(sock, msg, flags, seek);
 419 }
 420
 421 static ssize_t
 422 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 423                 struct bio_vec *bvec, unsigned long nr, size_t count,
 424                 size_t seek)
 425 {
 426         iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
 427         return xs_sock_recvmsg(sock, msg, flags, seek);
 428 }
 429
 430 static ssize_t
 431 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 432                 size_t count)
 433 {
 434         iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
 435         return xs_sock_recv_cmsg(sock, msg, flags);
 436 }
 437
 438 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 439 static void
 440 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 441 {
 442         struct bvec_iter bi = {
 443                 .bi_size = count,
 444         };
 445         struct bio_vec bv;
 446
 447         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 448         for_each_bvec(bv, bvec, bi, bi)
 449                 flush_dcache_page(bv.bv_page);
 450 }
 451 #else
 452 static inline void
 453 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 454 {
 455 }
 456 #endif
 457
 458 static ssize_t
 459 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 460                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 461 {
 462         size_t want, seek_init = seek, offset = 0;
 463         ssize_t ret;
 464
 465         want = min_t(size_t, count, buf->head[0].iov_len);
 466         if (seek < want) {
 467                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 468                 if (ret <= 0)
 469                         goto sock_err;
 470                 offset += ret;
 471                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 472                         goto out;
 473                 if (ret != want)
 474                         goto out;
 475                 seek = 0;
 476         } else {
 477                 seek -= want;
 478                 offset += want;
 479         }
 480
 481         want = xs_alloc_sparse_pages(
 482                 buf, min_t(size_t, count - offset, buf->page_len),
 483                 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
 484         if (seek < want) {
 485                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 486                                 xdr_buf_pagecount(buf),
 487                                 want + buf->page_base,
 488                                 seek + buf->page_base);
 489                 if (ret <= 0)
 490                         goto sock_err;
 491                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 492                 ret -= buf->page_base;
 493                 offset += ret;
 494                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 495                         goto out;
 496                 if (ret != want)
 497                         goto out;
 498                 seek = 0;
 499         } else {
 500                 seek -= want;
 501                 offset += want;
 502         }
 503
 504         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 505         if (seek < want) {
 506                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 507                 if (ret <= 0)
 508                         goto sock_err;
 509                 offset += ret;
 510                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 511                         goto out;
 512                 if (ret != want)
 513                         goto out;
 514         } else if (offset < seek_init)
 515                 offset = seek_init;
 516         ret = -EMSGSIZE;
 517 out:
 518         *read = offset - seek_init;
 519         return ret;
 520 sock_err:
 521         offset += seek;
 522         goto out;
 523 }
 524
 525 static void
 526 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 527 {
 528         if (!transport->recv.copied) {
 529                 if (buf->head[0].iov_len >= transport->recv.offset)
 530                         memcpy(buf->head[0].iov_base,
 531                                         &transport->recv.xid,
 532                                         transport->recv.offset);
 533                 transport->recv.copied = transport->recv.offset;
 534         }
 535 }
 536
 537 static bool
 538 xs_read_stream_request_done(struct sock_xprt *transport)
 539 {
 540         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 541 }
 542
 543 static void
 544 xs_read_stream_check_eor(struct sock_xprt *transport,
 545                 struct msghdr *msg)
 546 {
 547         if (xs_read_stream_request_done(transport))
 548                 msg->msg_flags |= MSG_EOR;
 549 }
 550
 551 static ssize_t
 552 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 553                 int flags, struct rpc_rqst *req)
 554 {
 555         struct xdr_buf *buf = &req->rq_private_buf;
 556         size_t want, read;
 557         ssize_t ret;
 558
 559         xs_read_header(transport, buf);
 560
 561         want = transport->recv.len - transport->recv.offset;
 562         if (want != 0) {
 563                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 564                                 transport->recv.copied + want,
 565                                 transport->recv.copied,
 566                                 &read);
 567                 transport->recv.offset += read;
 568                 transport->recv.copied += read;
 569         }
 570
 571         if (transport->recv.offset == transport->recv.len)
 572                 xs_read_stream_check_eor(transport, msg);
 573
 574         if (want == 0)
 575                 return 0;
 576
 577         switch (ret) {
 578         default:
 579                 break;
 580         case -EFAULT:
 581         case -EMSGSIZE:
 582                 msg->msg_flags |= MSG_TRUNC;
 583                 return read;
 584         case 0:
 585                 return -ESHUTDOWN;
 586         }
 587         return ret < 0 ? ret : read;
 588 }
 589
 590 static size_t
 591 xs_read_stream_headersize(bool isfrag)
 592 {
 593         if (isfrag)
 594                 return sizeof(__be32);
 595         return 3 * sizeof(__be32);
 596 }
 597
 598 static ssize_t
 599 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 600                 int flags, size_t want, size_t seek)
 601 {
 602         struct kvec kvec = {
 603                 .iov_base = &transport->recv.fraghdr,
 604                 .iov_len = want,
 605         };
 606         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 607 }
 608
 609 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 610 static ssize_t
 611 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 612 {
 613         struct rpc_xprt *xprt = &transport->xprt;
 614         struct rpc_rqst *req;
 615         ssize_t ret;
 616
 617         /* Is this transport associated with the backchannel? */
 618         if (!xprt->bc_serv)
 619                 return -ESHUTDOWN;
 620
 621         /* Look up and lock the request corresponding to the given XID */
 622         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 623         if (!req) {
 624                 printk(KERN_WARNING "Callback slot table overflowed\n");
 625                 return -ESHUTDOWN;
 626         }
 627         if (transport->recv.copied && !req->rq_private_buf.len)
 628                 return -ESHUTDOWN;
 629
 630         ret = xs_read_stream_request(transport, msg, flags, req);
 631         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 632                 xprt_complete_bc_request(req, transport->recv.copied);
 633         else
 634                 req->rq_private_buf.len = transport->recv.copied;
 635
 636         return ret;
 637 }
 638 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 639 static ssize_t
 640 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 641 {
 642         return -ESHUTDOWN;
 643 }
 644 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 645
 646 static ssize_t
 647 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 648 {
 649         struct rpc_xprt *xprt = &transport->xprt;
 650         struct rpc_rqst *req;
 651         ssize_t ret = 0;
 652
 653         /* Look up and lock the request corresponding to the given XID */
 654         spin_lock(&xprt->queue_lock);
 655         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 656         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 657                 msg->msg_flags |= MSG_TRUNC;
 658                 goto out;
 659         }
 660         xprt_pin_rqst(req);
 661         spin_unlock(&xprt->queue_lock);
 662
 663         ret = xs_read_stream_request(transport, msg, flags, req);
 664
 665         spin_lock(&xprt->queue_lock);
 666         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 667                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 668         else
 669                 req->rq_private_buf.len = transport->recv.copied;
 670         xprt_unpin_rqst(req);
 671 out:
 672         spin_unlock(&xprt->queue_lock);
 673         return ret;
 674 }
 675
 676 static ssize_t
 677 xs_read_stream(struct sock_xprt *transport, int flags)
 678 {
 679         struct msghdr msg = { 0 };
 680         size_t want, read = 0;
 681         ssize_t ret = 0;
 682
 683         if (transport->recv.len == 0) {
 684                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 685                 ret = xs_read_stream_header(transport, &msg, flags, want,
 686                                 transport->recv.offset);
 687                 if (ret <= 0)
 688                         goto out_err;
 689                 transport->recv.offset = ret;
 690                 if (transport->recv.offset != want)
 691                         return transport->recv.offset;
 692                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 693                         RPC_FRAGMENT_SIZE_MASK;
 694                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 695                 read = ret;
 696         }
 697
 698         switch (be32_to_cpu(transport->recv.calldir)) {
 699         default:
 700                 msg.msg_flags |= MSG_TRUNC;
 701                 break;
 702         case RPC_CALL:
 703                 ret = xs_read_stream_call(transport, &msg, flags);
 704                 break;
 705         case RPC_REPLY:
 706                 ret = xs_read_stream_reply(transport, &msg, flags);
 707         }
 708         if (msg.msg_flags & MSG_TRUNC) {
 709                 transport->recv.calldir = cpu_to_be32(-1);
 710                 transport->recv.copied = -1;
 711         }
 712         if (ret < 0)
 713                 goto out_err;
 714         read += ret;
 715         if (transport->recv.offset < transport->recv.len) {
 716                 if (!(msg.msg_flags & MSG_TRUNC))
 717                         return read;
 718                 msg.msg_flags = 0;
 719                 ret = xs_read_discard(transport->sock, &msg, flags,
 720                                 transport->recv.len - transport->recv.offset);
 721                 if (ret <= 0)
 722                         goto out_err;
 723                 transport->recv.offset += ret;
 724                 read += ret;
 725                 if (transport->recv.offset != transport->recv.len)
 726                         return read;
 727         }
 728         if (xs_read_stream_request_done(transport)) {
 729                 trace_xs_stream_read_request(transport);
 730                 transport->recv.copied = 0;
 731         }
 732         transport->recv.offset = 0;
 733         transport->recv.len = 0;
 734         return read;
 735 out_err:
 736         return ret != 0 ? ret : -ESHUTDOWN;
 737 }
 738
 739 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 740 {
 741         return transport->sock->ops->poll(transport->file, transport->sock,
 742                         NULL);
 743 }
 744
 745 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 746 {
 747         __poll_t events = xs_poll_socket(transport);
 748
 749         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 750 }
 751
 752 static void xs_poll_check_readable(struct sock_xprt *transport)
 753 {
 754
 755         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 756         if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
 757                 return;
 758         if (!xs_poll_socket_readable(transport))
 759                 return;
 760         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 761                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 762 }
 763
 764 static void xs_stream_data_receive(struct sock_xprt *transport)
 765 {
 766         size_t read = 0;
 767         ssize_t ret = 0;
 768
 769         mutex_lock(&transport->recv_mutex);
 770         if (transport->sock == NULL)
 771                 goto out;
 772         for (;;) {
 773                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 774                 if (ret < 0)
 775                         break;
 776                 read += ret;
 777                 cond_resched();
 778         }
 779         if (ret == -ESHUTDOWN)
 780                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 781         else
 782                 xs_poll_check_readable(transport);
 783 out:
 784         mutex_unlock(&transport->recv_mutex);
 785         trace_xs_stream_read_data(&transport->xprt, ret, read);
 786 }
 787
 788 static void xs_stream_data_receive_workfn(struct work_struct *work)
 789 {
 790         struct sock_xprt *transport =
 791                 container_of(work, struct sock_xprt, recv_worker);
 792         unsigned int pflags = memalloc_nofs_save();
 793
 794         xs_stream_data_receive(transport);
 795         memalloc_nofs_restore(pflags);
 796 }
 797
 798 static void
 799 xs_stream_reset_connect(struct sock_xprt *transport)
 800 {
 801         transport->recv.offset = 0;
 802         transport->recv.len = 0;
 803         transport->recv.copied = 0;
 804         transport->xmit.offset = 0;
 805 }
 806
 807 static void
 808 xs_stream_start_connect(struct sock_xprt *transport)
 809 {
 810         transport->xprt.stat.connect_count++;
 811         transport->xprt.stat.connect_start = jiffies;
 812 }
 813
 814 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 815
 816 /**
 817  * xs_nospace - handle transmit was incomplete
 818  * @req: pointer to RPC request
 819  * @transport: pointer to struct sock_xprt
 820  *
 821  */
 822 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 823 {
 824         struct rpc_xprt *xprt = &transport->xprt;
 825         struct sock *sk = transport->inet;
 826         int ret = -EAGAIN;
 827
 828         trace_rpc_socket_nospace(req, transport);
 829
 830         /* Protect against races with write_space */
 831         spin_lock(&xprt->transport_lock);
 832
 833         /* Don't race with disconnect */
 834         if (xprt_connected(xprt)) {
 835                 /* wait for more buffer space */
 836                 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
 837                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 838                 sk->sk_write_pending++;
 839                 xprt_wait_for_buffer_space(xprt);
 840         } else
 841                 ret = -ENOTCONN;
 842
 843         spin_unlock(&xprt->transport_lock);
 844         return ret;
 845 }
 846
 847 static int xs_sock_nospace(struct rpc_rqst *req)
 848 {
 849         struct sock_xprt *transport =
 850                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 851         struct sock *sk = transport->inet;
 852         int ret = -EAGAIN;
 853
 854         lock_sock(sk);
 855         if (!sock_writeable(sk))
 856                 ret = xs_nospace(req, transport);
 857         release_sock(sk);
 858         return ret;
 859 }
 860
 861 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
 862 {
 863         struct sock_xprt *transport =
 864                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 865         struct sock *sk = transport->inet;
 866         int ret = -EAGAIN;
 867
 868         if (vm_wait)
 869                 return -ENOBUFS;
 870         lock_sock(sk);
 871         if (!sk_stream_memory_free(sk))
 872                 ret = xs_nospace(req, transport);
 873         release_sock(sk);
 874         return ret;
 875 }
 876
 877 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
 878 {
 879         return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
 880 }
 881
 882 /*
 883  * Determine if the previous message in the stream was aborted before it
 884  * could complete transmission.
 885  */
 886 static bool
 887 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 888 {
 889         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 890 }
 891
 892 /*
 893  * Return the stream record marker field for a record of length < 2^31-1
 894  */
 895 static rpc_fraghdr
 896 xs_stream_record_marker(struct xdr_buf *xdr)
 897 {
 898         if (!xdr->len)
 899                 return 0;
 900         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 901 }
 902
 903 /**
 904  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 905  * @req: pointer to RPC request
 906  *
 907  * Return values:
 908  *        0:    The request has been sent
 909  *   EAGAIN:    The socket was blocked, please call again later to
 910  *              complete the request
 911  * ENOTCONN:    Caller needs to invoke connect logic then call again
 912  *    other:    Some other error occurred, the request was not sent
 913  */
 914 static int xs_local_send_request(struct rpc_rqst *req)
 915 {
 916         struct rpc_xprt *xprt = req->rq_xprt;
 917         struct sock_xprt *transport =
 918                                 container_of(xprt, struct sock_xprt, xprt);
 919         struct xdr_buf *xdr = &req->rq_snd_buf;
 920         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 921         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 922         struct msghdr msg = {
 923                 .msg_flags      = XS_SENDMSG_FLAGS,
 924         };
 925         bool vm_wait;
 926         unsigned int sent;
 927         int status;
 928
 929         /* Close the stream if the previous transmission was incomplete */
 930         if (xs_send_request_was_aborted(transport, req)) {
 931                 xprt_force_disconnect(xprt);
 932                 return -ENOTCONN;
 933         }
 934
 935         xs_pktdump("packet data:",
 936                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 937
 938         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
 939
 940         req->rq_xtime = ktime_get();
 941         status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 942                                    transport->xmit.offset, rm, &sent);
 943         dprintk("RPC:       %s(%u) = %d\n",
 944                         __func__, xdr->len - transport->xmit.offset, status);
 945
 946         if (likely(sent > 0) || status == 0) {
 947                 transport->xmit.offset += sent;
 948                 req->rq_bytes_sent = transport->xmit.offset;
 949                 if (likely(req->rq_bytes_sent >= msglen)) {
 950                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 951                         transport->xmit.offset = 0;
 952                         return 0;
 953                 }
 954                 status = -EAGAIN;
 955                 vm_wait = false;
 956         }
 957
 958         switch (status) {
 959         case -EAGAIN:
 960                 status = xs_stream_nospace(req, vm_wait);
 961                 break;
 962         default:
 963                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 964                         -status);
 965                 fallthrough;
 966         case -EPIPE:
 967                 xprt_force_disconnect(xprt);
 968                 status = -ENOTCONN;
 969         }
 970
 971         return status;
 972 }
 973
 974 /**
 975  * xs_udp_send_request - write an RPC request to a UDP socket
 976  * @req: pointer to RPC request
 977  *
 978  * Return values:
 979  *        0:    The request has been sent
 980  *   EAGAIN:    The socket was blocked, please call again later to
 981  *              complete the request
 982  * ENOTCONN:    Caller needs to invoke connect logic then call again
 983  *    other:    Some other error occurred, the request was not sent
 984  */
 985 static int xs_udp_send_request(struct rpc_rqst *req)
 986 {
 987         struct rpc_xprt *xprt = req->rq_xprt;
 988         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 989         struct xdr_buf *xdr = &req->rq_snd_buf;
 990         struct msghdr msg = {
 991                 .msg_name       = xs_addr(xprt),
 992                 .msg_namelen    = xprt->addrlen,
 993                 .msg_flags      = XS_SENDMSG_FLAGS,
 994         };
 995         unsigned int sent;
 996         int status;
 997
 998         xs_pktdump("packet data:",
 999                                 req->rq_svec->iov_base,
1000                                 req->rq_svec->iov_len);
1001
1002         if (!xprt_bound(xprt))
1003                 return -ENOTCONN;
1004
1005         if (!xprt_request_get_cong(xprt, req))
1006                 return -EBADSLT;
1007
1008         status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1009         if (status < 0)
1010                 return status;
1011         req->rq_xtime = ktime_get();
1012         status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1013
1014         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
1015                         xdr->len, status);
1016
1017         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1018         if (status == -EPERM)
1019                 goto process_status;
1020
1021         if (status == -EAGAIN && sock_writeable(transport->inet))
1022                 status = -ENOBUFS;
1023
1024         if (sent > 0 || status == 0) {
1025                 req->rq_xmit_bytes_sent += sent;
1026                 if (sent >= req->rq_slen)
1027                         return 0;
1028                 /* Still some bytes left; set up for a retry later. */
1029                 status = -EAGAIN;
1030         }
1031
1032 process_status:
1033         switch (status) {
1034         case -ENOTSOCK:
1035                 status = -ENOTCONN;
1036                 /* Should we call xs_close() here? */
1037                 break;
1038         case -EAGAIN:
1039                 status = xs_sock_nospace(req);
1040                 break;
1041         case -ENETUNREACH:
1042         case -ENOBUFS:
1043         case -EPIPE:
1044         case -ECONNREFUSED:
1045         case -EPERM:
1046                 /* When the server has died, an ICMP port unreachable message
1047                  * prompts ECONNREFUSED. */
1048                 break;
1049         default:
1050                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1051                         -status);
1052         }
1053
1054         return status;
1055 }
1056
1057 /**
1058  * xs_tcp_send_request - write an RPC request to a TCP socket
1059  * @req: pointer to RPC request
1060  *
1061  * Return values:
1062  *        0:    The request has been sent
1063  *   EAGAIN:    The socket was blocked, please call again later to
1064  *              complete the request
1065  * ENOTCONN:    Caller needs to invoke connect logic then call again
1066  *    other:    Some other error occurred, the request was not sent
1067  *
1068  * XXX: In the case of soft timeouts, should we eventually give up
1069  *      if sendmsg is not able to make progress?
1070  */
1071 static int xs_tcp_send_request(struct rpc_rqst *req)
1072 {
1073         struct rpc_xprt *xprt = req->rq_xprt;
1074         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1075         struct xdr_buf *xdr = &req->rq_snd_buf;
1076         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1077         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1078         struct msghdr msg = {
1079                 .msg_flags      = XS_SENDMSG_FLAGS,
1080         };
1081         bool vm_wait;
1082         unsigned int sent;
1083         int status;
1084
1085         /* Close the stream if the previous transmission was incomplete */
1086         if (xs_send_request_was_aborted(transport, req)) {
1087                 if (transport->sock != NULL)
1088                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1089                 return -ENOTCONN;
1090         }
1091         if (!transport->inet)
1092                 return -ENOTCONN;
1093
1094         xs_pktdump("packet data:",
1095                                 req->rq_svec->iov_base,
1096                                 req->rq_svec->iov_len);
1097
1098         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1099                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1100
1101         xs_set_srcport(transport, transport->sock);
1102
1103         /* Continue transmitting the packet/record. We must be careful
1104          * to cope with writespace callbacks arriving _after_ we have
1105          * called sendmsg(). */
1106         req->rq_xtime = ktime_get();
1107         tcp_sock_set_cork(transport->inet, true);
1108
1109         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1110
1111         do {
1112                 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1113                                            transport->xmit.offset, rm, &sent);
1114
1115                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1116                                 xdr->len - transport->xmit.offset, status);
1117
1118                 /* If we've sent the entire packet, immediately
1119                  * reset the count of bytes sent. */
1120                 transport->xmit.offset += sent;
1121                 req->rq_bytes_sent = transport->xmit.offset;
1122                 if (likely(req->rq_bytes_sent >= msglen)) {
1123                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1124                         transport->xmit.offset = 0;
1125                         if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1126                                 tcp_sock_set_cork(transport->inet, false);
1127                         return 0;
1128                 }
1129
1130                 WARN_ON_ONCE(sent == 0 && status == 0);
1131
1132                 if (sent > 0)
1133                         vm_wait = false;
1134
1135         } while (status == 0);
1136
1137         switch (status) {
1138         case -ENOTSOCK:
1139                 status = -ENOTCONN;
1140                 /* Should we call xs_close() here? */
1141                 break;
1142         case -EAGAIN:
1143                 status = xs_stream_nospace(req, vm_wait);
1144                 break;
1145         case -ECONNRESET:
1146         case -ECONNREFUSED:
1147         case -ENOTCONN:
1148         case -EADDRINUSE:
1149         case -ENOBUFS:
1150         case -EPIPE:
1151                 break;
1152         default:
1153                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1154                         -status);
1155         }
1156
1157         return status;
1158 }
1159
1160 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1161 {
1162         transport->old_data_ready = sk->sk_data_ready;
1163         transport->old_state_change = sk->sk_state_change;
1164         transport->old_write_space = sk->sk_write_space;
1165         transport->old_error_report = sk->sk_error_report;
1166 }
1167
1168 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1169 {
1170         sk->sk_data_ready = transport->old_data_ready;
1171         sk->sk_state_change = transport->old_state_change;
1172         sk->sk_write_space = transport->old_write_space;
1173         sk->sk_error_report = transport->old_error_report;
1174 }
1175
1176 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1177 {
1178         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1179
1180         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1181         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1182         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1183         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1184         clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1185 }
1186
1187 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1188 {
1189         set_bit(nr, &transport->sock_state);
1190         queue_work(xprtiod_workqueue, &transport->error_worker);
1191 }
1192
1193 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1194 {
1195         xprt->connect_cookie++;
1196         smp_mb__before_atomic();
1197         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1198         clear_bit(XPRT_CLOSING, &xprt->state);
1199         xs_sock_reset_state_flags(xprt);
1200         smp_mb__after_atomic();
1201 }
1202
1203 /**
1204  * xs_error_report - callback to handle TCP socket state errors
1205  * @sk: socket
1206  *
1207  * Note: we don't call sock_error() since there may be a rpc_task
1208  * using the socket, and so we don't want to clear sk->sk_err.
1209  */
1210 static void xs_error_report(struct sock *sk)
1211 {
1212         struct sock_xprt *transport;
1213         struct rpc_xprt *xprt;
1214
1215         if (!(xprt = xprt_from_sock(sk)))
1216                 return;
1217
1218         transport = container_of(xprt, struct sock_xprt, xprt);
1219         transport->xprt_err = -sk->sk_err;
1220         if (transport->xprt_err == 0)
1221                 return;
1222         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1223                         xprt, -transport->xprt_err);
1224         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1225
1226         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1227         smp_mb__before_atomic();
1228         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1229 }
1230
1231 static void xs_reset_transport(struct sock_xprt *transport)
1232 {
1233         struct socket *sock = transport->sock;
1234         struct sock *sk = transport->inet;
1235         struct rpc_xprt *xprt = &transport->xprt;
1236         struct file *filp = transport->file;
1237
1238         if (sk == NULL)
1239                 return;
1240         /*
1241          * Make sure we're calling this in a context from which it is safe
1242          * to call __fput_sync(). In practice that means rpciod and the
1243          * system workqueue.
1244          */
1245         if (!(current->flags & PF_WQ_WORKER)) {
1246                 WARN_ON_ONCE(1);
1247                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1248                 return;
1249         }
1250
1251         if (atomic_read(&transport->xprt.swapper))
1252                 sk_clear_memalloc(sk);
1253
1254         tls_handshake_cancel(sk);
1255
1256         kernel_sock_shutdown(sock, SHUT_RDWR);
1257
1258         mutex_lock(&transport->recv_mutex);
1259         lock_sock(sk);
1260         transport->inet = NULL;
1261         transport->sock = NULL;
1262         transport->file = NULL;
1263
1264         sk->sk_user_data = NULL;
1265
1266         xs_restore_old_callbacks(transport, sk);
1267         xprt_clear_connected(xprt);
1268         xs_sock_reset_connection_flags(xprt);
1269         /* Reset stream record info */
1270         xs_stream_reset_connect(transport);
1271         release_sock(sk);
1272         mutex_unlock(&transport->recv_mutex);
1273
1274         trace_rpc_socket_close(xprt, sock);
1275         __fput_sync(filp);
1276
1277         xprt_disconnect_done(xprt);
1278 }
1279
1280 /**
1281  * xs_close - close a socket
1282  * @xprt: transport
1283  *
1284  * This is used when all requests are complete; ie, no DRC state remains
1285  * on the server we want to save.
1286  *
1287  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1288  * xs_reset_transport() zeroing the socket from underneath a writer.
1289  */
1290 static void xs_close(struct rpc_xprt *xprt)
1291 {
1292         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1293
1294         dprintk("RPC:       xs_close xprt %p\n", xprt);
1295
1296         xs_reset_transport(transport);
1297         xprt->reestablish_timeout = 0;
1298 }
1299
1300 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1301 {
1302         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1303                 xprt);
1304         xprt_disconnect_done(xprt);
1305 }
1306
1307 static void xs_xprt_free(struct rpc_xprt *xprt)
1308 {
1309         xs_free_peer_addresses(xprt);
1310         xprt_free(xprt);
1311 }
1312
1313 /**
1314  * xs_destroy - prepare to shutdown a transport
1315  * @xprt: doomed transport
1316  *
1317  */
1318 static void xs_destroy(struct rpc_xprt *xprt)
1319 {
1320         struct sock_xprt *transport = container_of(xprt,
1321                         struct sock_xprt, xprt);
1322         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1323
1324         cancel_delayed_work_sync(&transport->connect_worker);
1325         xs_close(xprt);
1326         cancel_work_sync(&transport->recv_worker);
1327         cancel_work_sync(&transport->error_worker);
1328         xs_xprt_free(xprt);
1329         module_put(THIS_MODULE);
1330 }
1331
1332 /**
1333  * xs_udp_data_read_skb - receive callback for UDP sockets
1334  * @xprt: transport
1335  * @sk: socket
1336  * @skb: skbuff
1337  *
1338  */
1339 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1340                 struct sock *sk,
1341                 struct sk_buff *skb)
1342 {
1343         struct rpc_task *task;
1344         struct rpc_rqst *rovr;
1345         int repsize, copied;
1346         u32 _xid;
1347         __be32 *xp;
1348
1349         repsize = skb->len;
1350         if (repsize < 4) {
1351                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1352                 return;
1353         }
1354
1355         /* Copy the XID from the skb... */
1356         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1357         if (xp == NULL)
1358                 return;
1359
1360         /* Look up and lock the request corresponding to the given XID */
1361         spin_lock(&xprt->queue_lock);
1362         rovr = xprt_lookup_rqst(xprt, *xp);
1363         if (!rovr)
1364                 goto out_unlock;
1365         xprt_pin_rqst(rovr);
1366         xprt_update_rtt(rovr->rq_task);
1367         spin_unlock(&xprt->queue_lock);
1368         task = rovr->rq_task;
1369
1370         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1371                 copied = repsize;
1372
1373         /* Suck it into the iovec, verify checksum if not done by hw. */
1374         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1375                 spin_lock(&xprt->queue_lock);
1376                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1377                 goto out_unpin;
1378         }
1379
1380
1381         spin_lock(&xprt->transport_lock);
1382         xprt_adjust_cwnd(xprt, task, copied);
1383         spin_unlock(&xprt->transport_lock);
1384         spin_lock(&xprt->queue_lock);
1385         xprt_complete_rqst(task, copied);
1386         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1387 out_unpin:
1388         xprt_unpin_rqst(rovr);
1389  out_unlock:
1390         spin_unlock(&xprt->queue_lock);
1391 }
1392
1393 static void xs_udp_data_receive(struct sock_xprt *transport)
1394 {
1395         struct sk_buff *skb;
1396         struct sock *sk;
1397         int err;
1398
1399         mutex_lock(&transport->recv_mutex);
1400         sk = transport->inet;
1401         if (sk == NULL)
1402                 goto out;
1403         for (;;) {
1404                 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1405                 if (skb == NULL)
1406                         break;
1407                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1408                 consume_skb(skb);
1409                 cond_resched();
1410         }
1411         xs_poll_check_readable(transport);
1412 out:
1413         mutex_unlock(&transport->recv_mutex);
1414 }
1415
1416 static void xs_udp_data_receive_workfn(struct work_struct *work)
1417 {
1418         struct sock_xprt *transport =
1419                 container_of(work, struct sock_xprt, recv_worker);
1420         unsigned int pflags = memalloc_nofs_save();
1421
1422         xs_udp_data_receive(transport);
1423         memalloc_nofs_restore(pflags);
1424 }
1425
1426 /**
1427  * xs_data_ready - "data ready" callback for sockets
1428  * @sk: socket with data to read
1429  *
1430  */
1431 static void xs_data_ready(struct sock *sk)
1432 {
1433         struct rpc_xprt *xprt;
1434
1435         trace_sk_data_ready(sk);
1436
1437         xprt = xprt_from_sock(sk);
1438         if (xprt != NULL) {
1439                 struct sock_xprt *transport = container_of(xprt,
1440                                 struct sock_xprt, xprt);
1441
1442                 trace_xs_data_ready(xprt);
1443
1444                 transport->old_data_ready(sk);
1445
1446                 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1447                         return;
1448
1449                 /* Any data means we had a useful conversation, so
1450                  * then we don't need to delay the next reconnect
1451                  */
1452                 if (xprt->reestablish_timeout)
1453                         xprt->reestablish_timeout = 0;
1454                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1455                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1456         }
1457 }
1458
1459 /*
1460  * Helper function to force a TCP close if the server is sending
1461  * junk and/or it has put us in CLOSE_WAIT
1462  */
1463 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1464 {
1465         xprt_force_disconnect(xprt);
1466 }
1467
1468 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1469 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1470 {
1471         return PAGE_SIZE;
1472 }
1473 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1474
1475 /**
1476  * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1477  * @sk: socket whose state has changed
1478  *
1479  */
1480 static void xs_local_state_change(struct sock *sk)
1481 {
1482         struct rpc_xprt *xprt;
1483         struct sock_xprt *transport;
1484
1485         if (!(xprt = xprt_from_sock(sk)))
1486                 return;
1487         transport = container_of(xprt, struct sock_xprt, xprt);
1488         if (sk->sk_shutdown & SHUTDOWN_MASK) {
1489                 clear_bit(XPRT_CONNECTED, &xprt->state);
1490                 /* Trigger the socket release */
1491                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1492         }
1493 }
1494
1495 /**
1496  * xs_tcp_state_change - callback to handle TCP socket state changes
1497  * @sk: socket whose state has changed
1498  *
1499  */
1500 static void xs_tcp_state_change(struct sock *sk)
1501 {
1502         struct rpc_xprt *xprt;
1503         struct sock_xprt *transport;
1504
1505         if (!(xprt = xprt_from_sock(sk)))
1506                 return;
1507         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1508         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1509                         sk->sk_state, xprt_connected(xprt),
1510                         sock_flag(sk, SOCK_DEAD),
1511                         sock_flag(sk, SOCK_ZAPPED),
1512                         sk->sk_shutdown);
1513
1514         transport = container_of(xprt, struct sock_xprt, xprt);
1515         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1516         switch (sk->sk_state) {
1517         case TCP_ESTABLISHED:
1518                 if (!xprt_test_and_set_connected(xprt)) {
1519                         xprt->connect_cookie++;
1520                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1521                         xprt_clear_connecting(xprt);
1522
1523                         xprt->stat.connect_count++;
1524                         xprt->stat.connect_time += (long)jiffies -
1525                                                    xprt->stat.connect_start;
1526                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1527                 }
1528                 break;
1529         case TCP_FIN_WAIT1:
1530                 /* The client initiated a shutdown of the socket */
1531                 xprt->connect_cookie++;
1532                 xprt->reestablish_timeout = 0;
1533                 set_bit(XPRT_CLOSING, &xprt->state);
1534                 smp_mb__before_atomic();
1535                 clear_bit(XPRT_CONNECTED, &xprt->state);
1536                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1537                 smp_mb__after_atomic();
1538                 break;
1539         case TCP_CLOSE_WAIT:
1540                 /* The server initiated a shutdown of the socket */
1541                 xprt->connect_cookie++;
1542                 clear_bit(XPRT_CONNECTED, &xprt->state);
1543                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1544                 fallthrough;
1545         case TCP_CLOSING:
1546                 /*
1547                  * If the server closed down the connection, make sure that
1548                  * we back off before reconnecting
1549                  */
1550                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1551                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1552                 break;
1553         case TCP_LAST_ACK:
1554                 set_bit(XPRT_CLOSING, &xprt->state);
1555                 smp_mb__before_atomic();
1556                 clear_bit(XPRT_CONNECTED, &xprt->state);
1557                 smp_mb__after_atomic();
1558                 break;
1559         case TCP_CLOSE:
1560                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1561                                         &transport->sock_state))
1562                         xprt_clear_connecting(xprt);
1563                 clear_bit(XPRT_CLOSING, &xprt->state);
1564                 /* Trigger the socket release */
1565                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1566         }
1567 }
1568
1569 static void xs_write_space(struct sock *sk)
1570 {
1571         struct sock_xprt *transport;
1572         struct rpc_xprt *xprt;
1573
1574         if (!sk->sk_socket)
1575                 return;
1576         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1577
1578         if (unlikely(!(xprt = xprt_from_sock(sk))))
1579                 return;
1580         transport = container_of(xprt, struct sock_xprt, xprt);
1581         if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1582                 return;
1583         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1584         sk->sk_write_pending--;
1585 }
1586
1587 /**
1588  * xs_udp_write_space - callback invoked when socket buffer space
1589  *                             becomes available
1590  * @sk: socket whose state has changed
1591  *
1592  * Called when more output buffer space is available for this socket.
1593  * We try not to wake our writers until they can make "significant"
1594  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1595  * with a bunch of small requests.
1596  */
1597 static void xs_udp_write_space(struct sock *sk)
1598 {
1599         /* from net/core/sock.c:sock_def_write_space */
1600         if (sock_writeable(sk))
1601                 xs_write_space(sk);
1602 }
1603
1604 /**
1605  * xs_tcp_write_space - callback invoked when socket buffer space
1606  *                             becomes available
1607  * @sk: socket whose state has changed
1608  *
1609  * Called when more output buffer space is available for this socket.
1610  * We try not to wake our writers until they can make "significant"
1611  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1612  * with a bunch of small requests.
1613  */
1614 static void xs_tcp_write_space(struct sock *sk)
1615 {
1616         /* from net/core/stream.c:sk_stream_write_space */
1617         if (sk_stream_is_writeable(sk))
1618                 xs_write_space(sk);
1619 }
1620
1621 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1622 {
1623         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1624         struct sock *sk = transport->inet;
1625
1626         if (transport->rcvsize) {
1627                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1628                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1629         }
1630         if (transport->sndsize) {
1631                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1632                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1633                 sk->sk_write_space(sk);
1634         }
1635 }
1636
1637 /**
1638  * xs_udp_set_buffer_size - set send and receive limits
1639  * @xprt: generic transport
1640  * @sndsize: requested size of send buffer, in bytes
1641  * @rcvsize: requested size of receive buffer, in bytes
1642  *
1643  * Set socket send and receive buffer size limits.
1644  */
1645 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1646 {
1647         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1648
1649         transport->sndsize = 0;
1650         if (sndsize)
1651                 transport->sndsize = sndsize + 1024;
1652         transport->rcvsize = 0;
1653         if (rcvsize)
1654                 transport->rcvsize = rcvsize + 1024;
1655
1656         xs_udp_do_set_buffer_size(xprt);
1657 }
1658
1659 /**
1660  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1661  * @xprt: controlling transport
1662  * @task: task that timed out
1663  *
1664  * Adjust the congestion window after a retransmit timeout has occurred.
1665  */
1666 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1667 {
1668         spin_lock(&xprt->transport_lock);
1669         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1670         spin_unlock(&xprt->transport_lock);
1671 }
1672
1673 static int xs_get_random_port(void)
1674 {
1675         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1676         unsigned short range;
1677         unsigned short rand;
1678
1679         if (max < min)
1680                 return -EADDRINUSE;
1681         range = max - min + 1;
1682         rand = get_random_u32_below(range);
1683         return rand + min;
1684 }
1685
1686 static unsigned short xs_sock_getport(struct socket *sock)
1687 {
1688         struct sockaddr_storage buf;
1689         unsigned short port = 0;
1690
1691         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1692                 goto out;
1693         switch (buf.ss_family) {
1694         case AF_INET6:
1695                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1696                 break;
1697         case AF_INET:
1698                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1699         }
1700 out:
1701         return port;
1702 }
1703
1704 /**
1705  * xs_set_port - reset the port number in the remote endpoint address
1706  * @xprt: generic transport
1707  * @port: new port number
1708  *
1709  */
1710 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1711 {
1712         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1713
1714         rpc_set_port(xs_addr(xprt), port);
1715         xs_update_peer_port(xprt);
1716 }
1717
1718 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1719 {
1720         if (transport->srcport == 0 && transport->xprt.reuseport)
1721                 transport->srcport = xs_sock_getport(sock);
1722 }
1723
1724 static int xs_get_srcport(struct sock_xprt *transport)
1725 {
1726         int port = transport->srcport;
1727
1728         if (port == 0 && transport->xprt.resvport)
1729                 port = xs_get_random_port();
1730         return port;
1731 }
1732
1733 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1734 {
1735         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1736         unsigned short ret = 0;
1737         mutex_lock(&sock->recv_mutex);
1738         if (sock->sock)
1739                 ret = xs_sock_getport(sock->sock);
1740         mutex_unlock(&sock->recv_mutex);
1741         return ret;
1742 }
1743
1744 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1745 {
1746         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1747         union {
1748                 struct sockaddr sa;
1749                 struct sockaddr_storage st;
1750         } saddr;
1751         int ret = -ENOTCONN;
1752
1753         mutex_lock(&sock->recv_mutex);
1754         if (sock->sock) {
1755                 ret = kernel_getsockname(sock->sock, &saddr.sa);
1756                 if (ret >= 0)
1757                         ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1758         }
1759         mutex_unlock(&sock->recv_mutex);
1760         return ret;
1761 }
1762
1763 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1764 {
1765         if (transport->srcport != 0)
1766                 transport->srcport = 0;
1767         if (!transport->xprt.resvport)
1768                 return 0;
1769         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1770                 return xprt_max_resvport;
1771         return --port;
1772 }
1773 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1774 {
1775         struct sockaddr_storage myaddr;
1776         int err, nloop = 0;
1777         int port = xs_get_srcport(transport);
1778         unsigned short last;
1779
1780         /*
1781          * If we are asking for any ephemeral port (i.e. port == 0 &&
1782          * transport->xprt.resvport == 0), don't bind.  Let the local
1783          * port selection happen implicitly when the socket is used
1784          * (for example at connect time).
1785          *
1786          * This ensures that we can continue to establish TCP
1787          * connections even when all local ephemeral ports are already
1788          * a part of some TCP connection.  This makes no difference
1789          * for UDP sockets, but also doesn't harm them.
1790          *
1791          * If we're asking for any reserved port (i.e. port == 0 &&
1792          * transport->xprt.resvport == 1) xs_get_srcport above will
1793          * ensure that port is non-zero and we will bind as needed.
1794          */
1795         if (port <= 0)
1796                 return port;
1797
1798         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1799         do {
1800                 rpc_set_port((struct sockaddr *)&myaddr, port);
1801                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1802                                 transport->xprt.addrlen);
1803                 if (err == 0) {
1804                         if (transport->xprt.reuseport)
1805                                 transport->srcport = port;
1806                         break;
1807                 }
1808                 last = port;
1809                 port = xs_next_srcport(transport, port);
1810                 if (port > last)
1811                         nloop++;
1812         } while (err == -EADDRINUSE && nloop != 2);
1813
1814         if (myaddr.ss_family == AF_INET)
1815                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1816                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1817                                 port, err ? "failed" : "ok", err);
1818         else
1819                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1820                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1821                                 port, err ? "failed" : "ok", err);
1822         return err;
1823 }
1824
1825 /*
1826  * We don't support autobind on AF_LOCAL sockets
1827  */
1828 static void xs_local_rpcbind(struct rpc_task *task)
1829 {
1830         xprt_set_bound(task->tk_xprt);
1831 }
1832
1833 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1834 {
1835 }
1836
1837 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1838 static struct lock_class_key xs_key[3];
1839 static struct lock_class_key xs_slock_key[3];
1840
1841 static inline void xs_reclassify_socketu(struct socket *sock)
1842 {
1843         struct sock *sk = sock->sk;
1844
1845         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1846                 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1847 }
1848
1849 static inline void xs_reclassify_socket4(struct socket *sock)
1850 {
1851         struct sock *sk = sock->sk;
1852
1853         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1854                 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1855 }
1856
1857 static inline void xs_reclassify_socket6(struct socket *sock)
1858 {
1859         struct sock *sk = sock->sk;
1860
1861         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1862                 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1863 }
1864
1865 static inline void xs_reclassify_socket(int family, struct socket *sock)
1866 {
1867         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1868                 return;
1869
1870         switch (family) {
1871         case AF_LOCAL:
1872                 xs_reclassify_socketu(sock);
1873                 break;
1874         case AF_INET:
1875                 xs_reclassify_socket4(sock);
1876                 break;
1877         case AF_INET6:
1878                 xs_reclassify_socket6(sock);
1879                 break;
1880         }
1881 }
1882 #else
1883 static inline void xs_reclassify_socket(int family, struct socket *sock)
1884 {
1885 }
1886 #endif
1887
1888 static void xs_dummy_setup_socket(struct work_struct *work)
1889 {
1890 }
1891
1892 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1893                 struct sock_xprt *transport, int family, int type,
1894                 int protocol, bool reuseport)
1895 {
1896         struct file *filp;
1897         struct socket *sock;
1898         int err;
1899
1900         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1901         if (err < 0) {
1902                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1903                                 protocol, -err);
1904                 goto out;
1905         }
1906         xs_reclassify_socket(family, sock);
1907
1908         if (reuseport)
1909                 sock_set_reuseport(sock->sk);
1910
1911         err = xs_bind(transport, sock);
1912         if (err) {
1913                 sock_release(sock);
1914                 goto out;
1915         }
1916
1917         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1918         if (IS_ERR(filp))
1919                 return ERR_CAST(filp);
1920         transport->file = filp;
1921
1922         return sock;
1923 out:
1924         return ERR_PTR(err);
1925 }
1926
1927 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1928                                       struct socket *sock)
1929 {
1930         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1931                                                                         xprt);
1932
1933         if (!transport->inet) {
1934                 struct sock *sk = sock->sk;
1935
1936                 lock_sock(sk);
1937
1938                 xs_save_old_callbacks(transport, sk);
1939
1940                 sk->sk_user_data = xprt;
1941                 sk->sk_data_ready = xs_data_ready;
1942                 sk->sk_write_space = xs_udp_write_space;
1943                 sk->sk_state_change = xs_local_state_change;
1944                 sk->sk_error_report = xs_error_report;
1945                 sk->sk_use_task_frag = false;
1946
1947                 xprt_clear_connected(xprt);
1948
1949                 /* Reset to new socket */
1950                 transport->sock = sock;
1951                 transport->inet = sk;
1952
1953                 release_sock(sk);
1954         }
1955
1956         xs_stream_start_connect(transport);
1957
1958         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1959 }
1960
1961 /**
1962  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1963  * @transport: socket transport to connect
1964  */
1965 static int xs_local_setup_socket(struct sock_xprt *transport)
1966 {
1967         struct rpc_xprt *xprt = &transport->xprt;
1968         struct file *filp;
1969         struct socket *sock;
1970         int status;
1971
1972         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1973                                         SOCK_STREAM, 0, &sock, 1);
1974         if (status < 0) {
1975                 dprintk("RPC:       can't create AF_LOCAL "
1976                         "transport socket (%d).\n", -status);
1977                 goto out;
1978         }
1979         xs_reclassify_socket(AF_LOCAL, sock);
1980
1981         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1982         if (IS_ERR(filp)) {
1983                 status = PTR_ERR(filp);
1984                 goto out;
1985         }
1986         transport->file = filp;
1987
1988         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1989                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1990
1991         status = xs_local_finish_connecting(xprt, sock);
1992         trace_rpc_socket_connect(xprt, sock, status);
1993         switch (status) {
1994         case 0:
1995                 dprintk("RPC:       xprt %p connected to %s\n",
1996                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1997                 xprt->stat.connect_count++;
1998                 xprt->stat.connect_time += (long)jiffies -
1999                                            xprt->stat.connect_start;
2000                 xprt_set_connected(xprt);
2001                 break;
2002         case -ENOBUFS:
2003                 break;
2004         case -ENOENT:
2005                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2006                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2007                 break;
2008         case -ECONNREFUSED:
2009                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2010                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2011                 break;
2012         default:
2013                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2014                                 __func__, -status,
2015                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2016         }
2017
2018 out:
2019         xprt_clear_connecting(xprt);
2020         xprt_wake_pending_tasks(xprt, status);
2021         return status;
2022 }
2023
2024 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2025 {
2026         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2027         int ret;
2028
2029         if (transport->file)
2030                 goto force_disconnect;
2031
2032         if (RPC_IS_ASYNC(task)) {
2033                 /*
2034                  * We want the AF_LOCAL connect to be resolved in the
2035                  * filesystem namespace of the process making the rpc
2036                  * call.  Thus we connect synchronously.
2037                  *
2038                  * If we want to support asynchronous AF_LOCAL calls,
2039                  * we'll need to figure out how to pass a namespace to
2040                  * connect.
2041                  */
2042                 rpc_task_set_rpc_status(task, -ENOTCONN);
2043                 goto out_wake;
2044         }
2045         ret = xs_local_setup_socket(transport);
2046         if (ret && !RPC_IS_SOFTCONN(task))
2047                 msleep_interruptible(15000);
2048         return;
2049 force_disconnect:
2050         xprt_force_disconnect(xprt);
2051 out_wake:
2052         xprt_clear_connecting(xprt);
2053         xprt_wake_pending_tasks(xprt, -ENOTCONN);
2054 }
2055
2056 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2057 /*
2058  * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2059  * held, or when we otherwise know that we have exclusive access to the
2060  * socket, to guard against races with xs_reset_transport.
2061  */
2062 static void xs_set_memalloc(struct rpc_xprt *xprt)
2063 {
2064         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2065                         xprt);
2066
2067         /*
2068          * If there's no sock, then we have nothing to set. The
2069          * reconnecting process will get it for us.
2070          */
2071         if (!transport->inet)
2072                 return;
2073         if (atomic_read(&xprt->swapper))
2074                 sk_set_memalloc(transport->inet);
2075 }
2076
2077 /**
2078  * xs_enable_swap - Tag this transport as being used for swap.
2079  * @xprt: transport to tag
2080  *
2081  * Take a reference to this transport on behalf of the rpc_clnt, and
2082  * optionally mark it for swapping if it wasn't already.
2083  */
2084 static int
2085 xs_enable_swap(struct rpc_xprt *xprt)
2086 {
2087         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2088
2089         mutex_lock(&xs->recv_mutex);
2090         if (atomic_inc_return(&xprt->swapper) == 1 &&
2091             xs->inet)
2092                 sk_set_memalloc(xs->inet);
2093         mutex_unlock(&xs->recv_mutex);
2094         return 0;
2095 }
2096
2097 /**
2098  * xs_disable_swap - Untag this transport as being used for swap.
2099  * @xprt: transport to tag
2100  *
2101  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2102  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2103  */
2104 static void
2105 xs_disable_swap(struct rpc_xprt *xprt)
2106 {
2107         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2108
2109         mutex_lock(&xs->recv_mutex);
2110         if (atomic_dec_and_test(&xprt->swapper) &&
2111             xs->inet)
2112                 sk_clear_memalloc(xs->inet);
2113         mutex_unlock(&xs->recv_mutex);
2114 }
2115 #else
2116 static void xs_set_memalloc(struct rpc_xprt *xprt)
2117 {
2118 }
2119
2120 static int
2121 xs_enable_swap(struct rpc_xprt *xprt)
2122 {
2123         return -EINVAL;
2124 }
2125
2126 static void
2127 xs_disable_swap(struct rpc_xprt *xprt)
2128 {
2129 }
2130 #endif
2131
2132 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2133 {
2134         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2135
2136         if (!transport->inet) {
2137                 struct sock *sk = sock->sk;
2138
2139                 lock_sock(sk);
2140
2141                 xs_save_old_callbacks(transport, sk);
2142
2143                 sk->sk_user_data = xprt;
2144                 sk->sk_data_ready = xs_data_ready;
2145                 sk->sk_write_space = xs_udp_write_space;
2146                 sk->sk_use_task_frag = false;
2147
2148                 xprt_set_connected(xprt);
2149
2150                 /* Reset to new socket */
2151                 transport->sock = sock;
2152                 transport->inet = sk;
2153
2154                 xs_set_memalloc(xprt);
2155
2156                 release_sock(sk);
2157         }
2158         xs_udp_do_set_buffer_size(xprt);
2159
2160         xprt->stat.connect_start = jiffies;
2161 }
2162
2163 static void xs_udp_setup_socket(struct work_struct *work)
2164 {
2165         struct sock_xprt *transport =
2166                 container_of(work, struct sock_xprt, connect_worker.work);
2167         struct rpc_xprt *xprt = &transport->xprt;
2168         struct socket *sock;
2169         int status = -EIO;
2170         unsigned int pflags = current->flags;
2171
2172         if (atomic_read(&xprt->swapper))
2173                 current->flags |= PF_MEMALLOC;
2174         sock = xs_create_sock(xprt, transport,
2175                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2176                         IPPROTO_UDP, false);
2177         if (IS_ERR(sock))
2178                 goto out;
2179
2180         dprintk("RPC:       worker connecting xprt %p via %s to "
2181                                 "%s (port %s)\n", xprt,
2182                         xprt->address_strings[RPC_DISPLAY_PROTO],
2183                         xprt->address_strings[RPC_DISPLAY_ADDR],
2184                         xprt->address_strings[RPC_DISPLAY_PORT]);
2185
2186         xs_udp_finish_connecting(xprt, sock);
2187         trace_rpc_socket_connect(xprt, sock, 0);
2188         status = 0;
2189 out:
2190         xprt_clear_connecting(xprt);
2191         xprt_unlock_connect(xprt, transport);
2192         xprt_wake_pending_tasks(xprt, status);
2193         current_restore_flags(pflags, PF_MEMALLOC);
2194 }
2195
2196 /**
2197  * xs_tcp_shutdown - gracefully shut down a TCP socket
2198  * @xprt: transport
2199  *
2200  * Initiates a graceful shutdown of the TCP socket by calling the
2201  * equivalent of shutdown(SHUT_RDWR);
2202  */
2203 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2204 {
2205         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2206         struct socket *sock = transport->sock;
2207         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2208
2209         if (sock == NULL)
2210                 return;
2211         if (!xprt->reuseport) {
2212                 xs_close(xprt);
2213                 return;
2214         }
2215         switch (skst) {
2216         case TCP_FIN_WAIT1:
2217         case TCP_FIN_WAIT2:
2218         case TCP_LAST_ACK:
2219                 break;
2220         case TCP_ESTABLISHED:
2221         case TCP_CLOSE_WAIT:
2222                 kernel_sock_shutdown(sock, SHUT_RDWR);
2223                 trace_rpc_socket_shutdown(xprt, sock);
2224                 break;
2225         default:
2226                 xs_reset_transport(transport);
2227         }
2228 }
2229
2230 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2231                 struct socket *sock)
2232 {
2233         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2234         unsigned int keepidle;
2235         unsigned int keepcnt;
2236         unsigned int timeo;
2237
2238         spin_lock(&xprt->transport_lock);
2239         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2240         keepcnt = xprt->timeout->to_retries + 1;
2241         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2242                 (xprt->timeout->to_retries + 1);
2243         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2244         spin_unlock(&xprt->transport_lock);
2245
2246         /* TCP Keepalive options */
2247         sock_set_keepalive(sock->sk);
2248         tcp_sock_set_keepidle(sock->sk, keepidle);
2249         tcp_sock_set_keepintvl(sock->sk, keepidle);
2250         tcp_sock_set_keepcnt(sock->sk, keepcnt);
2251
2252         /* TCP user timeout (see RFC5482) */
2253         tcp_sock_set_user_timeout(sock->sk, timeo);
2254 }
2255
2256 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2257                 unsigned long connect_timeout,
2258                 unsigned long reconnect_timeout)
2259 {
2260         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2261         struct rpc_timeout to;
2262         unsigned long initval;
2263
2264         spin_lock(&xprt->transport_lock);
2265         if (reconnect_timeout < xprt->max_reconnect_timeout)
2266                 xprt->max_reconnect_timeout = reconnect_timeout;
2267         if (connect_timeout < xprt->connect_timeout) {
2268                 memcpy(&to, xprt->timeout, sizeof(to));
2269                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2270                 /* Arbitrary lower limit */
2271                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2272                         initval = XS_TCP_INIT_REEST_TO << 1;
2273                 to.to_initval = initval;
2274                 to.to_maxval = initval;
2275                 memcpy(&transport->tcp_timeout, &to,
2276                                 sizeof(transport->tcp_timeout));
2277                 xprt->timeout = &transport->tcp_timeout;
2278                 xprt->connect_timeout = connect_timeout;
2279         }
2280         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2281         spin_unlock(&xprt->transport_lock);
2282 }
2283
2284 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2285 {
2286         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2287
2288         if (!transport->inet) {
2289                 struct sock *sk = sock->sk;
2290
2291                 /* Avoid temporary address, they are bad for long-lived
2292                  * connections such as NFS mounts.
2293                  * RFC4941, section 3.6 suggests that:
2294                  *    Individual applications, which have specific
2295                  *    knowledge about the normal duration of connections,
2296                  *    MAY override this as appropriate.
2297                  */
2298                 if (xs_addr(xprt)->sa_family == PF_INET6) {
2299                         ip6_sock_set_addr_preferences(sk,
2300                                 IPV6_PREFER_SRC_PUBLIC);
2301                 }
2302
2303                 xs_tcp_set_socket_timeouts(xprt, sock);
2304                 tcp_sock_set_nodelay(sk);
2305
2306                 lock_sock(sk);
2307
2308                 xs_save_old_callbacks(transport, sk);
2309
2310                 sk->sk_user_data = xprt;
2311                 sk->sk_data_ready = xs_data_ready;
2312                 sk->sk_state_change = xs_tcp_state_change;
2313                 sk->sk_write_space = xs_tcp_write_space;
2314                 sk->sk_error_report = xs_error_report;
2315                 sk->sk_use_task_frag = false;
2316
2317                 /* socket options */
2318                 sock_reset_flag(sk, SOCK_LINGER);
2319
2320                 xprt_clear_connected(xprt);
2321
2322                 /* Reset to new socket */
2323                 transport->sock = sock;
2324                 transport->inet = sk;
2325
2326                 release_sock(sk);
2327         }
2328
2329         if (!xprt_bound(xprt))
2330                 return -ENOTCONN;
2331
2332         xs_set_memalloc(xprt);
2333
2334         xs_stream_start_connect(transport);
2335
2336         /* Tell the socket layer to start connecting... */
2337         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2338         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2339 }
2340
2341 /**
2342  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2343  * @work: queued work item
2344  *
2345  * Invoked by a work queue tasklet.
2346  */
2347 static void xs_tcp_setup_socket(struct work_struct *work)
2348 {
2349         struct sock_xprt *transport =
2350                 container_of(work, struct sock_xprt, connect_worker.work);
2351         struct socket *sock = transport->sock;
2352         struct rpc_xprt *xprt = &transport->xprt;
2353         int status;
2354         unsigned int pflags = current->flags;
2355
2356         if (atomic_read(&xprt->swapper))
2357                 current->flags |= PF_MEMALLOC;
2358
2359         if (xprt_connected(xprt))
2360                 goto out;
2361         if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2362                                &transport->sock_state) ||
2363             !sock) {
2364                 xs_reset_transport(transport);
2365                 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2366                                       SOCK_STREAM, IPPROTO_TCP, true);
2367                 if (IS_ERR(sock)) {
2368                         xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2369                         goto out;
2370                 }
2371         }
2372
2373         dprintk("RPC:       worker connecting xprt %p via %s to "
2374                                 "%s (port %s)\n", xprt,
2375                         xprt->address_strings[RPC_DISPLAY_PROTO],
2376                         xprt->address_strings[RPC_DISPLAY_ADDR],
2377                         xprt->address_strings[RPC_DISPLAY_PORT]);
2378
2379         status = xs_tcp_finish_connecting(xprt, sock);
2380         trace_rpc_socket_connect(xprt, sock, status);
2381         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2382                         xprt, -status, xprt_connected(xprt),
2383                         sock->sk->sk_state);
2384         switch (status) {
2385         case 0:
2386         case -EINPROGRESS:
2387                 /* SYN_SENT! */
2388                 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2389                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2390                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2391                 fallthrough;
2392         case -EALREADY:
2393                 goto out_unlock;
2394         case -EADDRNOTAVAIL:
2395                 /* Source port number is unavailable. Try a new one! */
2396                 transport->srcport = 0;
2397                 status = -EAGAIN;
2398                 break;
2399         case -EINVAL:
2400                 /* Happens, for instance, if the user specified a link
2401                  * local IPv6 address without a scope-id.
2402                  */
2403         case -ECONNREFUSED:
2404         case -ECONNRESET:
2405         case -ENETDOWN:
2406         case -ENETUNREACH:
2407         case -EHOSTUNREACH:
2408         case -EADDRINUSE:
2409         case -ENOBUFS:
2410                 break;
2411         default:
2412                 printk("%s: connect returned unhandled error %d\n",
2413                         __func__, status);
2414                 status = -EAGAIN;
2415         }
2416
2417         /* xs_tcp_force_close() wakes tasks with a fixed error code.
2418          * We need to wake them first to ensure the correct error code.
2419          */
2420         xprt_wake_pending_tasks(xprt, status);
2421         xs_tcp_force_close(xprt);
2422 out:
2423         xprt_clear_connecting(xprt);
2424 out_unlock:
2425         xprt_unlock_connect(xprt, transport);
2426         current_restore_flags(pflags, PF_MEMALLOC);
2427 }
2428
2429 /*
2430  * Transfer the connected socket to @upper_transport, then mark that
2431  * xprt CONNECTED.
2432  */
2433 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2434                                         struct sock_xprt *upper_transport)
2435 {
2436         struct sock_xprt *lower_transport =
2437                         container_of(lower_xprt, struct sock_xprt, xprt);
2438         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2439
2440         if (!upper_transport->inet) {
2441                 struct socket *sock = lower_transport->sock;
2442                 struct sock *sk = sock->sk;
2443
2444                 /* Avoid temporary address, they are bad for long-lived
2445                  * connections such as NFS mounts.
2446                  * RFC4941, section 3.6 suggests that:
2447                  *    Individual applications, which have specific
2448                  *    knowledge about the normal duration of connections,
2449                  *    MAY override this as appropriate.
2450                  */
2451                 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2452                         ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2453
2454                 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2455                 tcp_sock_set_nodelay(sk);
2456
2457                 lock_sock(sk);
2458
2459                 /* @sk is already connected, so it now has the RPC callbacks.
2460                  * Reach into @lower_transport to save the original ones.
2461                  */
2462                 upper_transport->old_data_ready = lower_transport->old_data_ready;
2463                 upper_transport->old_state_change = lower_transport->old_state_change;
2464                 upper_transport->old_write_space = lower_transport->old_write_space;
2465                 upper_transport->old_error_report = lower_transport->old_error_report;
2466                 sk->sk_user_data = upper_xprt;
2467
2468                 /* socket options */
2469                 sock_reset_flag(sk, SOCK_LINGER);
2470
2471                 xprt_clear_connected(upper_xprt);
2472
2473                 upper_transport->sock = sock;
2474                 upper_transport->inet = sk;
2475                 upper_transport->file = lower_transport->file;
2476
2477                 release_sock(sk);
2478
2479                 /* Reset lower_transport before shutting down its clnt */
2480                 mutex_lock(&lower_transport->recv_mutex);
2481                 lower_transport->inet = NULL;
2482                 lower_transport->sock = NULL;
2483                 lower_transport->file = NULL;
2484
2485                 xprt_clear_connected(lower_xprt);
2486                 xs_sock_reset_connection_flags(lower_xprt);
2487                 xs_stream_reset_connect(lower_transport);
2488                 mutex_unlock(&lower_transport->recv_mutex);
2489         }
2490
2491         if (!xprt_bound(upper_xprt))
2492                 return -ENOTCONN;
2493
2494         xs_set_memalloc(upper_xprt);
2495
2496         if (!xprt_test_and_set_connected(upper_xprt)) {
2497                 upper_xprt->connect_cookie++;
2498                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2499                 xprt_clear_connecting(upper_xprt);
2500
2501                 upper_xprt->stat.connect_count++;
2502                 upper_xprt->stat.connect_time += (long)jiffies -
2503                                            upper_xprt->stat.connect_start;
2504                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2505         }
2506         return 0;
2507 }
2508
2509 /**
2510  * xs_tls_handshake_done - TLS handshake completion handler
2511  * @data: address of xprt to wake
2512  * @status: status of handshake
2513  * @peerid: serial number of key containing the remote's identity
2514  *
2515  */
2516 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2517 {
2518         struct rpc_xprt *lower_xprt = data;
2519         struct sock_xprt *lower_transport =
2520                                 container_of(lower_xprt, struct sock_xprt, xprt);
2521
2522         lower_transport->xprt_err = status ? -EACCES : 0;
2523         complete(&lower_transport->handshake_done);
2524         xprt_put(lower_xprt);
2525 }
2526
2527 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2528 {
2529         struct sock_xprt *lower_transport =
2530                                 container_of(lower_xprt, struct sock_xprt, xprt);
2531         struct tls_handshake_args args = {
2532                 .ta_sock        = lower_transport->sock,
2533                 .ta_done        = xs_tls_handshake_done,
2534                 .ta_data        = xprt_get(lower_xprt),
2535                 .ta_peername    = lower_xprt->servername,
2536         };
2537         struct sock *sk = lower_transport->inet;
2538         int rc;
2539
2540         init_completion(&lower_transport->handshake_done);
2541         set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2542         lower_transport->xprt_err = -ETIMEDOUT;
2543         switch (xprtsec->policy) {
2544         case RPC_XPRTSEC_TLS_ANON:
2545                 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2546                 if (rc)
2547                         goto out_put_xprt;
2548                 break;
2549         case RPC_XPRTSEC_TLS_X509:
2550                 args.ta_my_cert = xprtsec->cert_serial;
2551                 args.ta_my_privkey = xprtsec->privkey_serial;
2552                 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2553                 if (rc)
2554                         goto out_put_xprt;
2555                 break;
2556         default:
2557                 rc = -EACCES;
2558                 goto out_put_xprt;
2559         }
2560
2561         rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2562                                                        XS_TLS_HANDSHAKE_TO);
2563         if (rc <= 0) {
2564                 if (!tls_handshake_cancel(sk)) {
2565                         if (rc == 0)
2566                                 rc = -ETIMEDOUT;
2567                         goto out_put_xprt;
2568                 }
2569         }
2570
2571         rc = lower_transport->xprt_err;
2572
2573 out:
2574         xs_stream_reset_connect(lower_transport);
2575         clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2576         return rc;
2577
2578 out_put_xprt:
2579         xprt_put(lower_xprt);
2580         goto out;
2581 }
2582
2583 /**
2584  * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2585  * @work: queued work item
2586  *
2587  * Invoked by a work queue tasklet.
2588  *
2589  * For RPC-with-TLS, there is a two-stage connection process.
2590  *
2591  * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2592  * been marked connected, the consumer knows that a TCP connection and
2593  * a TLS session have been established.
2594  *
2595  * A "lower-layer xprt", created in this function, handles the mechanics
2596  * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2597  * then driving the TLS handshake. Once all that is complete, the upper
2598  * layer xprt is marked connected.
2599  */
2600 static void xs_tcp_tls_setup_socket(struct work_struct *work)
2601 {
2602         struct sock_xprt *upper_transport =
2603                 container_of(work, struct sock_xprt, connect_worker.work);
2604         struct rpc_clnt *upper_clnt = upper_transport->clnt;
2605         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2606         struct rpc_create_args args = {
2607                 .net            = upper_xprt->xprt_net,
2608                 .protocol       = upper_xprt->prot,
2609                 .address        = (struct sockaddr *)&upper_xprt->addr,
2610                 .addrsize       = upper_xprt->addrlen,
2611                 .timeout        = upper_clnt->cl_timeout,
2612                 .servername     = upper_xprt->servername,
2613                 .program        = upper_clnt->cl_program,
2614                 .prognumber     = upper_clnt->cl_prog,
2615                 .version        = upper_clnt->cl_vers,
2616                 .authflavor     = RPC_AUTH_TLS,
2617                 .cred           = upper_clnt->cl_cred,
2618                 .xprtsec        = {
2619                         .policy         = RPC_XPRTSEC_NONE,
2620                 },
2621         };
2622         unsigned int pflags = current->flags;
2623         struct rpc_clnt *lower_clnt;
2624         struct rpc_xprt *lower_xprt;
2625         int status;
2626
2627         if (atomic_read(&upper_xprt->swapper))
2628                 current->flags |= PF_MEMALLOC;
2629
2630         xs_stream_start_connect(upper_transport);
2631
2632         /* This implicitly sends an RPC_AUTH_TLS probe */
2633         lower_clnt = rpc_create(&args);
2634         if (IS_ERR(lower_clnt)) {
2635                 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2636                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2637                 xprt_clear_connecting(upper_xprt);
2638                 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2639                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2640                 goto out_unlock;
2641         }
2642
2643         /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2644          * the lower xprt.
2645          */
2646         rcu_read_lock();
2647         lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2648         rcu_read_unlock();
2649         status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2650         if (status) {
2651                 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2652                 goto out_close;
2653         }
2654
2655         status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2656         if (status)
2657                 goto out_close;
2658
2659         trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2660         if (!xprt_test_and_set_connected(upper_xprt)) {
2661                 upper_xprt->connect_cookie++;
2662                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2663                 xprt_clear_connecting(upper_xprt);
2664
2665                 upper_xprt->stat.connect_count++;
2666                 upper_xprt->stat.connect_time += (long)jiffies -
2667                                            upper_xprt->stat.connect_start;
2668                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2669         }
2670         rpc_shutdown_client(lower_clnt);
2671
2672 out_unlock:
2673         current_restore_flags(pflags, PF_MEMALLOC);
2674         upper_transport->clnt = NULL;
2675         xprt_unlock_connect(upper_xprt, upper_transport);
2676         return;
2677
2678 out_close:
2679         rpc_shutdown_client(lower_clnt);
2680
2681         /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2682          * Wake them first here to ensure they get our tk_status code.
2683          */
2684         xprt_wake_pending_tasks(upper_xprt, status);
2685         xs_tcp_force_close(upper_xprt);
2686         xprt_clear_connecting(upper_xprt);
2687         goto out_unlock;
2688 }
2689
2690 /**
2691  * xs_connect - connect a socket to a remote endpoint
2692  * @xprt: pointer to transport structure
2693  * @task: address of RPC task that manages state of connect request
2694  *
2695  * TCP: If the remote end dropped the connection, delay reconnecting.
2696  *
2697  * UDP socket connects are synchronous, but we use a work queue anyway
2698  * to guarantee that even unprivileged user processes can set up a
2699  * socket on a privileged port.
2700  *
2701  * If a UDP socket connect fails, the delay behavior here prevents
2702  * retry floods (hard mounts).
2703  */
2704 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2705 {
2706         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2707         unsigned long delay = 0;
2708
2709         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2710
2711         if (transport->sock != NULL) {
2712                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2713                         "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2714
2715                 delay = xprt_reconnect_delay(xprt);
2716                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2717
2718         } else
2719                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2720
2721         transport->clnt = task->tk_client;
2722         queue_delayed_work(xprtiod_workqueue,
2723                         &transport->connect_worker,
2724                         delay);
2725 }
2726
2727 static void xs_wake_disconnect(struct sock_xprt *transport)
2728 {
2729         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2730                 xs_tcp_force_close(&transport->xprt);
2731 }
2732
2733 static void xs_wake_write(struct sock_xprt *transport)
2734 {
2735         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2736                 xprt_write_space(&transport->xprt);
2737 }
2738
2739 static void xs_wake_error(struct sock_xprt *transport)
2740 {
2741         int sockerr;
2742
2743         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2744                 return;
2745         mutex_lock(&transport->recv_mutex);
2746         if (transport->sock == NULL)
2747                 goto out;
2748         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2749                 goto out;
2750         sockerr = xchg(&transport->xprt_err, 0);
2751         if (sockerr < 0)
2752                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2753 out:
2754         mutex_unlock(&transport->recv_mutex);
2755 }
2756
2757 static void xs_wake_pending(struct sock_xprt *transport)
2758 {
2759         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2760                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2761 }
2762
2763 static void xs_error_handle(struct work_struct *work)
2764 {
2765         struct sock_xprt *transport = container_of(work,
2766                         struct sock_xprt, error_worker);
2767
2768         xs_wake_disconnect(transport);
2769         xs_wake_write(transport);
2770         xs_wake_error(transport);
2771         xs_wake_pending(transport);
2772 }
2773
2774 /**
2775  * xs_local_print_stats - display AF_LOCAL socket-specific stats
2776  * @xprt: rpc_xprt struct containing statistics
2777  * @seq: output file
2778  *
2779  */
2780 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2781 {
2782         long idle_time = 0;
2783
2784         if (xprt_connected(xprt))
2785                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2786
2787         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2788                         "%llu %llu %lu %llu %llu\n",
2789                         xprt->stat.bind_count,
2790                         xprt->stat.connect_count,
2791                         xprt->stat.connect_time / HZ,
2792                         idle_time,
2793                         xprt->stat.sends,
2794                         xprt->stat.recvs,
2795                         xprt->stat.bad_xids,
2796                         xprt->stat.req_u,
2797                         xprt->stat.bklog_u,
2798                         xprt->stat.max_slots,
2799                         xprt->stat.sending_u,
2800                         xprt->stat.pending_u);
2801 }
2802
2803 /**
2804  * xs_udp_print_stats - display UDP socket-specific stats
2805  * @xprt: rpc_xprt struct containing statistics
2806  * @seq: output file
2807  *
2808  */
2809 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2810 {
2811         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2812
2813         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2814                         "%lu %llu %llu\n",
2815                         transport->srcport,
2816                         xprt->stat.bind_count,
2817                         xprt->stat.sends,
2818                         xprt->stat.recvs,
2819                         xprt->stat.bad_xids,
2820                         xprt->stat.req_u,
2821                         xprt->stat.bklog_u,
2822                         xprt->stat.max_slots,
2823                         xprt->stat.sending_u,
2824                         xprt->stat.pending_u);
2825 }
2826
2827 /**
2828  * xs_tcp_print_stats - display TCP socket-specific stats
2829  * @xprt: rpc_xprt struct containing statistics
2830  * @seq: output file
2831  *
2832  */
2833 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2834 {
2835         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2836         long idle_time = 0;
2837
2838         if (xprt_connected(xprt))
2839                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2840
2841         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2842                         "%llu %llu %lu %llu %llu\n",
2843                         transport->srcport,
2844                         xprt->stat.bind_count,
2845                         xprt->stat.connect_count,
2846                         xprt->stat.connect_time / HZ,
2847                         idle_time,
2848                         xprt->stat.sends,
2849                         xprt->stat.recvs,
2850                         xprt->stat.bad_xids,
2851                         xprt->stat.req_u,
2852                         xprt->stat.bklog_u,
2853                         xprt->stat.max_slots,
2854                         xprt->stat.sending_u,
2855                         xprt->stat.pending_u);
2856 }
2857
2858 /*
2859  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2860  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2861  * to use the server side send routines.
2862  */
2863 static int bc_malloc(struct rpc_task *task)
2864 {
2865         struct rpc_rqst *rqst = task->tk_rqstp;
2866         size_t size = rqst->rq_callsize;
2867         struct page *page;
2868         struct rpc_buffer *buf;
2869
2870         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2871                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2872                           size);
2873                 return -EINVAL;
2874         }
2875
2876         page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2877         if (!page)
2878                 return -ENOMEM;
2879
2880         buf = page_address(page);
2881         buf->len = PAGE_SIZE;
2882
2883         rqst->rq_buffer = buf->data;
2884         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2885         return 0;
2886 }
2887
2888 /*
2889  * Free the space allocated in the bc_alloc routine
2890  */
2891 static void bc_free(struct rpc_task *task)
2892 {
2893         void *buffer = task->tk_rqstp->rq_buffer;
2894         struct rpc_buffer *buf;
2895
2896         buf = container_of(buffer, struct rpc_buffer, data);
2897         free_page((unsigned long)buf);
2898 }
2899
2900 static int bc_sendto(struct rpc_rqst *req)
2901 {
2902         struct xdr_buf *xdr = &req->rq_snd_buf;
2903         struct sock_xprt *transport =
2904                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2905         struct msghdr msg = {
2906                 .msg_flags      = 0,
2907         };
2908         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2909                                          (u32)xdr->len);
2910         unsigned int sent = 0;
2911         int err;
2912
2913         req->rq_xtime = ktime_get();
2914         err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2915         if (err < 0)
2916                 return err;
2917         err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2918         xdr_free_bvec(xdr);
2919         if (err < 0 || sent != (xdr->len + sizeof(marker)))
2920                 return -EAGAIN;
2921         return sent;
2922 }
2923
2924 /**
2925  * bc_send_request - Send a backchannel Call on a TCP socket
2926  * @req: rpc_rqst containing Call message to be sent
2927  *
2928  * xpt_mutex ensures @rqstp's whole message is written to the socket
2929  * without interruption.
2930  *
2931  * Return values:
2932  *   %0 if the message was sent successfully
2933  *   %ENOTCONN if the message was not sent
2934  */
2935 static int bc_send_request(struct rpc_rqst *req)
2936 {
2937         struct svc_xprt *xprt;
2938         int len;
2939
2940         /*
2941          * Get the server socket associated with this callback xprt
2942          */
2943         xprt = req->rq_xprt->bc_xprt;
2944
2945         /*
2946          * Grab the mutex to serialize data as the connection is shared
2947          * with the fore channel
2948          */
2949         mutex_lock(&xprt->xpt_mutex);
2950         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2951                 len = -ENOTCONN;
2952         else
2953                 len = bc_sendto(req);
2954         mutex_unlock(&xprt->xpt_mutex);
2955
2956         if (len > 0)
2957                 len = 0;
2958
2959         return len;
2960 }
2961
2962 /*
2963  * The close routine. Since this is client initiated, we do nothing
2964  */
2965
2966 static void bc_close(struct rpc_xprt *xprt)
2967 {
2968         xprt_disconnect_done(xprt);
2969 }
2970
2971 /*
2972  * The xprt destroy routine. Again, because this connection is client
2973  * initiated, we do nothing
2974  */
2975
2976 static void bc_destroy(struct rpc_xprt *xprt)
2977 {
2978         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2979
2980         xs_xprt_free(xprt);
2981         module_put(THIS_MODULE);
2982 }
2983
2984 static const struct rpc_xprt_ops xs_local_ops = {
2985         .reserve_xprt           = xprt_reserve_xprt,
2986         .release_xprt           = xprt_release_xprt,
2987         .alloc_slot             = xprt_alloc_slot,
2988         .free_slot              = xprt_free_slot,
2989         .rpcbind                = xs_local_rpcbind,
2990         .set_port               = xs_local_set_port,
2991         .connect                = xs_local_connect,
2992         .buf_alloc              = rpc_malloc,
2993         .buf_free               = rpc_free,
2994         .prepare_request        = xs_stream_prepare_request,
2995         .send_request           = xs_local_send_request,
2996         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2997         .close                  = xs_close,
2998         .destroy                = xs_destroy,
2999         .print_stats            = xs_local_print_stats,
3000         .enable_swap            = xs_enable_swap,
3001         .disable_swap           = xs_disable_swap,
3002 };
3003
3004 static const struct rpc_xprt_ops xs_udp_ops = {
3005         .set_buffer_size        = xs_udp_set_buffer_size,
3006         .reserve_xprt           = xprt_reserve_xprt_cong,
3007         .release_xprt           = xprt_release_xprt_cong,
3008         .alloc_slot             = xprt_alloc_slot,
3009         .free_slot              = xprt_free_slot,
3010         .rpcbind                = rpcb_getport_async,
3011         .set_port               = xs_set_port,
3012         .connect                = xs_connect,
3013         .get_srcaddr            = xs_sock_srcaddr,
3014         .get_srcport            = xs_sock_srcport,
3015         .buf_alloc              = rpc_malloc,
3016         .buf_free               = rpc_free,
3017         .send_request           = xs_udp_send_request,
3018         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3019         .timer                  = xs_udp_timer,
3020         .release_request        = xprt_release_rqst_cong,
3021         .close                  = xs_close,
3022         .destroy                = xs_destroy,
3023         .print_stats            = xs_udp_print_stats,
3024         .enable_swap            = xs_enable_swap,
3025         .disable_swap           = xs_disable_swap,
3026         .inject_disconnect      = xs_inject_disconnect,
3027 };
3028
3029 static const struct rpc_xprt_ops xs_tcp_ops = {
3030         .reserve_xprt           = xprt_reserve_xprt,
3031         .release_xprt           = xprt_release_xprt,
3032         .alloc_slot             = xprt_alloc_slot,
3033         .free_slot              = xprt_free_slot,
3034         .rpcbind                = rpcb_getport_async,
3035         .set_port               = xs_set_port,
3036         .connect                = xs_connect,
3037         .get_srcaddr            = xs_sock_srcaddr,
3038         .get_srcport            = xs_sock_srcport,
3039         .buf_alloc              = rpc_malloc,
3040         .buf_free               = rpc_free,
3041         .prepare_request        = xs_stream_prepare_request,
3042         .send_request           = xs_tcp_send_request,
3043         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3044         .close                  = xs_tcp_shutdown,
3045         .destroy                = xs_destroy,
3046         .set_connect_timeout    = xs_tcp_set_connect_timeout,
3047         .print_stats            = xs_tcp_print_stats,
3048         .enable_swap            = xs_enable_swap,
3049         .disable_swap           = xs_disable_swap,
3050         .inject_disconnect      = xs_inject_disconnect,
3051 #ifdef CONFIG_SUNRPC_BACKCHANNEL
3052         .bc_setup               = xprt_setup_bc,
3053         .bc_maxpayload          = xs_tcp_bc_maxpayload,
3054         .bc_num_slots           = xprt_bc_max_slots,
3055         .bc_free_rqst           = xprt_free_bc_rqst,
3056         .bc_destroy             = xprt_destroy_bc,
3057 #endif
3058 };
3059
3060 /*
3061  * The rpc_xprt_ops for the server backchannel
3062  */
3063
3064 static const struct rpc_xprt_ops bc_tcp_ops = {
3065         .reserve_xprt           = xprt_reserve_xprt,
3066         .release_xprt           = xprt_release_xprt,
3067         .alloc_slot             = xprt_alloc_slot,
3068         .free_slot              = xprt_free_slot,
3069         .buf_alloc              = bc_malloc,
3070         .buf_free               = bc_free,
3071         .send_request           = bc_send_request,
3072         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3073         .close                  = bc_close,
3074         .destroy                = bc_destroy,
3075         .print_stats            = xs_tcp_print_stats,
3076         .enable_swap            = xs_enable_swap,
3077         .disable_swap           = xs_disable_swap,
3078         .inject_disconnect      = xs_inject_disconnect,
3079 };
3080
3081 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3082 {
3083         static const struct sockaddr_in sin = {
3084                 .sin_family             = AF_INET,
3085                 .sin_addr.s_addr        = htonl(INADDR_ANY),
3086         };
3087         static const struct sockaddr_in6 sin6 = {
3088                 .sin6_family            = AF_INET6,
3089                 .sin6_addr              = IN6ADDR_ANY_INIT,
3090         };
3091
3092         switch (family) {
3093         case AF_LOCAL:
3094                 break;
3095         case AF_INET:
3096                 memcpy(sap, &sin, sizeof(sin));
3097                 break;
3098         case AF_INET6:
3099                 memcpy(sap, &sin6, sizeof(sin6));
3100                 break;
3101         default:
3102                 dprintk("RPC:       %s: Bad address family\n", __func__);
3103                 return -EAFNOSUPPORT;
3104         }
3105         return 0;
3106 }
3107
3108 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3109                                       unsigned int slot_table_size,
3110                                       unsigned int max_slot_table_size)
3111 {
3112         struct rpc_xprt *xprt;
3113         struct sock_xprt *new;
3114
3115         if (args->addrlen > sizeof(xprt->addr)) {
3116                 dprintk("RPC:       xs_setup_xprt: address too large\n");
3117                 return ERR_PTR(-EBADF);
3118         }
3119
3120         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3121                         max_slot_table_size);
3122         if (xprt == NULL) {
3123                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
3124                                 "rpc_xprt\n");
3125                 return ERR_PTR(-ENOMEM);
3126         }
3127
3128         new = container_of(xprt, struct sock_xprt, xprt);
3129         mutex_init(&new->recv_mutex);
3130         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3131         xprt->addrlen = args->addrlen;
3132         if (args->srcaddr)
3133                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3134         else {
3135                 int err;
3136                 err = xs_init_anyaddr(args->dstaddr->sa_family,
3137                                         (struct sockaddr *)&new->srcaddr);
3138                 if (err != 0) {
3139                         xprt_free(xprt);
3140                         return ERR_PTR(err);
3141                 }
3142         }
3143
3144         return xprt;
3145 }
3146
3147 static const struct rpc_timeout xs_local_default_timeout = {
3148         .to_initval = 10 * HZ,
3149         .to_maxval = 10 * HZ,
3150         .to_retries = 2,
3151 };
3152
3153 /**
3154  * xs_setup_local - Set up transport to use an AF_LOCAL socket
3155  * @args: rpc transport creation arguments
3156  *
3157  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3158  */
3159 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3160 {
3161         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3162         struct sock_xprt *transport;
3163         struct rpc_xprt *xprt;
3164         struct rpc_xprt *ret;
3165
3166         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3167                         xprt_max_tcp_slot_table_entries);
3168         if (IS_ERR(xprt))
3169                 return xprt;
3170         transport = container_of(xprt, struct sock_xprt, xprt);
3171
3172         xprt->prot = 0;
3173         xprt->xprt_class = &xs_local_transport;
3174         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3175
3176         xprt->bind_timeout = XS_BIND_TO;
3177         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3178         xprt->idle_timeout = XS_IDLE_DISC_TO;
3179
3180         xprt->ops = &xs_local_ops;
3181         xprt->timeout = &xs_local_default_timeout;
3182
3183         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3184         INIT_WORK(&transport->error_worker, xs_error_handle);
3185         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3186
3187         switch (sun->sun_family) {
3188         case AF_LOCAL:
3189                 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3190                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
3191                                         sun->sun_path);
3192                         ret = ERR_PTR(-EINVAL);
3193                         goto out_err;
3194                 }
3195                 xprt_set_bound(xprt);
3196                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3197                 break;
3198         default:
3199                 ret = ERR_PTR(-EAFNOSUPPORT);
3200                 goto out_err;
3201         }
3202
3203         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
3204                         xprt->address_strings[RPC_DISPLAY_ADDR]);
3205
3206         if (try_module_get(THIS_MODULE))
3207                 return xprt;
3208         ret = ERR_PTR(-EINVAL);
3209 out_err:
3210         xs_xprt_free(xprt);
3211         return ret;
3212 }
3213
3214 static const struct rpc_timeout xs_udp_default_timeout = {
3215         .to_initval = 5 * HZ,
3216         .to_maxval = 30 * HZ,
3217         .to_increment = 5 * HZ,
3218         .to_retries = 5,
3219 };
3220
3221 /**
3222  * xs_setup_udp - Set up transport to use a UDP socket
3223  * @args: rpc transport creation arguments
3224  *
3225  */
3226 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3227 {
3228         struct sockaddr *addr = args->dstaddr;
3229         struct rpc_xprt *xprt;
3230         struct sock_xprt *transport;
3231         struct rpc_xprt *ret;
3232
3233         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3234                         xprt_udp_slot_table_entries);
3235         if (IS_ERR(xprt))
3236                 return xprt;
3237         transport = container_of(xprt, struct sock_xprt, xprt);
3238
3239         xprt->prot = IPPROTO_UDP;
3240         xprt->xprt_class = &xs_udp_transport;
3241         /* XXX: header size can vary due to auth type, IPv6, etc. */
3242         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3243
3244         xprt->bind_timeout = XS_BIND_TO;
3245         xprt->reestablish_timeout = XS_UDP_REEST_TO;
3246         xprt->idle_timeout = XS_IDLE_DISC_TO;
3247
3248         xprt->ops = &xs_udp_ops;
3249
3250         xprt->timeout = &xs_udp_default_timeout;
3251
3252         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3253         INIT_WORK(&transport->error_worker, xs_error_handle);
3254         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3255
3256         switch (addr->sa_family) {
3257         case AF_INET:
3258                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3259                         xprt_set_bound(xprt);
3260
3261                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3262                 break;
3263         case AF_INET6:
3264                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3265                         xprt_set_bound(xprt);
3266
3267                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3268                 break;
3269         default:
3270                 ret = ERR_PTR(-EAFNOSUPPORT);
3271                 goto out_err;
3272         }
3273
3274         if (xprt_bound(xprt))
3275                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3276                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3277                                 xprt->address_strings[RPC_DISPLAY_PORT],
3278                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3279         else
3280                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3281                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3282                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3283
3284         if (try_module_get(THIS_MODULE))
3285                 return xprt;
3286         ret = ERR_PTR(-EINVAL);
3287 out_err:
3288         xs_xprt_free(xprt);
3289         return ret;
3290 }
3291
3292 static const struct rpc_timeout xs_tcp_default_timeout = {
3293         .to_initval = 60 * HZ,
3294         .to_maxval = 60 * HZ,
3295         .to_retries = 2,
3296 };
3297
3298 /**
3299  * xs_setup_tcp - Set up transport to use a TCP socket
3300  * @args: rpc transport creation arguments
3301  *
3302  */
3303 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3304 {
3305         struct sockaddr *addr = args->dstaddr;
3306         struct rpc_xprt *xprt;
3307         struct sock_xprt *transport;
3308         struct rpc_xprt *ret;
3309         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3310
3311         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3312                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3313
3314         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3315                         max_slot_table_size);
3316         if (IS_ERR(xprt))
3317                 return xprt;
3318         transport = container_of(xprt, struct sock_xprt, xprt);
3319
3320         xprt->prot = IPPROTO_TCP;
3321         xprt->xprt_class = &xs_tcp_transport;
3322         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3323
3324         xprt->bind_timeout = XS_BIND_TO;
3325         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3326         xprt->idle_timeout = XS_IDLE_DISC_TO;
3327
3328         xprt->ops = &xs_tcp_ops;
3329         xprt->timeout = &xs_tcp_default_timeout;
3330
3331         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3332         xprt->connect_timeout = xprt->timeout->to_initval *
3333                 (xprt->timeout->to_retries + 1);
3334
3335         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3336         INIT_WORK(&transport->error_worker, xs_error_handle);
3337         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3338
3339         switch (addr->sa_family) {
3340         case AF_INET:
3341                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3342                         xprt_set_bound(xprt);
3343
3344                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3345                 break;
3346         case AF_INET6:
3347                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3348                         xprt_set_bound(xprt);
3349
3350                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3351                 break;
3352         default:
3353                 ret = ERR_PTR(-EAFNOSUPPORT);
3354                 goto out_err;
3355         }
3356
3357         if (xprt_bound(xprt))
3358                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3359                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3360                                 xprt->address_strings[RPC_DISPLAY_PORT],
3361                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3362         else
3363                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3364                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3365                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3366
3367         if (try_module_get(THIS_MODULE))
3368                 return xprt;
3369         ret = ERR_PTR(-EINVAL);
3370 out_err:
3371         xs_xprt_free(xprt);
3372         return ret;
3373 }
3374
3375 /**
3376  * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3377  * @args: rpc transport creation arguments
3378  *
3379  */
3380 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3381 {
3382         struct sockaddr *addr = args->dstaddr;
3383         struct rpc_xprt *xprt;
3384         struct sock_xprt *transport;
3385         struct rpc_xprt *ret;
3386         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3387
3388         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3389                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3390
3391         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3392                              max_slot_table_size);
3393         if (IS_ERR(xprt))
3394                 return xprt;
3395         transport = container_of(xprt, struct sock_xprt, xprt);
3396
3397         xprt->prot = IPPROTO_TCP;
3398         xprt->xprt_class = &xs_tcp_transport;
3399         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3400
3401         xprt->bind_timeout = XS_BIND_TO;
3402         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3403         xprt->idle_timeout = XS_IDLE_DISC_TO;
3404
3405         xprt->ops = &xs_tcp_ops;
3406         xprt->timeout = &xs_tcp_default_timeout;
3407
3408         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3409         xprt->connect_timeout = xprt->timeout->to_initval *
3410                 (xprt->timeout->to_retries + 1);
3411
3412         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3413         INIT_WORK(&transport->error_worker, xs_error_handle);
3414
3415         switch (args->xprtsec.policy) {
3416         case RPC_XPRTSEC_TLS_ANON:
3417         case RPC_XPRTSEC_TLS_X509:
3418                 xprt->xprtsec = args->xprtsec;
3419                 INIT_DELAYED_WORK(&transport->connect_worker,
3420                                   xs_tcp_tls_setup_socket);
3421                 break;
3422         default:
3423                 ret = ERR_PTR(-EACCES);
3424                 goto out_err;
3425         }
3426
3427         switch (addr->sa_family) {
3428         case AF_INET:
3429                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3430                         xprt_set_bound(xprt);
3431
3432                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3433                 break;
3434         case AF_INET6:
3435                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3436                         xprt_set_bound(xprt);
3437
3438                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3439                 break;
3440         default:
3441                 ret = ERR_PTR(-EAFNOSUPPORT);
3442                 goto out_err;
3443         }
3444
3445         if (xprt_bound(xprt))
3446                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3447                         xprt->address_strings[RPC_DISPLAY_ADDR],
3448                         xprt->address_strings[RPC_DISPLAY_PORT],
3449                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3450         else
3451                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3452                         xprt->address_strings[RPC_DISPLAY_ADDR],
3453                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3454
3455         if (try_module_get(THIS_MODULE))
3456                 return xprt;
3457         ret = ERR_PTR(-EINVAL);
3458 out_err:
3459         xs_xprt_free(xprt);
3460         return ret;
3461 }
3462
3463 /**
3464  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3465  * @args: rpc transport creation arguments
3466  *
3467  */
3468 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3469 {
3470         struct sockaddr *addr = args->dstaddr;
3471         struct rpc_xprt *xprt;
3472         struct sock_xprt *transport;
3473         struct svc_sock *bc_sock;
3474         struct rpc_xprt *ret;
3475
3476         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3477                         xprt_tcp_slot_table_entries);
3478         if (IS_ERR(xprt))
3479                 return xprt;
3480         transport = container_of(xprt, struct sock_xprt, xprt);
3481
3482         xprt->prot = IPPROTO_TCP;
3483         xprt->xprt_class = &xs_bc_tcp_transport;
3484         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3485         xprt->timeout = &xs_tcp_default_timeout;
3486
3487         /* backchannel */
3488         xprt_set_bound(xprt);
3489         xprt->bind_timeout = 0;
3490         xprt->reestablish_timeout = 0;
3491         xprt->idle_timeout = 0;
3492
3493         xprt->ops = &bc_tcp_ops;
3494
3495         switch (addr->sa_family) {
3496         case AF_INET:
3497                 xs_format_peer_addresses(xprt, "tcp",
3498                                          RPCBIND_NETID_TCP);
3499                 break;
3500         case AF_INET6:
3501                 xs_format_peer_addresses(xprt, "tcp",
3502                                    RPCBIND_NETID_TCP6);
3503                 break;
3504         default:
3505                 ret = ERR_PTR(-EAFNOSUPPORT);
3506                 goto out_err;
3507         }
3508
3509         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3510                         xprt->address_strings[RPC_DISPLAY_ADDR],
3511                         xprt->address_strings[RPC_DISPLAY_PORT],
3512                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3513
3514         /*
3515          * Once we've associated a backchannel xprt with a connection,
3516          * we want to keep it around as long as the connection lasts,
3517          * in case we need to start using it for a backchannel again;
3518          * this reference won't be dropped until bc_xprt is destroyed.
3519          */
3520         xprt_get(xprt);
3521         args->bc_xprt->xpt_bc_xprt = xprt;
3522         xprt->bc_xprt = args->bc_xprt;
3523         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3524         transport->sock = bc_sock->sk_sock;
3525         transport->inet = bc_sock->sk_sk;
3526
3527         /*
3528          * Since we don't want connections for the backchannel, we set
3529          * the xprt status to connected
3530          */
3531         xprt_set_connected(xprt);
3532
3533         if (try_module_get(THIS_MODULE))
3534                 return xprt;
3535
3536         args->bc_xprt->xpt_bc_xprt = NULL;
3537         args->bc_xprt->xpt_bc_xps = NULL;
3538         xprt_put(xprt);
3539         ret = ERR_PTR(-EINVAL);
3540 out_err:
3541         xs_xprt_free(xprt);
3542         return ret;
3543 }
3544
3545 static struct xprt_class        xs_local_transport = {
3546         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3547         .name           = "named UNIX socket",
3548         .owner          = THIS_MODULE,
3549         .ident          = XPRT_TRANSPORT_LOCAL,
3550         .setup          = xs_setup_local,
3551         .netid          = { "" },
3552 };
3553
3554 static struct xprt_class        xs_udp_transport = {
3555         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3556         .name           = "udp",
3557         .owner          = THIS_MODULE,
3558         .ident          = XPRT_TRANSPORT_UDP,
3559         .setup          = xs_setup_udp,
3560         .netid          = { "udp", "udp6", "" },
3561 };
3562
3563 static struct xprt_class        xs_tcp_transport = {
3564         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3565         .name           = "tcp",
3566         .owner          = THIS_MODULE,
3567         .ident          = XPRT_TRANSPORT_TCP,
3568         .setup          = xs_setup_tcp,
3569         .netid          = { "tcp", "tcp6", "" },
3570 };
3571
3572 static struct xprt_class        xs_tcp_tls_transport = {
3573         .list           = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3574         .name           = "tcp-with-tls",
3575         .owner          = THIS_MODULE,
3576         .ident          = XPRT_TRANSPORT_TCP_TLS,
3577         .setup          = xs_setup_tcp_tls,
3578         .netid          = { "tcp", "tcp6", "" },
3579 };
3580
3581 static struct xprt_class        xs_bc_tcp_transport = {
3582         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3583         .name           = "tcp NFSv4.1 backchannel",
3584         .owner          = THIS_MODULE,
3585         .ident          = XPRT_TRANSPORT_BC_TCP,
3586         .setup          = xs_setup_bc_tcp,
3587         .netid          = { "" },
3588 };
3589
3590 /**
3591  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3592  *
3593  */
3594 int init_socket_xprt(void)
3595 {
3596         if (!sunrpc_table_header)
3597                 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3598
3599         xprt_register_transport(&xs_local_transport);
3600         xprt_register_transport(&xs_udp_transport);
3601         xprt_register_transport(&xs_tcp_transport);
3602         xprt_register_transport(&xs_tcp_tls_transport);
3603         xprt_register_transport(&xs_bc_tcp_transport);
3604
3605         return 0;
3606 }
3607
3608 /**
3609  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3610  *
3611  */
3612 void cleanup_socket_xprt(void)
3613 {
3614         if (sunrpc_table_header) {
3615                 unregister_sysctl_table(sunrpc_table_header);
3616                 sunrpc_table_header = NULL;
3617         }
3618
3619         xprt_unregister_transport(&xs_local_transport);
3620         xprt_unregister_transport(&xs_udp_transport);
3621         xprt_unregister_transport(&xs_tcp_transport);
3622         xprt_unregister_transport(&xs_tcp_tls_transport);
3623         xprt_unregister_transport(&xs_bc_tcp_transport);
3624 }
3625
3626 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3627 {
3628         return param_set_uint_minmax(val, kp,
3629                         RPC_MIN_RESVPORT,
3630                         RPC_MAX_RESVPORT);
3631 }
3632
3633 static const struct kernel_param_ops param_ops_portnr = {
3634         .set = param_set_portnr,
3635         .get = param_get_uint,
3636 };
3637
3638 #define param_check_portnr(name, p) \
3639         __param_check(name, p, unsigned int);
3640
3641 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3642 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3643
3644 static int param_set_slot_table_size(const char *val,
3645                                      const struct kernel_param *kp)
3646 {
3647         return param_set_uint_minmax(val, kp,
3648                         RPC_MIN_SLOT_TABLE,
3649                         RPC_MAX_SLOT_TABLE);
3650 }
3651
3652 static const struct kernel_param_ops param_ops_slot_table_size = {
3653         .set = param_set_slot_table_size,
3654         .get = param_get_uint,
3655 };
3656
3657 #define param_check_slot_table_size(name, p) \
3658         __param_check(name, p, unsigned int);
3659
3660 static int param_set_max_slot_table_size(const char *val,
3661                                      const struct kernel_param *kp)
3662 {
3663         return param_set_uint_minmax(val, kp,
3664                         RPC_MIN_SLOT_TABLE,
3665                         RPC_MAX_SLOT_TABLE_LIMIT);
3666 }
3667
3668 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3669         .set = param_set_max_slot_table_size,
3670         .get = param_get_uint,
3671 };
3672
3673 #define param_check_max_slot_table_size(name, p) \
3674         __param_check(name, p, unsigned int);
3675
3676 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3677                    slot_table_size, 0644);
3678 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3679                    max_slot_table_size, 0644);
3680 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3681                    slot_table_size, 0644);