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_prot.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         u8 content_type = tls_get_record_type(sock->sk, cmsg);
 364         u8 level, description;
 365
 366         switch (content_type) {
 367         case 0:
 368                 break;
 369         case TLS_RECORD_TYPE_DATA:
 370                 /* TLS sets EOR at the end of each application data
 371                  * record, even though there might be more frames
 372                  * waiting to be decrypted.
 373                  */
 374                 msg->msg_flags &= ~MSG_EOR;
 375                 break;
 376         case TLS_RECORD_TYPE_ALERT:
 377                 tls_alert_recv(sock->sk, msg, &level, &description);
 378                 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
 379                         -EACCES : -EAGAIN;
 380                 break;
 381         default:
 382                 /* discard this record type */
 383                 ret = -EAGAIN;
 384         }
 385         return ret;
 386 }
 387
 388 static int
 389 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
 390 {
 391         union {
 392                 struct cmsghdr  cmsg;
 393                 u8              buf[CMSG_SPACE(sizeof(u8))];
 394         } u;
 395         int ret;
 396
 397         msg->msg_control = &u;
 398         msg->msg_controllen = sizeof(u);
 399         ret = sock_recvmsg(sock, msg, flags);
 400         if (msg->msg_controllen != sizeof(u))
 401                 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
 402         return ret;
 403 }
 404
 405 static ssize_t
 406 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 407 {
 408         ssize_t ret;
 409         if (seek != 0)
 410                 iov_iter_advance(&msg->msg_iter, seek);
 411         ret = xs_sock_recv_cmsg(sock, msg, flags);
 412         return ret > 0 ? ret + seek : ret;
 413 }
 414
 415 static ssize_t
 416 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 417                 struct kvec *kvec, size_t count, size_t seek)
 418 {
 419         iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
 420         return xs_sock_recvmsg(sock, msg, flags, seek);
 421 }
 422
 423 static ssize_t
 424 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 425                 struct bio_vec *bvec, unsigned long nr, size_t count,
 426                 size_t seek)
 427 {
 428         iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
 429         return xs_sock_recvmsg(sock, msg, flags, seek);
 430 }
 431
 432 static ssize_t
 433 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 434                 size_t count)
 435 {
 436         iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
 437         return xs_sock_recv_cmsg(sock, msg, flags);
 438 }
 439
 440 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 441 static void
 442 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 443 {
 444         struct bvec_iter bi = {
 445                 .bi_size = count,
 446         };
 447         struct bio_vec bv;
 448
 449         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 450         for_each_bvec(bv, bvec, bi, bi)
 451                 flush_dcache_page(bv.bv_page);
 452 }
 453 #else
 454 static inline void
 455 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 456 {
 457 }
 458 #endif
 459
 460 static ssize_t
 461 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 462                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 463 {
 464         size_t want, seek_init = seek, offset = 0;
 465         ssize_t ret;
 466
 467         want = min_t(size_t, count, buf->head[0].iov_len);
 468         if (seek < want) {
 469                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 470                 if (ret <= 0)
 471                         goto sock_err;
 472                 offset += ret;
 473                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 474                         goto out;
 475                 if (ret != want)
 476                         goto out;
 477                 seek = 0;
 478         } else {
 479                 seek -= want;
 480                 offset += want;
 481         }
 482
 483         want = xs_alloc_sparse_pages(
 484                 buf, min_t(size_t, count - offset, buf->page_len),
 485                 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
 486         if (seek < want) {
 487                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 488                                 xdr_buf_pagecount(buf),
 489                                 want + buf->page_base,
 490                                 seek + buf->page_base);
 491                 if (ret <= 0)
 492                         goto sock_err;
 493                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 494                 ret -= buf->page_base;
 495                 offset += ret;
 496                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 497                         goto out;
 498                 if (ret != want)
 499                         goto out;
 500                 seek = 0;
 501         } else {
 502                 seek -= want;
 503                 offset += want;
 504         }
 505
 506         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 507         if (seek < want) {
 508                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 509                 if (ret <= 0)
 510                         goto sock_err;
 511                 offset += ret;
 512                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 513                         goto out;
 514                 if (ret != want)
 515                         goto out;
 516         } else if (offset < seek_init)
 517                 offset = seek_init;
 518         ret = -EMSGSIZE;
 519 out:
 520         *read = offset - seek_init;
 521         return ret;
 522 sock_err:
 523         offset += seek;
 524         goto out;
 525 }
 526
 527 static void
 528 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 529 {
 530         if (!transport->recv.copied) {
 531                 if (buf->head[0].iov_len >= transport->recv.offset)
 532                         memcpy(buf->head[0].iov_base,
 533                                         &transport->recv.xid,
 534                                         transport->recv.offset);
 535                 transport->recv.copied = transport->recv.offset;
 536         }
 537 }
 538
 539 static bool
 540 xs_read_stream_request_done(struct sock_xprt *transport)
 541 {
 542         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 543 }
 544
 545 static void
 546 xs_read_stream_check_eor(struct sock_xprt *transport,
 547                 struct msghdr *msg)
 548 {
 549         if (xs_read_stream_request_done(transport))
 550                 msg->msg_flags |= MSG_EOR;
 551 }
 552
 553 static ssize_t
 554 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 555                 int flags, struct rpc_rqst *req)
 556 {
 557         struct xdr_buf *buf = &req->rq_private_buf;
 558         size_t want, read;
 559         ssize_t ret;
 560
 561         xs_read_header(transport, buf);
 562
 563         want = transport->recv.len - transport->recv.offset;
 564         if (want != 0) {
 565                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 566                                 transport->recv.copied + want,
 567                                 transport->recv.copied,
 568                                 &read);
 569                 transport->recv.offset += read;
 570                 transport->recv.copied += read;
 571         }
 572
 573         if (transport->recv.offset == transport->recv.len)
 574                 xs_read_stream_check_eor(transport, msg);
 575
 576         if (want == 0)
 577                 return 0;
 578
 579         switch (ret) {
 580         default:
 581                 break;
 582         case -EFAULT:
 583         case -EMSGSIZE:
 584                 msg->msg_flags |= MSG_TRUNC;
 585                 return read;
 586         case 0:
 587                 return -ESHUTDOWN;
 588         }
 589         return ret < 0 ? ret : read;
 590 }
 591
 592 static size_t
 593 xs_read_stream_headersize(bool isfrag)
 594 {
 595         if (isfrag)
 596                 return sizeof(__be32);
 597         return 3 * sizeof(__be32);
 598 }
 599
 600 static ssize_t
 601 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 602                 int flags, size_t want, size_t seek)
 603 {
 604         struct kvec kvec = {
 605                 .iov_base = &transport->recv.fraghdr,
 606                 .iov_len = want,
 607         };
 608         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 609 }
 610
 611 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 612 static ssize_t
 613 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 614 {
 615         struct rpc_xprt *xprt = &transport->xprt;
 616         struct rpc_rqst *req;
 617         ssize_t ret;
 618
 619         /* Is this transport associated with the backchannel? */
 620         if (!xprt->bc_serv)
 621                 return -ESHUTDOWN;
 622
 623         /* Look up and lock the request corresponding to the given XID */
 624         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 625         if (!req) {
 626                 printk(KERN_WARNING "Callback slot table overflowed\n");
 627                 return -ESHUTDOWN;
 628         }
 629         if (transport->recv.copied && !req->rq_private_buf.len)
 630                 return -ESHUTDOWN;
 631
 632         ret = xs_read_stream_request(transport, msg, flags, req);
 633         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 634                 xprt_complete_bc_request(req, transport->recv.copied);
 635         else
 636                 req->rq_private_buf.len = transport->recv.copied;
 637
 638         return ret;
 639 }
 640 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 641 static ssize_t
 642 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 643 {
 644         return -ESHUTDOWN;
 645 }
 646 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 647
 648 static ssize_t
 649 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 650 {
 651         struct rpc_xprt *xprt = &transport->xprt;
 652         struct rpc_rqst *req;
 653         ssize_t ret = 0;
 654
 655         /* Look up and lock the request corresponding to the given XID */
 656         spin_lock(&xprt->queue_lock);
 657         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 658         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 659                 msg->msg_flags |= MSG_TRUNC;
 660                 goto out;
 661         }
 662         xprt_pin_rqst(req);
 663         spin_unlock(&xprt->queue_lock);
 664
 665         ret = xs_read_stream_request(transport, msg, flags, req);
 666
 667         spin_lock(&xprt->queue_lock);
 668         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 669                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 670         else
 671                 req->rq_private_buf.len = transport->recv.copied;
 672         xprt_unpin_rqst(req);
 673 out:
 674         spin_unlock(&xprt->queue_lock);
 675         return ret;
 676 }
 677
 678 static ssize_t
 679 xs_read_stream(struct sock_xprt *transport, int flags)
 680 {
 681         struct msghdr msg = { 0 };
 682         size_t want, read = 0;
 683         ssize_t ret = 0;
 684
 685         if (transport->recv.len == 0) {
 686                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 687                 ret = xs_read_stream_header(transport, &msg, flags, want,
 688                                 transport->recv.offset);
 689                 if (ret <= 0)
 690                         goto out_err;
 691                 transport->recv.offset = ret;
 692                 if (transport->recv.offset != want)
 693                         return transport->recv.offset;
 694                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 695                         RPC_FRAGMENT_SIZE_MASK;
 696                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 697                 read = ret;
 698         }
 699
 700         switch (be32_to_cpu(transport->recv.calldir)) {
 701         default:
 702                 msg.msg_flags |= MSG_TRUNC;
 703                 break;
 704         case RPC_CALL:
 705                 ret = xs_read_stream_call(transport, &msg, flags);
 706                 break;
 707         case RPC_REPLY:
 708                 ret = xs_read_stream_reply(transport, &msg, flags);
 709         }
 710         if (msg.msg_flags & MSG_TRUNC) {
 711                 transport->recv.calldir = cpu_to_be32(-1);
 712                 transport->recv.copied = -1;
 713         }
 714         if (ret < 0)
 715                 goto out_err;
 716         read += ret;
 717         if (transport->recv.offset < transport->recv.len) {
 718                 if (!(msg.msg_flags & MSG_TRUNC))
 719                         return read;
 720                 msg.msg_flags = 0;
 721                 ret = xs_read_discard(transport->sock, &msg, flags,
 722                                 transport->recv.len - transport->recv.offset);
 723                 if (ret <= 0)
 724                         goto out_err;
 725                 transport->recv.offset += ret;
 726                 read += ret;
 727                 if (transport->recv.offset != transport->recv.len)
 728                         return read;
 729         }
 730         if (xs_read_stream_request_done(transport)) {
 731                 trace_xs_stream_read_request(transport);
 732                 transport->recv.copied = 0;
 733         }
 734         transport->recv.offset = 0;
 735         transport->recv.len = 0;
 736         return read;
 737 out_err:
 738         return ret != 0 ? ret : -ESHUTDOWN;
 739 }
 740
 741 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 742 {
 743         return transport->sock->ops->poll(transport->file, transport->sock,
 744                         NULL);
 745 }
 746
 747 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 748 {
 749         __poll_t events = xs_poll_socket(transport);
 750
 751         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 752 }
 753
 754 static void xs_poll_check_readable(struct sock_xprt *transport)
 755 {
 756
 757         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 758         if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
 759                 return;
 760         if (!xs_poll_socket_readable(transport))
 761                 return;
 762         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 763                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 764 }
 765
 766 static void xs_stream_data_receive(struct sock_xprt *transport)
 767 {
 768         size_t read = 0;
 769         ssize_t ret = 0;
 770
 771         mutex_lock(&transport->recv_mutex);
 772         if (transport->sock == NULL)
 773                 goto out;
 774         for (;;) {
 775                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 776                 if (ret < 0)
 777                         break;
 778                 read += ret;
 779                 cond_resched();
 780         }
 781         if (ret == -ESHUTDOWN)
 782                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 783         else if (ret == -EACCES)
 784                 xprt_wake_pending_tasks(&transport->xprt, -EACCES);
 785         else
 786                 xs_poll_check_readable(transport);
 787 out:
 788         mutex_unlock(&transport->recv_mutex);
 789         trace_xs_stream_read_data(&transport->xprt, ret, read);
 790 }
 791
 792 static void xs_stream_data_receive_workfn(struct work_struct *work)
 793 {
 794         struct sock_xprt *transport =
 795                 container_of(work, struct sock_xprt, recv_worker);
 796         unsigned int pflags = memalloc_nofs_save();
 797
 798         xs_stream_data_receive(transport);
 799         memalloc_nofs_restore(pflags);
 800 }
 801
 802 static void
 803 xs_stream_reset_connect(struct sock_xprt *transport)
 804 {
 805         transport->recv.offset = 0;
 806         transport->recv.len = 0;
 807         transport->recv.copied = 0;
 808         transport->xmit.offset = 0;
 809 }
 810
 811 static void
 812 xs_stream_start_connect(struct sock_xprt *transport)
 813 {
 814         transport->xprt.stat.connect_count++;
 815         transport->xprt.stat.connect_start = jiffies;
 816 }
 817
 818 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 819
 820 /**
 821  * xs_nospace - handle transmit was incomplete
 822  * @req: pointer to RPC request
 823  * @transport: pointer to struct sock_xprt
 824  *
 825  */
 826 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 827 {
 828         struct rpc_xprt *xprt = &transport->xprt;
 829         struct sock *sk = transport->inet;
 830         int ret = -EAGAIN;
 831
 832         trace_rpc_socket_nospace(req, transport);
 833
 834         /* Protect against races with write_space */
 835         spin_lock(&xprt->transport_lock);
 836
 837         /* Don't race with disconnect */
 838         if (xprt_connected(xprt)) {
 839                 /* wait for more buffer space */
 840                 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
 841                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 842                 sk->sk_write_pending++;
 843                 xprt_wait_for_buffer_space(xprt);
 844         } else
 845                 ret = -ENOTCONN;
 846
 847         spin_unlock(&xprt->transport_lock);
 848         return ret;
 849 }
 850
 851 static int xs_sock_nospace(struct rpc_rqst *req)
 852 {
 853         struct sock_xprt *transport =
 854                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 855         struct sock *sk = transport->inet;
 856         int ret = -EAGAIN;
 857
 858         lock_sock(sk);
 859         if (!sock_writeable(sk))
 860                 ret = xs_nospace(req, transport);
 861         release_sock(sk);
 862         return ret;
 863 }
 864
 865 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
 866 {
 867         struct sock_xprt *transport =
 868                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 869         struct sock *sk = transport->inet;
 870         int ret = -EAGAIN;
 871
 872         if (vm_wait)
 873                 return -ENOBUFS;
 874         lock_sock(sk);
 875         if (!sk_stream_memory_free(sk))
 876                 ret = xs_nospace(req, transport);
 877         release_sock(sk);
 878         return ret;
 879 }
 880
 881 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
 882 {
 883         return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
 884 }
 885
 886 /*
 887  * Determine if the previous message in the stream was aborted before it
 888  * could complete transmission.
 889  */
 890 static bool
 891 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 892 {
 893         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 894 }
 895
 896 /*
 897  * Return the stream record marker field for a record of length < 2^31-1
 898  */
 899 static rpc_fraghdr
 900 xs_stream_record_marker(struct xdr_buf *xdr)
 901 {
 902         if (!xdr->len)
 903                 return 0;
 904         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 905 }
 906
 907 /**
 908  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 909  * @req: pointer to RPC request
 910  *
 911  * Return values:
 912  *        0:    The request has been sent
 913  *   EAGAIN:    The socket was blocked, please call again later to
 914  *              complete the request
 915  * ENOTCONN:    Caller needs to invoke connect logic then call again
 916  *    other:    Some other error occurred, the request was not sent
 917  */
 918 static int xs_local_send_request(struct rpc_rqst *req)
 919 {
 920         struct rpc_xprt *xprt = req->rq_xprt;
 921         struct sock_xprt *transport =
 922                                 container_of(xprt, struct sock_xprt, xprt);
 923         struct xdr_buf *xdr = &req->rq_snd_buf;
 924         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 925         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 926         struct msghdr msg = {
 927                 .msg_flags      = XS_SENDMSG_FLAGS,
 928         };
 929         bool vm_wait;
 930         unsigned int sent;
 931         int status;
 932
 933         /* Close the stream if the previous transmission was incomplete */
 934         if (xs_send_request_was_aborted(transport, req)) {
 935                 xprt_force_disconnect(xprt);
 936                 return -ENOTCONN;
 937         }
 938
 939         xs_pktdump("packet data:",
 940                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 941
 942         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
 943
 944         req->rq_xtime = ktime_get();
 945         status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 946                                    transport->xmit.offset, rm, &sent);
 947         dprintk("RPC:       %s(%u) = %d\n",
 948                         __func__, xdr->len - transport->xmit.offset, status);
 949
 950         if (likely(sent > 0) || status == 0) {
 951                 transport->xmit.offset += sent;
 952                 req->rq_bytes_sent = transport->xmit.offset;
 953                 if (likely(req->rq_bytes_sent >= msglen)) {
 954                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 955                         transport->xmit.offset = 0;
 956                         return 0;
 957                 }
 958                 status = -EAGAIN;
 959                 vm_wait = false;
 960         }
 961
 962         switch (status) {
 963         case -EAGAIN:
 964                 status = xs_stream_nospace(req, vm_wait);
 965                 break;
 966         default:
 967                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 968                         -status);
 969                 fallthrough;
 970         case -EPIPE:
 971                 xprt_force_disconnect(xprt);
 972                 status = -ENOTCONN;
 973         }
 974
 975         return status;
 976 }
 977
 978 /**
 979  * xs_udp_send_request - write an RPC request to a UDP socket
 980  * @req: pointer to RPC request
 981  *
 982  * Return values:
 983  *        0:    The request has been sent
 984  *   EAGAIN:    The socket was blocked, please call again later to
 985  *              complete the request
 986  * ENOTCONN:    Caller needs to invoke connect logic then call again
 987  *    other:    Some other error occurred, the request was not sent
 988  */
 989 static int xs_udp_send_request(struct rpc_rqst *req)
 990 {
 991         struct rpc_xprt *xprt = req->rq_xprt;
 992         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 993         struct xdr_buf *xdr = &req->rq_snd_buf;
 994         struct msghdr msg = {
 995                 .msg_name       = xs_addr(xprt),
 996                 .msg_namelen    = xprt->addrlen,
 997                 .msg_flags      = XS_SENDMSG_FLAGS,
 998         };
 999         unsigned int sent;
1000         int status;
1001
1002         xs_pktdump("packet data:",
1003                                 req->rq_svec->iov_base,
1004                                 req->rq_svec->iov_len);
1005
1006         if (!xprt_bound(xprt))
1007                 return -ENOTCONN;
1008
1009         if (!xprt_request_get_cong(xprt, req))
1010                 return -EBADSLT;
1011
1012         status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1013         if (status < 0)
1014                 return status;
1015         req->rq_xtime = ktime_get();
1016         status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1017
1018         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
1019                         xdr->len, status);
1020
1021         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1022         if (status == -EPERM)
1023                 goto process_status;
1024
1025         if (status == -EAGAIN && sock_writeable(transport->inet))
1026                 status = -ENOBUFS;
1027
1028         if (sent > 0 || status == 0) {
1029                 req->rq_xmit_bytes_sent += sent;
1030                 if (sent >= req->rq_slen)
1031                         return 0;
1032                 /* Still some bytes left; set up for a retry later. */
1033                 status = -EAGAIN;
1034         }
1035
1036 process_status:
1037         switch (status) {
1038         case -ENOTSOCK:
1039                 status = -ENOTCONN;
1040                 /* Should we call xs_close() here? */
1041                 break;
1042         case -EAGAIN:
1043                 status = xs_sock_nospace(req);
1044                 break;
1045         case -ENETUNREACH:
1046         case -ENOBUFS:
1047         case -EPIPE:
1048         case -ECONNREFUSED:
1049         case -EPERM:
1050                 /* When the server has died, an ICMP port unreachable message
1051                  * prompts ECONNREFUSED. */
1052                 break;
1053         default:
1054                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1055                         -status);
1056         }
1057
1058         return status;
1059 }
1060
1061 /**
1062  * xs_tcp_send_request - write an RPC request to a TCP socket
1063  * @req: pointer to RPC request
1064  *
1065  * Return values:
1066  *        0:    The request has been sent
1067  *   EAGAIN:    The socket was blocked, please call again later to
1068  *              complete the request
1069  * ENOTCONN:    Caller needs to invoke connect logic then call again
1070  *    other:    Some other error occurred, the request was not sent
1071  *
1072  * XXX: In the case of soft timeouts, should we eventually give up
1073  *      if sendmsg is not able to make progress?
1074  */
1075 static int xs_tcp_send_request(struct rpc_rqst *req)
1076 {
1077         struct rpc_xprt *xprt = req->rq_xprt;
1078         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1079         struct xdr_buf *xdr = &req->rq_snd_buf;
1080         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1081         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1082         struct msghdr msg = {
1083                 .msg_flags      = XS_SENDMSG_FLAGS,
1084         };
1085         bool vm_wait;
1086         unsigned int sent;
1087         int status;
1088
1089         /* Close the stream if the previous transmission was incomplete */
1090         if (xs_send_request_was_aborted(transport, req)) {
1091                 if (transport->sock != NULL)
1092                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1093                 return -ENOTCONN;
1094         }
1095         if (!transport->inet)
1096                 return -ENOTCONN;
1097
1098         xs_pktdump("packet data:",
1099                                 req->rq_svec->iov_base,
1100                                 req->rq_svec->iov_len);
1101
1102         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1103                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1104
1105         xs_set_srcport(transport, transport->sock);
1106
1107         /* Continue transmitting the packet/record. We must be careful
1108          * to cope with writespace callbacks arriving _after_ we have
1109          * called sendmsg(). */
1110         req->rq_xtime = ktime_get();
1111         tcp_sock_set_cork(transport->inet, true);
1112
1113         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1114
1115         do {
1116                 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1117                                            transport->xmit.offset, rm, &sent);
1118
1119                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1120                                 xdr->len - transport->xmit.offset, status);
1121
1122                 /* If we've sent the entire packet, immediately
1123                  * reset the count of bytes sent. */
1124                 transport->xmit.offset += sent;
1125                 req->rq_bytes_sent = transport->xmit.offset;
1126                 if (likely(req->rq_bytes_sent >= msglen)) {
1127                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1128                         transport->xmit.offset = 0;
1129                         if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1130                                 tcp_sock_set_cork(transport->inet, false);
1131                         return 0;
1132                 }
1133
1134                 WARN_ON_ONCE(sent == 0 && status == 0);
1135
1136                 if (sent > 0)
1137                         vm_wait = false;
1138
1139         } while (status == 0);
1140
1141         switch (status) {
1142         case -ENOTSOCK:
1143                 status = -ENOTCONN;
1144                 /* Should we call xs_close() here? */
1145                 break;
1146         case -EAGAIN:
1147                 status = xs_stream_nospace(req, vm_wait);
1148                 break;
1149         case -ECONNRESET:
1150         case -ECONNREFUSED:
1151         case -ENOTCONN:
1152         case -EADDRINUSE:
1153         case -ENOBUFS:
1154         case -EPIPE:
1155                 break;
1156         default:
1157                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1158                         -status);
1159         }
1160
1161         return status;
1162 }
1163
1164 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1165 {
1166         transport->old_data_ready = sk->sk_data_ready;
1167         transport->old_state_change = sk->sk_state_change;
1168         transport->old_write_space = sk->sk_write_space;
1169         transport->old_error_report = sk->sk_error_report;
1170 }
1171
1172 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1173 {
1174         sk->sk_data_ready = transport->old_data_ready;
1175         sk->sk_state_change = transport->old_state_change;
1176         sk->sk_write_space = transport->old_write_space;
1177         sk->sk_error_report = transport->old_error_report;
1178 }
1179
1180 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1181 {
1182         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1183
1184         transport->xprt_err = 0;
1185         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1186         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1187         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1188         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1189         clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1190 }
1191
1192 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1193 {
1194         set_bit(nr, &transport->sock_state);
1195         queue_work(xprtiod_workqueue, &transport->error_worker);
1196 }
1197
1198 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1199 {
1200         xprt->connect_cookie++;
1201         smp_mb__before_atomic();
1202         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1203         clear_bit(XPRT_CLOSING, &xprt->state);
1204         xs_sock_reset_state_flags(xprt);
1205         smp_mb__after_atomic();
1206 }
1207
1208 /**
1209  * xs_error_report - callback to handle TCP socket state errors
1210  * @sk: socket
1211  *
1212  * Note: we don't call sock_error() since there may be a rpc_task
1213  * using the socket, and so we don't want to clear sk->sk_err.
1214  */
1215 static void xs_error_report(struct sock *sk)
1216 {
1217         struct sock_xprt *transport;
1218         struct rpc_xprt *xprt;
1219
1220         if (!(xprt = xprt_from_sock(sk)))
1221                 return;
1222
1223         transport = container_of(xprt, struct sock_xprt, xprt);
1224         transport->xprt_err = -sk->sk_err;
1225         if (transport->xprt_err == 0)
1226                 return;
1227         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1228                         xprt, -transport->xprt_err);
1229         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1230
1231         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1232         smp_mb__before_atomic();
1233         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1234 }
1235
1236 static void xs_reset_transport(struct sock_xprt *transport)
1237 {
1238         struct socket *sock = transport->sock;
1239         struct sock *sk = transport->inet;
1240         struct rpc_xprt *xprt = &transport->xprt;
1241         struct file *filp = transport->file;
1242
1243         if (sk == NULL)
1244                 return;
1245         /*
1246          * Make sure we're calling this in a context from which it is safe
1247          * to call __fput_sync(). In practice that means rpciod and the
1248          * system workqueue.
1249          */
1250         if (!(current->flags & PF_WQ_WORKER)) {
1251                 WARN_ON_ONCE(1);
1252                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1253                 return;
1254         }
1255
1256         if (atomic_read(&transport->xprt.swapper))
1257                 sk_clear_memalloc(sk);
1258
1259         tls_handshake_cancel(sk);
1260
1261         kernel_sock_shutdown(sock, SHUT_RDWR);
1262
1263         mutex_lock(&transport->recv_mutex);
1264         lock_sock(sk);
1265         transport->inet = NULL;
1266         transport->sock = NULL;
1267         transport->file = NULL;
1268
1269         sk->sk_user_data = NULL;
1270
1271         xs_restore_old_callbacks(transport, sk);
1272         xprt_clear_connected(xprt);
1273         xs_sock_reset_connection_flags(xprt);
1274         /* Reset stream record info */
1275         xs_stream_reset_connect(transport);
1276         release_sock(sk);
1277         mutex_unlock(&transport->recv_mutex);
1278
1279         trace_rpc_socket_close(xprt, sock);
1280         __fput_sync(filp);
1281
1282         xprt_disconnect_done(xprt);
1283 }
1284
1285 /**
1286  * xs_close - close a socket
1287  * @xprt: transport
1288  *
1289  * This is used when all requests are complete; ie, no DRC state remains
1290  * on the server we want to save.
1291  *
1292  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1293  * xs_reset_transport() zeroing the socket from underneath a writer.
1294  */
1295 static void xs_close(struct rpc_xprt *xprt)
1296 {
1297         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1298
1299         dprintk("RPC:       xs_close xprt %p\n", xprt);
1300
1301         if (transport->sock)
1302                 tls_handshake_close(transport->sock);
1303         xs_reset_transport(transport);
1304         xprt->reestablish_timeout = 0;
1305 }
1306
1307 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1308 {
1309         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1310                 xprt);
1311         xprt_disconnect_done(xprt);
1312 }
1313
1314 static void xs_xprt_free(struct rpc_xprt *xprt)
1315 {
1316         xs_free_peer_addresses(xprt);
1317         xprt_free(xprt);
1318 }
1319
1320 /**
1321  * xs_destroy - prepare to shutdown a transport
1322  * @xprt: doomed transport
1323  *
1324  */
1325 static void xs_destroy(struct rpc_xprt *xprt)
1326 {
1327         struct sock_xprt *transport = container_of(xprt,
1328                         struct sock_xprt, xprt);
1329         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1330
1331         cancel_delayed_work_sync(&transport->connect_worker);
1332         xs_close(xprt);
1333         cancel_work_sync(&transport->recv_worker);
1334         cancel_work_sync(&transport->error_worker);
1335         xs_xprt_free(xprt);
1336         module_put(THIS_MODULE);
1337 }
1338
1339 /**
1340  * xs_udp_data_read_skb - receive callback for UDP sockets
1341  * @xprt: transport
1342  * @sk: socket
1343  * @skb: skbuff
1344  *
1345  */
1346 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1347                 struct sock *sk,
1348                 struct sk_buff *skb)
1349 {
1350         struct rpc_task *task;
1351         struct rpc_rqst *rovr;
1352         int repsize, copied;
1353         u32 _xid;
1354         __be32 *xp;
1355
1356         repsize = skb->len;
1357         if (repsize < 4) {
1358                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1359                 return;
1360         }
1361
1362         /* Copy the XID from the skb... */
1363         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1364         if (xp == NULL)
1365                 return;
1366
1367         /* Look up and lock the request corresponding to the given XID */
1368         spin_lock(&xprt->queue_lock);
1369         rovr = xprt_lookup_rqst(xprt, *xp);
1370         if (!rovr)
1371                 goto out_unlock;
1372         xprt_pin_rqst(rovr);
1373         xprt_update_rtt(rovr->rq_task);
1374         spin_unlock(&xprt->queue_lock);
1375         task = rovr->rq_task;
1376
1377         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1378                 copied = repsize;
1379
1380         /* Suck it into the iovec, verify checksum if not done by hw. */
1381         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1382                 spin_lock(&xprt->queue_lock);
1383                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1384                 goto out_unpin;
1385         }
1386
1387
1388         spin_lock(&xprt->transport_lock);
1389         xprt_adjust_cwnd(xprt, task, copied);
1390         spin_unlock(&xprt->transport_lock);
1391         spin_lock(&xprt->queue_lock);
1392         xprt_complete_rqst(task, copied);
1393         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1394 out_unpin:
1395         xprt_unpin_rqst(rovr);
1396  out_unlock:
1397         spin_unlock(&xprt->queue_lock);
1398 }
1399
1400 static void xs_udp_data_receive(struct sock_xprt *transport)
1401 {
1402         struct sk_buff *skb;
1403         struct sock *sk;
1404         int err;
1405
1406         mutex_lock(&transport->recv_mutex);
1407         sk = transport->inet;
1408         if (sk == NULL)
1409                 goto out;
1410         for (;;) {
1411                 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1412                 if (skb == NULL)
1413                         break;
1414                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1415                 consume_skb(skb);
1416                 cond_resched();
1417         }
1418         xs_poll_check_readable(transport);
1419 out:
1420         mutex_unlock(&transport->recv_mutex);
1421 }
1422
1423 static void xs_udp_data_receive_workfn(struct work_struct *work)
1424 {
1425         struct sock_xprt *transport =
1426                 container_of(work, struct sock_xprt, recv_worker);
1427         unsigned int pflags = memalloc_nofs_save();
1428
1429         xs_udp_data_receive(transport);
1430         memalloc_nofs_restore(pflags);
1431 }
1432
1433 /**
1434  * xs_data_ready - "data ready" callback for sockets
1435  * @sk: socket with data to read
1436  *
1437  */
1438 static void xs_data_ready(struct sock *sk)
1439 {
1440         struct rpc_xprt *xprt;
1441
1442         trace_sk_data_ready(sk);
1443
1444         xprt = xprt_from_sock(sk);
1445         if (xprt != NULL) {
1446                 struct sock_xprt *transport = container_of(xprt,
1447                                 struct sock_xprt, xprt);
1448
1449                 trace_xs_data_ready(xprt);
1450
1451                 transport->old_data_ready(sk);
1452
1453                 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1454                         return;
1455
1456                 /* Any data means we had a useful conversation, so
1457                  * then we don't need to delay the next reconnect
1458                  */
1459                 if (xprt->reestablish_timeout)
1460                         xprt->reestablish_timeout = 0;
1461                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1462                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1463         }
1464 }
1465
1466 /*
1467  * Helper function to force a TCP close if the server is sending
1468  * junk and/or it has put us in CLOSE_WAIT
1469  */
1470 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1471 {
1472         xprt_force_disconnect(xprt);
1473 }
1474
1475 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1476 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1477 {
1478         return PAGE_SIZE;
1479 }
1480 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1481
1482 /**
1483  * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1484  * @sk: socket whose state has changed
1485  *
1486  */
1487 static void xs_local_state_change(struct sock *sk)
1488 {
1489         struct rpc_xprt *xprt;
1490         struct sock_xprt *transport;
1491
1492         if (!(xprt = xprt_from_sock(sk)))
1493                 return;
1494         transport = container_of(xprt, struct sock_xprt, xprt);
1495         if (sk->sk_shutdown & SHUTDOWN_MASK) {
1496                 clear_bit(XPRT_CONNECTED, &xprt->state);
1497                 /* Trigger the socket release */
1498                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1499         }
1500 }
1501
1502 /**
1503  * xs_tcp_state_change - callback to handle TCP socket state changes
1504  * @sk: socket whose state has changed
1505  *
1506  */
1507 static void xs_tcp_state_change(struct sock *sk)
1508 {
1509         struct rpc_xprt *xprt;
1510         struct sock_xprt *transport;
1511
1512         if (!(xprt = xprt_from_sock(sk)))
1513                 return;
1514         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1515         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1516                         sk->sk_state, xprt_connected(xprt),
1517                         sock_flag(sk, SOCK_DEAD),
1518                         sock_flag(sk, SOCK_ZAPPED),
1519                         sk->sk_shutdown);
1520
1521         transport = container_of(xprt, struct sock_xprt, xprt);
1522         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1523         switch (sk->sk_state) {
1524         case TCP_ESTABLISHED:
1525                 if (!xprt_test_and_set_connected(xprt)) {
1526                         xprt->connect_cookie++;
1527                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1528                         xprt_clear_connecting(xprt);
1529
1530                         xprt->stat.connect_count++;
1531                         xprt->stat.connect_time += (long)jiffies -
1532                                                    xprt->stat.connect_start;
1533                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1534                 }
1535                 break;
1536         case TCP_FIN_WAIT1:
1537                 /* The client initiated a shutdown of the socket */
1538                 xprt->connect_cookie++;
1539                 xprt->reestablish_timeout = 0;
1540                 set_bit(XPRT_CLOSING, &xprt->state);
1541                 smp_mb__before_atomic();
1542                 clear_bit(XPRT_CONNECTED, &xprt->state);
1543                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1544                 smp_mb__after_atomic();
1545                 break;
1546         case TCP_CLOSE_WAIT:
1547                 /* The server initiated a shutdown of the socket */
1548                 xprt->connect_cookie++;
1549                 clear_bit(XPRT_CONNECTED, &xprt->state);
1550                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1551                 fallthrough;
1552         case TCP_CLOSING:
1553                 /*
1554                  * If the server closed down the connection, make sure that
1555                  * we back off before reconnecting
1556                  */
1557                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1558                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1559                 break;
1560         case TCP_LAST_ACK:
1561                 set_bit(XPRT_CLOSING, &xprt->state);
1562                 smp_mb__before_atomic();
1563                 clear_bit(XPRT_CONNECTED, &xprt->state);
1564                 smp_mb__after_atomic();
1565                 break;
1566         case TCP_CLOSE:
1567                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1568                                         &transport->sock_state))
1569                         xprt_clear_connecting(xprt);
1570                 clear_bit(XPRT_CLOSING, &xprt->state);
1571                 /* Trigger the socket release */
1572                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1573         }
1574 }
1575
1576 static void xs_write_space(struct sock *sk)
1577 {
1578         struct sock_xprt *transport;
1579         struct rpc_xprt *xprt;
1580
1581         if (!sk->sk_socket)
1582                 return;
1583         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1584
1585         if (unlikely(!(xprt = xprt_from_sock(sk))))
1586                 return;
1587         transport = container_of(xprt, struct sock_xprt, xprt);
1588         if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1589                 return;
1590         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1591         sk->sk_write_pending--;
1592 }
1593
1594 /**
1595  * xs_udp_write_space - callback invoked when socket buffer space
1596  *                             becomes available
1597  * @sk: socket whose state has changed
1598  *
1599  * Called when more output buffer space is available for this socket.
1600  * We try not to wake our writers until they can make "significant"
1601  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1602  * with a bunch of small requests.
1603  */
1604 static void xs_udp_write_space(struct sock *sk)
1605 {
1606         /* from net/core/sock.c:sock_def_write_space */
1607         if (sock_writeable(sk))
1608                 xs_write_space(sk);
1609 }
1610
1611 /**
1612  * xs_tcp_write_space - callback invoked when socket buffer space
1613  *                             becomes available
1614  * @sk: socket whose state has changed
1615  *
1616  * Called when more output buffer space is available for this socket.
1617  * We try not to wake our writers until they can make "significant"
1618  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1619  * with a bunch of small requests.
1620  */
1621 static void xs_tcp_write_space(struct sock *sk)
1622 {
1623         /* from net/core/stream.c:sk_stream_write_space */
1624         if (sk_stream_is_writeable(sk))
1625                 xs_write_space(sk);
1626 }
1627
1628 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1629 {
1630         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1631         struct sock *sk = transport->inet;
1632
1633         if (transport->rcvsize) {
1634                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1635                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1636         }
1637         if (transport->sndsize) {
1638                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1639                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1640                 sk->sk_write_space(sk);
1641         }
1642 }
1643
1644 /**
1645  * xs_udp_set_buffer_size - set send and receive limits
1646  * @xprt: generic transport
1647  * @sndsize: requested size of send buffer, in bytes
1648  * @rcvsize: requested size of receive buffer, in bytes
1649  *
1650  * Set socket send and receive buffer size limits.
1651  */
1652 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1653 {
1654         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1655
1656         transport->sndsize = 0;
1657         if (sndsize)
1658                 transport->sndsize = sndsize + 1024;
1659         transport->rcvsize = 0;
1660         if (rcvsize)
1661                 transport->rcvsize = rcvsize + 1024;
1662
1663         xs_udp_do_set_buffer_size(xprt);
1664 }
1665
1666 /**
1667  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1668  * @xprt: controlling transport
1669  * @task: task that timed out
1670  *
1671  * Adjust the congestion window after a retransmit timeout has occurred.
1672  */
1673 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1674 {
1675         spin_lock(&xprt->transport_lock);
1676         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1677         spin_unlock(&xprt->transport_lock);
1678 }
1679
1680 static int xs_get_random_port(void)
1681 {
1682         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1683         unsigned short range;
1684         unsigned short rand;
1685
1686         if (max < min)
1687                 return -EADDRINUSE;
1688         range = max - min + 1;
1689         rand = get_random_u32_below(range);
1690         return rand + min;
1691 }
1692
1693 static unsigned short xs_sock_getport(struct socket *sock)
1694 {
1695         struct sockaddr_storage buf;
1696         unsigned short port = 0;
1697
1698         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1699                 goto out;
1700         switch (buf.ss_family) {
1701         case AF_INET6:
1702                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1703                 break;
1704         case AF_INET:
1705                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1706         }
1707 out:
1708         return port;
1709 }
1710
1711 /**
1712  * xs_set_port - reset the port number in the remote endpoint address
1713  * @xprt: generic transport
1714  * @port: new port number
1715  *
1716  */
1717 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1718 {
1719         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1720
1721         rpc_set_port(xs_addr(xprt), port);
1722         xs_update_peer_port(xprt);
1723 }
1724
1725 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1726 {
1727         if (transport->srcport == 0 && transport->xprt.reuseport)
1728                 transport->srcport = xs_sock_getport(sock);
1729 }
1730
1731 static int xs_get_srcport(struct sock_xprt *transport)
1732 {
1733         int port = transport->srcport;
1734
1735         if (port == 0 && transport->xprt.resvport)
1736                 port = xs_get_random_port();
1737         return port;
1738 }
1739
1740 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1741 {
1742         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1743         unsigned short ret = 0;
1744         mutex_lock(&sock->recv_mutex);
1745         if (sock->sock)
1746                 ret = xs_sock_getport(sock->sock);
1747         mutex_unlock(&sock->recv_mutex);
1748         return ret;
1749 }
1750
1751 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1752 {
1753         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1754         union {
1755                 struct sockaddr sa;
1756                 struct sockaddr_storage st;
1757         } saddr;
1758         int ret = -ENOTCONN;
1759
1760         mutex_lock(&sock->recv_mutex);
1761         if (sock->sock) {
1762                 ret = kernel_getsockname(sock->sock, &saddr.sa);
1763                 if (ret >= 0)
1764                         ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1765         }
1766         mutex_unlock(&sock->recv_mutex);
1767         return ret;
1768 }
1769
1770 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1771 {
1772         if (transport->srcport != 0)
1773                 transport->srcport = 0;
1774         if (!transport->xprt.resvport)
1775                 return 0;
1776         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1777                 return xprt_max_resvport;
1778         return --port;
1779 }
1780 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1781 {
1782         struct sockaddr_storage myaddr;
1783         int err, nloop = 0;
1784         int port = xs_get_srcport(transport);
1785         unsigned short last;
1786
1787         /*
1788          * If we are asking for any ephemeral port (i.e. port == 0 &&
1789          * transport->xprt.resvport == 0), don't bind.  Let the local
1790          * port selection happen implicitly when the socket is used
1791          * (for example at connect time).
1792          *
1793          * This ensures that we can continue to establish TCP
1794          * connections even when all local ephemeral ports are already
1795          * a part of some TCP connection.  This makes no difference
1796          * for UDP sockets, but also doesn't harm them.
1797          *
1798          * If we're asking for any reserved port (i.e. port == 0 &&
1799          * transport->xprt.resvport == 1) xs_get_srcport above will
1800          * ensure that port is non-zero and we will bind as needed.
1801          */
1802         if (port <= 0)
1803                 return port;
1804
1805         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1806         do {
1807                 rpc_set_port((struct sockaddr *)&myaddr, port);
1808                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1809                                 transport->xprt.addrlen);
1810                 if (err == 0) {
1811                         if (transport->xprt.reuseport)
1812                                 transport->srcport = port;
1813                         break;
1814                 }
1815                 last = port;
1816                 port = xs_next_srcport(transport, port);
1817                 if (port > last)
1818                         nloop++;
1819         } while (err == -EADDRINUSE && nloop != 2);
1820
1821         if (myaddr.ss_family == AF_INET)
1822                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1823                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1824                                 port, err ? "failed" : "ok", err);
1825         else
1826                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1827                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1828                                 port, err ? "failed" : "ok", err);
1829         return err;
1830 }
1831
1832 /*
1833  * We don't support autobind on AF_LOCAL sockets
1834  */
1835 static void xs_local_rpcbind(struct rpc_task *task)
1836 {
1837         xprt_set_bound(task->tk_xprt);
1838 }
1839
1840 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1841 {
1842 }
1843
1844 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1845 static struct lock_class_key xs_key[3];
1846 static struct lock_class_key xs_slock_key[3];
1847
1848 static inline void xs_reclassify_socketu(struct socket *sock)
1849 {
1850         struct sock *sk = sock->sk;
1851
1852         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1853                 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1854 }
1855
1856 static inline void xs_reclassify_socket4(struct socket *sock)
1857 {
1858         struct sock *sk = sock->sk;
1859
1860         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1861                 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1862 }
1863
1864 static inline void xs_reclassify_socket6(struct socket *sock)
1865 {
1866         struct sock *sk = sock->sk;
1867
1868         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1869                 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1870 }
1871
1872 static inline void xs_reclassify_socket(int family, struct socket *sock)
1873 {
1874         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1875                 return;
1876
1877         switch (family) {
1878         case AF_LOCAL:
1879                 xs_reclassify_socketu(sock);
1880                 break;
1881         case AF_INET:
1882                 xs_reclassify_socket4(sock);
1883                 break;
1884         case AF_INET6:
1885                 xs_reclassify_socket6(sock);
1886                 break;
1887         }
1888 }
1889 #else
1890 static inline void xs_reclassify_socket(int family, struct socket *sock)
1891 {
1892 }
1893 #endif
1894
1895 static void xs_dummy_setup_socket(struct work_struct *work)
1896 {
1897 }
1898
1899 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1900                 struct sock_xprt *transport, int family, int type,
1901                 int protocol, bool reuseport)
1902 {
1903         struct file *filp;
1904         struct socket *sock;
1905         int err;
1906
1907         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1908         if (err < 0) {
1909                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1910                                 protocol, -err);
1911                 goto out;
1912         }
1913         xs_reclassify_socket(family, sock);
1914
1915         if (reuseport)
1916                 sock_set_reuseport(sock->sk);
1917
1918         err = xs_bind(transport, sock);
1919         if (err) {
1920                 sock_release(sock);
1921                 goto out;
1922         }
1923
1924         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1925         if (IS_ERR(filp))
1926                 return ERR_CAST(filp);
1927         transport->file = filp;
1928
1929         return sock;
1930 out:
1931         return ERR_PTR(err);
1932 }
1933
1934 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1935                                       struct socket *sock)
1936 {
1937         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1938                                                                         xprt);
1939
1940         if (!transport->inet) {
1941                 struct sock *sk = sock->sk;
1942
1943                 lock_sock(sk);
1944
1945                 xs_save_old_callbacks(transport, sk);
1946
1947                 sk->sk_user_data = xprt;
1948                 sk->sk_data_ready = xs_data_ready;
1949                 sk->sk_write_space = xs_udp_write_space;
1950                 sk->sk_state_change = xs_local_state_change;
1951                 sk->sk_error_report = xs_error_report;
1952                 sk->sk_use_task_frag = false;
1953
1954                 xprt_clear_connected(xprt);
1955
1956                 /* Reset to new socket */
1957                 transport->sock = sock;
1958                 transport->inet = sk;
1959
1960                 release_sock(sk);
1961         }
1962
1963         xs_stream_start_connect(transport);
1964
1965         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1966 }
1967
1968 /**
1969  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1970  * @transport: socket transport to connect
1971  */
1972 static int xs_local_setup_socket(struct sock_xprt *transport)
1973 {
1974         struct rpc_xprt *xprt = &transport->xprt;
1975         struct file *filp;
1976         struct socket *sock;
1977         int status;
1978
1979         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1980                                         SOCK_STREAM, 0, &sock, 1);
1981         if (status < 0) {
1982                 dprintk("RPC:       can't create AF_LOCAL "
1983                         "transport socket (%d).\n", -status);
1984                 goto out;
1985         }
1986         xs_reclassify_socket(AF_LOCAL, sock);
1987
1988         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1989         if (IS_ERR(filp)) {
1990                 status = PTR_ERR(filp);
1991                 goto out;
1992         }
1993         transport->file = filp;
1994
1995         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1996                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1997
1998         status = xs_local_finish_connecting(xprt, sock);
1999         trace_rpc_socket_connect(xprt, sock, status);
2000         switch (status) {
2001         case 0:
2002                 dprintk("RPC:       xprt %p connected to %s\n",
2003                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2004                 xprt->stat.connect_count++;
2005                 xprt->stat.connect_time += (long)jiffies -
2006                                            xprt->stat.connect_start;
2007                 xprt_set_connected(xprt);
2008                 break;
2009         case -ENOBUFS:
2010                 break;
2011         case -ENOENT:
2012                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2013                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2014                 break;
2015         case -ECONNREFUSED:
2016                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2017                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2018                 break;
2019         default:
2020                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2021                                 __func__, -status,
2022                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2023         }
2024
2025 out:
2026         xprt_clear_connecting(xprt);
2027         xprt_wake_pending_tasks(xprt, status);
2028         return status;
2029 }
2030
2031 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2032 {
2033         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2034         int ret;
2035
2036         if (transport->file)
2037                 goto force_disconnect;
2038
2039         if (RPC_IS_ASYNC(task)) {
2040                 /*
2041                  * We want the AF_LOCAL connect to be resolved in the
2042                  * filesystem namespace of the process making the rpc
2043                  * call.  Thus we connect synchronously.
2044                  *
2045                  * If we want to support asynchronous AF_LOCAL calls,
2046                  * we'll need to figure out how to pass a namespace to
2047                  * connect.
2048                  */
2049                 rpc_task_set_rpc_status(task, -ENOTCONN);
2050                 goto out_wake;
2051         }
2052         ret = xs_local_setup_socket(transport);
2053         if (ret && !RPC_IS_SOFTCONN(task))
2054                 msleep_interruptible(15000);
2055         return;
2056 force_disconnect:
2057         xprt_force_disconnect(xprt);
2058 out_wake:
2059         xprt_clear_connecting(xprt);
2060         xprt_wake_pending_tasks(xprt, -ENOTCONN);
2061 }
2062
2063 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2064 /*
2065  * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2066  * held, or when we otherwise know that we have exclusive access to the
2067  * socket, to guard against races with xs_reset_transport.
2068  */
2069 static void xs_set_memalloc(struct rpc_xprt *xprt)
2070 {
2071         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2072                         xprt);
2073
2074         /*
2075          * If there's no sock, then we have nothing to set. The
2076          * reconnecting process will get it for us.
2077          */
2078         if (!transport->inet)
2079                 return;
2080         if (atomic_read(&xprt->swapper))
2081                 sk_set_memalloc(transport->inet);
2082 }
2083
2084 /**
2085  * xs_enable_swap - Tag this transport as being used for swap.
2086  * @xprt: transport to tag
2087  *
2088  * Take a reference to this transport on behalf of the rpc_clnt, and
2089  * optionally mark it for swapping if it wasn't already.
2090  */
2091 static int
2092 xs_enable_swap(struct rpc_xprt *xprt)
2093 {
2094         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2095
2096         mutex_lock(&xs->recv_mutex);
2097         if (atomic_inc_return(&xprt->swapper) == 1 &&
2098             xs->inet)
2099                 sk_set_memalloc(xs->inet);
2100         mutex_unlock(&xs->recv_mutex);
2101         return 0;
2102 }
2103
2104 /**
2105  * xs_disable_swap - Untag this transport as being used for swap.
2106  * @xprt: transport to tag
2107  *
2108  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2109  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2110  */
2111 static void
2112 xs_disable_swap(struct rpc_xprt *xprt)
2113 {
2114         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2115
2116         mutex_lock(&xs->recv_mutex);
2117         if (atomic_dec_and_test(&xprt->swapper) &&
2118             xs->inet)
2119                 sk_clear_memalloc(xs->inet);
2120         mutex_unlock(&xs->recv_mutex);
2121 }
2122 #else
2123 static void xs_set_memalloc(struct rpc_xprt *xprt)
2124 {
2125 }
2126
2127 static int
2128 xs_enable_swap(struct rpc_xprt *xprt)
2129 {
2130         return -EINVAL;
2131 }
2132
2133 static void
2134 xs_disable_swap(struct rpc_xprt *xprt)
2135 {
2136 }
2137 #endif
2138
2139 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2140 {
2141         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2142
2143         if (!transport->inet) {
2144                 struct sock *sk = sock->sk;
2145
2146                 lock_sock(sk);
2147
2148                 xs_save_old_callbacks(transport, sk);
2149
2150                 sk->sk_user_data = xprt;
2151                 sk->sk_data_ready = xs_data_ready;
2152                 sk->sk_write_space = xs_udp_write_space;
2153                 sk->sk_use_task_frag = false;
2154
2155                 xprt_set_connected(xprt);
2156
2157                 /* Reset to new socket */
2158                 transport->sock = sock;
2159                 transport->inet = sk;
2160
2161                 xs_set_memalloc(xprt);
2162
2163                 release_sock(sk);
2164         }
2165         xs_udp_do_set_buffer_size(xprt);
2166
2167         xprt->stat.connect_start = jiffies;
2168 }
2169
2170 static void xs_udp_setup_socket(struct work_struct *work)
2171 {
2172         struct sock_xprt *transport =
2173                 container_of(work, struct sock_xprt, connect_worker.work);
2174         struct rpc_xprt *xprt = &transport->xprt;
2175         struct socket *sock;
2176         int status = -EIO;
2177         unsigned int pflags = current->flags;
2178
2179         if (atomic_read(&xprt->swapper))
2180                 current->flags |= PF_MEMALLOC;
2181         sock = xs_create_sock(xprt, transport,
2182                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2183                         IPPROTO_UDP, false);
2184         if (IS_ERR(sock))
2185                 goto out;
2186
2187         dprintk("RPC:       worker connecting xprt %p via %s to "
2188                                 "%s (port %s)\n", xprt,
2189                         xprt->address_strings[RPC_DISPLAY_PROTO],
2190                         xprt->address_strings[RPC_DISPLAY_ADDR],
2191                         xprt->address_strings[RPC_DISPLAY_PORT]);
2192
2193         xs_udp_finish_connecting(xprt, sock);
2194         trace_rpc_socket_connect(xprt, sock, 0);
2195         status = 0;
2196 out:
2197         xprt_clear_connecting(xprt);
2198         xprt_unlock_connect(xprt, transport);
2199         xprt_wake_pending_tasks(xprt, status);
2200         current_restore_flags(pflags, PF_MEMALLOC);
2201 }
2202
2203 /**
2204  * xs_tcp_shutdown - gracefully shut down a TCP socket
2205  * @xprt: transport
2206  *
2207  * Initiates a graceful shutdown of the TCP socket by calling the
2208  * equivalent of shutdown(SHUT_RDWR);
2209  */
2210 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2211 {
2212         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2213         struct socket *sock = transport->sock;
2214         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2215
2216         if (sock == NULL)
2217                 return;
2218         if (!xprt->reuseport) {
2219                 xs_close(xprt);
2220                 return;
2221         }
2222         switch (skst) {
2223         case TCP_FIN_WAIT1:
2224         case TCP_FIN_WAIT2:
2225         case TCP_LAST_ACK:
2226                 break;
2227         case TCP_ESTABLISHED:
2228         case TCP_CLOSE_WAIT:
2229                 kernel_sock_shutdown(sock, SHUT_RDWR);
2230                 trace_rpc_socket_shutdown(xprt, sock);
2231                 break;
2232         default:
2233                 xs_reset_transport(transport);
2234         }
2235 }
2236
2237 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2238                 struct socket *sock)
2239 {
2240         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2241         struct net *net = sock_net(sock->sk);
2242         unsigned long connect_timeout;
2243         unsigned long syn_retries;
2244         unsigned int keepidle;
2245         unsigned int keepcnt;
2246         unsigned int timeo;
2247         unsigned long t;
2248
2249         spin_lock(&xprt->transport_lock);
2250         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2251         keepcnt = xprt->timeout->to_retries + 1;
2252         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2253                 (xprt->timeout->to_retries + 1);
2254         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2255         spin_unlock(&xprt->transport_lock);
2256
2257         /* TCP Keepalive options */
2258         sock_set_keepalive(sock->sk);
2259         tcp_sock_set_keepidle(sock->sk, keepidle);
2260         tcp_sock_set_keepintvl(sock->sk, keepidle);
2261         tcp_sock_set_keepcnt(sock->sk, keepcnt);
2262
2263         /* TCP user timeout (see RFC5482) */
2264         tcp_sock_set_user_timeout(sock->sk, timeo);
2265
2266         /* Connect timeout */
2267         connect_timeout = max_t(unsigned long,
2268                                 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2269         syn_retries = max_t(unsigned long,
2270                             READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2271         for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2272                 ;
2273         if (t <= syn_retries)
2274                 tcp_sock_set_syncnt(sock->sk, t - 1);
2275 }
2276
2277 static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2278                                           unsigned long connect_timeout)
2279 {
2280         struct sock_xprt *transport =
2281                 container_of(xprt, struct sock_xprt, xprt);
2282         struct rpc_timeout to;
2283         unsigned long initval;
2284
2285         memcpy(&to, xprt->timeout, sizeof(to));
2286         /* Arbitrary lower limit */
2287         initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2288         to.to_initval = initval;
2289         to.to_maxval = initval;
2290         to.to_retries = 0;
2291         memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2292         xprt->timeout = &transport->tcp_timeout;
2293         xprt->connect_timeout = connect_timeout;
2294 }
2295
2296 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2297                 unsigned long connect_timeout,
2298                 unsigned long reconnect_timeout)
2299 {
2300         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2301
2302         spin_lock(&xprt->transport_lock);
2303         if (reconnect_timeout < xprt->max_reconnect_timeout)
2304                 xprt->max_reconnect_timeout = reconnect_timeout;
2305         if (connect_timeout < xprt->connect_timeout)
2306                 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2307         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2308         spin_unlock(&xprt->transport_lock);
2309 }
2310
2311 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2312 {
2313         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314
2315         if (!transport->inet) {
2316                 struct sock *sk = sock->sk;
2317
2318                 /* Avoid temporary address, they are bad for long-lived
2319                  * connections such as NFS mounts.
2320                  * RFC4941, section 3.6 suggests that:
2321                  *    Individual applications, which have specific
2322                  *    knowledge about the normal duration of connections,
2323                  *    MAY override this as appropriate.
2324                  */
2325                 if (xs_addr(xprt)->sa_family == PF_INET6) {
2326                         ip6_sock_set_addr_preferences(sk,
2327                                 IPV6_PREFER_SRC_PUBLIC);
2328                 }
2329
2330                 xs_tcp_set_socket_timeouts(xprt, sock);
2331                 tcp_sock_set_nodelay(sk);
2332
2333                 lock_sock(sk);
2334
2335                 xs_save_old_callbacks(transport, sk);
2336
2337                 sk->sk_user_data = xprt;
2338                 sk->sk_data_ready = xs_data_ready;
2339                 sk->sk_state_change = xs_tcp_state_change;
2340                 sk->sk_write_space = xs_tcp_write_space;
2341                 sk->sk_error_report = xs_error_report;
2342                 sk->sk_use_task_frag = false;
2343
2344                 /* socket options */
2345                 sock_reset_flag(sk, SOCK_LINGER);
2346
2347                 xprt_clear_connected(xprt);
2348
2349                 /* Reset to new socket */
2350                 transport->sock = sock;
2351                 transport->inet = sk;
2352
2353                 release_sock(sk);
2354         }
2355
2356         if (!xprt_bound(xprt))
2357                 return -ENOTCONN;
2358
2359         xs_set_memalloc(xprt);
2360
2361         xs_stream_start_connect(transport);
2362
2363         /* Tell the socket layer to start connecting... */
2364         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2365         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2366 }
2367
2368 /**
2369  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2370  * @work: queued work item
2371  *
2372  * Invoked by a work queue tasklet.
2373  */
2374 static void xs_tcp_setup_socket(struct work_struct *work)
2375 {
2376         struct sock_xprt *transport =
2377                 container_of(work, struct sock_xprt, connect_worker.work);
2378         struct socket *sock = transport->sock;
2379         struct rpc_xprt *xprt = &transport->xprt;
2380         int status;
2381         unsigned int pflags = current->flags;
2382
2383         if (atomic_read(&xprt->swapper))
2384                 current->flags |= PF_MEMALLOC;
2385
2386         if (xprt_connected(xprt))
2387                 goto out;
2388         if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2389                                &transport->sock_state) ||
2390             !sock) {
2391                 xs_reset_transport(transport);
2392                 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2393                                       SOCK_STREAM, IPPROTO_TCP, true);
2394                 if (IS_ERR(sock)) {
2395                         xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2396                         goto out;
2397                 }
2398         }
2399
2400         dprintk("RPC:       worker connecting xprt %p via %s to "
2401                                 "%s (port %s)\n", xprt,
2402                         xprt->address_strings[RPC_DISPLAY_PROTO],
2403                         xprt->address_strings[RPC_DISPLAY_ADDR],
2404                         xprt->address_strings[RPC_DISPLAY_PORT]);
2405
2406         status = xs_tcp_finish_connecting(xprt, sock);
2407         trace_rpc_socket_connect(xprt, sock, status);
2408         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2409                         xprt, -status, xprt_connected(xprt),
2410                         sock->sk->sk_state);
2411         switch (status) {
2412         case 0:
2413         case -EINPROGRESS:
2414                 /* SYN_SENT! */
2415                 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2416                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2417                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2418                 fallthrough;
2419         case -EALREADY:
2420                 goto out_unlock;
2421         case -EADDRNOTAVAIL:
2422                 /* Source port number is unavailable. Try a new one! */
2423                 transport->srcport = 0;
2424                 status = -EAGAIN;
2425                 break;
2426         case -EINVAL:
2427                 /* Happens, for instance, if the user specified a link
2428                  * local IPv6 address without a scope-id.
2429                  */
2430         case -ECONNREFUSED:
2431         case -ECONNRESET:
2432         case -ENETDOWN:
2433         case -ENETUNREACH:
2434         case -EHOSTUNREACH:
2435         case -EADDRINUSE:
2436         case -ENOBUFS:
2437                 break;
2438         default:
2439                 printk("%s: connect returned unhandled error %d\n",
2440                         __func__, status);
2441                 status = -EAGAIN;
2442         }
2443
2444         /* xs_tcp_force_close() wakes tasks with a fixed error code.
2445          * We need to wake them first to ensure the correct error code.
2446          */
2447         xprt_wake_pending_tasks(xprt, status);
2448         xs_tcp_force_close(xprt);
2449 out:
2450         xprt_clear_connecting(xprt);
2451 out_unlock:
2452         xprt_unlock_connect(xprt, transport);
2453         current_restore_flags(pflags, PF_MEMALLOC);
2454 }
2455
2456 /*
2457  * Transfer the connected socket to @upper_transport, then mark that
2458  * xprt CONNECTED.
2459  */
2460 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2461                                         struct sock_xprt *upper_transport)
2462 {
2463         struct sock_xprt *lower_transport =
2464                         container_of(lower_xprt, struct sock_xprt, xprt);
2465         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2466
2467         if (!upper_transport->inet) {
2468                 struct socket *sock = lower_transport->sock;
2469                 struct sock *sk = sock->sk;
2470
2471                 /* Avoid temporary address, they are bad for long-lived
2472                  * connections such as NFS mounts.
2473                  * RFC4941, section 3.6 suggests that:
2474                  *    Individual applications, which have specific
2475                  *    knowledge about the normal duration of connections,
2476                  *    MAY override this as appropriate.
2477                  */
2478                 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2479                         ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2480
2481                 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2482                 tcp_sock_set_nodelay(sk);
2483
2484                 lock_sock(sk);
2485
2486                 /* @sk is already connected, so it now has the RPC callbacks.
2487                  * Reach into @lower_transport to save the original ones.
2488                  */
2489                 upper_transport->old_data_ready = lower_transport->old_data_ready;
2490                 upper_transport->old_state_change = lower_transport->old_state_change;
2491                 upper_transport->old_write_space = lower_transport->old_write_space;
2492                 upper_transport->old_error_report = lower_transport->old_error_report;
2493                 sk->sk_user_data = upper_xprt;
2494
2495                 /* socket options */
2496                 sock_reset_flag(sk, SOCK_LINGER);
2497
2498                 xprt_clear_connected(upper_xprt);
2499
2500                 upper_transport->sock = sock;
2501                 upper_transport->inet = sk;
2502                 upper_transport->file = lower_transport->file;
2503
2504                 release_sock(sk);
2505
2506                 /* Reset lower_transport before shutting down its clnt */
2507                 mutex_lock(&lower_transport->recv_mutex);
2508                 lower_transport->inet = NULL;
2509                 lower_transport->sock = NULL;
2510                 lower_transport->file = NULL;
2511
2512                 xprt_clear_connected(lower_xprt);
2513                 xs_sock_reset_connection_flags(lower_xprt);
2514                 xs_stream_reset_connect(lower_transport);
2515                 mutex_unlock(&lower_transport->recv_mutex);
2516         }
2517
2518         if (!xprt_bound(upper_xprt))
2519                 return -ENOTCONN;
2520
2521         xs_set_memalloc(upper_xprt);
2522
2523         if (!xprt_test_and_set_connected(upper_xprt)) {
2524                 upper_xprt->connect_cookie++;
2525                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2526                 xprt_clear_connecting(upper_xprt);
2527
2528                 upper_xprt->stat.connect_count++;
2529                 upper_xprt->stat.connect_time += (long)jiffies -
2530                                            upper_xprt->stat.connect_start;
2531                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2532         }
2533         return 0;
2534 }
2535
2536 /**
2537  * xs_tls_handshake_done - TLS handshake completion handler
2538  * @data: address of xprt to wake
2539  * @status: status of handshake
2540  * @peerid: serial number of key containing the remote's identity
2541  *
2542  */
2543 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2544 {
2545         struct rpc_xprt *lower_xprt = data;
2546         struct sock_xprt *lower_transport =
2547                                 container_of(lower_xprt, struct sock_xprt, xprt);
2548
2549         lower_transport->xprt_err = status ? -EACCES : 0;
2550         complete(&lower_transport->handshake_done);
2551         xprt_put(lower_xprt);
2552 }
2553
2554 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2555 {
2556         struct sock_xprt *lower_transport =
2557                                 container_of(lower_xprt, struct sock_xprt, xprt);
2558         struct tls_handshake_args args = {
2559                 .ta_sock        = lower_transport->sock,
2560                 .ta_done        = xs_tls_handshake_done,
2561                 .ta_data        = xprt_get(lower_xprt),
2562                 .ta_peername    = lower_xprt->servername,
2563         };
2564         struct sock *sk = lower_transport->inet;
2565         int rc;
2566
2567         init_completion(&lower_transport->handshake_done);
2568         set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2569         lower_transport->xprt_err = -ETIMEDOUT;
2570         switch (xprtsec->policy) {
2571         case RPC_XPRTSEC_TLS_ANON:
2572                 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2573                 if (rc)
2574                         goto out_put_xprt;
2575                 break;
2576         case RPC_XPRTSEC_TLS_X509:
2577                 args.ta_my_cert = xprtsec->cert_serial;
2578                 args.ta_my_privkey = xprtsec->privkey_serial;
2579                 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2580                 if (rc)
2581                         goto out_put_xprt;
2582                 break;
2583         default:
2584                 rc = -EACCES;
2585                 goto out_put_xprt;
2586         }
2587
2588         rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2589                                                        XS_TLS_HANDSHAKE_TO);
2590         if (rc <= 0) {
2591                 if (!tls_handshake_cancel(sk)) {
2592                         if (rc == 0)
2593                                 rc = -ETIMEDOUT;
2594                         goto out_put_xprt;
2595                 }
2596         }
2597
2598         rc = lower_transport->xprt_err;
2599
2600 out:
2601         xs_stream_reset_connect(lower_transport);
2602         clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2603         return rc;
2604
2605 out_put_xprt:
2606         xprt_put(lower_xprt);
2607         goto out;
2608 }
2609
2610 /**
2611  * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2612  * @work: queued work item
2613  *
2614  * Invoked by a work queue tasklet.
2615  *
2616  * For RPC-with-TLS, there is a two-stage connection process.
2617  *
2618  * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2619  * been marked connected, the consumer knows that a TCP connection and
2620  * a TLS session have been established.
2621  *
2622  * A "lower-layer xprt", created in this function, handles the mechanics
2623  * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2624  * then driving the TLS handshake. Once all that is complete, the upper
2625  * layer xprt is marked connected.
2626  */
2627 static void xs_tcp_tls_setup_socket(struct work_struct *work)
2628 {
2629         struct sock_xprt *upper_transport =
2630                 container_of(work, struct sock_xprt, connect_worker.work);
2631         struct rpc_clnt *upper_clnt = upper_transport->clnt;
2632         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2633         struct rpc_create_args args = {
2634                 .net            = upper_xprt->xprt_net,
2635                 .protocol       = upper_xprt->prot,
2636                 .address        = (struct sockaddr *)&upper_xprt->addr,
2637                 .addrsize       = upper_xprt->addrlen,
2638                 .timeout        = upper_clnt->cl_timeout,
2639                 .servername     = upper_xprt->servername,
2640                 .program        = upper_clnt->cl_program,
2641                 .prognumber     = upper_clnt->cl_prog,
2642                 .version        = upper_clnt->cl_vers,
2643                 .authflavor     = RPC_AUTH_TLS,
2644                 .cred           = upper_clnt->cl_cred,
2645                 .xprtsec        = {
2646                         .policy         = RPC_XPRTSEC_NONE,
2647                 },
2648         };
2649         unsigned int pflags = current->flags;
2650         struct rpc_clnt *lower_clnt;
2651         struct rpc_xprt *lower_xprt;
2652         int status;
2653
2654         if (atomic_read(&upper_xprt->swapper))
2655                 current->flags |= PF_MEMALLOC;
2656
2657         xs_stream_start_connect(upper_transport);
2658
2659         /* This implicitly sends an RPC_AUTH_TLS probe */
2660         lower_clnt = rpc_create(&args);
2661         if (IS_ERR(lower_clnt)) {
2662                 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2663                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2664                 xprt_clear_connecting(upper_xprt);
2665                 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2666                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2667                 goto out_unlock;
2668         }
2669
2670         /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2671          * the lower xprt.
2672          */
2673         rcu_read_lock();
2674         lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2675         rcu_read_unlock();
2676
2677         if (wait_on_bit_lock(&lower_xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2678                 goto out_unlock;
2679
2680         status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2681         if (status) {
2682                 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2683                 goto out_close;
2684         }
2685
2686         status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2687         if (status)
2688                 goto out_close;
2689         xprt_release_write(lower_xprt, NULL);
2690
2691         trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2692         if (!xprt_test_and_set_connected(upper_xprt)) {
2693                 upper_xprt->connect_cookie++;
2694                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2695                 xprt_clear_connecting(upper_xprt);
2696
2697                 upper_xprt->stat.connect_count++;
2698                 upper_xprt->stat.connect_time += (long)jiffies -
2699                                            upper_xprt->stat.connect_start;
2700                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2701         }
2702         rpc_shutdown_client(lower_clnt);
2703
2704 out_unlock:
2705         current_restore_flags(pflags, PF_MEMALLOC);
2706         upper_transport->clnt = NULL;
2707         xprt_unlock_connect(upper_xprt, upper_transport);
2708         return;
2709
2710 out_close:
2711         xprt_release_write(lower_xprt, NULL);
2712         rpc_shutdown_client(lower_clnt);
2713
2714         /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2715          * Wake them first here to ensure they get our tk_status code.
2716          */
2717         xprt_wake_pending_tasks(upper_xprt, status);
2718         xs_tcp_force_close(upper_xprt);
2719         xprt_clear_connecting(upper_xprt);
2720         goto out_unlock;
2721 }
2722
2723 /**
2724  * xs_connect - connect a socket to a remote endpoint
2725  * @xprt: pointer to transport structure
2726  * @task: address of RPC task that manages state of connect request
2727  *
2728  * TCP: If the remote end dropped the connection, delay reconnecting.
2729  *
2730  * UDP socket connects are synchronous, but we use a work queue anyway
2731  * to guarantee that even unprivileged user processes can set up a
2732  * socket on a privileged port.
2733  *
2734  * If a UDP socket connect fails, the delay behavior here prevents
2735  * retry floods (hard mounts).
2736  */
2737 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2738 {
2739         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2740         unsigned long delay = 0;
2741
2742         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2743
2744         if (transport->sock != NULL) {
2745                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2746                         "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2747
2748                 delay = xprt_reconnect_delay(xprt);
2749                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2750
2751         } else
2752                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2753
2754         transport->clnt = task->tk_client;
2755         queue_delayed_work(xprtiod_workqueue,
2756                         &transport->connect_worker,
2757                         delay);
2758 }
2759
2760 static void xs_wake_disconnect(struct sock_xprt *transport)
2761 {
2762         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2763                 xs_tcp_force_close(&transport->xprt);
2764 }
2765
2766 static void xs_wake_write(struct sock_xprt *transport)
2767 {
2768         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2769                 xprt_write_space(&transport->xprt);
2770 }
2771
2772 static void xs_wake_error(struct sock_xprt *transport)
2773 {
2774         int sockerr;
2775
2776         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2777                 return;
2778         sockerr = xchg(&transport->xprt_err, 0);
2779         if (sockerr < 0) {
2780                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2781                 xs_tcp_force_close(&transport->xprt);
2782         }
2783 }
2784
2785 static void xs_wake_pending(struct sock_xprt *transport)
2786 {
2787         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2788                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2789 }
2790
2791 static void xs_error_handle(struct work_struct *work)
2792 {
2793         struct sock_xprt *transport = container_of(work,
2794                         struct sock_xprt, error_worker);
2795
2796         xs_wake_disconnect(transport);
2797         xs_wake_write(transport);
2798         xs_wake_error(transport);
2799         xs_wake_pending(transport);
2800 }
2801
2802 /**
2803  * xs_local_print_stats - display AF_LOCAL socket-specific stats
2804  * @xprt: rpc_xprt struct containing statistics
2805  * @seq: output file
2806  *
2807  */
2808 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2809 {
2810         long idle_time = 0;
2811
2812         if (xprt_connected(xprt))
2813                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2814
2815         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2816                         "%llu %llu %lu %llu %llu\n",
2817                         xprt->stat.bind_count,
2818                         xprt->stat.connect_count,
2819                         xprt->stat.connect_time / HZ,
2820                         idle_time,
2821                         xprt->stat.sends,
2822                         xprt->stat.recvs,
2823                         xprt->stat.bad_xids,
2824                         xprt->stat.req_u,
2825                         xprt->stat.bklog_u,
2826                         xprt->stat.max_slots,
2827                         xprt->stat.sending_u,
2828                         xprt->stat.pending_u);
2829 }
2830
2831 /**
2832  * xs_udp_print_stats - display UDP socket-specific stats
2833  * @xprt: rpc_xprt struct containing statistics
2834  * @seq: output file
2835  *
2836  */
2837 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2838 {
2839         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2840
2841         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2842                         "%lu %llu %llu\n",
2843                         transport->srcport,
2844                         xprt->stat.bind_count,
2845                         xprt->stat.sends,
2846                         xprt->stat.recvs,
2847                         xprt->stat.bad_xids,
2848                         xprt->stat.req_u,
2849                         xprt->stat.bklog_u,
2850                         xprt->stat.max_slots,
2851                         xprt->stat.sending_u,
2852                         xprt->stat.pending_u);
2853 }
2854
2855 /**
2856  * xs_tcp_print_stats - display TCP socket-specific stats
2857  * @xprt: rpc_xprt struct containing statistics
2858  * @seq: output file
2859  *
2860  */
2861 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2862 {
2863         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2864         long idle_time = 0;
2865
2866         if (xprt_connected(xprt))
2867                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2868
2869         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2870                         "%llu %llu %lu %llu %llu\n",
2871                         transport->srcport,
2872                         xprt->stat.bind_count,
2873                         xprt->stat.connect_count,
2874                         xprt->stat.connect_time / HZ,
2875                         idle_time,
2876                         xprt->stat.sends,
2877                         xprt->stat.recvs,
2878                         xprt->stat.bad_xids,
2879                         xprt->stat.req_u,
2880                         xprt->stat.bklog_u,
2881                         xprt->stat.max_slots,
2882                         xprt->stat.sending_u,
2883                         xprt->stat.pending_u);
2884 }
2885
2886 /*
2887  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2888  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2889  * to use the server side send routines.
2890  */
2891 static int bc_malloc(struct rpc_task *task)
2892 {
2893         struct rpc_rqst *rqst = task->tk_rqstp;
2894         size_t size = rqst->rq_callsize;
2895         struct page *page;
2896         struct rpc_buffer *buf;
2897
2898         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2899                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2900                           size);
2901                 return -EINVAL;
2902         }
2903
2904         page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2905         if (!page)
2906                 return -ENOMEM;
2907
2908         buf = page_address(page);
2909         buf->len = PAGE_SIZE;
2910
2911         rqst->rq_buffer = buf->data;
2912         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2913         return 0;
2914 }
2915
2916 /*
2917  * Free the space allocated in the bc_alloc routine
2918  */
2919 static void bc_free(struct rpc_task *task)
2920 {
2921         void *buffer = task->tk_rqstp->rq_buffer;
2922         struct rpc_buffer *buf;
2923
2924         buf = container_of(buffer, struct rpc_buffer, data);
2925         free_page((unsigned long)buf);
2926 }
2927
2928 static int bc_sendto(struct rpc_rqst *req)
2929 {
2930         struct xdr_buf *xdr = &req->rq_snd_buf;
2931         struct sock_xprt *transport =
2932                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2933         struct msghdr msg = {
2934                 .msg_flags      = 0,
2935         };
2936         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2937                                          (u32)xdr->len);
2938         unsigned int sent = 0;
2939         int err;
2940
2941         req->rq_xtime = ktime_get();
2942         err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2943         if (err < 0)
2944                 return err;
2945         err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2946         xdr_free_bvec(xdr);
2947         if (err < 0 || sent != (xdr->len + sizeof(marker)))
2948                 return -EAGAIN;
2949         return sent;
2950 }
2951
2952 /**
2953  * bc_send_request - Send a backchannel Call on a TCP socket
2954  * @req: rpc_rqst containing Call message to be sent
2955  *
2956  * xpt_mutex ensures @rqstp's whole message is written to the socket
2957  * without interruption.
2958  *
2959  * Return values:
2960  *   %0 if the message was sent successfully
2961  *   %ENOTCONN if the message was not sent
2962  */
2963 static int bc_send_request(struct rpc_rqst *req)
2964 {
2965         struct svc_xprt *xprt;
2966         int len;
2967
2968         /*
2969          * Get the server socket associated with this callback xprt
2970          */
2971         xprt = req->rq_xprt->bc_xprt;
2972
2973         /*
2974          * Grab the mutex to serialize data as the connection is shared
2975          * with the fore channel
2976          */
2977         mutex_lock(&xprt->xpt_mutex);
2978         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2979                 len = -ENOTCONN;
2980         else
2981                 len = bc_sendto(req);
2982         mutex_unlock(&xprt->xpt_mutex);
2983
2984         if (len > 0)
2985                 len = 0;
2986
2987         return len;
2988 }
2989
2990 /*
2991  * The close routine. Since this is client initiated, we do nothing
2992  */
2993
2994 static void bc_close(struct rpc_xprt *xprt)
2995 {
2996         xprt_disconnect_done(xprt);
2997 }
2998
2999 /*
3000  * The xprt destroy routine. Again, because this connection is client
3001  * initiated, we do nothing
3002  */
3003
3004 static void bc_destroy(struct rpc_xprt *xprt)
3005 {
3006         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
3007
3008         xs_xprt_free(xprt);
3009         module_put(THIS_MODULE);
3010 }
3011
3012 static const struct rpc_xprt_ops xs_local_ops = {
3013         .reserve_xprt           = xprt_reserve_xprt,
3014         .release_xprt           = xprt_release_xprt,
3015         .alloc_slot             = xprt_alloc_slot,
3016         .free_slot              = xprt_free_slot,
3017         .rpcbind                = xs_local_rpcbind,
3018         .set_port               = xs_local_set_port,
3019         .connect                = xs_local_connect,
3020         .buf_alloc              = rpc_malloc,
3021         .buf_free               = rpc_free,
3022         .prepare_request        = xs_stream_prepare_request,
3023         .send_request           = xs_local_send_request,
3024         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3025         .close                  = xs_close,
3026         .destroy                = xs_destroy,
3027         .print_stats            = xs_local_print_stats,
3028         .enable_swap            = xs_enable_swap,
3029         .disable_swap           = xs_disable_swap,
3030 };
3031
3032 static const struct rpc_xprt_ops xs_udp_ops = {
3033         .set_buffer_size        = xs_udp_set_buffer_size,
3034         .reserve_xprt           = xprt_reserve_xprt_cong,
3035         .release_xprt           = xprt_release_xprt_cong,
3036         .alloc_slot             = xprt_alloc_slot,
3037         .free_slot              = xprt_free_slot,
3038         .rpcbind                = rpcb_getport_async,
3039         .set_port               = xs_set_port,
3040         .connect                = xs_connect,
3041         .get_srcaddr            = xs_sock_srcaddr,
3042         .get_srcport            = xs_sock_srcport,
3043         .buf_alloc              = rpc_malloc,
3044         .buf_free               = rpc_free,
3045         .send_request           = xs_udp_send_request,
3046         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3047         .timer                  = xs_udp_timer,
3048         .release_request        = xprt_release_rqst_cong,
3049         .close                  = xs_close,
3050         .destroy                = xs_destroy,
3051         .print_stats            = xs_udp_print_stats,
3052         .enable_swap            = xs_enable_swap,
3053         .disable_swap           = xs_disable_swap,
3054         .inject_disconnect      = xs_inject_disconnect,
3055 };
3056
3057 static const struct rpc_xprt_ops xs_tcp_ops = {
3058         .reserve_xprt           = xprt_reserve_xprt,
3059         .release_xprt           = xprt_release_xprt,
3060         .alloc_slot             = xprt_alloc_slot,
3061         .free_slot              = xprt_free_slot,
3062         .rpcbind                = rpcb_getport_async,
3063         .set_port               = xs_set_port,
3064         .connect                = xs_connect,
3065         .get_srcaddr            = xs_sock_srcaddr,
3066         .get_srcport            = xs_sock_srcport,
3067         .buf_alloc              = rpc_malloc,
3068         .buf_free               = rpc_free,
3069         .prepare_request        = xs_stream_prepare_request,
3070         .send_request           = xs_tcp_send_request,
3071         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3072         .close                  = xs_tcp_shutdown,
3073         .destroy                = xs_destroy,
3074         .set_connect_timeout    = xs_tcp_set_connect_timeout,
3075         .print_stats            = xs_tcp_print_stats,
3076         .enable_swap            = xs_enable_swap,
3077         .disable_swap           = xs_disable_swap,
3078         .inject_disconnect      = xs_inject_disconnect,
3079 #ifdef CONFIG_SUNRPC_BACKCHANNEL
3080         .bc_setup               = xprt_setup_bc,
3081         .bc_maxpayload          = xs_tcp_bc_maxpayload,
3082         .bc_num_slots           = xprt_bc_max_slots,
3083         .bc_free_rqst           = xprt_free_bc_rqst,
3084         .bc_destroy             = xprt_destroy_bc,
3085 #endif
3086 };
3087
3088 /*
3089  * The rpc_xprt_ops for the server backchannel
3090  */
3091
3092 static const struct rpc_xprt_ops bc_tcp_ops = {
3093         .reserve_xprt           = xprt_reserve_xprt,
3094         .release_xprt           = xprt_release_xprt,
3095         .alloc_slot             = xprt_alloc_slot,
3096         .free_slot              = xprt_free_slot,
3097         .buf_alloc              = bc_malloc,
3098         .buf_free               = bc_free,
3099         .send_request           = bc_send_request,
3100         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3101         .close                  = bc_close,
3102         .destroy                = bc_destroy,
3103         .print_stats            = xs_tcp_print_stats,
3104         .enable_swap            = xs_enable_swap,
3105         .disable_swap           = xs_disable_swap,
3106         .inject_disconnect      = xs_inject_disconnect,
3107 };
3108
3109 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3110 {
3111         static const struct sockaddr_in sin = {
3112                 .sin_family             = AF_INET,
3113                 .sin_addr.s_addr        = htonl(INADDR_ANY),
3114         };
3115         static const struct sockaddr_in6 sin6 = {
3116                 .sin6_family            = AF_INET6,
3117                 .sin6_addr              = IN6ADDR_ANY_INIT,
3118         };
3119
3120         switch (family) {
3121         case AF_LOCAL:
3122                 break;
3123         case AF_INET:
3124                 memcpy(sap, &sin, sizeof(sin));
3125                 break;
3126         case AF_INET6:
3127                 memcpy(sap, &sin6, sizeof(sin6));
3128                 break;
3129         default:
3130                 dprintk("RPC:       %s: Bad address family\n", __func__);
3131                 return -EAFNOSUPPORT;
3132         }
3133         return 0;
3134 }
3135
3136 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3137                                       unsigned int slot_table_size,
3138                                       unsigned int max_slot_table_size)
3139 {
3140         struct rpc_xprt *xprt;
3141         struct sock_xprt *new;
3142
3143         if (args->addrlen > sizeof(xprt->addr)) {
3144                 dprintk("RPC:       xs_setup_xprt: address too large\n");
3145                 return ERR_PTR(-EBADF);
3146         }
3147
3148         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3149                         max_slot_table_size);
3150         if (xprt == NULL) {
3151                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
3152                                 "rpc_xprt\n");
3153                 return ERR_PTR(-ENOMEM);
3154         }
3155
3156         new = container_of(xprt, struct sock_xprt, xprt);
3157         mutex_init(&new->recv_mutex);
3158         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3159         xprt->addrlen = args->addrlen;
3160         if (args->srcaddr)
3161                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3162         else {
3163                 int err;
3164                 err = xs_init_anyaddr(args->dstaddr->sa_family,
3165                                         (struct sockaddr *)&new->srcaddr);
3166                 if (err != 0) {
3167                         xprt_free(xprt);
3168                         return ERR_PTR(err);
3169                 }
3170         }
3171
3172         return xprt;
3173 }
3174
3175 static const struct rpc_timeout xs_local_default_timeout = {
3176         .to_initval = 10 * HZ,
3177         .to_maxval = 10 * HZ,
3178         .to_retries = 2,
3179 };
3180
3181 /**
3182  * xs_setup_local - Set up transport to use an AF_LOCAL socket
3183  * @args: rpc transport creation arguments
3184  *
3185  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3186  */
3187 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3188 {
3189         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3190         struct sock_xprt *transport;
3191         struct rpc_xprt *xprt;
3192         struct rpc_xprt *ret;
3193
3194         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3195                         xprt_max_tcp_slot_table_entries);
3196         if (IS_ERR(xprt))
3197                 return xprt;
3198         transport = container_of(xprt, struct sock_xprt, xprt);
3199
3200         xprt->prot = 0;
3201         xprt->xprt_class = &xs_local_transport;
3202         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3203
3204         xprt->bind_timeout = XS_BIND_TO;
3205         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3206         xprt->idle_timeout = XS_IDLE_DISC_TO;
3207
3208         xprt->ops = &xs_local_ops;
3209         xprt->timeout = &xs_local_default_timeout;
3210
3211         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3212         INIT_WORK(&transport->error_worker, xs_error_handle);
3213         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3214
3215         switch (sun->sun_family) {
3216         case AF_LOCAL:
3217                 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3218                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
3219                                         sun->sun_path);
3220                         ret = ERR_PTR(-EINVAL);
3221                         goto out_err;
3222                 }
3223                 xprt_set_bound(xprt);
3224                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3225                 break;
3226         default:
3227                 ret = ERR_PTR(-EAFNOSUPPORT);
3228                 goto out_err;
3229         }
3230
3231         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
3232                         xprt->address_strings[RPC_DISPLAY_ADDR]);
3233
3234         if (try_module_get(THIS_MODULE))
3235                 return xprt;
3236         ret = ERR_PTR(-EINVAL);
3237 out_err:
3238         xs_xprt_free(xprt);
3239         return ret;
3240 }
3241
3242 static const struct rpc_timeout xs_udp_default_timeout = {
3243         .to_initval = 5 * HZ,
3244         .to_maxval = 30 * HZ,
3245         .to_increment = 5 * HZ,
3246         .to_retries = 5,
3247 };
3248
3249 /**
3250  * xs_setup_udp - Set up transport to use a UDP socket
3251  * @args: rpc transport creation arguments
3252  *
3253  */
3254 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3255 {
3256         struct sockaddr *addr = args->dstaddr;
3257         struct rpc_xprt *xprt;
3258         struct sock_xprt *transport;
3259         struct rpc_xprt *ret;
3260
3261         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3262                         xprt_udp_slot_table_entries);
3263         if (IS_ERR(xprt))
3264                 return xprt;
3265         transport = container_of(xprt, struct sock_xprt, xprt);
3266
3267         xprt->prot = IPPROTO_UDP;
3268         xprt->xprt_class = &xs_udp_transport;
3269         /* XXX: header size can vary due to auth type, IPv6, etc. */
3270         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3271
3272         xprt->bind_timeout = XS_BIND_TO;
3273         xprt->reestablish_timeout = XS_UDP_REEST_TO;
3274         xprt->idle_timeout = XS_IDLE_DISC_TO;
3275
3276         xprt->ops = &xs_udp_ops;
3277
3278         xprt->timeout = &xs_udp_default_timeout;
3279
3280         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3281         INIT_WORK(&transport->error_worker, xs_error_handle);
3282         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3283
3284         switch (addr->sa_family) {
3285         case AF_INET:
3286                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3287                         xprt_set_bound(xprt);
3288
3289                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3290                 break;
3291         case AF_INET6:
3292                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3293                         xprt_set_bound(xprt);
3294
3295                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3296                 break;
3297         default:
3298                 ret = ERR_PTR(-EAFNOSUPPORT);
3299                 goto out_err;
3300         }
3301
3302         if (xprt_bound(xprt))
3303                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3304                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3305                                 xprt->address_strings[RPC_DISPLAY_PORT],
3306                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3307         else
3308                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3309                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3310                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3311
3312         if (try_module_get(THIS_MODULE))
3313                 return xprt;
3314         ret = ERR_PTR(-EINVAL);
3315 out_err:
3316         xs_xprt_free(xprt);
3317         return ret;
3318 }
3319
3320 static const struct rpc_timeout xs_tcp_default_timeout = {
3321         .to_initval = 60 * HZ,
3322         .to_maxval = 60 * HZ,
3323         .to_retries = 2,
3324 };
3325
3326 /**
3327  * xs_setup_tcp - Set up transport to use a TCP socket
3328  * @args: rpc transport creation arguments
3329  *
3330  */
3331 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3332 {
3333         struct sockaddr *addr = args->dstaddr;
3334         struct rpc_xprt *xprt;
3335         struct sock_xprt *transport;
3336         struct rpc_xprt *ret;
3337         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3338
3339         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3340                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3341
3342         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3343                         max_slot_table_size);
3344         if (IS_ERR(xprt))
3345                 return xprt;
3346         transport = container_of(xprt, struct sock_xprt, xprt);
3347
3348         xprt->prot = IPPROTO_TCP;
3349         xprt->xprt_class = &xs_tcp_transport;
3350         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3351
3352         xprt->bind_timeout = XS_BIND_TO;
3353         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3354         xprt->idle_timeout = XS_IDLE_DISC_TO;
3355
3356         xprt->ops = &xs_tcp_ops;
3357         xprt->timeout = &xs_tcp_default_timeout;
3358
3359         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3360         if (args->reconnect_timeout)
3361                 xprt->max_reconnect_timeout = args->reconnect_timeout;
3362
3363         xprt->connect_timeout = xprt->timeout->to_initval *
3364                 (xprt->timeout->to_retries + 1);
3365         if (args->connect_timeout)
3366                 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3367
3368         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3369         INIT_WORK(&transport->error_worker, xs_error_handle);
3370         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3371
3372         switch (addr->sa_family) {
3373         case AF_INET:
3374                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3375                         xprt_set_bound(xprt);
3376
3377                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3378                 break;
3379         case AF_INET6:
3380                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3381                         xprt_set_bound(xprt);
3382
3383                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3384                 break;
3385         default:
3386                 ret = ERR_PTR(-EAFNOSUPPORT);
3387                 goto out_err;
3388         }
3389
3390         if (xprt_bound(xprt))
3391                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3392                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3393                                 xprt->address_strings[RPC_DISPLAY_PORT],
3394                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3395         else
3396                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3397                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3398                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3399
3400         if (try_module_get(THIS_MODULE))
3401                 return xprt;
3402         ret = ERR_PTR(-EINVAL);
3403 out_err:
3404         xs_xprt_free(xprt);
3405         return ret;
3406 }
3407
3408 /**
3409  * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3410  * @args: rpc transport creation arguments
3411  *
3412  */
3413 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3414 {
3415         struct sockaddr *addr = args->dstaddr;
3416         struct rpc_xprt *xprt;
3417         struct sock_xprt *transport;
3418         struct rpc_xprt *ret;
3419         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3420
3421         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3422                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3423
3424         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3425                              max_slot_table_size);
3426         if (IS_ERR(xprt))
3427                 return xprt;
3428         transport = container_of(xprt, struct sock_xprt, xprt);
3429
3430         xprt->prot = IPPROTO_TCP;
3431         xprt->xprt_class = &xs_tcp_transport;
3432         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3433
3434         xprt->bind_timeout = XS_BIND_TO;
3435         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3436         xprt->idle_timeout = XS_IDLE_DISC_TO;
3437
3438         xprt->ops = &xs_tcp_ops;
3439         xprt->timeout = &xs_tcp_default_timeout;
3440
3441         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3442         xprt->connect_timeout = xprt->timeout->to_initval *
3443                 (xprt->timeout->to_retries + 1);
3444
3445         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3446         INIT_WORK(&transport->error_worker, xs_error_handle);
3447
3448         switch (args->xprtsec.policy) {
3449         case RPC_XPRTSEC_TLS_ANON:
3450         case RPC_XPRTSEC_TLS_X509:
3451                 xprt->xprtsec = args->xprtsec;
3452                 INIT_DELAYED_WORK(&transport->connect_worker,
3453                                   xs_tcp_tls_setup_socket);
3454                 break;
3455         default:
3456                 ret = ERR_PTR(-EACCES);
3457                 goto out_err;
3458         }
3459
3460         switch (addr->sa_family) {
3461         case AF_INET:
3462                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3463                         xprt_set_bound(xprt);
3464
3465                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3466                 break;
3467         case AF_INET6:
3468                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3469                         xprt_set_bound(xprt);
3470
3471                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3472                 break;
3473         default:
3474                 ret = ERR_PTR(-EAFNOSUPPORT);
3475                 goto out_err;
3476         }
3477
3478         if (xprt_bound(xprt))
3479                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3480                         xprt->address_strings[RPC_DISPLAY_ADDR],
3481                         xprt->address_strings[RPC_DISPLAY_PORT],
3482                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3483         else
3484                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3485                         xprt->address_strings[RPC_DISPLAY_ADDR],
3486                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3487
3488         if (try_module_get(THIS_MODULE))
3489                 return xprt;
3490         ret = ERR_PTR(-EINVAL);
3491 out_err:
3492         xs_xprt_free(xprt);
3493         return ret;
3494 }
3495
3496 /**
3497  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3498  * @args: rpc transport creation arguments
3499  *
3500  */
3501 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3502 {
3503         struct sockaddr *addr = args->dstaddr;
3504         struct rpc_xprt *xprt;
3505         struct sock_xprt *transport;
3506         struct svc_sock *bc_sock;
3507         struct rpc_xprt *ret;
3508
3509         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3510                         xprt_tcp_slot_table_entries);
3511         if (IS_ERR(xprt))
3512                 return xprt;
3513         transport = container_of(xprt, struct sock_xprt, xprt);
3514
3515         xprt->prot = IPPROTO_TCP;
3516         xprt->xprt_class = &xs_bc_tcp_transport;
3517         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3518         xprt->timeout = &xs_tcp_default_timeout;
3519
3520         /* backchannel */
3521         xprt_set_bound(xprt);
3522         xprt->bind_timeout = 0;
3523         xprt->reestablish_timeout = 0;
3524         xprt->idle_timeout = 0;
3525
3526         xprt->ops = &bc_tcp_ops;
3527
3528         switch (addr->sa_family) {
3529         case AF_INET:
3530                 xs_format_peer_addresses(xprt, "tcp",
3531                                          RPCBIND_NETID_TCP);
3532                 break;
3533         case AF_INET6:
3534                 xs_format_peer_addresses(xprt, "tcp",
3535                                    RPCBIND_NETID_TCP6);
3536                 break;
3537         default:
3538                 ret = ERR_PTR(-EAFNOSUPPORT);
3539                 goto out_err;
3540         }
3541
3542         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3543                         xprt->address_strings[RPC_DISPLAY_ADDR],
3544                         xprt->address_strings[RPC_DISPLAY_PORT],
3545                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3546
3547         /*
3548          * Once we've associated a backchannel xprt with a connection,
3549          * we want to keep it around as long as the connection lasts,
3550          * in case we need to start using it for a backchannel again;
3551          * this reference won't be dropped until bc_xprt is destroyed.
3552          */
3553         xprt_get(xprt);
3554         args->bc_xprt->xpt_bc_xprt = xprt;
3555         xprt->bc_xprt = args->bc_xprt;
3556         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3557         transport->sock = bc_sock->sk_sock;
3558         transport->inet = bc_sock->sk_sk;
3559
3560         /*
3561          * Since we don't want connections for the backchannel, we set
3562          * the xprt status to connected
3563          */
3564         xprt_set_connected(xprt);
3565
3566         if (try_module_get(THIS_MODULE))
3567                 return xprt;
3568
3569         args->bc_xprt->xpt_bc_xprt = NULL;
3570         args->bc_xprt->xpt_bc_xps = NULL;
3571         xprt_put(xprt);
3572         ret = ERR_PTR(-EINVAL);
3573 out_err:
3574         xs_xprt_free(xprt);
3575         return ret;
3576 }
3577
3578 static struct xprt_class        xs_local_transport = {
3579         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3580         .name           = "named UNIX socket",
3581         .owner          = THIS_MODULE,
3582         .ident          = XPRT_TRANSPORT_LOCAL,
3583         .setup          = xs_setup_local,
3584         .netid          = { "" },
3585 };
3586
3587 static struct xprt_class        xs_udp_transport = {
3588         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3589         .name           = "udp",
3590         .owner          = THIS_MODULE,
3591         .ident          = XPRT_TRANSPORT_UDP,
3592         .setup          = xs_setup_udp,
3593         .netid          = { "udp", "udp6", "" },
3594 };
3595
3596 static struct xprt_class        xs_tcp_transport = {
3597         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3598         .name           = "tcp",
3599         .owner          = THIS_MODULE,
3600         .ident          = XPRT_TRANSPORT_TCP,
3601         .setup          = xs_setup_tcp,
3602         .netid          = { "tcp", "tcp6", "" },
3603 };
3604
3605 static struct xprt_class        xs_tcp_tls_transport = {
3606         .list           = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3607         .name           = "tcp-with-tls",
3608         .owner          = THIS_MODULE,
3609         .ident          = XPRT_TRANSPORT_TCP_TLS,
3610         .setup          = xs_setup_tcp_tls,
3611         .netid          = { "tcp", "tcp6", "" },
3612 };
3613
3614 static struct xprt_class        xs_bc_tcp_transport = {
3615         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3616         .name           = "tcp NFSv4.1 backchannel",
3617         .owner          = THIS_MODULE,
3618         .ident          = XPRT_TRANSPORT_BC_TCP,
3619         .setup          = xs_setup_bc_tcp,
3620         .netid          = { "" },
3621 };
3622
3623 /**
3624  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3625  *
3626  */
3627 int init_socket_xprt(void)
3628 {
3629         if (!sunrpc_table_header)
3630                 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3631
3632         xprt_register_transport(&xs_local_transport);
3633         xprt_register_transport(&xs_udp_transport);
3634         xprt_register_transport(&xs_tcp_transport);
3635         xprt_register_transport(&xs_tcp_tls_transport);
3636         xprt_register_transport(&xs_bc_tcp_transport);
3637
3638         return 0;
3639 }
3640
3641 /**
3642  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3643  *
3644  */
3645 void cleanup_socket_xprt(void)
3646 {
3647         if (sunrpc_table_header) {
3648                 unregister_sysctl_table(sunrpc_table_header);
3649                 sunrpc_table_header = NULL;
3650         }
3651
3652         xprt_unregister_transport(&xs_local_transport);
3653         xprt_unregister_transport(&xs_udp_transport);
3654         xprt_unregister_transport(&xs_tcp_transport);
3655         xprt_unregister_transport(&xs_tcp_tls_transport);
3656         xprt_unregister_transport(&xs_bc_tcp_transport);
3657 }
3658
3659 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3660 {
3661         return param_set_uint_minmax(val, kp,
3662                         RPC_MIN_RESVPORT,
3663                         RPC_MAX_RESVPORT);
3664 }
3665
3666 static const struct kernel_param_ops param_ops_portnr = {
3667         .set = param_set_portnr,
3668         .get = param_get_uint,
3669 };
3670
3671 #define param_check_portnr(name, p) \
3672         __param_check(name, p, unsigned int);
3673
3674 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3675 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3676
3677 static int param_set_slot_table_size(const char *val,
3678                                      const struct kernel_param *kp)
3679 {
3680         return param_set_uint_minmax(val, kp,
3681                         RPC_MIN_SLOT_TABLE,
3682                         RPC_MAX_SLOT_TABLE);
3683 }
3684
3685 static const struct kernel_param_ops param_ops_slot_table_size = {
3686         .set = param_set_slot_table_size,
3687         .get = param_get_uint,
3688 };
3689
3690 #define param_check_slot_table_size(name, p) \
3691         __param_check(name, p, unsigned int);
3692
3693 static int param_set_max_slot_table_size(const char *val,
3694                                      const struct kernel_param *kp)
3695 {
3696         return param_set_uint_minmax(val, kp,
3697                         RPC_MIN_SLOT_TABLE,
3698                         RPC_MAX_SLOT_TABLE_LIMIT);
3699 }
3700
3701 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3702         .set = param_set_max_slot_table_size,
3703         .get = param_get_uint,
3704 };
3705
3706 #define param_check_max_slot_table_size(name, p) \
3707         __param_check(name, p, unsigned int);
3708
3709 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3710                    slot_table_size, 0644);
3711 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3712                    max_slot_table_size, 0644);
3713 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3714                    slot_table_size, 0644);