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/handshake.h>
  52
  53 #include <linux/bvec.h>
  54 #include <linux/highmem.h>
  55 #include <linux/uio.h>
  56 #include <linux/sched/mm.h>
  57
  58 #include <trace/events/sock.h>
  59 #include <trace/events/sunrpc.h>
  60
  61 #include "socklib.h"
  62 #include "sunrpc.h"
  63
  64 static void xs_close(struct rpc_xprt *xprt);
  65 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
  66 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  67                 struct socket *sock);
  68
  69 /*
  70  * xprtsock tunables
  71  */
  72 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  73 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  74 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  75
  76 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  77 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  78
  79 #define XS_TCP_LINGER_TO        (15U * HZ)
  80 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  81
  82 /*
  83  * We can register our own files under /proc/sys/sunrpc by
  84  * calling register_sysctl() again.  The files in that
  85  * directory become the union of all files registered there.
  86  *
  87  * We simply need to make sure that we don't collide with
  88  * someone else's file names!
  89  */
  90
  91 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  92 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  93 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  94 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  95 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  96
  97 static struct ctl_table_header *sunrpc_table_header;
  98
  99 static struct xprt_class xs_local_transport;
 100 static struct xprt_class xs_udp_transport;
 101 static struct xprt_class xs_tcp_transport;
 102 static struct xprt_class xs_tcp_tls_transport;
 103 static struct xprt_class xs_bc_tcp_transport;
 104
 105 /*
 106  * FIXME: changing the UDP slot table size should also resize the UDP
 107  *        socket buffers for existing UDP transports
 108  */
 109 static struct ctl_table xs_tunables_table[] = {
 110         {
 111                 .procname       = "udp_slot_table_entries",
 112                 .data           = &xprt_udp_slot_table_entries,
 113                 .maxlen         = sizeof(unsigned int),
 114                 .mode           = 0644,
 115                 .proc_handler   = proc_dointvec_minmax,
 116                 .extra1         = &min_slot_table_size,
 117                 .extra2         = &max_slot_table_size
 118         },
 119         {
 120                 .procname       = "tcp_slot_table_entries",
 121                 .data           = &xprt_tcp_slot_table_entries,
 122                 .maxlen         = sizeof(unsigned int),
 123                 .mode           = 0644,
 124                 .proc_handler   = proc_dointvec_minmax,
 125                 .extra1         = &min_slot_table_size,
 126                 .extra2         = &max_slot_table_size
 127         },
 128         {
 129                 .procname       = "tcp_max_slot_table_entries",
 130                 .data           = &xprt_max_tcp_slot_table_entries,
 131                 .maxlen         = sizeof(unsigned int),
 132                 .mode           = 0644,
 133                 .proc_handler   = proc_dointvec_minmax,
 134                 .extra1         = &min_slot_table_size,
 135                 .extra2         = &max_tcp_slot_table_limit
 136         },
 137         {
 138                 .procname       = "min_resvport",
 139                 .data           = &xprt_min_resvport,
 140                 .maxlen         = sizeof(unsigned int),
 141                 .mode           = 0644,
 142                 .proc_handler   = proc_dointvec_minmax,
 143                 .extra1         = &xprt_min_resvport_limit,
 144                 .extra2         = &xprt_max_resvport_limit
 145         },
 146         {
 147                 .procname       = "max_resvport",
 148                 .data           = &xprt_max_resvport,
 149                 .maxlen         = sizeof(unsigned int),
 150                 .mode           = 0644,
 151                 .proc_handler   = proc_dointvec_minmax,
 152                 .extra1         = &xprt_min_resvport_limit,
 153                 .extra2         = &xprt_max_resvport_limit
 154         },
 155         {
 156                 .procname       = "tcp_fin_timeout",
 157                 .data           = &xs_tcp_fin_timeout,
 158                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 159                 .mode           = 0644,
 160                 .proc_handler   = proc_dointvec_jiffies,
 161         },
 162         { },
 163 };
 164
 165 /*
 166  * Wait duration for a reply from the RPC portmapper.
 167  */
 168 #define XS_BIND_TO              (60U * HZ)
 169
 170 /*
 171  * Delay if a UDP socket connect error occurs.  This is most likely some
 172  * kind of resource problem on the local host.
 173  */
 174 #define XS_UDP_REEST_TO         (2U * HZ)
 175
 176 /*
 177  * The reestablish timeout allows clients to delay for a bit before attempting
 178  * to reconnect to a server that just dropped our connection.
 179  *
 180  * We implement an exponential backoff when trying to reestablish a TCP
 181  * transport connection with the server.  Some servers like to drop a TCP
 182  * connection when they are overworked, so we start with a short timeout and
 183  * increase over time if the server is down or not responding.
 184  */
 185 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 186
 187 /*
 188  * TCP idle timeout; client drops the transport socket if it is idle
 189  * for this long.  Note that we also timeout UDP sockets to prevent
 190  * holding port numbers when there is no RPC traffic.
 191  */
 192 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 193
 194 /*
 195  * TLS handshake timeout.
 196  */
 197 #define XS_TLS_HANDSHAKE_TO     (10U * HZ)
 198
 199 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 200 # undef  RPC_DEBUG_DATA
 201 # define RPCDBG_FACILITY        RPCDBG_TRANS
 202 #endif
 203
 204 #ifdef RPC_DEBUG_DATA
 205 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 206 {
 207         u8 *buf = (u8 *) packet;
 208         int j;
 209
 210         dprintk("RPC:       %s\n", msg);
 211         for (j = 0; j < count && j < 128; j += 4) {
 212                 if (!(j & 31)) {
 213                         if (j)
 214                                 dprintk("\n");
 215                         dprintk("0x%04x ", j);
 216                 }
 217                 dprintk("%02x%02x%02x%02x ",
 218                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 219         }
 220         dprintk("\n");
 221 }
 222 #else
 223 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 224 {
 225         /* NOP */
 226 }
 227 #endif
 228
 229 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 230 {
 231         return (struct rpc_xprt *) sk->sk_user_data;
 232 }
 233
 234 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_un *) &xprt->addr;
 242 }
 243
 244 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 245 {
 246         return (struct sockaddr_in *) &xprt->addr;
 247 }
 248
 249 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 250 {
 251         return (struct sockaddr_in6 *) &xprt->addr;
 252 }
 253
 254 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 255 {
 256         struct sockaddr *sap = xs_addr(xprt);
 257         struct sockaddr_in6 *sin6;
 258         struct sockaddr_in *sin;
 259         struct sockaddr_un *sun;
 260         char buf[128];
 261
 262         switch (sap->sa_family) {
 263         case AF_LOCAL:
 264                 sun = xs_addr_un(xprt);
 265                 if (sun->sun_path[0]) {
 266                         strscpy(buf, sun->sun_path, sizeof(buf));
 267                 } else {
 268                         buf[0] = '@';
 269                         strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1);
 270                 }
 271                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 272                                                 kstrdup(buf, GFP_KERNEL);
 273                 break;
 274         case AF_INET:
 275                 (void)rpc_ntop(sap, buf, sizeof(buf));
 276                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 277                                                 kstrdup(buf, GFP_KERNEL);
 278                 sin = xs_addr_in(xprt);
 279                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 280                 break;
 281         case AF_INET6:
 282                 (void)rpc_ntop(sap, buf, sizeof(buf));
 283                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 284                                                 kstrdup(buf, GFP_KERNEL);
 285                 sin6 = xs_addr_in6(xprt);
 286                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 287                 break;
 288         default:
 289                 BUG();
 290         }
 291
 292         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 293 }
 294
 295 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 296 {
 297         struct sockaddr *sap = xs_addr(xprt);
 298         char buf[128];
 299
 300         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 301         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 302
 303         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 304         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 305 }
 306
 307 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 308                                      const char *protocol,
 309                                      const char *netid)
 310 {
 311         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 312         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 313         xs_format_common_peer_addresses(xprt);
 314         xs_format_common_peer_ports(xprt);
 315 }
 316
 317 static void xs_update_peer_port(struct rpc_xprt *xprt)
 318 {
 319         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 320         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 321
 322         xs_format_common_peer_ports(xprt);
 323 }
 324
 325 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 326 {
 327         unsigned int i;
 328
 329         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 330                 switch (i) {
 331                 case RPC_DISPLAY_PROTO:
 332                 case RPC_DISPLAY_NETID:
 333                         continue;
 334                 default:
 335                         kfree(xprt->address_strings[i]);
 336                 }
 337 }
 338
 339 static size_t
 340 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 341 {
 342         size_t i,n;
 343
 344         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 345                 return want;
 346         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 347         for (i = 0; i < n; i++) {
 348                 if (buf->pages[i])
 349                         continue;
 350                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 351                 if (!buf->pages[i]) {
 352                         i *= PAGE_SIZE;
 353                         return i > buf->page_base ? i - buf->page_base : 0;
 354                 }
 355         }
 356         return want;
 357 }
 358
 359 static int
 360 xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
 361                      struct cmsghdr *cmsg, int ret)
 362 {
 363         if (cmsg->cmsg_level == SOL_TLS &&
 364             cmsg->cmsg_type == TLS_GET_RECORD_TYPE) {
 365                 u8 content_type = *((u8 *)CMSG_DATA(cmsg));
 366
 367                 switch (content_type) {
 368                 case TLS_RECORD_TYPE_DATA:
 369                         /* TLS sets EOR at the end of each application data
 370                          * record, even though there might be more frames
 371                          * waiting to be decrypted.
 372                          */
 373                         msg->msg_flags &= ~MSG_EOR;
 374                         break;
 375                 case TLS_RECORD_TYPE_ALERT:
 376                         ret = -ENOTCONN;
 377                         break;
 378                 default:
 379                         ret = -EAGAIN;
 380                 }
 381         }
 382         return ret;
 383 }
 384
 385 static int
 386 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
 387 {
 388         union {
 389                 struct cmsghdr  cmsg;
 390                 u8              buf[CMSG_SPACE(sizeof(u8))];
 391         } u;
 392         int ret;
 393
 394         msg->msg_control = &u;
 395         msg->msg_controllen = sizeof(u);
 396         ret = sock_recvmsg(sock, msg, flags);
 397         if (msg->msg_controllen != sizeof(u))
 398                 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
 399         return ret;
 400 }
 401
 402 static ssize_t
 403 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 404 {
 405         ssize_t ret;
 406         if (seek != 0)
 407                 iov_iter_advance(&msg->msg_iter, seek);
 408         ret = xs_sock_recv_cmsg(sock, msg, flags);
 409         return ret > 0 ? ret + seek : ret;
 410 }
 411
 412 static ssize_t
 413 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 414                 struct kvec *kvec, size_t count, size_t seek)
 415 {
 416         iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
 417         return xs_sock_recvmsg(sock, msg, flags, seek);
 418 }
 419
 420 static ssize_t
 421 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 422                 struct bio_vec *bvec, unsigned long nr, size_t count,
 423                 size_t seek)
 424 {
 425         iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
 426         return xs_sock_recvmsg(sock, msg, flags, seek);
 427 }
 428
 429 static ssize_t
 430 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 431                 size_t count)
 432 {
 433         iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
 434         return xs_sock_recv_cmsg(sock, msg, flags);
 435 }
 436
 437 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 438 static void
 439 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 440 {
 441         struct bvec_iter bi = {
 442                 .bi_size = count,
 443         };
 444         struct bio_vec bv;
 445
 446         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 447         for_each_bvec(bv, bvec, bi, bi)
 448                 flush_dcache_page(bv.bv_page);
 449 }
 450 #else
 451 static inline void
 452 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 453 {
 454 }
 455 #endif
 456
 457 static ssize_t
 458 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 459                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 460 {
 461         size_t want, seek_init = seek, offset = 0;
 462         ssize_t ret;
 463
 464         want = min_t(size_t, count, buf->head[0].iov_len);
 465         if (seek < want) {
 466                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 467                 if (ret <= 0)
 468                         goto sock_err;
 469                 offset += ret;
 470                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 471                         goto out;
 472                 if (ret != want)
 473                         goto out;
 474                 seek = 0;
 475         } else {
 476                 seek -= want;
 477                 offset += want;
 478         }
 479
 480         want = xs_alloc_sparse_pages(
 481                 buf, min_t(size_t, count - offset, buf->page_len),
 482                 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
 483         if (seek < want) {
 484                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 485                                 xdr_buf_pagecount(buf),
 486                                 want + buf->page_base,
 487                                 seek + buf->page_base);
 488                 if (ret <= 0)
 489                         goto sock_err;
 490                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 491                 ret -= buf->page_base;
 492                 offset += ret;
 493                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 494                         goto out;
 495                 if (ret != want)
 496                         goto out;
 497                 seek = 0;
 498         } else {
 499                 seek -= want;
 500                 offset += want;
 501         }
 502
 503         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 504         if (seek < want) {
 505                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 506                 if (ret <= 0)
 507                         goto sock_err;
 508                 offset += ret;
 509                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 510                         goto out;
 511                 if (ret != want)
 512                         goto out;
 513         } else if (offset < seek_init)
 514                 offset = seek_init;
 515         ret = -EMSGSIZE;
 516 out:
 517         *read = offset - seek_init;
 518         return ret;
 519 sock_err:
 520         offset += seek;
 521         goto out;
 522 }
 523
 524 static void
 525 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 526 {
 527         if (!transport->recv.copied) {
 528                 if (buf->head[0].iov_len >= transport->recv.offset)
 529                         memcpy(buf->head[0].iov_base,
 530                                         &transport->recv.xid,
 531                                         transport->recv.offset);
 532                 transport->recv.copied = transport->recv.offset;
 533         }
 534 }
 535
 536 static bool
 537 xs_read_stream_request_done(struct sock_xprt *transport)
 538 {
 539         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 540 }
 541
 542 static void
 543 xs_read_stream_check_eor(struct sock_xprt *transport,
 544                 struct msghdr *msg)
 545 {
 546         if (xs_read_stream_request_done(transport))
 547                 msg->msg_flags |= MSG_EOR;
 548 }
 549
 550 static ssize_t
 551 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 552                 int flags, struct rpc_rqst *req)
 553 {
 554         struct xdr_buf *buf = &req->rq_private_buf;
 555         size_t want, read;
 556         ssize_t ret;
 557
 558         xs_read_header(transport, buf);
 559
 560         want = transport->recv.len - transport->recv.offset;
 561         if (want != 0) {
 562                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 563                                 transport->recv.copied + want,
 564                                 transport->recv.copied,
 565                                 &read);
 566                 transport->recv.offset += read;
 567                 transport->recv.copied += read;
 568         }
 569
 570         if (transport->recv.offset == transport->recv.len)
 571                 xs_read_stream_check_eor(transport, msg);
 572
 573         if (want == 0)
 574                 return 0;
 575
 576         switch (ret) {
 577         default:
 578                 break;
 579         case -EFAULT:
 580         case -EMSGSIZE:
 581                 msg->msg_flags |= MSG_TRUNC;
 582                 return read;
 583         case 0:
 584                 return -ESHUTDOWN;
 585         }
 586         return ret < 0 ? ret : read;
 587 }
 588
 589 static size_t
 590 xs_read_stream_headersize(bool isfrag)
 591 {
 592         if (isfrag)
 593                 return sizeof(__be32);
 594         return 3 * sizeof(__be32);
 595 }
 596
 597 static ssize_t
 598 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 599                 int flags, size_t want, size_t seek)
 600 {
 601         struct kvec kvec = {
 602                 .iov_base = &transport->recv.fraghdr,
 603                 .iov_len = want,
 604         };
 605         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 606 }
 607
 608 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 609 static ssize_t
 610 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 611 {
 612         struct rpc_xprt *xprt = &transport->xprt;
 613         struct rpc_rqst *req;
 614         ssize_t ret;
 615
 616         /* Is this transport associated with the backchannel? */
 617         if (!xprt->bc_serv)
 618                 return -ESHUTDOWN;
 619
 620         /* Look up and lock the request corresponding to the given XID */
 621         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 622         if (!req) {
 623                 printk(KERN_WARNING "Callback slot table overflowed\n");
 624                 return -ESHUTDOWN;
 625         }
 626         if (transport->recv.copied && !req->rq_private_buf.len)
 627                 return -ESHUTDOWN;
 628
 629         ret = xs_read_stream_request(transport, msg, flags, req);
 630         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 631                 xprt_complete_bc_request(req, transport->recv.copied);
 632         else
 633                 req->rq_private_buf.len = transport->recv.copied;
 634
 635         return ret;
 636 }
 637 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 638 static ssize_t
 639 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 640 {
 641         return -ESHUTDOWN;
 642 }
 643 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 644
 645 static ssize_t
 646 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 647 {
 648         struct rpc_xprt *xprt = &transport->xprt;
 649         struct rpc_rqst *req;
 650         ssize_t ret = 0;
 651
 652         /* Look up and lock the request corresponding to the given XID */
 653         spin_lock(&xprt->queue_lock);
 654         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 655         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 656                 msg->msg_flags |= MSG_TRUNC;
 657                 goto out;
 658         }
 659         xprt_pin_rqst(req);
 660         spin_unlock(&xprt->queue_lock);
 661
 662         ret = xs_read_stream_request(transport, msg, flags, req);
 663
 664         spin_lock(&xprt->queue_lock);
 665         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 666                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 667         else
 668                 req->rq_private_buf.len = transport->recv.copied;
 669         xprt_unpin_rqst(req);
 670 out:
 671         spin_unlock(&xprt->queue_lock);
 672         return ret;
 673 }
 674
 675 static ssize_t
 676 xs_read_stream(struct sock_xprt *transport, int flags)
 677 {
 678         struct msghdr msg = { 0 };
 679         size_t want, read = 0;
 680         ssize_t ret = 0;
 681
 682         if (transport->recv.len == 0) {
 683                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 684                 ret = xs_read_stream_header(transport, &msg, flags, want,
 685                                 transport->recv.offset);
 686                 if (ret <= 0)
 687                         goto out_err;
 688                 transport->recv.offset = ret;
 689                 if (transport->recv.offset != want)
 690                         return transport->recv.offset;
 691                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 692                         RPC_FRAGMENT_SIZE_MASK;
 693                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 694                 read = ret;
 695         }
 696
 697         switch (be32_to_cpu(transport->recv.calldir)) {
 698         default:
 699                 msg.msg_flags |= MSG_TRUNC;
 700                 break;
 701         case RPC_CALL:
 702                 ret = xs_read_stream_call(transport, &msg, flags);
 703                 break;
 704         case RPC_REPLY:
 705                 ret = xs_read_stream_reply(transport, &msg, flags);
 706         }
 707         if (msg.msg_flags & MSG_TRUNC) {
 708                 transport->recv.calldir = cpu_to_be32(-1);
 709                 transport->recv.copied = -1;
 710         }
 711         if (ret < 0)
 712                 goto out_err;
 713         read += ret;
 714         if (transport->recv.offset < transport->recv.len) {
 715                 if (!(msg.msg_flags & MSG_TRUNC))
 716                         return read;
 717                 msg.msg_flags = 0;
 718                 ret = xs_read_discard(transport->sock, &msg, flags,
 719                                 transport->recv.len - transport->recv.offset);
 720                 if (ret <= 0)
 721                         goto out_err;
 722                 transport->recv.offset += ret;
 723                 read += ret;
 724                 if (transport->recv.offset != transport->recv.len)
 725                         return read;
 726         }
 727         if (xs_read_stream_request_done(transport)) {
 728                 trace_xs_stream_read_request(transport);
 729                 transport->recv.copied = 0;
 730         }
 731         transport->recv.offset = 0;
 732         transport->recv.len = 0;
 733         return read;
 734 out_err:
 735         return ret != 0 ? ret : -ESHUTDOWN;
 736 }
 737
 738 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 739 {
 740         return transport->sock->ops->poll(transport->file, transport->sock,
 741                         NULL);
 742 }
 743
 744 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 745 {
 746         __poll_t events = xs_poll_socket(transport);
 747
 748         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 749 }
 750
 751 static void xs_poll_check_readable(struct sock_xprt *transport)
 752 {
 753
 754         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 755         if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
 756                 return;
 757         if (!xs_poll_socket_readable(transport))
 758                 return;
 759         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 760                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 761 }
 762
 763 static void xs_stream_data_receive(struct sock_xprt *transport)
 764 {
 765         size_t read = 0;
 766         ssize_t ret = 0;
 767
 768         mutex_lock(&transport->recv_mutex);
 769         if (transport->sock == NULL)
 770                 goto out;
 771         for (;;) {
 772                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 773                 if (ret < 0)
 774                         break;
 775                 read += ret;
 776                 cond_resched();
 777         }
 778         if (ret == -ESHUTDOWN)
 779                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 780         else
 781                 xs_poll_check_readable(transport);
 782 out:
 783         mutex_unlock(&transport->recv_mutex);
 784         trace_xs_stream_read_data(&transport->xprt, ret, read);
 785 }
 786
 787 static void xs_stream_data_receive_workfn(struct work_struct *work)
 788 {
 789         struct sock_xprt *transport =
 790                 container_of(work, struct sock_xprt, recv_worker);
 791         unsigned int pflags = memalloc_nofs_save();
 792
 793         xs_stream_data_receive(transport);
 794         memalloc_nofs_restore(pflags);
 795 }
 796
 797 static void
 798 xs_stream_reset_connect(struct sock_xprt *transport)
 799 {
 800         transport->recv.offset = 0;
 801         transport->recv.len = 0;
 802         transport->recv.copied = 0;
 803         transport->xmit.offset = 0;
 804 }
 805
 806 static void
 807 xs_stream_start_connect(struct sock_xprt *transport)
 808 {
 809         transport->xprt.stat.connect_count++;
 810         transport->xprt.stat.connect_start = jiffies;
 811 }
 812
 813 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 814
 815 /**
 816  * xs_nospace - handle transmit was incomplete
 817  * @req: pointer to RPC request
 818  * @transport: pointer to struct sock_xprt
 819  *
 820  */
 821 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 822 {
 823         struct rpc_xprt *xprt = &transport->xprt;
 824         struct sock *sk = transport->inet;
 825         int ret = -EAGAIN;
 826
 827         trace_rpc_socket_nospace(req, transport);
 828
 829         /* Protect against races with write_space */
 830         spin_lock(&xprt->transport_lock);
 831
 832         /* Don't race with disconnect */
 833         if (xprt_connected(xprt)) {
 834                 /* wait for more buffer space */
 835                 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
 836                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 837                 sk->sk_write_pending++;
 838                 xprt_wait_for_buffer_space(xprt);
 839         } else
 840                 ret = -ENOTCONN;
 841
 842         spin_unlock(&xprt->transport_lock);
 843         return ret;
 844 }
 845
 846 static int xs_sock_nospace(struct rpc_rqst *req)
 847 {
 848         struct sock_xprt *transport =
 849                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 850         struct sock *sk = transport->inet;
 851         int ret = -EAGAIN;
 852
 853         lock_sock(sk);
 854         if (!sock_writeable(sk))
 855                 ret = xs_nospace(req, transport);
 856         release_sock(sk);
 857         return ret;
 858 }
 859
 860 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
 861 {
 862         struct sock_xprt *transport =
 863                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 864         struct sock *sk = transport->inet;
 865         int ret = -EAGAIN;
 866
 867         if (vm_wait)
 868                 return -ENOBUFS;
 869         lock_sock(sk);
 870         if (!sk_stream_memory_free(sk))
 871                 ret = xs_nospace(req, transport);
 872         release_sock(sk);
 873         return ret;
 874 }
 875
 876 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
 877 {
 878         return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
 879 }
 880
 881 /*
 882  * Determine if the previous message in the stream was aborted before it
 883  * could complete transmission.
 884  */
 885 static bool
 886 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 887 {
 888         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 889 }
 890
 891 /*
 892  * Return the stream record marker field for a record of length < 2^31-1
 893  */
 894 static rpc_fraghdr
 895 xs_stream_record_marker(struct xdr_buf *xdr)
 896 {
 897         if (!xdr->len)
 898                 return 0;
 899         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 900 }
 901
 902 /**
 903  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 904  * @req: pointer to RPC request
 905  *
 906  * Return values:
 907  *        0:    The request has been sent
 908  *   EAGAIN:    The socket was blocked, please call again later to
 909  *              complete the request
 910  * ENOTCONN:    Caller needs to invoke connect logic then call again
 911  *    other:    Some other error occurred, the request was not sent
 912  */
 913 static int xs_local_send_request(struct rpc_rqst *req)
 914 {
 915         struct rpc_xprt *xprt = req->rq_xprt;
 916         struct sock_xprt *transport =
 917                                 container_of(xprt, struct sock_xprt, xprt);
 918         struct xdr_buf *xdr = &req->rq_snd_buf;
 919         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 920         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 921         struct msghdr msg = {
 922                 .msg_flags      = XS_SENDMSG_FLAGS,
 923         };
 924         bool vm_wait;
 925         unsigned int sent;
 926         int status;
 927
 928         /* Close the stream if the previous transmission was incomplete */
 929         if (xs_send_request_was_aborted(transport, req)) {
 930                 xprt_force_disconnect(xprt);
 931                 return -ENOTCONN;
 932         }
 933
 934         xs_pktdump("packet data:",
 935                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 936
 937         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
 938
 939         req->rq_xtime = ktime_get();
 940         status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 941                                    transport->xmit.offset, rm, &sent);
 942         dprintk("RPC:       %s(%u) = %d\n",
 943                         __func__, xdr->len - transport->xmit.offset, status);
 944
 945         if (likely(sent > 0) || status == 0) {
 946                 transport->xmit.offset += sent;
 947                 req->rq_bytes_sent = transport->xmit.offset;
 948                 if (likely(req->rq_bytes_sent >= msglen)) {
 949                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 950                         transport->xmit.offset = 0;
 951                         return 0;
 952                 }
 953                 status = -EAGAIN;
 954                 vm_wait = false;
 955         }
 956
 957         switch (status) {
 958         case -EAGAIN:
 959                 status = xs_stream_nospace(req, vm_wait);
 960                 break;
 961         default:
 962                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 963                         -status);
 964                 fallthrough;
 965         case -EPIPE:
 966                 xprt_force_disconnect(xprt);
 967                 status = -ENOTCONN;
 968         }
 969
 970         return status;
 971 }
 972
 973 /**
 974  * xs_udp_send_request - write an RPC request to a UDP socket
 975  * @req: pointer to RPC request
 976  *
 977  * Return values:
 978  *        0:    The request has been sent
 979  *   EAGAIN:    The socket was blocked, please call again later to
 980  *              complete the request
 981  * ENOTCONN:    Caller needs to invoke connect logic then call again
 982  *    other:    Some other error occurred, the request was not sent
 983  */
 984 static int xs_udp_send_request(struct rpc_rqst *req)
 985 {
 986         struct rpc_xprt *xprt = req->rq_xprt;
 987         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 988         struct xdr_buf *xdr = &req->rq_snd_buf;
 989         struct msghdr msg = {
 990                 .msg_name       = xs_addr(xprt),
 991                 .msg_namelen    = xprt->addrlen,
 992                 .msg_flags      = XS_SENDMSG_FLAGS,
 993         };
 994         unsigned int sent;
 995         int status;
 996
 997         xs_pktdump("packet data:",
 998                                 req->rq_svec->iov_base,
 999                                 req->rq_svec->iov_len);
1000
1001         if (!xprt_bound(xprt))
1002                 return -ENOTCONN;
1003
1004         if (!xprt_request_get_cong(xprt, req))
1005                 return -EBADSLT;
1006
1007         status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1008         if (status < 0)
1009                 return status;
1010         req->rq_xtime = ktime_get();
1011         status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1012
1013         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
1014                         xdr->len, status);
1015
1016         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1017         if (status == -EPERM)
1018                 goto process_status;
1019
1020         if (status == -EAGAIN && sock_writeable(transport->inet))
1021                 status = -ENOBUFS;
1022
1023         if (sent > 0 || status == 0) {
1024                 req->rq_xmit_bytes_sent += sent;
1025                 if (sent >= req->rq_slen)
1026                         return 0;
1027                 /* Still some bytes left; set up for a retry later. */
1028                 status = -EAGAIN;
1029         }
1030
1031 process_status:
1032         switch (status) {
1033         case -ENOTSOCK:
1034                 status = -ENOTCONN;
1035                 /* Should we call xs_close() here? */
1036                 break;
1037         case -EAGAIN:
1038                 status = xs_sock_nospace(req);
1039                 break;
1040         case -ENETUNREACH:
1041         case -ENOBUFS:
1042         case -EPIPE:
1043         case -ECONNREFUSED:
1044         case -EPERM:
1045                 /* When the server has died, an ICMP port unreachable message
1046                  * prompts ECONNREFUSED. */
1047                 break;
1048         default:
1049                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1050                         -status);
1051         }
1052
1053         return status;
1054 }
1055
1056 /**
1057  * xs_tcp_send_request - write an RPC request to a TCP socket
1058  * @req: pointer to RPC request
1059  *
1060  * Return values:
1061  *        0:    The request has been sent
1062  *   EAGAIN:    The socket was blocked, please call again later to
1063  *              complete the request
1064  * ENOTCONN:    Caller needs to invoke connect logic then call again
1065  *    other:    Some other error occurred, the request was not sent
1066  *
1067  * XXX: In the case of soft timeouts, should we eventually give up
1068  *      if sendmsg is not able to make progress?
1069  */
1070 static int xs_tcp_send_request(struct rpc_rqst *req)
1071 {
1072         struct rpc_xprt *xprt = req->rq_xprt;
1073         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1074         struct xdr_buf *xdr = &req->rq_snd_buf;
1075         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1076         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1077         struct msghdr msg = {
1078                 .msg_flags      = XS_SENDMSG_FLAGS,
1079         };
1080         bool vm_wait;
1081         unsigned int sent;
1082         int status;
1083
1084         /* Close the stream if the previous transmission was incomplete */
1085         if (xs_send_request_was_aborted(transport, req)) {
1086                 if (transport->sock != NULL)
1087                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1088                 return -ENOTCONN;
1089         }
1090         if (!transport->inet)
1091                 return -ENOTCONN;
1092
1093         xs_pktdump("packet data:",
1094                                 req->rq_svec->iov_base,
1095                                 req->rq_svec->iov_len);
1096
1097         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1098                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1099
1100         xs_set_srcport(transport, transport->sock);
1101
1102         /* Continue transmitting the packet/record. We must be careful
1103          * to cope with writespace callbacks arriving _after_ we have
1104          * called sendmsg(). */
1105         req->rq_xtime = ktime_get();
1106         tcp_sock_set_cork(transport->inet, true);
1107
1108         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1109
1110         do {
1111                 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1112                                            transport->xmit.offset, rm, &sent);
1113
1114                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1115                                 xdr->len - transport->xmit.offset, status);
1116
1117                 /* If we've sent the entire packet, immediately
1118                  * reset the count of bytes sent. */
1119                 transport->xmit.offset += sent;
1120                 req->rq_bytes_sent = transport->xmit.offset;
1121                 if (likely(req->rq_bytes_sent >= msglen)) {
1122                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1123                         transport->xmit.offset = 0;
1124                         if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1125                                 tcp_sock_set_cork(transport->inet, false);
1126                         return 0;
1127                 }
1128
1129                 WARN_ON_ONCE(sent == 0 && status == 0);
1130
1131                 if (sent > 0)
1132                         vm_wait = false;
1133
1134         } while (status == 0);
1135
1136         switch (status) {
1137         case -ENOTSOCK:
1138                 status = -ENOTCONN;
1139                 /* Should we call xs_close() here? */
1140                 break;
1141         case -EAGAIN:
1142                 status = xs_stream_nospace(req, vm_wait);
1143                 break;
1144         case -ECONNRESET:
1145         case -ECONNREFUSED:
1146         case -ENOTCONN:
1147         case -EADDRINUSE:
1148         case -ENOBUFS:
1149         case -EPIPE:
1150                 break;
1151         default:
1152                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1153                         -status);
1154         }
1155
1156         return status;
1157 }
1158
1159 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1160 {
1161         transport->old_data_ready = sk->sk_data_ready;
1162         transport->old_state_change = sk->sk_state_change;
1163         transport->old_write_space = sk->sk_write_space;
1164         transport->old_error_report = sk->sk_error_report;
1165 }
1166
1167 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1168 {
1169         sk->sk_data_ready = transport->old_data_ready;
1170         sk->sk_state_change = transport->old_state_change;
1171         sk->sk_write_space = transport->old_write_space;
1172         sk->sk_error_report = transport->old_error_report;
1173 }
1174
1175 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1176 {
1177         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1178
1179         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1180         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1181         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1182         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1183         clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1184 }
1185
1186 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1187 {
1188         set_bit(nr, &transport->sock_state);
1189         queue_work(xprtiod_workqueue, &transport->error_worker);
1190 }
1191
1192 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1193 {
1194         xprt->connect_cookie++;
1195         smp_mb__before_atomic();
1196         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1197         clear_bit(XPRT_CLOSING, &xprt->state);
1198         xs_sock_reset_state_flags(xprt);
1199         smp_mb__after_atomic();
1200 }
1201
1202 /**
1203  * xs_error_report - callback to handle TCP socket state errors
1204  * @sk: socket
1205  *
1206  * Note: we don't call sock_error() since there may be a rpc_task
1207  * using the socket, and so we don't want to clear sk->sk_err.
1208  */
1209 static void xs_error_report(struct sock *sk)
1210 {
1211         struct sock_xprt *transport;
1212         struct rpc_xprt *xprt;
1213
1214         if (!(xprt = xprt_from_sock(sk)))
1215                 return;
1216
1217         transport = container_of(xprt, struct sock_xprt, xprt);
1218         transport->xprt_err = -sk->sk_err;
1219         if (transport->xprt_err == 0)
1220                 return;
1221         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1222                         xprt, -transport->xprt_err);
1223         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1224
1225         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1226         smp_mb__before_atomic();
1227         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1228 }
1229
1230 static void xs_reset_transport(struct sock_xprt *transport)
1231 {
1232         struct socket *sock = transport->sock;
1233         struct sock *sk = transport->inet;
1234         struct rpc_xprt *xprt = &transport->xprt;
1235         struct file *filp = transport->file;
1236
1237         if (sk == NULL)
1238                 return;
1239         /*
1240          * Make sure we're calling this in a context from which it is safe
1241          * to call __fput_sync(). In practice that means rpciod and the
1242          * system workqueue.
1243          */
1244         if (!(current->flags & PF_WQ_WORKER)) {
1245                 WARN_ON_ONCE(1);
1246                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1247                 return;
1248         }
1249
1250         if (atomic_read(&transport->xprt.swapper))
1251                 sk_clear_memalloc(sk);
1252
1253         tls_handshake_cancel(sk);
1254
1255         kernel_sock_shutdown(sock, SHUT_RDWR);
1256
1257         mutex_lock(&transport->recv_mutex);
1258         lock_sock(sk);
1259         transport->inet = NULL;
1260         transport->sock = NULL;
1261         transport->file = NULL;
1262
1263         sk->sk_user_data = NULL;
1264
1265         xs_restore_old_callbacks(transport, sk);
1266         xprt_clear_connected(xprt);
1267         xs_sock_reset_connection_flags(xprt);
1268         /* Reset stream record info */
1269         xs_stream_reset_connect(transport);
1270         release_sock(sk);
1271         mutex_unlock(&transport->recv_mutex);
1272
1273         trace_rpc_socket_close(xprt, sock);
1274         __fput_sync(filp);
1275
1276         xprt_disconnect_done(xprt);
1277 }
1278
1279 /**
1280  * xs_close - close a socket
1281  * @xprt: transport
1282  *
1283  * This is used when all requests are complete; ie, no DRC state remains
1284  * on the server we want to save.
1285  *
1286  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1287  * xs_reset_transport() zeroing the socket from underneath a writer.
1288  */
1289 static void xs_close(struct rpc_xprt *xprt)
1290 {
1291         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1292
1293         dprintk("RPC:       xs_close xprt %p\n", xprt);
1294
1295         xs_reset_transport(transport);
1296         xprt->reestablish_timeout = 0;
1297 }
1298
1299 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1300 {
1301         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1302                 xprt);
1303         xprt_disconnect_done(xprt);
1304 }
1305
1306 static void xs_xprt_free(struct rpc_xprt *xprt)
1307 {
1308         xs_free_peer_addresses(xprt);
1309         xprt_free(xprt);
1310 }
1311
1312 /**
1313  * xs_destroy - prepare to shutdown a transport
1314  * @xprt: doomed transport
1315  *
1316  */
1317 static void xs_destroy(struct rpc_xprt *xprt)
1318 {
1319         struct sock_xprt *transport = container_of(xprt,
1320                         struct sock_xprt, xprt);
1321         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1322
1323         cancel_delayed_work_sync(&transport->connect_worker);
1324         xs_close(xprt);
1325         cancel_work_sync(&transport->recv_worker);
1326         cancel_work_sync(&transport->error_worker);
1327         xs_xprt_free(xprt);
1328         module_put(THIS_MODULE);
1329 }
1330
1331 /**
1332  * xs_udp_data_read_skb - receive callback for UDP sockets
1333  * @xprt: transport
1334  * @sk: socket
1335  * @skb: skbuff
1336  *
1337  */
1338 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1339                 struct sock *sk,
1340                 struct sk_buff *skb)
1341 {
1342         struct rpc_task *task;
1343         struct rpc_rqst *rovr;
1344         int repsize, copied;
1345         u32 _xid;
1346         __be32 *xp;
1347
1348         repsize = skb->len;
1349         if (repsize < 4) {
1350                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1351                 return;
1352         }
1353
1354         /* Copy the XID from the skb... */
1355         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1356         if (xp == NULL)
1357                 return;
1358
1359         /* Look up and lock the request corresponding to the given XID */
1360         spin_lock(&xprt->queue_lock);
1361         rovr = xprt_lookup_rqst(xprt, *xp);
1362         if (!rovr)
1363                 goto out_unlock;
1364         xprt_pin_rqst(rovr);
1365         xprt_update_rtt(rovr->rq_task);
1366         spin_unlock(&xprt->queue_lock);
1367         task = rovr->rq_task;
1368
1369         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1370                 copied = repsize;
1371
1372         /* Suck it into the iovec, verify checksum if not done by hw. */
1373         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1374                 spin_lock(&xprt->queue_lock);
1375                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1376                 goto out_unpin;
1377         }
1378
1379
1380         spin_lock(&xprt->transport_lock);
1381         xprt_adjust_cwnd(xprt, task, copied);
1382         spin_unlock(&xprt->transport_lock);
1383         spin_lock(&xprt->queue_lock);
1384         xprt_complete_rqst(task, copied);
1385         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1386 out_unpin:
1387         xprt_unpin_rqst(rovr);
1388  out_unlock:
1389         spin_unlock(&xprt->queue_lock);
1390 }
1391
1392 static void xs_udp_data_receive(struct sock_xprt *transport)
1393 {
1394         struct sk_buff *skb;
1395         struct sock *sk;
1396         int err;
1397
1398         mutex_lock(&transport->recv_mutex);
1399         sk = transport->inet;
1400         if (sk == NULL)
1401                 goto out;
1402         for (;;) {
1403                 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1404                 if (skb == NULL)
1405                         break;
1406                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1407                 consume_skb(skb);
1408                 cond_resched();
1409         }
1410         xs_poll_check_readable(transport);
1411 out:
1412         mutex_unlock(&transport->recv_mutex);
1413 }
1414
1415 static void xs_udp_data_receive_workfn(struct work_struct *work)
1416 {
1417         struct sock_xprt *transport =
1418                 container_of(work, struct sock_xprt, recv_worker);
1419         unsigned int pflags = memalloc_nofs_save();
1420
1421         xs_udp_data_receive(transport);
1422         memalloc_nofs_restore(pflags);
1423 }
1424
1425 /**
1426  * xs_data_ready - "data ready" callback for sockets
1427  * @sk: socket with data to read
1428  *
1429  */
1430 static void xs_data_ready(struct sock *sk)
1431 {
1432         struct rpc_xprt *xprt;
1433
1434         trace_sk_data_ready(sk);
1435
1436         xprt = xprt_from_sock(sk);
1437         if (xprt != NULL) {
1438                 struct sock_xprt *transport = container_of(xprt,
1439                                 struct sock_xprt, xprt);
1440
1441                 trace_xs_data_ready(xprt);
1442
1443                 transport->old_data_ready(sk);
1444
1445                 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1446                         return;
1447
1448                 /* Any data means we had a useful conversation, so
1449                  * then we don't need to delay the next reconnect
1450                  */
1451                 if (xprt->reestablish_timeout)
1452                         xprt->reestablish_timeout = 0;
1453                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1454                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1455         }
1456 }
1457
1458 /*
1459  * Helper function to force a TCP close if the server is sending
1460  * junk and/or it has put us in CLOSE_WAIT
1461  */
1462 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1463 {
1464         xprt_force_disconnect(xprt);
1465 }
1466
1467 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1468 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1469 {
1470         return PAGE_SIZE;
1471 }
1472 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1473
1474 /**
1475  * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1476  * @sk: socket whose state has changed
1477  *
1478  */
1479 static void xs_local_state_change(struct sock *sk)
1480 {
1481         struct rpc_xprt *xprt;
1482         struct sock_xprt *transport;
1483
1484         if (!(xprt = xprt_from_sock(sk)))
1485                 return;
1486         transport = container_of(xprt, struct sock_xprt, xprt);
1487         if (sk->sk_shutdown & SHUTDOWN_MASK) {
1488                 clear_bit(XPRT_CONNECTED, &xprt->state);
1489                 /* Trigger the socket release */
1490                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1491         }
1492 }
1493
1494 /**
1495  * xs_tcp_state_change - callback to handle TCP socket state changes
1496  * @sk: socket whose state has changed
1497  *
1498  */
1499 static void xs_tcp_state_change(struct sock *sk)
1500 {
1501         struct rpc_xprt *xprt;
1502         struct sock_xprt *transport;
1503
1504         if (!(xprt = xprt_from_sock(sk)))
1505                 return;
1506         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1507         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1508                         sk->sk_state, xprt_connected(xprt),
1509                         sock_flag(sk, SOCK_DEAD),
1510                         sock_flag(sk, SOCK_ZAPPED),
1511                         sk->sk_shutdown);
1512
1513         transport = container_of(xprt, struct sock_xprt, xprt);
1514         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1515         switch (sk->sk_state) {
1516         case TCP_ESTABLISHED:
1517                 if (!xprt_test_and_set_connected(xprt)) {
1518                         xprt->connect_cookie++;
1519                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1520                         xprt_clear_connecting(xprt);
1521
1522                         xprt->stat.connect_count++;
1523                         xprt->stat.connect_time += (long)jiffies -
1524                                                    xprt->stat.connect_start;
1525                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1526                 }
1527                 break;
1528         case TCP_FIN_WAIT1:
1529                 /* The client initiated a shutdown of the socket */
1530                 xprt->connect_cookie++;
1531                 xprt->reestablish_timeout = 0;
1532                 set_bit(XPRT_CLOSING, &xprt->state);
1533                 smp_mb__before_atomic();
1534                 clear_bit(XPRT_CONNECTED, &xprt->state);
1535                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1536                 smp_mb__after_atomic();
1537                 break;
1538         case TCP_CLOSE_WAIT:
1539                 /* The server initiated a shutdown of the socket */
1540                 xprt->connect_cookie++;
1541                 clear_bit(XPRT_CONNECTED, &xprt->state);
1542                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1543                 fallthrough;
1544         case TCP_CLOSING:
1545                 /*
1546                  * If the server closed down the connection, make sure that
1547                  * we back off before reconnecting
1548                  */
1549                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1550                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1551                 break;
1552         case TCP_LAST_ACK:
1553                 set_bit(XPRT_CLOSING, &xprt->state);
1554                 smp_mb__before_atomic();
1555                 clear_bit(XPRT_CONNECTED, &xprt->state);
1556                 smp_mb__after_atomic();
1557                 break;
1558         case TCP_CLOSE:
1559                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1560                                         &transport->sock_state))
1561                         xprt_clear_connecting(xprt);
1562                 clear_bit(XPRT_CLOSING, &xprt->state);
1563                 /* Trigger the socket release */
1564                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1565         }
1566 }
1567
1568 static void xs_write_space(struct sock *sk)
1569 {
1570         struct sock_xprt *transport;
1571         struct rpc_xprt *xprt;
1572
1573         if (!sk->sk_socket)
1574                 return;
1575         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1576
1577         if (unlikely(!(xprt = xprt_from_sock(sk))))
1578                 return;
1579         transport = container_of(xprt, struct sock_xprt, xprt);
1580         if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1581                 return;
1582         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1583         sk->sk_write_pending--;
1584 }
1585
1586 /**
1587  * xs_udp_write_space - callback invoked when socket buffer space
1588  *                             becomes available
1589  * @sk: socket whose state has changed
1590  *
1591  * Called when more output buffer space is available for this socket.
1592  * We try not to wake our writers until they can make "significant"
1593  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1594  * with a bunch of small requests.
1595  */
1596 static void xs_udp_write_space(struct sock *sk)
1597 {
1598         /* from net/core/sock.c:sock_def_write_space */
1599         if (sock_writeable(sk))
1600                 xs_write_space(sk);
1601 }
1602
1603 /**
1604  * xs_tcp_write_space - callback invoked when socket buffer space
1605  *                             becomes available
1606  * @sk: socket whose state has changed
1607  *
1608  * Called when more output buffer space is available for this socket.
1609  * We try not to wake our writers until they can make "significant"
1610  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1611  * with a bunch of small requests.
1612  */
1613 static void xs_tcp_write_space(struct sock *sk)
1614 {
1615         /* from net/core/stream.c:sk_stream_write_space */
1616         if (sk_stream_is_writeable(sk))
1617                 xs_write_space(sk);
1618 }
1619
1620 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1621 {
1622         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1623         struct sock *sk = transport->inet;
1624
1625         if (transport->rcvsize) {
1626                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1627                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1628         }
1629         if (transport->sndsize) {
1630                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1631                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1632                 sk->sk_write_space(sk);
1633         }
1634 }
1635
1636 /**
1637  * xs_udp_set_buffer_size - set send and receive limits
1638  * @xprt: generic transport
1639  * @sndsize: requested size of send buffer, in bytes
1640  * @rcvsize: requested size of receive buffer, in bytes
1641  *
1642  * Set socket send and receive buffer size limits.
1643  */
1644 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1645 {
1646         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1647
1648         transport->sndsize = 0;
1649         if (sndsize)
1650                 transport->sndsize = sndsize + 1024;
1651         transport->rcvsize = 0;
1652         if (rcvsize)
1653                 transport->rcvsize = rcvsize + 1024;
1654
1655         xs_udp_do_set_buffer_size(xprt);
1656 }
1657
1658 /**
1659  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1660  * @xprt: controlling transport
1661  * @task: task that timed out
1662  *
1663  * Adjust the congestion window after a retransmit timeout has occurred.
1664  */
1665 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1666 {
1667         spin_lock(&xprt->transport_lock);
1668         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1669         spin_unlock(&xprt->transport_lock);
1670 }
1671
1672 static int xs_get_random_port(void)
1673 {
1674         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1675         unsigned short range;
1676         unsigned short rand;
1677
1678         if (max < min)
1679                 return -EADDRINUSE;
1680         range = max - min + 1;
1681         rand = get_random_u32_below(range);
1682         return rand + min;
1683 }
1684
1685 static unsigned short xs_sock_getport(struct socket *sock)
1686 {
1687         struct sockaddr_storage buf;
1688         unsigned short port = 0;
1689
1690         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1691                 goto out;
1692         switch (buf.ss_family) {
1693         case AF_INET6:
1694                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1695                 break;
1696         case AF_INET:
1697                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1698         }
1699 out:
1700         return port;
1701 }
1702
1703 /**
1704  * xs_set_port - reset the port number in the remote endpoint address
1705  * @xprt: generic transport
1706  * @port: new port number
1707  *
1708  */
1709 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1710 {
1711         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1712
1713         rpc_set_port(xs_addr(xprt), port);
1714         xs_update_peer_port(xprt);
1715 }
1716
1717 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1718 {
1719         if (transport->srcport == 0 && transport->xprt.reuseport)
1720                 transport->srcport = xs_sock_getport(sock);
1721 }
1722
1723 static int xs_get_srcport(struct sock_xprt *transport)
1724 {
1725         int port = transport->srcport;
1726
1727         if (port == 0 && transport->xprt.resvport)
1728                 port = xs_get_random_port();
1729         return port;
1730 }
1731
1732 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1733 {
1734         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1735         unsigned short ret = 0;
1736         mutex_lock(&sock->recv_mutex);
1737         if (sock->sock)
1738                 ret = xs_sock_getport(sock->sock);
1739         mutex_unlock(&sock->recv_mutex);
1740         return ret;
1741 }
1742
1743 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1744 {
1745         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1746         union {
1747                 struct sockaddr sa;
1748                 struct sockaddr_storage st;
1749         } saddr;
1750         int ret = -ENOTCONN;
1751
1752         mutex_lock(&sock->recv_mutex);
1753         if (sock->sock) {
1754                 ret = kernel_getsockname(sock->sock, &saddr.sa);
1755                 if (ret >= 0)
1756                         ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1757         }
1758         mutex_unlock(&sock->recv_mutex);
1759         return ret;
1760 }
1761
1762 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1763 {
1764         if (transport->srcport != 0)
1765                 transport->srcport = 0;
1766         if (!transport->xprt.resvport)
1767                 return 0;
1768         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1769                 return xprt_max_resvport;
1770         return --port;
1771 }
1772 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1773 {
1774         struct sockaddr_storage myaddr;
1775         int err, nloop = 0;
1776         int port = xs_get_srcport(transport);
1777         unsigned short last;
1778
1779         /*
1780          * If we are asking for any ephemeral port (i.e. port == 0 &&
1781          * transport->xprt.resvport == 0), don't bind.  Let the local
1782          * port selection happen implicitly when the socket is used
1783          * (for example at connect time).
1784          *
1785          * This ensures that we can continue to establish TCP
1786          * connections even when all local ephemeral ports are already
1787          * a part of some TCP connection.  This makes no difference
1788          * for UDP sockets, but also doesn't harm them.
1789          *
1790          * If we're asking for any reserved port (i.e. port == 0 &&
1791          * transport->xprt.resvport == 1) xs_get_srcport above will
1792          * ensure that port is non-zero and we will bind as needed.
1793          */
1794         if (port <= 0)
1795                 return port;
1796
1797         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1798         do {
1799                 rpc_set_port((struct sockaddr *)&myaddr, port);
1800                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1801                                 transport->xprt.addrlen);
1802                 if (err == 0) {
1803                         if (transport->xprt.reuseport)
1804                                 transport->srcport = port;
1805                         break;
1806                 }
1807                 last = port;
1808                 port = xs_next_srcport(transport, port);
1809                 if (port > last)
1810                         nloop++;
1811         } while (err == -EADDRINUSE && nloop != 2);
1812
1813         if (myaddr.ss_family == AF_INET)
1814                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1815                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1816                                 port, err ? "failed" : "ok", err);
1817         else
1818                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1819                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1820                                 port, err ? "failed" : "ok", err);
1821         return err;
1822 }
1823
1824 /*
1825  * We don't support autobind on AF_LOCAL sockets
1826  */
1827 static void xs_local_rpcbind(struct rpc_task *task)
1828 {
1829         xprt_set_bound(task->tk_xprt);
1830 }
1831
1832 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1833 {
1834 }
1835
1836 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1837 static struct lock_class_key xs_key[3];
1838 static struct lock_class_key xs_slock_key[3];
1839
1840 static inline void xs_reclassify_socketu(struct socket *sock)
1841 {
1842         struct sock *sk = sock->sk;
1843
1844         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1845                 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1846 }
1847
1848 static inline void xs_reclassify_socket4(struct socket *sock)
1849 {
1850         struct sock *sk = sock->sk;
1851
1852         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1853                 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1854 }
1855
1856 static inline void xs_reclassify_socket6(struct socket *sock)
1857 {
1858         struct sock *sk = sock->sk;
1859
1860         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1861                 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1862 }
1863
1864 static inline void xs_reclassify_socket(int family, struct socket *sock)
1865 {
1866         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1867                 return;
1868
1869         switch (family) {
1870         case AF_LOCAL:
1871                 xs_reclassify_socketu(sock);
1872                 break;
1873         case AF_INET:
1874                 xs_reclassify_socket4(sock);
1875                 break;
1876         case AF_INET6:
1877                 xs_reclassify_socket6(sock);
1878                 break;
1879         }
1880 }
1881 #else
1882 static inline void xs_reclassify_socket(int family, struct socket *sock)
1883 {
1884 }
1885 #endif
1886
1887 static void xs_dummy_setup_socket(struct work_struct *work)
1888 {
1889 }
1890
1891 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1892                 struct sock_xprt *transport, int family, int type,
1893                 int protocol, bool reuseport)
1894 {
1895         struct file *filp;
1896         struct socket *sock;
1897         int err;
1898
1899         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1900         if (err < 0) {
1901                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1902                                 protocol, -err);
1903                 goto out;
1904         }
1905         xs_reclassify_socket(family, sock);
1906
1907         if (reuseport)
1908                 sock_set_reuseport(sock->sk);
1909
1910         err = xs_bind(transport, sock);
1911         if (err) {
1912                 sock_release(sock);
1913                 goto out;
1914         }
1915
1916         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1917         if (IS_ERR(filp))
1918                 return ERR_CAST(filp);
1919         transport->file = filp;
1920
1921         return sock;
1922 out:
1923         return ERR_PTR(err);
1924 }
1925
1926 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1927                                       struct socket *sock)
1928 {
1929         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1930                                                                         xprt);
1931
1932         if (!transport->inet) {
1933                 struct sock *sk = sock->sk;
1934
1935                 lock_sock(sk);
1936
1937                 xs_save_old_callbacks(transport, sk);
1938
1939                 sk->sk_user_data = xprt;
1940                 sk->sk_data_ready = xs_data_ready;
1941                 sk->sk_write_space = xs_udp_write_space;
1942                 sk->sk_state_change = xs_local_state_change;
1943                 sk->sk_error_report = xs_error_report;
1944                 sk->sk_use_task_frag = false;
1945
1946                 xprt_clear_connected(xprt);
1947
1948                 /* Reset to new socket */
1949                 transport->sock = sock;
1950                 transport->inet = sk;
1951
1952                 release_sock(sk);
1953         }
1954
1955         xs_stream_start_connect(transport);
1956
1957         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1958 }
1959
1960 /**
1961  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1962  * @transport: socket transport to connect
1963  */
1964 static int xs_local_setup_socket(struct sock_xprt *transport)
1965 {
1966         struct rpc_xprt *xprt = &transport->xprt;
1967         struct file *filp;
1968         struct socket *sock;
1969         int status;
1970
1971         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1972                                         SOCK_STREAM, 0, &sock, 1);
1973         if (status < 0) {
1974                 dprintk("RPC:       can't create AF_LOCAL "
1975                         "transport socket (%d).\n", -status);
1976                 goto out;
1977         }
1978         xs_reclassify_socket(AF_LOCAL, sock);
1979
1980         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1981         if (IS_ERR(filp)) {
1982                 status = PTR_ERR(filp);
1983                 goto out;
1984         }
1985         transport->file = filp;
1986
1987         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1988                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1989
1990         status = xs_local_finish_connecting(xprt, sock);
1991         trace_rpc_socket_connect(xprt, sock, status);
1992         switch (status) {
1993         case 0:
1994                 dprintk("RPC:       xprt %p connected to %s\n",
1995                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1996                 xprt->stat.connect_count++;
1997                 xprt->stat.connect_time += (long)jiffies -
1998                                            xprt->stat.connect_start;
1999                 xprt_set_connected(xprt);
2000                 break;
2001         case -ENOBUFS:
2002                 break;
2003         case -ENOENT:
2004                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2005                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2006                 break;
2007         case -ECONNREFUSED:
2008                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2009                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2010                 break;
2011         default:
2012                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2013                                 __func__, -status,
2014                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2015         }
2016
2017 out:
2018         xprt_clear_connecting(xprt);
2019         xprt_wake_pending_tasks(xprt, status);
2020         return status;
2021 }
2022
2023 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2024 {
2025         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2026         int ret;
2027
2028         if (transport->file)
2029                 goto force_disconnect;
2030
2031         if (RPC_IS_ASYNC(task)) {
2032                 /*
2033                  * We want the AF_LOCAL connect to be resolved in the
2034                  * filesystem namespace of the process making the rpc
2035                  * call.  Thus we connect synchronously.
2036                  *
2037                  * If we want to support asynchronous AF_LOCAL calls,
2038                  * we'll need to figure out how to pass a namespace to
2039                  * connect.
2040                  */
2041                 rpc_task_set_rpc_status(task, -ENOTCONN);
2042                 goto out_wake;
2043         }
2044         ret = xs_local_setup_socket(transport);
2045         if (ret && !RPC_IS_SOFTCONN(task))
2046                 msleep_interruptible(15000);
2047         return;
2048 force_disconnect:
2049         xprt_force_disconnect(xprt);
2050 out_wake:
2051         xprt_clear_connecting(xprt);
2052         xprt_wake_pending_tasks(xprt, -ENOTCONN);
2053 }
2054
2055 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2056 /*
2057  * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2058  * held, or when we otherwise know that we have exclusive access to the
2059  * socket, to guard against races with xs_reset_transport.
2060  */
2061 static void xs_set_memalloc(struct rpc_xprt *xprt)
2062 {
2063         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2064                         xprt);
2065
2066         /*
2067          * If there's no sock, then we have nothing to set. The
2068          * reconnecting process will get it for us.
2069          */
2070         if (!transport->inet)
2071                 return;
2072         if (atomic_read(&xprt->swapper))
2073                 sk_set_memalloc(transport->inet);
2074 }
2075
2076 /**
2077  * xs_enable_swap - Tag this transport as being used for swap.
2078  * @xprt: transport to tag
2079  *
2080  * Take a reference to this transport on behalf of the rpc_clnt, and
2081  * optionally mark it for swapping if it wasn't already.
2082  */
2083 static int
2084 xs_enable_swap(struct rpc_xprt *xprt)
2085 {
2086         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2087
2088         mutex_lock(&xs->recv_mutex);
2089         if (atomic_inc_return(&xprt->swapper) == 1 &&
2090             xs->inet)
2091                 sk_set_memalloc(xs->inet);
2092         mutex_unlock(&xs->recv_mutex);
2093         return 0;
2094 }
2095
2096 /**
2097  * xs_disable_swap - Untag this transport as being used for swap.
2098  * @xprt: transport to tag
2099  *
2100  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2101  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2102  */
2103 static void
2104 xs_disable_swap(struct rpc_xprt *xprt)
2105 {
2106         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2107
2108         mutex_lock(&xs->recv_mutex);
2109         if (atomic_dec_and_test(&xprt->swapper) &&
2110             xs->inet)
2111                 sk_clear_memalloc(xs->inet);
2112         mutex_unlock(&xs->recv_mutex);
2113 }
2114 #else
2115 static void xs_set_memalloc(struct rpc_xprt *xprt)
2116 {
2117 }
2118
2119 static int
2120 xs_enable_swap(struct rpc_xprt *xprt)
2121 {
2122         return -EINVAL;
2123 }
2124
2125 static void
2126 xs_disable_swap(struct rpc_xprt *xprt)
2127 {
2128 }
2129 #endif
2130
2131 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2132 {
2133         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2134
2135         if (!transport->inet) {
2136                 struct sock *sk = sock->sk;
2137
2138                 lock_sock(sk);
2139
2140                 xs_save_old_callbacks(transport, sk);
2141
2142                 sk->sk_user_data = xprt;
2143                 sk->sk_data_ready = xs_data_ready;
2144                 sk->sk_write_space = xs_udp_write_space;
2145                 sk->sk_use_task_frag = false;
2146
2147                 xprt_set_connected(xprt);
2148
2149                 /* Reset to new socket */
2150                 transport->sock = sock;
2151                 transport->inet = sk;
2152
2153                 xs_set_memalloc(xprt);
2154
2155                 release_sock(sk);
2156         }
2157         xs_udp_do_set_buffer_size(xprt);
2158
2159         xprt->stat.connect_start = jiffies;
2160 }
2161
2162 static void xs_udp_setup_socket(struct work_struct *work)
2163 {
2164         struct sock_xprt *transport =
2165                 container_of(work, struct sock_xprt, connect_worker.work);
2166         struct rpc_xprt *xprt = &transport->xprt;
2167         struct socket *sock;
2168         int status = -EIO;
2169         unsigned int pflags = current->flags;
2170
2171         if (atomic_read(&xprt->swapper))
2172                 current->flags |= PF_MEMALLOC;
2173         sock = xs_create_sock(xprt, transport,
2174                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2175                         IPPROTO_UDP, false);
2176         if (IS_ERR(sock))
2177                 goto out;
2178
2179         dprintk("RPC:       worker connecting xprt %p via %s to "
2180                                 "%s (port %s)\n", xprt,
2181                         xprt->address_strings[RPC_DISPLAY_PROTO],
2182                         xprt->address_strings[RPC_DISPLAY_ADDR],
2183                         xprt->address_strings[RPC_DISPLAY_PORT]);
2184
2185         xs_udp_finish_connecting(xprt, sock);
2186         trace_rpc_socket_connect(xprt, sock, 0);
2187         status = 0;
2188 out:
2189         xprt_clear_connecting(xprt);
2190         xprt_unlock_connect(xprt, transport);
2191         xprt_wake_pending_tasks(xprt, status);
2192         current_restore_flags(pflags, PF_MEMALLOC);
2193 }
2194
2195 /**
2196  * xs_tcp_shutdown - gracefully shut down a TCP socket
2197  * @xprt: transport
2198  *
2199  * Initiates a graceful shutdown of the TCP socket by calling the
2200  * equivalent of shutdown(SHUT_RDWR);
2201  */
2202 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2203 {
2204         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2205         struct socket *sock = transport->sock;
2206         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2207
2208         if (sock == NULL)
2209                 return;
2210         if (!xprt->reuseport) {
2211                 xs_close(xprt);
2212                 return;
2213         }
2214         switch (skst) {
2215         case TCP_FIN_WAIT1:
2216         case TCP_FIN_WAIT2:
2217         case TCP_LAST_ACK:
2218                 break;
2219         case TCP_ESTABLISHED:
2220         case TCP_CLOSE_WAIT:
2221                 kernel_sock_shutdown(sock, SHUT_RDWR);
2222                 trace_rpc_socket_shutdown(xprt, sock);
2223                 break;
2224         default:
2225                 xs_reset_transport(transport);
2226         }
2227 }
2228
2229 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2230                 struct socket *sock)
2231 {
2232         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2233         struct net *net = sock_net(sock->sk);
2234         unsigned long connect_timeout;
2235         unsigned long syn_retries;
2236         unsigned int keepidle;
2237         unsigned int keepcnt;
2238         unsigned int timeo;
2239         unsigned long t;
2240
2241         spin_lock(&xprt->transport_lock);
2242         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2243         keepcnt = xprt->timeout->to_retries + 1;
2244         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2245                 (xprt->timeout->to_retries + 1);
2246         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2247         spin_unlock(&xprt->transport_lock);
2248
2249         /* TCP Keepalive options */
2250         sock_set_keepalive(sock->sk);
2251         tcp_sock_set_keepidle(sock->sk, keepidle);
2252         tcp_sock_set_keepintvl(sock->sk, keepidle);
2253         tcp_sock_set_keepcnt(sock->sk, keepcnt);
2254
2255         /* TCP user timeout (see RFC5482) */
2256         tcp_sock_set_user_timeout(sock->sk, timeo);
2257
2258         /* Connect timeout */
2259         connect_timeout = max_t(unsigned long,
2260                                 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2261         syn_retries = max_t(unsigned long,
2262                             READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2263         for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2264                 ;
2265         if (t <= syn_retries)
2266                 tcp_sock_set_syncnt(sock->sk, t - 1);
2267 }
2268
2269 static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2270                                           unsigned long connect_timeout)
2271 {
2272         struct sock_xprt *transport =
2273                 container_of(xprt, struct sock_xprt, xprt);
2274         struct rpc_timeout to;
2275         unsigned long initval;
2276
2277         memcpy(&to, xprt->timeout, sizeof(to));
2278         /* Arbitrary lower limit */
2279         initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2280         to.to_initval = initval;
2281         to.to_maxval = initval;
2282         to.to_retries = 0;
2283         memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2284         xprt->timeout = &transport->tcp_timeout;
2285         xprt->connect_timeout = connect_timeout;
2286 }
2287
2288 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2289                 unsigned long connect_timeout,
2290                 unsigned long reconnect_timeout)
2291 {
2292         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2293
2294         spin_lock(&xprt->transport_lock);
2295         if (reconnect_timeout < xprt->max_reconnect_timeout)
2296                 xprt->max_reconnect_timeout = reconnect_timeout;
2297         if (connect_timeout < xprt->connect_timeout)
2298                 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2299         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2300         spin_unlock(&xprt->transport_lock);
2301 }
2302
2303 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2304 {
2305         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2306
2307         if (!transport->inet) {
2308                 struct sock *sk = sock->sk;
2309
2310                 /* Avoid temporary address, they are bad for long-lived
2311                  * connections such as NFS mounts.
2312                  * RFC4941, section 3.6 suggests that:
2313                  *    Individual applications, which have specific
2314                  *    knowledge about the normal duration of connections,
2315                  *    MAY override this as appropriate.
2316                  */
2317                 if (xs_addr(xprt)->sa_family == PF_INET6) {
2318                         ip6_sock_set_addr_preferences(sk,
2319                                 IPV6_PREFER_SRC_PUBLIC);
2320                 }
2321
2322                 xs_tcp_set_socket_timeouts(xprt, sock);
2323                 tcp_sock_set_nodelay(sk);
2324
2325                 lock_sock(sk);
2326
2327                 xs_save_old_callbacks(transport, sk);
2328
2329                 sk->sk_user_data = xprt;
2330                 sk->sk_data_ready = xs_data_ready;
2331                 sk->sk_state_change = xs_tcp_state_change;
2332                 sk->sk_write_space = xs_tcp_write_space;
2333                 sk->sk_error_report = xs_error_report;
2334                 sk->sk_use_task_frag = false;
2335
2336                 /* socket options */
2337                 sock_reset_flag(sk, SOCK_LINGER);
2338
2339                 xprt_clear_connected(xprt);
2340
2341                 /* Reset to new socket */
2342                 transport->sock = sock;
2343                 transport->inet = sk;
2344
2345                 release_sock(sk);
2346         }
2347
2348         if (!xprt_bound(xprt))
2349                 return -ENOTCONN;
2350
2351         xs_set_memalloc(xprt);
2352
2353         xs_stream_start_connect(transport);
2354
2355         /* Tell the socket layer to start connecting... */
2356         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2357         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2358 }
2359
2360 /**
2361  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2362  * @work: queued work item
2363  *
2364  * Invoked by a work queue tasklet.
2365  */
2366 static void xs_tcp_setup_socket(struct work_struct *work)
2367 {
2368         struct sock_xprt *transport =
2369                 container_of(work, struct sock_xprt, connect_worker.work);
2370         struct socket *sock = transport->sock;
2371         struct rpc_xprt *xprt = &transport->xprt;
2372         int status;
2373         unsigned int pflags = current->flags;
2374
2375         if (atomic_read(&xprt->swapper))
2376                 current->flags |= PF_MEMALLOC;
2377
2378         if (xprt_connected(xprt))
2379                 goto out;
2380         if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2381                                &transport->sock_state) ||
2382             !sock) {
2383                 xs_reset_transport(transport);
2384                 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2385                                       SOCK_STREAM, IPPROTO_TCP, true);
2386                 if (IS_ERR(sock)) {
2387                         xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2388                         goto out;
2389                 }
2390         }
2391
2392         dprintk("RPC:       worker connecting xprt %p via %s to "
2393                                 "%s (port %s)\n", xprt,
2394                         xprt->address_strings[RPC_DISPLAY_PROTO],
2395                         xprt->address_strings[RPC_DISPLAY_ADDR],
2396                         xprt->address_strings[RPC_DISPLAY_PORT]);
2397
2398         status = xs_tcp_finish_connecting(xprt, sock);
2399         trace_rpc_socket_connect(xprt, sock, status);
2400         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2401                         xprt, -status, xprt_connected(xprt),
2402                         sock->sk->sk_state);
2403         switch (status) {
2404         case 0:
2405         case -EINPROGRESS:
2406                 /* SYN_SENT! */
2407                 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2408                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2409                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2410                 fallthrough;
2411         case -EALREADY:
2412                 goto out_unlock;
2413         case -EADDRNOTAVAIL:
2414                 /* Source port number is unavailable. Try a new one! */
2415                 transport->srcport = 0;
2416                 status = -EAGAIN;
2417                 break;
2418         case -EINVAL:
2419                 /* Happens, for instance, if the user specified a link
2420                  * local IPv6 address without a scope-id.
2421                  */
2422         case -ECONNREFUSED:
2423         case -ECONNRESET:
2424         case -ENETDOWN:
2425         case -ENETUNREACH:
2426         case -EHOSTUNREACH:
2427         case -EADDRINUSE:
2428         case -ENOBUFS:
2429                 break;
2430         default:
2431                 printk("%s: connect returned unhandled error %d\n",
2432                         __func__, status);
2433                 status = -EAGAIN;
2434         }
2435
2436         /* xs_tcp_force_close() wakes tasks with a fixed error code.
2437          * We need to wake them first to ensure the correct error code.
2438          */
2439         xprt_wake_pending_tasks(xprt, status);
2440         xs_tcp_force_close(xprt);
2441 out:
2442         xprt_clear_connecting(xprt);
2443 out_unlock:
2444         xprt_unlock_connect(xprt, transport);
2445         current_restore_flags(pflags, PF_MEMALLOC);
2446 }
2447
2448 /*
2449  * Transfer the connected socket to @upper_transport, then mark that
2450  * xprt CONNECTED.
2451  */
2452 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2453                                         struct sock_xprt *upper_transport)
2454 {
2455         struct sock_xprt *lower_transport =
2456                         container_of(lower_xprt, struct sock_xprt, xprt);
2457         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2458
2459         if (!upper_transport->inet) {
2460                 struct socket *sock = lower_transport->sock;
2461                 struct sock *sk = sock->sk;
2462
2463                 /* Avoid temporary address, they are bad for long-lived
2464                  * connections such as NFS mounts.
2465                  * RFC4941, section 3.6 suggests that:
2466                  *    Individual applications, which have specific
2467                  *    knowledge about the normal duration of connections,
2468                  *    MAY override this as appropriate.
2469                  */
2470                 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2471                         ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2472
2473                 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2474                 tcp_sock_set_nodelay(sk);
2475
2476                 lock_sock(sk);
2477
2478                 /* @sk is already connected, so it now has the RPC callbacks.
2479                  * Reach into @lower_transport to save the original ones.
2480                  */
2481                 upper_transport->old_data_ready = lower_transport->old_data_ready;
2482                 upper_transport->old_state_change = lower_transport->old_state_change;
2483                 upper_transport->old_write_space = lower_transport->old_write_space;
2484                 upper_transport->old_error_report = lower_transport->old_error_report;
2485                 sk->sk_user_data = upper_xprt;
2486
2487                 /* socket options */
2488                 sock_reset_flag(sk, SOCK_LINGER);
2489
2490                 xprt_clear_connected(upper_xprt);
2491
2492                 upper_transport->sock = sock;
2493                 upper_transport->inet = sk;
2494                 upper_transport->file = lower_transport->file;
2495
2496                 release_sock(sk);
2497
2498                 /* Reset lower_transport before shutting down its clnt */
2499                 mutex_lock(&lower_transport->recv_mutex);
2500                 lower_transport->inet = NULL;
2501                 lower_transport->sock = NULL;
2502                 lower_transport->file = NULL;
2503
2504                 xprt_clear_connected(lower_xprt);
2505                 xs_sock_reset_connection_flags(lower_xprt);
2506                 xs_stream_reset_connect(lower_transport);
2507                 mutex_unlock(&lower_transport->recv_mutex);
2508         }
2509
2510         if (!xprt_bound(upper_xprt))
2511                 return -ENOTCONN;
2512
2513         xs_set_memalloc(upper_xprt);
2514
2515         if (!xprt_test_and_set_connected(upper_xprt)) {
2516                 upper_xprt->connect_cookie++;
2517                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2518                 xprt_clear_connecting(upper_xprt);
2519
2520                 upper_xprt->stat.connect_count++;
2521                 upper_xprt->stat.connect_time += (long)jiffies -
2522                                            upper_xprt->stat.connect_start;
2523                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2524         }
2525         return 0;
2526 }
2527
2528 /**
2529  * xs_tls_handshake_done - TLS handshake completion handler
2530  * @data: address of xprt to wake
2531  * @status: status of handshake
2532  * @peerid: serial number of key containing the remote's identity
2533  *
2534  */
2535 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2536 {
2537         struct rpc_xprt *lower_xprt = data;
2538         struct sock_xprt *lower_transport =
2539                                 container_of(lower_xprt, struct sock_xprt, xprt);
2540
2541         lower_transport->xprt_err = status ? -EACCES : 0;
2542         complete(&lower_transport->handshake_done);
2543         xprt_put(lower_xprt);
2544 }
2545
2546 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2547 {
2548         struct sock_xprt *lower_transport =
2549                                 container_of(lower_xprt, struct sock_xprt, xprt);
2550         struct tls_handshake_args args = {
2551                 .ta_sock        = lower_transport->sock,
2552                 .ta_done        = xs_tls_handshake_done,
2553                 .ta_data        = xprt_get(lower_xprt),
2554                 .ta_peername    = lower_xprt->servername,
2555         };
2556         struct sock *sk = lower_transport->inet;
2557         int rc;
2558
2559         init_completion(&lower_transport->handshake_done);
2560         set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2561         lower_transport->xprt_err = -ETIMEDOUT;
2562         switch (xprtsec->policy) {
2563         case RPC_XPRTSEC_TLS_ANON:
2564                 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2565                 if (rc)
2566                         goto out_put_xprt;
2567                 break;
2568         case RPC_XPRTSEC_TLS_X509:
2569                 args.ta_my_cert = xprtsec->cert_serial;
2570                 args.ta_my_privkey = xprtsec->privkey_serial;
2571                 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2572                 if (rc)
2573                         goto out_put_xprt;
2574                 break;
2575         default:
2576                 rc = -EACCES;
2577                 goto out_put_xprt;
2578         }
2579
2580         rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2581                                                        XS_TLS_HANDSHAKE_TO);
2582         if (rc <= 0) {
2583                 if (!tls_handshake_cancel(sk)) {
2584                         if (rc == 0)
2585                                 rc = -ETIMEDOUT;
2586                         goto out_put_xprt;
2587                 }
2588         }
2589
2590         rc = lower_transport->xprt_err;
2591
2592 out:
2593         xs_stream_reset_connect(lower_transport);
2594         clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2595         return rc;
2596
2597 out_put_xprt:
2598         xprt_put(lower_xprt);
2599         goto out;
2600 }
2601
2602 /**
2603  * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2604  * @work: queued work item
2605  *
2606  * Invoked by a work queue tasklet.
2607  *
2608  * For RPC-with-TLS, there is a two-stage connection process.
2609  *
2610  * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2611  * been marked connected, the consumer knows that a TCP connection and
2612  * a TLS session have been established.
2613  *
2614  * A "lower-layer xprt", created in this function, handles the mechanics
2615  * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2616  * then driving the TLS handshake. Once all that is complete, the upper
2617  * layer xprt is marked connected.
2618  */
2619 static void xs_tcp_tls_setup_socket(struct work_struct *work)
2620 {
2621         struct sock_xprt *upper_transport =
2622                 container_of(work, struct sock_xprt, connect_worker.work);
2623         struct rpc_clnt *upper_clnt = upper_transport->clnt;
2624         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2625         struct rpc_create_args args = {
2626                 .net            = upper_xprt->xprt_net,
2627                 .protocol       = upper_xprt->prot,
2628                 .address        = (struct sockaddr *)&upper_xprt->addr,
2629                 .addrsize       = upper_xprt->addrlen,
2630                 .timeout        = upper_clnt->cl_timeout,
2631                 .servername     = upper_xprt->servername,
2632                 .program        = upper_clnt->cl_program,
2633                 .prognumber     = upper_clnt->cl_prog,
2634                 .version        = upper_clnt->cl_vers,
2635                 .authflavor     = RPC_AUTH_TLS,
2636                 .cred           = upper_clnt->cl_cred,
2637                 .xprtsec        = {
2638                         .policy         = RPC_XPRTSEC_NONE,
2639                 },
2640         };
2641         unsigned int pflags = current->flags;
2642         struct rpc_clnt *lower_clnt;
2643         struct rpc_xprt *lower_xprt;
2644         int status;
2645
2646         if (atomic_read(&upper_xprt->swapper))
2647                 current->flags |= PF_MEMALLOC;
2648
2649         xs_stream_start_connect(upper_transport);
2650
2651         /* This implicitly sends an RPC_AUTH_TLS probe */
2652         lower_clnt = rpc_create(&args);
2653         if (IS_ERR(lower_clnt)) {
2654                 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2655                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2656                 xprt_clear_connecting(upper_xprt);
2657                 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2658                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2659                 goto out_unlock;
2660         }
2661
2662         /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2663          * the lower xprt.
2664          */
2665         rcu_read_lock();
2666         lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2667         rcu_read_unlock();
2668         status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2669         if (status) {
2670                 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2671                 goto out_close;
2672         }
2673
2674         status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2675         if (status)
2676                 goto out_close;
2677
2678         trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2679         if (!xprt_test_and_set_connected(upper_xprt)) {
2680                 upper_xprt->connect_cookie++;
2681                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2682                 xprt_clear_connecting(upper_xprt);
2683
2684                 upper_xprt->stat.connect_count++;
2685                 upper_xprt->stat.connect_time += (long)jiffies -
2686                                            upper_xprt->stat.connect_start;
2687                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2688         }
2689         rpc_shutdown_client(lower_clnt);
2690
2691 out_unlock:
2692         current_restore_flags(pflags, PF_MEMALLOC);
2693         upper_transport->clnt = NULL;
2694         xprt_unlock_connect(upper_xprt, upper_transport);
2695         return;
2696
2697 out_close:
2698         rpc_shutdown_client(lower_clnt);
2699
2700         /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2701          * Wake them first here to ensure they get our tk_status code.
2702          */
2703         xprt_wake_pending_tasks(upper_xprt, status);
2704         xs_tcp_force_close(upper_xprt);
2705         xprt_clear_connecting(upper_xprt);
2706         goto out_unlock;
2707 }
2708
2709 /**
2710  * xs_connect - connect a socket to a remote endpoint
2711  * @xprt: pointer to transport structure
2712  * @task: address of RPC task that manages state of connect request
2713  *
2714  * TCP: If the remote end dropped the connection, delay reconnecting.
2715  *
2716  * UDP socket connects are synchronous, but we use a work queue anyway
2717  * to guarantee that even unprivileged user processes can set up a
2718  * socket on a privileged port.
2719  *
2720  * If a UDP socket connect fails, the delay behavior here prevents
2721  * retry floods (hard mounts).
2722  */
2723 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2724 {
2725         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2726         unsigned long delay = 0;
2727
2728         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2729
2730         if (transport->sock != NULL) {
2731                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2732                         "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2733
2734                 delay = xprt_reconnect_delay(xprt);
2735                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2736
2737         } else
2738                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2739
2740         transport->clnt = task->tk_client;
2741         queue_delayed_work(xprtiod_workqueue,
2742                         &transport->connect_worker,
2743                         delay);
2744 }
2745
2746 static void xs_wake_disconnect(struct sock_xprt *transport)
2747 {
2748         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2749                 xs_tcp_force_close(&transport->xprt);
2750 }
2751
2752 static void xs_wake_write(struct sock_xprt *transport)
2753 {
2754         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2755                 xprt_write_space(&transport->xprt);
2756 }
2757
2758 static void xs_wake_error(struct sock_xprt *transport)
2759 {
2760         int sockerr;
2761
2762         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2763                 return;
2764         mutex_lock(&transport->recv_mutex);
2765         if (transport->sock == NULL)
2766                 goto out;
2767         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2768                 goto out;
2769         sockerr = xchg(&transport->xprt_err, 0);
2770         if (sockerr < 0)
2771                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2772 out:
2773         mutex_unlock(&transport->recv_mutex);
2774 }
2775
2776 static void xs_wake_pending(struct sock_xprt *transport)
2777 {
2778         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2779                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2780 }
2781
2782 static void xs_error_handle(struct work_struct *work)
2783 {
2784         struct sock_xprt *transport = container_of(work,
2785                         struct sock_xprt, error_worker);
2786
2787         xs_wake_disconnect(transport);
2788         xs_wake_write(transport);
2789         xs_wake_error(transport);
2790         xs_wake_pending(transport);
2791 }
2792
2793 /**
2794  * xs_local_print_stats - display AF_LOCAL socket-specific stats
2795  * @xprt: rpc_xprt struct containing statistics
2796  * @seq: output file
2797  *
2798  */
2799 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2800 {
2801         long idle_time = 0;
2802
2803         if (xprt_connected(xprt))
2804                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2805
2806         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2807                         "%llu %llu %lu %llu %llu\n",
2808                         xprt->stat.bind_count,
2809                         xprt->stat.connect_count,
2810                         xprt->stat.connect_time / HZ,
2811                         idle_time,
2812                         xprt->stat.sends,
2813                         xprt->stat.recvs,
2814                         xprt->stat.bad_xids,
2815                         xprt->stat.req_u,
2816                         xprt->stat.bklog_u,
2817                         xprt->stat.max_slots,
2818                         xprt->stat.sending_u,
2819                         xprt->stat.pending_u);
2820 }
2821
2822 /**
2823  * xs_udp_print_stats - display UDP socket-specific stats
2824  * @xprt: rpc_xprt struct containing statistics
2825  * @seq: output file
2826  *
2827  */
2828 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2829 {
2830         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2831
2832         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2833                         "%lu %llu %llu\n",
2834                         transport->srcport,
2835                         xprt->stat.bind_count,
2836                         xprt->stat.sends,
2837                         xprt->stat.recvs,
2838                         xprt->stat.bad_xids,
2839                         xprt->stat.req_u,
2840                         xprt->stat.bklog_u,
2841                         xprt->stat.max_slots,
2842                         xprt->stat.sending_u,
2843                         xprt->stat.pending_u);
2844 }
2845
2846 /**
2847  * xs_tcp_print_stats - display TCP socket-specific stats
2848  * @xprt: rpc_xprt struct containing statistics
2849  * @seq: output file
2850  *
2851  */
2852 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2853 {
2854         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2855         long idle_time = 0;
2856
2857         if (xprt_connected(xprt))
2858                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2859
2860         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2861                         "%llu %llu %lu %llu %llu\n",
2862                         transport->srcport,
2863                         xprt->stat.bind_count,
2864                         xprt->stat.connect_count,
2865                         xprt->stat.connect_time / HZ,
2866                         idle_time,
2867                         xprt->stat.sends,
2868                         xprt->stat.recvs,
2869                         xprt->stat.bad_xids,
2870                         xprt->stat.req_u,
2871                         xprt->stat.bklog_u,
2872                         xprt->stat.max_slots,
2873                         xprt->stat.sending_u,
2874                         xprt->stat.pending_u);
2875 }
2876
2877 /*
2878  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2879  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2880  * to use the server side send routines.
2881  */
2882 static int bc_malloc(struct rpc_task *task)
2883 {
2884         struct rpc_rqst *rqst = task->tk_rqstp;
2885         size_t size = rqst->rq_callsize;
2886         struct page *page;
2887         struct rpc_buffer *buf;
2888
2889         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2890                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2891                           size);
2892                 return -EINVAL;
2893         }
2894
2895         page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2896         if (!page)
2897                 return -ENOMEM;
2898
2899         buf = page_address(page);
2900         buf->len = PAGE_SIZE;
2901
2902         rqst->rq_buffer = buf->data;
2903         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2904         return 0;
2905 }
2906
2907 /*
2908  * Free the space allocated in the bc_alloc routine
2909  */
2910 static void bc_free(struct rpc_task *task)
2911 {
2912         void *buffer = task->tk_rqstp->rq_buffer;
2913         struct rpc_buffer *buf;
2914
2915         buf = container_of(buffer, struct rpc_buffer, data);
2916         free_page((unsigned long)buf);
2917 }
2918
2919 static int bc_sendto(struct rpc_rqst *req)
2920 {
2921         struct xdr_buf *xdr = &req->rq_snd_buf;
2922         struct sock_xprt *transport =
2923                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2924         struct msghdr msg = {
2925                 .msg_flags      = 0,
2926         };
2927         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2928                                          (u32)xdr->len);
2929         unsigned int sent = 0;
2930         int err;
2931
2932         req->rq_xtime = ktime_get();
2933         err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2934         if (err < 0)
2935                 return err;
2936         err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2937         xdr_free_bvec(xdr);
2938         if (err < 0 || sent != (xdr->len + sizeof(marker)))
2939                 return -EAGAIN;
2940         return sent;
2941 }
2942
2943 /**
2944  * bc_send_request - Send a backchannel Call on a TCP socket
2945  * @req: rpc_rqst containing Call message to be sent
2946  *
2947  * xpt_mutex ensures @rqstp's whole message is written to the socket
2948  * without interruption.
2949  *
2950  * Return values:
2951  *   %0 if the message was sent successfully
2952  *   %ENOTCONN if the message was not sent
2953  */
2954 static int bc_send_request(struct rpc_rqst *req)
2955 {
2956         struct svc_xprt *xprt;
2957         int len;
2958
2959         /*
2960          * Get the server socket associated with this callback xprt
2961          */
2962         xprt = req->rq_xprt->bc_xprt;
2963
2964         /*
2965          * Grab the mutex to serialize data as the connection is shared
2966          * with the fore channel
2967          */
2968         mutex_lock(&xprt->xpt_mutex);
2969         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2970                 len = -ENOTCONN;
2971         else
2972                 len = bc_sendto(req);
2973         mutex_unlock(&xprt->xpt_mutex);
2974
2975         if (len > 0)
2976                 len = 0;
2977
2978         return len;
2979 }
2980
2981 /*
2982  * The close routine. Since this is client initiated, we do nothing
2983  */
2984
2985 static void bc_close(struct rpc_xprt *xprt)
2986 {
2987         xprt_disconnect_done(xprt);
2988 }
2989
2990 /*
2991  * The xprt destroy routine. Again, because this connection is client
2992  * initiated, we do nothing
2993  */
2994
2995 static void bc_destroy(struct rpc_xprt *xprt)
2996 {
2997         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2998
2999         xs_xprt_free(xprt);
3000         module_put(THIS_MODULE);
3001 }
3002
3003 static const struct rpc_xprt_ops xs_local_ops = {
3004         .reserve_xprt           = xprt_reserve_xprt,
3005         .release_xprt           = xprt_release_xprt,
3006         .alloc_slot             = xprt_alloc_slot,
3007         .free_slot              = xprt_free_slot,
3008         .rpcbind                = xs_local_rpcbind,
3009         .set_port               = xs_local_set_port,
3010         .connect                = xs_local_connect,
3011         .buf_alloc              = rpc_malloc,
3012         .buf_free               = rpc_free,
3013         .prepare_request        = xs_stream_prepare_request,
3014         .send_request           = xs_local_send_request,
3015         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3016         .close                  = xs_close,
3017         .destroy                = xs_destroy,
3018         .print_stats            = xs_local_print_stats,
3019         .enable_swap            = xs_enable_swap,
3020         .disable_swap           = xs_disable_swap,
3021 };
3022
3023 static const struct rpc_xprt_ops xs_udp_ops = {
3024         .set_buffer_size        = xs_udp_set_buffer_size,
3025         .reserve_xprt           = xprt_reserve_xprt_cong,
3026         .release_xprt           = xprt_release_xprt_cong,
3027         .alloc_slot             = xprt_alloc_slot,
3028         .free_slot              = xprt_free_slot,
3029         .rpcbind                = rpcb_getport_async,
3030         .set_port               = xs_set_port,
3031         .connect                = xs_connect,
3032         .get_srcaddr            = xs_sock_srcaddr,
3033         .get_srcport            = xs_sock_srcport,
3034         .buf_alloc              = rpc_malloc,
3035         .buf_free               = rpc_free,
3036         .send_request           = xs_udp_send_request,
3037         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3038         .timer                  = xs_udp_timer,
3039         .release_request        = xprt_release_rqst_cong,
3040         .close                  = xs_close,
3041         .destroy                = xs_destroy,
3042         .print_stats            = xs_udp_print_stats,
3043         .enable_swap            = xs_enable_swap,
3044         .disable_swap           = xs_disable_swap,
3045         .inject_disconnect      = xs_inject_disconnect,
3046 };
3047
3048 static const struct rpc_xprt_ops xs_tcp_ops = {
3049         .reserve_xprt           = xprt_reserve_xprt,
3050         .release_xprt           = xprt_release_xprt,
3051         .alloc_slot             = xprt_alloc_slot,
3052         .free_slot              = xprt_free_slot,
3053         .rpcbind                = rpcb_getport_async,
3054         .set_port               = xs_set_port,
3055         .connect                = xs_connect,
3056         .get_srcaddr            = xs_sock_srcaddr,
3057         .get_srcport            = xs_sock_srcport,
3058         .buf_alloc              = rpc_malloc,
3059         .buf_free               = rpc_free,
3060         .prepare_request        = xs_stream_prepare_request,
3061         .send_request           = xs_tcp_send_request,
3062         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3063         .close                  = xs_tcp_shutdown,
3064         .destroy                = xs_destroy,
3065         .set_connect_timeout    = xs_tcp_set_connect_timeout,
3066         .print_stats            = xs_tcp_print_stats,
3067         .enable_swap            = xs_enable_swap,
3068         .disable_swap           = xs_disable_swap,
3069         .inject_disconnect      = xs_inject_disconnect,
3070 #ifdef CONFIG_SUNRPC_BACKCHANNEL
3071         .bc_setup               = xprt_setup_bc,
3072         .bc_maxpayload          = xs_tcp_bc_maxpayload,
3073         .bc_num_slots           = xprt_bc_max_slots,
3074         .bc_free_rqst           = xprt_free_bc_rqst,
3075         .bc_destroy             = xprt_destroy_bc,
3076 #endif
3077 };
3078
3079 /*
3080  * The rpc_xprt_ops for the server backchannel
3081  */
3082
3083 static const struct rpc_xprt_ops bc_tcp_ops = {
3084         .reserve_xprt           = xprt_reserve_xprt,
3085         .release_xprt           = xprt_release_xprt,
3086         .alloc_slot             = xprt_alloc_slot,
3087         .free_slot              = xprt_free_slot,
3088         .buf_alloc              = bc_malloc,
3089         .buf_free               = bc_free,
3090         .send_request           = bc_send_request,
3091         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3092         .close                  = bc_close,
3093         .destroy                = bc_destroy,
3094         .print_stats            = xs_tcp_print_stats,
3095         .enable_swap            = xs_enable_swap,
3096         .disable_swap           = xs_disable_swap,
3097         .inject_disconnect      = xs_inject_disconnect,
3098 };
3099
3100 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3101 {
3102         static const struct sockaddr_in sin = {
3103                 .sin_family             = AF_INET,
3104                 .sin_addr.s_addr        = htonl(INADDR_ANY),
3105         };
3106         static const struct sockaddr_in6 sin6 = {
3107                 .sin6_family            = AF_INET6,
3108                 .sin6_addr              = IN6ADDR_ANY_INIT,
3109         };
3110
3111         switch (family) {
3112         case AF_LOCAL:
3113                 break;
3114         case AF_INET:
3115                 memcpy(sap, &sin, sizeof(sin));
3116                 break;
3117         case AF_INET6:
3118                 memcpy(sap, &sin6, sizeof(sin6));
3119                 break;
3120         default:
3121                 dprintk("RPC:       %s: Bad address family\n", __func__);
3122                 return -EAFNOSUPPORT;
3123         }
3124         return 0;
3125 }
3126
3127 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3128                                       unsigned int slot_table_size,
3129                                       unsigned int max_slot_table_size)
3130 {
3131         struct rpc_xprt *xprt;
3132         struct sock_xprt *new;
3133
3134         if (args->addrlen > sizeof(xprt->addr)) {
3135                 dprintk("RPC:       xs_setup_xprt: address too large\n");
3136                 return ERR_PTR(-EBADF);
3137         }
3138
3139         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3140                         max_slot_table_size);
3141         if (xprt == NULL) {
3142                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
3143                                 "rpc_xprt\n");
3144                 return ERR_PTR(-ENOMEM);
3145         }
3146
3147         new = container_of(xprt, struct sock_xprt, xprt);
3148         mutex_init(&new->recv_mutex);
3149         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3150         xprt->addrlen = args->addrlen;
3151         if (args->srcaddr)
3152                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3153         else {
3154                 int err;
3155                 err = xs_init_anyaddr(args->dstaddr->sa_family,
3156                                         (struct sockaddr *)&new->srcaddr);
3157                 if (err != 0) {
3158                         xprt_free(xprt);
3159                         return ERR_PTR(err);
3160                 }
3161         }
3162
3163         return xprt;
3164 }
3165
3166 static const struct rpc_timeout xs_local_default_timeout = {
3167         .to_initval = 10 * HZ,
3168         .to_maxval = 10 * HZ,
3169         .to_retries = 2,
3170 };
3171
3172 /**
3173  * xs_setup_local - Set up transport to use an AF_LOCAL socket
3174  * @args: rpc transport creation arguments
3175  *
3176  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3177  */
3178 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3179 {
3180         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3181         struct sock_xprt *transport;
3182         struct rpc_xprt *xprt;
3183         struct rpc_xprt *ret;
3184
3185         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3186                         xprt_max_tcp_slot_table_entries);
3187         if (IS_ERR(xprt))
3188                 return xprt;
3189         transport = container_of(xprt, struct sock_xprt, xprt);
3190
3191         xprt->prot = 0;
3192         xprt->xprt_class = &xs_local_transport;
3193         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3194
3195         xprt->bind_timeout = XS_BIND_TO;
3196         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3197         xprt->idle_timeout = XS_IDLE_DISC_TO;
3198
3199         xprt->ops = &xs_local_ops;
3200         xprt->timeout = &xs_local_default_timeout;
3201
3202         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3203         INIT_WORK(&transport->error_worker, xs_error_handle);
3204         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3205
3206         switch (sun->sun_family) {
3207         case AF_LOCAL:
3208                 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3209                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
3210                                         sun->sun_path);
3211                         ret = ERR_PTR(-EINVAL);
3212                         goto out_err;
3213                 }
3214                 xprt_set_bound(xprt);
3215                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3216                 break;
3217         default:
3218                 ret = ERR_PTR(-EAFNOSUPPORT);
3219                 goto out_err;
3220         }
3221
3222         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
3223                         xprt->address_strings[RPC_DISPLAY_ADDR]);
3224
3225         if (try_module_get(THIS_MODULE))
3226                 return xprt;
3227         ret = ERR_PTR(-EINVAL);
3228 out_err:
3229         xs_xprt_free(xprt);
3230         return ret;
3231 }
3232
3233 static const struct rpc_timeout xs_udp_default_timeout = {
3234         .to_initval = 5 * HZ,
3235         .to_maxval = 30 * HZ,
3236         .to_increment = 5 * HZ,
3237         .to_retries = 5,
3238 };
3239
3240 /**
3241  * xs_setup_udp - Set up transport to use a UDP socket
3242  * @args: rpc transport creation arguments
3243  *
3244  */
3245 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3246 {
3247         struct sockaddr *addr = args->dstaddr;
3248         struct rpc_xprt *xprt;
3249         struct sock_xprt *transport;
3250         struct rpc_xprt *ret;
3251
3252         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3253                         xprt_udp_slot_table_entries);
3254         if (IS_ERR(xprt))
3255                 return xprt;
3256         transport = container_of(xprt, struct sock_xprt, xprt);
3257
3258         xprt->prot = IPPROTO_UDP;
3259         xprt->xprt_class = &xs_udp_transport;
3260         /* XXX: header size can vary due to auth type, IPv6, etc. */
3261         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3262
3263         xprt->bind_timeout = XS_BIND_TO;
3264         xprt->reestablish_timeout = XS_UDP_REEST_TO;
3265         xprt->idle_timeout = XS_IDLE_DISC_TO;
3266
3267         xprt->ops = &xs_udp_ops;
3268
3269         xprt->timeout = &xs_udp_default_timeout;
3270
3271         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3272         INIT_WORK(&transport->error_worker, xs_error_handle);
3273         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3274
3275         switch (addr->sa_family) {
3276         case AF_INET:
3277                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3278                         xprt_set_bound(xprt);
3279
3280                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3281                 break;
3282         case AF_INET6:
3283                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3284                         xprt_set_bound(xprt);
3285
3286                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3287                 break;
3288         default:
3289                 ret = ERR_PTR(-EAFNOSUPPORT);
3290                 goto out_err;
3291         }
3292
3293         if (xprt_bound(xprt))
3294                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3295                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3296                                 xprt->address_strings[RPC_DISPLAY_PORT],
3297                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3298         else
3299                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3300                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3301                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3302
3303         if (try_module_get(THIS_MODULE))
3304                 return xprt;
3305         ret = ERR_PTR(-EINVAL);
3306 out_err:
3307         xs_xprt_free(xprt);
3308         return ret;
3309 }
3310
3311 static const struct rpc_timeout xs_tcp_default_timeout = {
3312         .to_initval = 60 * HZ,
3313         .to_maxval = 60 * HZ,
3314         .to_retries = 2,
3315 };
3316
3317 /**
3318  * xs_setup_tcp - Set up transport to use a TCP socket
3319  * @args: rpc transport creation arguments
3320  *
3321  */
3322 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3323 {
3324         struct sockaddr *addr = args->dstaddr;
3325         struct rpc_xprt *xprt;
3326         struct sock_xprt *transport;
3327         struct rpc_xprt *ret;
3328         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3329
3330         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3331                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3332
3333         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3334                         max_slot_table_size);
3335         if (IS_ERR(xprt))
3336                 return xprt;
3337         transport = container_of(xprt, struct sock_xprt, xprt);
3338
3339         xprt->prot = IPPROTO_TCP;
3340         xprt->xprt_class = &xs_tcp_transport;
3341         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3342
3343         xprt->bind_timeout = XS_BIND_TO;
3344         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3345         xprt->idle_timeout = XS_IDLE_DISC_TO;
3346
3347         xprt->ops = &xs_tcp_ops;
3348         xprt->timeout = &xs_tcp_default_timeout;
3349
3350         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3351         if (args->reconnect_timeout)
3352                 xprt->max_reconnect_timeout = args->reconnect_timeout;
3353
3354         xprt->connect_timeout = xprt->timeout->to_initval *
3355                 (xprt->timeout->to_retries + 1);
3356         if (args->connect_timeout)
3357                 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3358
3359         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3360         INIT_WORK(&transport->error_worker, xs_error_handle);
3361         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3362
3363         switch (addr->sa_family) {
3364         case AF_INET:
3365                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3366                         xprt_set_bound(xprt);
3367
3368                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3369                 break;
3370         case AF_INET6:
3371                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3372                         xprt_set_bound(xprt);
3373
3374                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3375                 break;
3376         default:
3377                 ret = ERR_PTR(-EAFNOSUPPORT);
3378                 goto out_err;
3379         }
3380
3381         if (xprt_bound(xprt))
3382                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3383                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3384                                 xprt->address_strings[RPC_DISPLAY_PORT],
3385                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3386         else
3387                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3388                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3389                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3390
3391         if (try_module_get(THIS_MODULE))
3392                 return xprt;
3393         ret = ERR_PTR(-EINVAL);
3394 out_err:
3395         xs_xprt_free(xprt);
3396         return ret;
3397 }
3398
3399 /**
3400  * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3401  * @args: rpc transport creation arguments
3402  *
3403  */
3404 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3405 {
3406         struct sockaddr *addr = args->dstaddr;
3407         struct rpc_xprt *xprt;
3408         struct sock_xprt *transport;
3409         struct rpc_xprt *ret;
3410         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3411
3412         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3413                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3414
3415         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3416                              max_slot_table_size);
3417         if (IS_ERR(xprt))
3418                 return xprt;
3419         transport = container_of(xprt, struct sock_xprt, xprt);
3420
3421         xprt->prot = IPPROTO_TCP;
3422         xprt->xprt_class = &xs_tcp_transport;
3423         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3424
3425         xprt->bind_timeout = XS_BIND_TO;
3426         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3427         xprt->idle_timeout = XS_IDLE_DISC_TO;
3428
3429         xprt->ops = &xs_tcp_ops;
3430         xprt->timeout = &xs_tcp_default_timeout;
3431
3432         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3433         xprt->connect_timeout = xprt->timeout->to_initval *
3434                 (xprt->timeout->to_retries + 1);
3435
3436         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3437         INIT_WORK(&transport->error_worker, xs_error_handle);
3438
3439         switch (args->xprtsec.policy) {
3440         case RPC_XPRTSEC_TLS_ANON:
3441         case RPC_XPRTSEC_TLS_X509:
3442                 xprt->xprtsec = args->xprtsec;
3443                 INIT_DELAYED_WORK(&transport->connect_worker,
3444                                   xs_tcp_tls_setup_socket);
3445                 break;
3446         default:
3447                 ret = ERR_PTR(-EACCES);
3448                 goto out_err;
3449         }
3450
3451         switch (addr->sa_family) {
3452         case AF_INET:
3453                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3454                         xprt_set_bound(xprt);
3455
3456                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3457                 break;
3458         case AF_INET6:
3459                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3460                         xprt_set_bound(xprt);
3461
3462                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3463                 break;
3464         default:
3465                 ret = ERR_PTR(-EAFNOSUPPORT);
3466                 goto out_err;
3467         }
3468
3469         if (xprt_bound(xprt))
3470                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3471                         xprt->address_strings[RPC_DISPLAY_ADDR],
3472                         xprt->address_strings[RPC_DISPLAY_PORT],
3473                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3474         else
3475                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3476                         xprt->address_strings[RPC_DISPLAY_ADDR],
3477                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3478
3479         if (try_module_get(THIS_MODULE))
3480                 return xprt;
3481         ret = ERR_PTR(-EINVAL);
3482 out_err:
3483         xs_xprt_free(xprt);
3484         return ret;
3485 }
3486
3487 /**
3488  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3489  * @args: rpc transport creation arguments
3490  *
3491  */
3492 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3493 {
3494         struct sockaddr *addr = args->dstaddr;
3495         struct rpc_xprt *xprt;
3496         struct sock_xprt *transport;
3497         struct svc_sock *bc_sock;
3498         struct rpc_xprt *ret;
3499
3500         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3501                         xprt_tcp_slot_table_entries);
3502         if (IS_ERR(xprt))
3503                 return xprt;
3504         transport = container_of(xprt, struct sock_xprt, xprt);
3505
3506         xprt->prot = IPPROTO_TCP;
3507         xprt->xprt_class = &xs_bc_tcp_transport;
3508         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3509         xprt->timeout = &xs_tcp_default_timeout;
3510
3511         /* backchannel */
3512         xprt_set_bound(xprt);
3513         xprt->bind_timeout = 0;
3514         xprt->reestablish_timeout = 0;
3515         xprt->idle_timeout = 0;
3516
3517         xprt->ops = &bc_tcp_ops;
3518
3519         switch (addr->sa_family) {
3520         case AF_INET:
3521                 xs_format_peer_addresses(xprt, "tcp",
3522                                          RPCBIND_NETID_TCP);
3523                 break;
3524         case AF_INET6:
3525                 xs_format_peer_addresses(xprt, "tcp",
3526                                    RPCBIND_NETID_TCP6);
3527                 break;
3528         default:
3529                 ret = ERR_PTR(-EAFNOSUPPORT);
3530                 goto out_err;
3531         }
3532
3533         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3534                         xprt->address_strings[RPC_DISPLAY_ADDR],
3535                         xprt->address_strings[RPC_DISPLAY_PORT],
3536                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3537
3538         /*
3539          * Once we've associated a backchannel xprt with a connection,
3540          * we want to keep it around as long as the connection lasts,
3541          * in case we need to start using it for a backchannel again;
3542          * this reference won't be dropped until bc_xprt is destroyed.
3543          */
3544         xprt_get(xprt);
3545         args->bc_xprt->xpt_bc_xprt = xprt;
3546         xprt->bc_xprt = args->bc_xprt;
3547         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3548         transport->sock = bc_sock->sk_sock;
3549         transport->inet = bc_sock->sk_sk;
3550
3551         /*
3552          * Since we don't want connections for the backchannel, we set
3553          * the xprt status to connected
3554          */
3555         xprt_set_connected(xprt);
3556
3557         if (try_module_get(THIS_MODULE))
3558                 return xprt;
3559
3560         args->bc_xprt->xpt_bc_xprt = NULL;
3561         args->bc_xprt->xpt_bc_xps = NULL;
3562         xprt_put(xprt);
3563         ret = ERR_PTR(-EINVAL);
3564 out_err:
3565         xs_xprt_free(xprt);
3566         return ret;
3567 }
3568
3569 static struct xprt_class        xs_local_transport = {
3570         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3571         .name           = "named UNIX socket",
3572         .owner          = THIS_MODULE,
3573         .ident          = XPRT_TRANSPORT_LOCAL,
3574         .setup          = xs_setup_local,
3575         .netid          = { "" },
3576 };
3577
3578 static struct xprt_class        xs_udp_transport = {
3579         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3580         .name           = "udp",
3581         .owner          = THIS_MODULE,
3582         .ident          = XPRT_TRANSPORT_UDP,
3583         .setup          = xs_setup_udp,
3584         .netid          = { "udp", "udp6", "" },
3585 };
3586
3587 static struct xprt_class        xs_tcp_transport = {
3588         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3589         .name           = "tcp",
3590         .owner          = THIS_MODULE,
3591         .ident          = XPRT_TRANSPORT_TCP,
3592         .setup          = xs_setup_tcp,
3593         .netid          = { "tcp", "tcp6", "" },
3594 };
3595
3596 static struct xprt_class        xs_tcp_tls_transport = {
3597         .list           = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3598         .name           = "tcp-with-tls",
3599         .owner          = THIS_MODULE,
3600         .ident          = XPRT_TRANSPORT_TCP_TLS,
3601         .setup          = xs_setup_tcp_tls,
3602         .netid          = { "tcp", "tcp6", "" },
3603 };
3604
3605 static struct xprt_class        xs_bc_tcp_transport = {
3606         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3607         .name           = "tcp NFSv4.1 backchannel",
3608         .owner          = THIS_MODULE,
3609         .ident          = XPRT_TRANSPORT_BC_TCP,
3610         .setup          = xs_setup_bc_tcp,
3611         .netid          = { "" },
3612 };
3613
3614 /**
3615  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3616  *
3617  */
3618 int init_socket_xprt(void)
3619 {
3620         if (!sunrpc_table_header)
3621                 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3622
3623         xprt_register_transport(&xs_local_transport);
3624         xprt_register_transport(&xs_udp_transport);
3625         xprt_register_transport(&xs_tcp_transport);
3626         xprt_register_transport(&xs_tcp_tls_transport);
3627         xprt_register_transport(&xs_bc_tcp_transport);
3628
3629         return 0;
3630 }
3631
3632 /**
3633  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3634  *
3635  */
3636 void cleanup_socket_xprt(void)
3637 {
3638         if (sunrpc_table_header) {
3639                 unregister_sysctl_table(sunrpc_table_header);
3640                 sunrpc_table_header = NULL;
3641         }
3642
3643         xprt_unregister_transport(&xs_local_transport);
3644         xprt_unregister_transport(&xs_udp_transport);
3645         xprt_unregister_transport(&xs_tcp_transport);
3646         xprt_unregister_transport(&xs_tcp_tls_transport);
3647         xprt_unregister_transport(&xs_bc_tcp_transport);
3648 }
3649
3650 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3651 {
3652         return param_set_uint_minmax(val, kp,
3653                         RPC_MIN_RESVPORT,
3654                         RPC_MAX_RESVPORT);
3655 }
3656
3657 static const struct kernel_param_ops param_ops_portnr = {
3658         .set = param_set_portnr,
3659         .get = param_get_uint,
3660 };
3661
3662 #define param_check_portnr(name, p) \
3663         __param_check(name, p, unsigned int);
3664
3665 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3666 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3667
3668 static int param_set_slot_table_size(const char *val,
3669                                      const struct kernel_param *kp)
3670 {
3671         return param_set_uint_minmax(val, kp,
3672                         RPC_MIN_SLOT_TABLE,
3673                         RPC_MAX_SLOT_TABLE);
3674 }
3675
3676 static const struct kernel_param_ops param_ops_slot_table_size = {
3677         .set = param_set_slot_table_size,
3678         .get = param_get_uint,
3679 };
3680
3681 #define param_check_slot_table_size(name, p) \
3682         __param_check(name, p, unsigned int);
3683
3684 static int param_set_max_slot_table_size(const char *val,
3685                                      const struct kernel_param *kp)
3686 {
3687         return param_set_uint_minmax(val, kp,
3688                         RPC_MIN_SLOT_TABLE,
3689                         RPC_MAX_SLOT_TABLE_LIMIT);
3690 }
3691
3692 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3693         .set = param_set_max_slot_table_size,
3694         .get = param_get_uint,
3695 };
3696
3697 #define param_check_max_slot_table_size(name, p) \
3698         __param_check(name, p, unsigned int);
3699
3700 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3701                    slot_table_size, 0644);
3702 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3703                    max_slot_table_size, 0644);
3704 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3705                    slot_table_size, 0644);