net/sunrpc/xprtsock.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/net/sunrpc/xprtsock.c
   4  *
   5  * Client-side transport implementation for sockets.
   6  *
   7  * TCP callback races fixes (C) 1998 Red Hat
   8  * TCP send fixes (C) 1998 Red Hat
   9  * TCP NFS related read + write fixes
  10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  11  *
  12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  13  * Fix behaviour when socket buffer is full.
  14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  15  *
  16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  17  *
  18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  19  *   <gilles.quillard@bull.net>
  20  */
  21
  22 #include <linux/types.h>
  23 #include <linux/string.h>
  24 #include <linux/slab.h>
  25 #include <linux/module.h>
  26 #include <linux/capability.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/errno.h>
  29 #include <linux/socket.h>
  30 #include <linux/in.h>
  31 #include <linux/net.h>
  32 #include <linux/mm.h>
  33 #include <linux/un.h>
  34 #include <linux/udp.h>
  35 #include <linux/tcp.h>
  36 #include <linux/sunrpc/clnt.h>
  37 #include <linux/sunrpc/addr.h>
  38 #include <linux/sunrpc/sched.h>
  39 #include <linux/sunrpc/svcsock.h>
  40 #include <linux/sunrpc/xprtsock.h>
  41 #include <linux/file.h>
  42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  43 #include <linux/sunrpc/bc_xprt.h>
  44 #endif
  45
  46 #include <net/sock.h>
  47 #include <net/checksum.h>
  48 #include <net/udp.h>
  49 #include <net/tcp.h>
  50 #include <net/tls_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         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1185         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1186         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1187         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1188         clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1189 }
1190
1191 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1192 {
1193         set_bit(nr, &transport->sock_state);
1194         queue_work(xprtiod_workqueue, &transport->error_worker);
1195 }
1196
1197 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1198 {
1199         xprt->connect_cookie++;
1200         smp_mb__before_atomic();
1201         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1202         clear_bit(XPRT_CLOSING, &xprt->state);
1203         xs_sock_reset_state_flags(xprt);
1204         smp_mb__after_atomic();
1205 }
1206
1207 /**
1208  * xs_error_report - callback to handle TCP socket state errors
1209  * @sk: socket
1210  *
1211  * Note: we don't call sock_error() since there may be a rpc_task
1212  * using the socket, and so we don't want to clear sk->sk_err.
1213  */
1214 static void xs_error_report(struct sock *sk)
1215 {
1216         struct sock_xprt *transport;
1217         struct rpc_xprt *xprt;
1218
1219         if (!(xprt = xprt_from_sock(sk)))
1220                 return;
1221
1222         transport = container_of(xprt, struct sock_xprt, xprt);
1223         transport->xprt_err = -sk->sk_err;
1224         if (transport->xprt_err == 0)
1225                 return;
1226         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1227                         xprt, -transport->xprt_err);
1228         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1229
1230         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1231         smp_mb__before_atomic();
1232         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1233 }
1234
1235 static void xs_reset_transport(struct sock_xprt *transport)
1236 {
1237         struct socket *sock = transport->sock;
1238         struct sock *sk = transport->inet;
1239         struct rpc_xprt *xprt = &transport->xprt;
1240         struct file *filp = transport->file;
1241
1242         if (sk == NULL)
1243                 return;
1244         /*
1245          * Make sure we're calling this in a context from which it is safe
1246          * to call __fput_sync(). In practice that means rpciod and the
1247          * system workqueue.
1248          */
1249         if (!(current->flags & PF_WQ_WORKER)) {
1250                 WARN_ON_ONCE(1);
1251                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1252                 return;
1253         }
1254
1255         if (atomic_read(&transport->xprt.swapper))
1256                 sk_clear_memalloc(sk);
1257
1258         tls_handshake_cancel(sk);
1259
1260         kernel_sock_shutdown(sock, SHUT_RDWR);
1261
1262         mutex_lock(&transport->recv_mutex);
1263         lock_sock(sk);
1264         transport->inet = NULL;
1265         transport->sock = NULL;
1266         transport->file = NULL;
1267
1268         sk->sk_user_data = NULL;
1269
1270         xs_restore_old_callbacks(transport, sk);
1271         xprt_clear_connected(xprt);
1272         xs_sock_reset_connection_flags(xprt);
1273         /* Reset stream record info */
1274         xs_stream_reset_connect(transport);
1275         release_sock(sk);
1276         mutex_unlock(&transport->recv_mutex);
1277
1278         trace_rpc_socket_close(xprt, sock);
1279         __fput_sync(filp);
1280
1281         xprt_disconnect_done(xprt);
1282 }
1283
1284 /**
1285  * xs_close - close a socket
1286  * @xprt: transport
1287  *
1288  * This is used when all requests are complete; ie, no DRC state remains
1289  * on the server we want to save.
1290  *
1291  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1292  * xs_reset_transport() zeroing the socket from underneath a writer.
1293  */
1294 static void xs_close(struct rpc_xprt *xprt)
1295 {
1296         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1297
1298         dprintk("RPC:       xs_close xprt %p\n", xprt);
1299
1300         if (transport->sock)
1301                 tls_handshake_close(transport->sock);
1302         xs_reset_transport(transport);
1303         xprt->reestablish_timeout = 0;
1304 }
1305
1306 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1307 {
1308         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1309                 xprt);
1310         xprt_disconnect_done(xprt);
1311 }
1312
1313 static void xs_xprt_free(struct rpc_xprt *xprt)
1314 {
1315         xs_free_peer_addresses(xprt);
1316         xprt_free(xprt);
1317 }
1318
1319 /**
1320  * xs_destroy - prepare to shutdown a transport
1321  * @xprt: doomed transport
1322  *
1323  */
1324 static void xs_destroy(struct rpc_xprt *xprt)
1325 {
1326         struct sock_xprt *transport = container_of(xprt,
1327                         struct sock_xprt, xprt);
1328         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1329
1330         cancel_delayed_work_sync(&transport->connect_worker);
1331         xs_close(xprt);
1332         cancel_work_sync(&transport->recv_worker);
1333         cancel_work_sync(&transport->error_worker);
1334         xs_xprt_free(xprt);
1335         module_put(THIS_MODULE);
1336 }
1337
1338 /**
1339  * xs_udp_data_read_skb - receive callback for UDP sockets
1340  * @xprt: transport
1341  * @sk: socket
1342  * @skb: skbuff
1343  *
1344  */
1345 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1346                 struct sock *sk,
1347                 struct sk_buff *skb)
1348 {
1349         struct rpc_task *task;
1350         struct rpc_rqst *rovr;
1351         int repsize, copied;
1352         u32 _xid;
1353         __be32 *xp;
1354
1355         repsize = skb->len;
1356         if (repsize < 4) {
1357                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1358                 return;
1359         }
1360
1361         /* Copy the XID from the skb... */
1362         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1363         if (xp == NULL)
1364                 return;
1365
1366         /* Look up and lock the request corresponding to the given XID */
1367         spin_lock(&xprt->queue_lock);
1368         rovr = xprt_lookup_rqst(xprt, *xp);
1369         if (!rovr)
1370                 goto out_unlock;
1371         xprt_pin_rqst(rovr);
1372         xprt_update_rtt(rovr->rq_task);
1373         spin_unlock(&xprt->queue_lock);
1374         task = rovr->rq_task;
1375
1376         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1377                 copied = repsize;
1378
1379         /* Suck it into the iovec, verify checksum if not done by hw. */
1380         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1381                 spin_lock(&xprt->queue_lock);
1382                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1383                 goto out_unpin;
1384         }
1385
1386
1387         spin_lock(&xprt->transport_lock);
1388         xprt_adjust_cwnd(xprt, task, copied);
1389         spin_unlock(&xprt->transport_lock);
1390         spin_lock(&xprt->queue_lock);
1391         xprt_complete_rqst(task, copied);
1392         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1393 out_unpin:
1394         xprt_unpin_rqst(rovr);
1395  out_unlock:
1396         spin_unlock(&xprt->queue_lock);
1397 }
1398
1399 static void xs_udp_data_receive(struct sock_xprt *transport)
1400 {
1401         struct sk_buff *skb;
1402         struct sock *sk;
1403         int err;
1404
1405         mutex_lock(&transport->recv_mutex);
1406         sk = transport->inet;
1407         if (sk == NULL)
1408                 goto out;
1409         for (;;) {
1410                 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1411                 if (skb == NULL)
1412                         break;
1413                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1414                 consume_skb(skb);
1415                 cond_resched();
1416         }
1417         xs_poll_check_readable(transport);
1418 out:
1419         mutex_unlock(&transport->recv_mutex);
1420 }
1421
1422 static void xs_udp_data_receive_workfn(struct work_struct *work)
1423 {
1424         struct sock_xprt *transport =
1425                 container_of(work, struct sock_xprt, recv_worker);
1426         unsigned int pflags = memalloc_nofs_save();
1427
1428         xs_udp_data_receive(transport);
1429         memalloc_nofs_restore(pflags);
1430 }
1431
1432 /**
1433  * xs_data_ready - "data ready" callback for sockets
1434  * @sk: socket with data to read
1435  *
1436  */
1437 static void xs_data_ready(struct sock *sk)
1438 {
1439         struct rpc_xprt *xprt;
1440
1441         trace_sk_data_ready(sk);
1442
1443         xprt = xprt_from_sock(sk);
1444         if (xprt != NULL) {
1445                 struct sock_xprt *transport = container_of(xprt,
1446                                 struct sock_xprt, xprt);
1447
1448                 trace_xs_data_ready(xprt);
1449
1450                 transport->old_data_ready(sk);
1451
1452                 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1453                         return;
1454
1455                 /* Any data means we had a useful conversation, so
1456                  * then we don't need to delay the next reconnect
1457                  */
1458                 if (xprt->reestablish_timeout)
1459                         xprt->reestablish_timeout = 0;
1460                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1461                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1462         }
1463 }
1464
1465 /*
1466  * Helper function to force a TCP close if the server is sending
1467  * junk and/or it has put us in CLOSE_WAIT
1468  */
1469 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1470 {
1471         xprt_force_disconnect(xprt);
1472 }
1473
1474 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1475 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1476 {
1477         return PAGE_SIZE;
1478 }
1479 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1480
1481 /**
1482  * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1483  * @sk: socket whose state has changed
1484  *
1485  */
1486 static void xs_local_state_change(struct sock *sk)
1487 {
1488         struct rpc_xprt *xprt;
1489         struct sock_xprt *transport;
1490
1491         if (!(xprt = xprt_from_sock(sk)))
1492                 return;
1493         transport = container_of(xprt, struct sock_xprt, xprt);
1494         if (sk->sk_shutdown & SHUTDOWN_MASK) {
1495                 clear_bit(XPRT_CONNECTED, &xprt->state);
1496                 /* Trigger the socket release */
1497                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1498         }
1499 }
1500
1501 /**
1502  * xs_tcp_state_change - callback to handle TCP socket state changes
1503  * @sk: socket whose state has changed
1504  *
1505  */
1506 static void xs_tcp_state_change(struct sock *sk)
1507 {
1508         struct rpc_xprt *xprt;
1509         struct sock_xprt *transport;
1510
1511         if (!(xprt = xprt_from_sock(sk)))
1512                 return;
1513         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1514         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1515                         sk->sk_state, xprt_connected(xprt),
1516                         sock_flag(sk, SOCK_DEAD),
1517                         sock_flag(sk, SOCK_ZAPPED),
1518                         sk->sk_shutdown);
1519
1520         transport = container_of(xprt, struct sock_xprt, xprt);
1521         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1522         switch (sk->sk_state) {
1523         case TCP_ESTABLISHED:
1524                 if (!xprt_test_and_set_connected(xprt)) {
1525                         xprt->connect_cookie++;
1526                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1527                         xprt_clear_connecting(xprt);
1528
1529                         xprt->stat.connect_count++;
1530                         xprt->stat.connect_time += (long)jiffies -
1531                                                    xprt->stat.connect_start;
1532                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1533                 }
1534                 break;
1535         case TCP_FIN_WAIT1:
1536                 /* The client initiated a shutdown of the socket */
1537                 xprt->connect_cookie++;
1538                 xprt->reestablish_timeout = 0;
1539                 set_bit(XPRT_CLOSING, &xprt->state);
1540                 smp_mb__before_atomic();
1541                 clear_bit(XPRT_CONNECTED, &xprt->state);
1542                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1543                 smp_mb__after_atomic();
1544                 break;
1545         case TCP_CLOSE_WAIT:
1546                 /* The server initiated a shutdown of the socket */
1547                 xprt->connect_cookie++;
1548                 clear_bit(XPRT_CONNECTED, &xprt->state);
1549                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1550                 fallthrough;
1551         case TCP_CLOSING:
1552                 /*
1553                  * If the server closed down the connection, make sure that
1554                  * we back off before reconnecting
1555                  */
1556                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1557                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1558                 break;
1559         case TCP_LAST_ACK:
1560                 set_bit(XPRT_CLOSING, &xprt->state);
1561                 smp_mb__before_atomic();
1562                 clear_bit(XPRT_CONNECTED, &xprt->state);
1563                 smp_mb__after_atomic();
1564                 break;
1565         case TCP_CLOSE:
1566                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1567                                         &transport->sock_state))
1568                         xprt_clear_connecting(xprt);
1569                 clear_bit(XPRT_CLOSING, &xprt->state);
1570                 /* Trigger the socket release */
1571                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1572         }
1573 }
1574
1575 static void xs_write_space(struct sock *sk)
1576 {
1577         struct sock_xprt *transport;
1578         struct rpc_xprt *xprt;
1579
1580         if (!sk->sk_socket)
1581                 return;
1582         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1583
1584         if (unlikely(!(xprt = xprt_from_sock(sk))))
1585                 return;
1586         transport = container_of(xprt, struct sock_xprt, xprt);
1587         if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1588                 return;
1589         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1590         sk->sk_write_pending--;
1591 }
1592
1593 /**
1594  * xs_udp_write_space - callback invoked when socket buffer space
1595  *                             becomes available
1596  * @sk: socket whose state has changed
1597  *
1598  * Called when more output buffer space is available for this socket.
1599  * We try not to wake our writers until they can make "significant"
1600  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1601  * with a bunch of small requests.
1602  */
1603 static void xs_udp_write_space(struct sock *sk)
1604 {
1605         /* from net/core/sock.c:sock_def_write_space */
1606         if (sock_writeable(sk))
1607                 xs_write_space(sk);
1608 }
1609
1610 /**
1611  * xs_tcp_write_space - callback invoked when socket buffer space
1612  *                             becomes available
1613  * @sk: socket whose state has changed
1614  *
1615  * Called when more output buffer space is available for this socket.
1616  * We try not to wake our writers until they can make "significant"
1617  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1618  * with a bunch of small requests.
1619  */
1620 static void xs_tcp_write_space(struct sock *sk)
1621 {
1622         /* from net/core/stream.c:sk_stream_write_space */
1623         if (sk_stream_is_writeable(sk))
1624                 xs_write_space(sk);
1625 }
1626
1627 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1628 {
1629         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1630         struct sock *sk = transport->inet;
1631
1632         if (transport->rcvsize) {
1633                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1634                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1635         }
1636         if (transport->sndsize) {
1637                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1638                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1639                 sk->sk_write_space(sk);
1640         }
1641 }
1642
1643 /**
1644  * xs_udp_set_buffer_size - set send and receive limits
1645  * @xprt: generic transport
1646  * @sndsize: requested size of send buffer, in bytes
1647  * @rcvsize: requested size of receive buffer, in bytes
1648  *
1649  * Set socket send and receive buffer size limits.
1650  */
1651 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1652 {
1653         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1654
1655         transport->sndsize = 0;
1656         if (sndsize)
1657                 transport->sndsize = sndsize + 1024;
1658         transport->rcvsize = 0;
1659         if (rcvsize)
1660                 transport->rcvsize = rcvsize + 1024;
1661
1662         xs_udp_do_set_buffer_size(xprt);
1663 }
1664
1665 /**
1666  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1667  * @xprt: controlling transport
1668  * @task: task that timed out
1669  *
1670  * Adjust the congestion window after a retransmit timeout has occurred.
1671  */
1672 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1673 {
1674         spin_lock(&xprt->transport_lock);
1675         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1676         spin_unlock(&xprt->transport_lock);
1677 }
1678
1679 static int xs_get_random_port(void)
1680 {
1681         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1682         unsigned short range;
1683         unsigned short rand;
1684
1685         if (max < min)
1686                 return -EADDRINUSE;
1687         range = max - min + 1;
1688         rand = get_random_u32_below(range);
1689         return rand + min;
1690 }
1691
1692 static unsigned short xs_sock_getport(struct socket *sock)
1693 {
1694         struct sockaddr_storage buf;
1695         unsigned short port = 0;
1696
1697         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1698                 goto out;
1699         switch (buf.ss_family) {
1700         case AF_INET6:
1701                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1702                 break;
1703         case AF_INET:
1704                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1705         }
1706 out:
1707         return port;
1708 }
1709
1710 /**
1711  * xs_set_port - reset the port number in the remote endpoint address
1712  * @xprt: generic transport
1713  * @port: new port number
1714  *
1715  */
1716 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1717 {
1718         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1719
1720         rpc_set_port(xs_addr(xprt), port);
1721         xs_update_peer_port(xprt);
1722 }
1723
1724 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1725 {
1726         if (transport->srcport == 0 && transport->xprt.reuseport)
1727                 transport->srcport = xs_sock_getport(sock);
1728 }
1729
1730 static int xs_get_srcport(struct sock_xprt *transport)
1731 {
1732         int port = transport->srcport;
1733
1734         if (port == 0 && transport->xprt.resvport)
1735                 port = xs_get_random_port();
1736         return port;
1737 }
1738
1739 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1740 {
1741         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1742         unsigned short ret = 0;
1743         mutex_lock(&sock->recv_mutex);
1744         if (sock->sock)
1745                 ret = xs_sock_getport(sock->sock);
1746         mutex_unlock(&sock->recv_mutex);
1747         return ret;
1748 }
1749
1750 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1751 {
1752         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1753         union {
1754                 struct sockaddr sa;
1755                 struct sockaddr_storage st;
1756         } saddr;
1757         int ret = -ENOTCONN;
1758
1759         mutex_lock(&sock->recv_mutex);
1760         if (sock->sock) {
1761                 ret = kernel_getsockname(sock->sock, &saddr.sa);
1762                 if (ret >= 0)
1763                         ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1764         }
1765         mutex_unlock(&sock->recv_mutex);
1766         return ret;
1767 }
1768
1769 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1770 {
1771         if (transport->srcport != 0)
1772                 transport->srcport = 0;
1773         if (!transport->xprt.resvport)
1774                 return 0;
1775         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1776                 return xprt_max_resvport;
1777         return --port;
1778 }
1779 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1780 {
1781         struct sockaddr_storage myaddr;
1782         int err, nloop = 0;
1783         int port = xs_get_srcport(transport);
1784         unsigned short last;
1785
1786         /*
1787          * If we are asking for any ephemeral port (i.e. port == 0 &&
1788          * transport->xprt.resvport == 0), don't bind.  Let the local
1789          * port selection happen implicitly when the socket is used
1790          * (for example at connect time).
1791          *
1792          * This ensures that we can continue to establish TCP
1793          * connections even when all local ephemeral ports are already
1794          * a part of some TCP connection.  This makes no difference
1795          * for UDP sockets, but also doesn't harm them.
1796          *
1797          * If we're asking for any reserved port (i.e. port == 0 &&
1798          * transport->xprt.resvport == 1) xs_get_srcport above will
1799          * ensure that port is non-zero and we will bind as needed.
1800          */
1801         if (port <= 0)
1802                 return port;
1803
1804         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1805         do {
1806                 rpc_set_port((struct sockaddr *)&myaddr, port);
1807                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1808                                 transport->xprt.addrlen);
1809                 if (err == 0) {
1810                         if (transport->xprt.reuseport)
1811                                 transport->srcport = port;
1812                         break;
1813                 }
1814                 last = port;
1815                 port = xs_next_srcport(transport, port);
1816                 if (port > last)
1817                         nloop++;
1818         } while (err == -EADDRINUSE && nloop != 2);
1819
1820         if (myaddr.ss_family == AF_INET)
1821                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1822                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1823                                 port, err ? "failed" : "ok", err);
1824         else
1825                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1826                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1827                                 port, err ? "failed" : "ok", err);
1828         return err;
1829 }
1830
1831 /*
1832  * We don't support autobind on AF_LOCAL sockets
1833  */
1834 static void xs_local_rpcbind(struct rpc_task *task)
1835 {
1836         xprt_set_bound(task->tk_xprt);
1837 }
1838
1839 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1840 {
1841 }
1842
1843 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1844 static struct lock_class_key xs_key[3];
1845 static struct lock_class_key xs_slock_key[3];
1846
1847 static inline void xs_reclassify_socketu(struct socket *sock)
1848 {
1849         struct sock *sk = sock->sk;
1850
1851         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1852                 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1853 }
1854
1855 static inline void xs_reclassify_socket4(struct socket *sock)
1856 {
1857         struct sock *sk = sock->sk;
1858
1859         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1860                 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1861 }
1862
1863 static inline void xs_reclassify_socket6(struct socket *sock)
1864 {
1865         struct sock *sk = sock->sk;
1866
1867         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1868                 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1869 }
1870
1871 static inline void xs_reclassify_socket(int family, struct socket *sock)
1872 {
1873         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1874                 return;
1875
1876         switch (family) {
1877         case AF_LOCAL:
1878                 xs_reclassify_socketu(sock);
1879                 break;
1880         case AF_INET:
1881                 xs_reclassify_socket4(sock);
1882                 break;
1883         case AF_INET6:
1884                 xs_reclassify_socket6(sock);
1885                 break;
1886         }
1887 }
1888 #else
1889 static inline void xs_reclassify_socket(int family, struct socket *sock)
1890 {
1891 }
1892 #endif
1893
1894 static void xs_dummy_setup_socket(struct work_struct *work)
1895 {
1896 }
1897
1898 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1899                 struct sock_xprt *transport, int family, int type,
1900                 int protocol, bool reuseport)
1901 {
1902         struct file *filp;
1903         struct socket *sock;
1904         int err;
1905
1906         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1907         if (err < 0) {
1908                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1909                                 protocol, -err);
1910                 goto out;
1911         }
1912         xs_reclassify_socket(family, sock);
1913
1914         if (reuseport)
1915                 sock_set_reuseport(sock->sk);
1916
1917         err = xs_bind(transport, sock);
1918         if (err) {
1919                 sock_release(sock);
1920                 goto out;
1921         }
1922
1923         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1924         if (IS_ERR(filp))
1925                 return ERR_CAST(filp);
1926         transport->file = filp;
1927
1928         return sock;
1929 out:
1930         return ERR_PTR(err);
1931 }
1932
1933 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1934                                       struct socket *sock)
1935 {
1936         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1937                                                                         xprt);
1938
1939         if (!transport->inet) {
1940                 struct sock *sk = sock->sk;
1941
1942                 lock_sock(sk);
1943
1944                 xs_save_old_callbacks(transport, sk);
1945
1946                 sk->sk_user_data = xprt;
1947                 sk->sk_data_ready = xs_data_ready;
1948                 sk->sk_write_space = xs_udp_write_space;
1949                 sk->sk_state_change = xs_local_state_change;
1950                 sk->sk_error_report = xs_error_report;
1951                 sk->sk_use_task_frag = false;
1952
1953                 xprt_clear_connected(xprt);
1954
1955                 /* Reset to new socket */
1956                 transport->sock = sock;
1957                 transport->inet = sk;
1958
1959                 release_sock(sk);
1960         }
1961
1962         xs_stream_start_connect(transport);
1963
1964         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1965 }
1966
1967 /**
1968  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1969  * @transport: socket transport to connect
1970  */
1971 static int xs_local_setup_socket(struct sock_xprt *transport)
1972 {
1973         struct rpc_xprt *xprt = &transport->xprt;
1974         struct file *filp;
1975         struct socket *sock;
1976         int status;
1977
1978         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1979                                         SOCK_STREAM, 0, &sock, 1);
1980         if (status < 0) {
1981                 dprintk("RPC:       can't create AF_LOCAL "
1982                         "transport socket (%d).\n", -status);
1983                 goto out;
1984         }
1985         xs_reclassify_socket(AF_LOCAL, sock);
1986
1987         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1988         if (IS_ERR(filp)) {
1989                 status = PTR_ERR(filp);
1990                 goto out;
1991         }
1992         transport->file = filp;
1993
1994         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1995                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1996
1997         status = xs_local_finish_connecting(xprt, sock);
1998         trace_rpc_socket_connect(xprt, sock, status);
1999         switch (status) {
2000         case 0:
2001                 dprintk("RPC:       xprt %p connected to %s\n",
2002                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2003                 xprt->stat.connect_count++;
2004                 xprt->stat.connect_time += (long)jiffies -
2005                                            xprt->stat.connect_start;
2006                 xprt_set_connected(xprt);
2007                 break;
2008         case -ENOBUFS:
2009                 break;
2010         case -ENOENT:
2011                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2012                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2013                 break;
2014         case -ECONNREFUSED:
2015                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2016                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2017                 break;
2018         default:
2019                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2020                                 __func__, -status,
2021                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2022         }
2023
2024 out:
2025         xprt_clear_connecting(xprt);
2026         xprt_wake_pending_tasks(xprt, status);
2027         return status;
2028 }
2029
2030 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2031 {
2032         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2033         int ret;
2034
2035         if (transport->file)
2036                 goto force_disconnect;
2037
2038         if (RPC_IS_ASYNC(task)) {
2039                 /*
2040                  * We want the AF_LOCAL connect to be resolved in the
2041                  * filesystem namespace of the process making the rpc
2042                  * call.  Thus we connect synchronously.
2043                  *
2044                  * If we want to support asynchronous AF_LOCAL calls,
2045                  * we'll need to figure out how to pass a namespace to
2046                  * connect.
2047                  */
2048                 rpc_task_set_rpc_status(task, -ENOTCONN);
2049                 goto out_wake;
2050         }
2051         ret = xs_local_setup_socket(transport);
2052         if (ret && !RPC_IS_SOFTCONN(task))
2053                 msleep_interruptible(15000);
2054         return;
2055 force_disconnect:
2056         xprt_force_disconnect(xprt);
2057 out_wake:
2058         xprt_clear_connecting(xprt);
2059         xprt_wake_pending_tasks(xprt, -ENOTCONN);
2060 }
2061
2062 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2063 /*
2064  * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2065  * held, or when we otherwise know that we have exclusive access to the
2066  * socket, to guard against races with xs_reset_transport.
2067  */
2068 static void xs_set_memalloc(struct rpc_xprt *xprt)
2069 {
2070         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2071                         xprt);
2072
2073         /*
2074          * If there's no sock, then we have nothing to set. The
2075          * reconnecting process will get it for us.
2076          */
2077         if (!transport->inet)
2078                 return;
2079         if (atomic_read(&xprt->swapper))
2080                 sk_set_memalloc(transport->inet);
2081 }
2082
2083 /**
2084  * xs_enable_swap - Tag this transport as being used for swap.
2085  * @xprt: transport to tag
2086  *
2087  * Take a reference to this transport on behalf of the rpc_clnt, and
2088  * optionally mark it for swapping if it wasn't already.
2089  */
2090 static int
2091 xs_enable_swap(struct rpc_xprt *xprt)
2092 {
2093         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2094
2095         mutex_lock(&xs->recv_mutex);
2096         if (atomic_inc_return(&xprt->swapper) == 1 &&
2097             xs->inet)
2098                 sk_set_memalloc(xs->inet);
2099         mutex_unlock(&xs->recv_mutex);
2100         return 0;
2101 }
2102
2103 /**
2104  * xs_disable_swap - Untag this transport as being used for swap.
2105  * @xprt: transport to tag
2106  *
2107  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2108  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2109  */
2110 static void
2111 xs_disable_swap(struct rpc_xprt *xprt)
2112 {
2113         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2114
2115         mutex_lock(&xs->recv_mutex);
2116         if (atomic_dec_and_test(&xprt->swapper) &&
2117             xs->inet)
2118                 sk_clear_memalloc(xs->inet);
2119         mutex_unlock(&xs->recv_mutex);
2120 }
2121 #else
2122 static void xs_set_memalloc(struct rpc_xprt *xprt)
2123 {
2124 }
2125
2126 static int
2127 xs_enable_swap(struct rpc_xprt *xprt)
2128 {
2129         return -EINVAL;
2130 }
2131
2132 static void
2133 xs_disable_swap(struct rpc_xprt *xprt)
2134 {
2135 }
2136 #endif
2137
2138 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2139 {
2140         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2141
2142         if (!transport->inet) {
2143                 struct sock *sk = sock->sk;
2144
2145                 lock_sock(sk);
2146
2147                 xs_save_old_callbacks(transport, sk);
2148
2149                 sk->sk_user_data = xprt;
2150                 sk->sk_data_ready = xs_data_ready;
2151                 sk->sk_write_space = xs_udp_write_space;
2152                 sk->sk_use_task_frag = false;
2153
2154                 xprt_set_connected(xprt);
2155
2156                 /* Reset to new socket */
2157                 transport->sock = sock;
2158                 transport->inet = sk;
2159
2160                 xs_set_memalloc(xprt);
2161
2162                 release_sock(sk);
2163         }
2164         xs_udp_do_set_buffer_size(xprt);
2165
2166         xprt->stat.connect_start = jiffies;
2167 }
2168
2169 static void xs_udp_setup_socket(struct work_struct *work)
2170 {
2171         struct sock_xprt *transport =
2172                 container_of(work, struct sock_xprt, connect_worker.work);
2173         struct rpc_xprt *xprt = &transport->xprt;
2174         struct socket *sock;
2175         int status = -EIO;
2176         unsigned int pflags = current->flags;
2177
2178         if (atomic_read(&xprt->swapper))
2179                 current->flags |= PF_MEMALLOC;
2180         sock = xs_create_sock(xprt, transport,
2181                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2182                         IPPROTO_UDP, false);
2183         if (IS_ERR(sock))
2184                 goto out;
2185
2186         dprintk("RPC:       worker connecting xprt %p via %s to "
2187                                 "%s (port %s)\n", xprt,
2188                         xprt->address_strings[RPC_DISPLAY_PROTO],
2189                         xprt->address_strings[RPC_DISPLAY_ADDR],
2190                         xprt->address_strings[RPC_DISPLAY_PORT]);
2191
2192         xs_udp_finish_connecting(xprt, sock);
2193         trace_rpc_socket_connect(xprt, sock, 0);
2194         status = 0;
2195 out:
2196         xprt_clear_connecting(xprt);
2197         xprt_unlock_connect(xprt, transport);
2198         xprt_wake_pending_tasks(xprt, status);
2199         current_restore_flags(pflags, PF_MEMALLOC);
2200 }
2201
2202 /**
2203  * xs_tcp_shutdown - gracefully shut down a TCP socket
2204  * @xprt: transport
2205  *
2206  * Initiates a graceful shutdown of the TCP socket by calling the
2207  * equivalent of shutdown(SHUT_RDWR);
2208  */
2209 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2210 {
2211         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2212         struct socket *sock = transport->sock;
2213         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2214
2215         if (sock == NULL)
2216                 return;
2217         if (!xprt->reuseport) {
2218                 xs_close(xprt);
2219                 return;
2220         }
2221         switch (skst) {
2222         case TCP_FIN_WAIT1:
2223         case TCP_FIN_WAIT2:
2224         case TCP_LAST_ACK:
2225                 break;
2226         case TCP_ESTABLISHED:
2227         case TCP_CLOSE_WAIT:
2228                 kernel_sock_shutdown(sock, SHUT_RDWR);
2229                 trace_rpc_socket_shutdown(xprt, sock);
2230                 break;
2231         default:
2232                 xs_reset_transport(transport);
2233         }
2234 }
2235
2236 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2237                 struct socket *sock)
2238 {
2239         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2240         struct net *net = sock_net(sock->sk);
2241         unsigned long connect_timeout;
2242         unsigned long syn_retries;
2243         unsigned int keepidle;
2244         unsigned int keepcnt;
2245         unsigned int timeo;
2246         unsigned long t;
2247
2248         spin_lock(&xprt->transport_lock);
2249         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2250         keepcnt = xprt->timeout->to_retries + 1;
2251         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2252                 (xprt->timeout->to_retries + 1);
2253         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2254         spin_unlock(&xprt->transport_lock);
2255
2256         /* TCP Keepalive options */
2257         sock_set_keepalive(sock->sk);
2258         tcp_sock_set_keepidle(sock->sk, keepidle);
2259         tcp_sock_set_keepintvl(sock->sk, keepidle);
2260         tcp_sock_set_keepcnt(sock->sk, keepcnt);
2261
2262         /* TCP user timeout (see RFC5482) */
2263         tcp_sock_set_user_timeout(sock->sk, timeo);
2264
2265         /* Connect timeout */
2266         connect_timeout = max_t(unsigned long,
2267                                 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2268         syn_retries = max_t(unsigned long,
2269                             READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2270         for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2271                 ;
2272         if (t <= syn_retries)
2273                 tcp_sock_set_syncnt(sock->sk, t - 1);
2274 }
2275
2276 static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2277                                           unsigned long connect_timeout)
2278 {
2279         struct sock_xprt *transport =
2280                 container_of(xprt, struct sock_xprt, xprt);
2281         struct rpc_timeout to;
2282         unsigned long initval;
2283
2284         memcpy(&to, xprt->timeout, sizeof(to));
2285         /* Arbitrary lower limit */
2286         initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2287         to.to_initval = initval;
2288         to.to_maxval = initval;
2289         to.to_retries = 0;
2290         memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2291         xprt->timeout = &transport->tcp_timeout;
2292         xprt->connect_timeout = connect_timeout;
2293 }
2294
2295 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2296                 unsigned long connect_timeout,
2297                 unsigned long reconnect_timeout)
2298 {
2299         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2300
2301         spin_lock(&xprt->transport_lock);
2302         if (reconnect_timeout < xprt->max_reconnect_timeout)
2303                 xprt->max_reconnect_timeout = reconnect_timeout;
2304         if (connect_timeout < xprt->connect_timeout)
2305                 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2306         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2307         spin_unlock(&xprt->transport_lock);
2308 }
2309
2310 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2311 {
2312         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2313
2314         if (!transport->inet) {
2315                 struct sock *sk = sock->sk;
2316
2317                 /* Avoid temporary address, they are bad for long-lived
2318                  * connections such as NFS mounts.
2319                  * RFC4941, section 3.6 suggests that:
2320                  *    Individual applications, which have specific
2321                  *    knowledge about the normal duration of connections,
2322                  *    MAY override this as appropriate.
2323                  */
2324                 if (xs_addr(xprt)->sa_family == PF_INET6) {
2325                         ip6_sock_set_addr_preferences(sk,
2326                                 IPV6_PREFER_SRC_PUBLIC);
2327                 }
2328
2329                 xs_tcp_set_socket_timeouts(xprt, sock);
2330                 tcp_sock_set_nodelay(sk);
2331
2332                 lock_sock(sk);
2333
2334                 xs_save_old_callbacks(transport, sk);
2335
2336                 sk->sk_user_data = xprt;
2337                 sk->sk_data_ready = xs_data_ready;
2338                 sk->sk_state_change = xs_tcp_state_change;
2339                 sk->sk_write_space = xs_tcp_write_space;
2340                 sk->sk_error_report = xs_error_report;
2341                 sk->sk_use_task_frag = false;
2342
2343                 /* socket options */
2344                 sock_reset_flag(sk, SOCK_LINGER);
2345
2346                 xprt_clear_connected(xprt);
2347
2348                 /* Reset to new socket */
2349                 transport->sock = sock;
2350                 transport->inet = sk;
2351
2352                 release_sock(sk);
2353         }
2354
2355         if (!xprt_bound(xprt))
2356                 return -ENOTCONN;
2357
2358         xs_set_memalloc(xprt);
2359
2360         xs_stream_start_connect(transport);
2361
2362         /* Tell the socket layer to start connecting... */
2363         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2364         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2365 }
2366
2367 /**
2368  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2369  * @work: queued work item
2370  *
2371  * Invoked by a work queue tasklet.
2372  */
2373 static void xs_tcp_setup_socket(struct work_struct *work)
2374 {
2375         struct sock_xprt *transport =
2376                 container_of(work, struct sock_xprt, connect_worker.work);
2377         struct socket *sock = transport->sock;
2378         struct rpc_xprt *xprt = &transport->xprt;
2379         int status;
2380         unsigned int pflags = current->flags;
2381
2382         if (atomic_read(&xprt->swapper))
2383                 current->flags |= PF_MEMALLOC;
2384
2385         if (xprt_connected(xprt))
2386                 goto out;
2387         if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2388                                &transport->sock_state) ||
2389             !sock) {
2390                 xs_reset_transport(transport);
2391                 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2392                                       SOCK_STREAM, IPPROTO_TCP, true);
2393                 if (IS_ERR(sock)) {
2394                         xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2395                         goto out;
2396                 }
2397         }
2398
2399         dprintk("RPC:       worker connecting xprt %p via %s to "
2400                                 "%s (port %s)\n", xprt,
2401                         xprt->address_strings[RPC_DISPLAY_PROTO],
2402                         xprt->address_strings[RPC_DISPLAY_ADDR],
2403                         xprt->address_strings[RPC_DISPLAY_PORT]);
2404
2405         status = xs_tcp_finish_connecting(xprt, sock);
2406         trace_rpc_socket_connect(xprt, sock, status);
2407         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2408                         xprt, -status, xprt_connected(xprt),
2409                         sock->sk->sk_state);
2410         switch (status) {
2411         case 0:
2412         case -EINPROGRESS:
2413                 /* SYN_SENT! */
2414                 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2415                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2416                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2417                 fallthrough;
2418         case -EALREADY:
2419                 goto out_unlock;
2420         case -EADDRNOTAVAIL:
2421                 /* Source port number is unavailable. Try a new one! */
2422                 transport->srcport = 0;
2423                 status = -EAGAIN;
2424                 break;
2425         case -EINVAL:
2426                 /* Happens, for instance, if the user specified a link
2427                  * local IPv6 address without a scope-id.
2428                  */
2429         case -ECONNREFUSED:
2430         case -ECONNRESET:
2431         case -ENETDOWN:
2432         case -ENETUNREACH:
2433         case -EHOSTUNREACH:
2434         case -EADDRINUSE:
2435         case -ENOBUFS:
2436                 break;
2437         default:
2438                 printk("%s: connect returned unhandled error %d\n",
2439                         __func__, status);
2440                 status = -EAGAIN;
2441         }
2442
2443         /* xs_tcp_force_close() wakes tasks with a fixed error code.
2444          * We need to wake them first to ensure the correct error code.
2445          */
2446         xprt_wake_pending_tasks(xprt, status);
2447         xs_tcp_force_close(xprt);
2448 out:
2449         xprt_clear_connecting(xprt);
2450 out_unlock:
2451         xprt_unlock_connect(xprt, transport);
2452         current_restore_flags(pflags, PF_MEMALLOC);
2453 }
2454
2455 /*
2456  * Transfer the connected socket to @upper_transport, then mark that
2457  * xprt CONNECTED.
2458  */
2459 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2460                                         struct sock_xprt *upper_transport)
2461 {
2462         struct sock_xprt *lower_transport =
2463                         container_of(lower_xprt, struct sock_xprt, xprt);
2464         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2465
2466         if (!upper_transport->inet) {
2467                 struct socket *sock = lower_transport->sock;
2468                 struct sock *sk = sock->sk;
2469
2470                 /* Avoid temporary address, they are bad for long-lived
2471                  * connections such as NFS mounts.
2472                  * RFC4941, section 3.6 suggests that:
2473                  *    Individual applications, which have specific
2474                  *    knowledge about the normal duration of connections,
2475                  *    MAY override this as appropriate.
2476                  */
2477                 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2478                         ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2479
2480                 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2481                 tcp_sock_set_nodelay(sk);
2482
2483                 lock_sock(sk);
2484
2485                 /* @sk is already connected, so it now has the RPC callbacks.
2486                  * Reach into @lower_transport to save the original ones.
2487                  */
2488                 upper_transport->old_data_ready = lower_transport->old_data_ready;
2489                 upper_transport->old_state_change = lower_transport->old_state_change;
2490                 upper_transport->old_write_space = lower_transport->old_write_space;
2491                 upper_transport->old_error_report = lower_transport->old_error_report;
2492                 sk->sk_user_data = upper_xprt;
2493
2494                 /* socket options */
2495                 sock_reset_flag(sk, SOCK_LINGER);
2496
2497                 xprt_clear_connected(upper_xprt);
2498
2499                 upper_transport->sock = sock;
2500                 upper_transport->inet = sk;
2501                 upper_transport->file = lower_transport->file;
2502
2503                 release_sock(sk);
2504
2505                 /* Reset lower_transport before shutting down its clnt */
2506                 mutex_lock(&lower_transport->recv_mutex);
2507                 lower_transport->inet = NULL;
2508                 lower_transport->sock = NULL;
2509                 lower_transport->file = NULL;
2510
2511                 xprt_clear_connected(lower_xprt);
2512                 xs_sock_reset_connection_flags(lower_xprt);
2513                 xs_stream_reset_connect(lower_transport);
2514                 mutex_unlock(&lower_transport->recv_mutex);
2515         }
2516
2517         if (!xprt_bound(upper_xprt))
2518                 return -ENOTCONN;
2519
2520         xs_set_memalloc(upper_xprt);
2521
2522         if (!xprt_test_and_set_connected(upper_xprt)) {
2523                 upper_xprt->connect_cookie++;
2524                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2525                 xprt_clear_connecting(upper_xprt);
2526
2527                 upper_xprt->stat.connect_count++;
2528                 upper_xprt->stat.connect_time += (long)jiffies -
2529                                            upper_xprt->stat.connect_start;
2530                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2531         }
2532         return 0;
2533 }
2534
2535 /**
2536  * xs_tls_handshake_done - TLS handshake completion handler
2537  * @data: address of xprt to wake
2538  * @status: status of handshake
2539  * @peerid: serial number of key containing the remote's identity
2540  *
2541  */
2542 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2543 {
2544         struct rpc_xprt *lower_xprt = data;
2545         struct sock_xprt *lower_transport =
2546                                 container_of(lower_xprt, struct sock_xprt, xprt);
2547
2548         lower_transport->xprt_err = status ? -EACCES : 0;
2549         complete(&lower_transport->handshake_done);
2550         xprt_put(lower_xprt);
2551 }
2552
2553 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2554 {
2555         struct sock_xprt *lower_transport =
2556                                 container_of(lower_xprt, struct sock_xprt, xprt);
2557         struct tls_handshake_args args = {
2558                 .ta_sock        = lower_transport->sock,
2559                 .ta_done        = xs_tls_handshake_done,
2560                 .ta_data        = xprt_get(lower_xprt),
2561                 .ta_peername    = lower_xprt->servername,
2562         };
2563         struct sock *sk = lower_transport->inet;
2564         int rc;
2565
2566         init_completion(&lower_transport->handshake_done);
2567         set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2568         lower_transport->xprt_err = -ETIMEDOUT;
2569         switch (xprtsec->policy) {
2570         case RPC_XPRTSEC_TLS_ANON:
2571                 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2572                 if (rc)
2573                         goto out_put_xprt;
2574                 break;
2575         case RPC_XPRTSEC_TLS_X509:
2576                 args.ta_my_cert = xprtsec->cert_serial;
2577                 args.ta_my_privkey = xprtsec->privkey_serial;
2578                 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2579                 if (rc)
2580                         goto out_put_xprt;
2581                 break;
2582         default:
2583                 rc = -EACCES;
2584                 goto out_put_xprt;
2585         }
2586
2587         rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2588                                                        XS_TLS_HANDSHAKE_TO);
2589         if (rc <= 0) {
2590                 if (!tls_handshake_cancel(sk)) {
2591                         if (rc == 0)
2592                                 rc = -ETIMEDOUT;
2593                         goto out_put_xprt;
2594                 }
2595         }
2596
2597         rc = lower_transport->xprt_err;
2598
2599 out:
2600         xs_stream_reset_connect(lower_transport);
2601         clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2602         return rc;
2603
2604 out_put_xprt:
2605         xprt_put(lower_xprt);
2606         goto out;
2607 }
2608
2609 /**
2610  * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2611  * @work: queued work item
2612  *
2613  * Invoked by a work queue tasklet.
2614  *
2615  * For RPC-with-TLS, there is a two-stage connection process.
2616  *
2617  * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2618  * been marked connected, the consumer knows that a TCP connection and
2619  * a TLS session have been established.
2620  *
2621  * A "lower-layer xprt", created in this function, handles the mechanics
2622  * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2623  * then driving the TLS handshake. Once all that is complete, the upper
2624  * layer xprt is marked connected.
2625  */
2626 static void xs_tcp_tls_setup_socket(struct work_struct *work)
2627 {
2628         struct sock_xprt *upper_transport =
2629                 container_of(work, struct sock_xprt, connect_worker.work);
2630         struct rpc_clnt *upper_clnt = upper_transport->clnt;
2631         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2632         struct rpc_create_args args = {
2633                 .net            = upper_xprt->xprt_net,
2634                 .protocol       = upper_xprt->prot,
2635                 .address        = (struct sockaddr *)&upper_xprt->addr,
2636                 .addrsize       = upper_xprt->addrlen,
2637                 .timeout        = upper_clnt->cl_timeout,
2638                 .servername     = upper_xprt->servername,
2639                 .program        = upper_clnt->cl_program,
2640                 .prognumber     = upper_clnt->cl_prog,
2641                 .version        = upper_clnt->cl_vers,
2642                 .authflavor     = RPC_AUTH_TLS,
2643                 .cred           = upper_clnt->cl_cred,
2644                 .xprtsec        = {
2645                         .policy         = RPC_XPRTSEC_NONE,
2646                 },
2647         };
2648         unsigned int pflags = current->flags;
2649         struct rpc_clnt *lower_clnt;
2650         struct rpc_xprt *lower_xprt;
2651         int status;
2652
2653         if (atomic_read(&upper_xprt->swapper))
2654                 current->flags |= PF_MEMALLOC;
2655
2656         xs_stream_start_connect(upper_transport);
2657
2658         /* This implicitly sends an RPC_AUTH_TLS probe */
2659         lower_clnt = rpc_create(&args);
2660         if (IS_ERR(lower_clnt)) {
2661                 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2662                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2663                 xprt_clear_connecting(upper_xprt);
2664                 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2665                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2666                 goto out_unlock;
2667         }
2668
2669         /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2670          * the lower xprt.
2671          */
2672         rcu_read_lock();
2673         lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2674         rcu_read_unlock();
2675         status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2676         if (status) {
2677                 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2678                 goto out_close;
2679         }
2680
2681         status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2682         if (status)
2683                 goto out_close;
2684
2685         trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2686         if (!xprt_test_and_set_connected(upper_xprt)) {
2687                 upper_xprt->connect_cookie++;
2688                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2689                 xprt_clear_connecting(upper_xprt);
2690
2691                 upper_xprt->stat.connect_count++;
2692                 upper_xprt->stat.connect_time += (long)jiffies -
2693                                            upper_xprt->stat.connect_start;
2694                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2695         }
2696         rpc_shutdown_client(lower_clnt);
2697
2698 out_unlock:
2699         current_restore_flags(pflags, PF_MEMALLOC);
2700         upper_transport->clnt = NULL;
2701         xprt_unlock_connect(upper_xprt, upper_transport);
2702         return;
2703
2704 out_close:
2705         rpc_shutdown_client(lower_clnt);
2706
2707         /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2708          * Wake them first here to ensure they get our tk_status code.
2709          */
2710         xprt_wake_pending_tasks(upper_xprt, status);
2711         xs_tcp_force_close(upper_xprt);
2712         xprt_clear_connecting(upper_xprt);
2713         goto out_unlock;
2714 }
2715
2716 /**
2717  * xs_connect - connect a socket to a remote endpoint
2718  * @xprt: pointer to transport structure
2719  * @task: address of RPC task that manages state of connect request
2720  *
2721  * TCP: If the remote end dropped the connection, delay reconnecting.
2722  *
2723  * UDP socket connects are synchronous, but we use a work queue anyway
2724  * to guarantee that even unprivileged user processes can set up a
2725  * socket on a privileged port.
2726  *
2727  * If a UDP socket connect fails, the delay behavior here prevents
2728  * retry floods (hard mounts).
2729  */
2730 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2731 {
2732         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2733         unsigned long delay = 0;
2734
2735         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2736
2737         if (transport->sock != NULL) {
2738                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2739                         "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2740
2741                 delay = xprt_reconnect_delay(xprt);
2742                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2743
2744         } else
2745                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2746
2747         transport->clnt = task->tk_client;
2748         queue_delayed_work(xprtiod_workqueue,
2749                         &transport->connect_worker,
2750                         delay);
2751 }
2752
2753 static void xs_wake_disconnect(struct sock_xprt *transport)
2754 {
2755         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2756                 xs_tcp_force_close(&transport->xprt);
2757 }
2758
2759 static void xs_wake_write(struct sock_xprt *transport)
2760 {
2761         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2762                 xprt_write_space(&transport->xprt);
2763 }
2764
2765 static void xs_wake_error(struct sock_xprt *transport)
2766 {
2767         int sockerr;
2768
2769         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2770                 return;
2771         mutex_lock(&transport->recv_mutex);
2772         if (transport->sock == NULL)
2773                 goto out;
2774         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2775                 goto out;
2776         sockerr = xchg(&transport->xprt_err, 0);
2777         if (sockerr < 0)
2778                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2779 out:
2780         mutex_unlock(&transport->recv_mutex);
2781 }
2782
2783 static void xs_wake_pending(struct sock_xprt *transport)
2784 {
2785         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2786                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2787 }
2788
2789 static void xs_error_handle(struct work_struct *work)
2790 {
2791         struct sock_xprt *transport = container_of(work,
2792                         struct sock_xprt, error_worker);
2793
2794         xs_wake_disconnect(transport);
2795         xs_wake_write(transport);
2796         xs_wake_error(transport);
2797         xs_wake_pending(transport);
2798 }
2799
2800 /**
2801  * xs_local_print_stats - display AF_LOCAL socket-specific stats
2802  * @xprt: rpc_xprt struct containing statistics
2803  * @seq: output file
2804  *
2805  */
2806 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2807 {
2808         long idle_time = 0;
2809
2810         if (xprt_connected(xprt))
2811                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2812
2813         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2814                         "%llu %llu %lu %llu %llu\n",
2815                         xprt->stat.bind_count,
2816                         xprt->stat.connect_count,
2817                         xprt->stat.connect_time / HZ,
2818                         idle_time,
2819                         xprt->stat.sends,
2820                         xprt->stat.recvs,
2821                         xprt->stat.bad_xids,
2822                         xprt->stat.req_u,
2823                         xprt->stat.bklog_u,
2824                         xprt->stat.max_slots,
2825                         xprt->stat.sending_u,
2826                         xprt->stat.pending_u);
2827 }
2828
2829 /**
2830  * xs_udp_print_stats - display UDP socket-specific stats
2831  * @xprt: rpc_xprt struct containing statistics
2832  * @seq: output file
2833  *
2834  */
2835 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2836 {
2837         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2838
2839         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2840                         "%lu %llu %llu\n",
2841                         transport->srcport,
2842                         xprt->stat.bind_count,
2843                         xprt->stat.sends,
2844                         xprt->stat.recvs,
2845                         xprt->stat.bad_xids,
2846                         xprt->stat.req_u,
2847                         xprt->stat.bklog_u,
2848                         xprt->stat.max_slots,
2849                         xprt->stat.sending_u,
2850                         xprt->stat.pending_u);
2851 }
2852
2853 /**
2854  * xs_tcp_print_stats - display TCP socket-specific stats
2855  * @xprt: rpc_xprt struct containing statistics
2856  * @seq: output file
2857  *
2858  */
2859 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2860 {
2861         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2862         long idle_time = 0;
2863
2864         if (xprt_connected(xprt))
2865                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2866
2867         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2868                         "%llu %llu %lu %llu %llu\n",
2869                         transport->srcport,
2870                         xprt->stat.bind_count,
2871                         xprt->stat.connect_count,
2872                         xprt->stat.connect_time / HZ,
2873                         idle_time,
2874                         xprt->stat.sends,
2875                         xprt->stat.recvs,
2876                         xprt->stat.bad_xids,
2877                         xprt->stat.req_u,
2878                         xprt->stat.bklog_u,
2879                         xprt->stat.max_slots,
2880                         xprt->stat.sending_u,
2881                         xprt->stat.pending_u);
2882 }
2883
2884 /*
2885  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2886  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2887  * to use the server side send routines.
2888  */
2889 static int bc_malloc(struct rpc_task *task)
2890 {
2891         struct rpc_rqst *rqst = task->tk_rqstp;
2892         size_t size = rqst->rq_callsize;
2893         struct page *page;
2894         struct rpc_buffer *buf;
2895
2896         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2897                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2898                           size);
2899                 return -EINVAL;
2900         }
2901
2902         page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2903         if (!page)
2904                 return -ENOMEM;
2905
2906         buf = page_address(page);
2907         buf->len = PAGE_SIZE;
2908
2909         rqst->rq_buffer = buf->data;
2910         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2911         return 0;
2912 }
2913
2914 /*
2915  * Free the space allocated in the bc_alloc routine
2916  */
2917 static void bc_free(struct rpc_task *task)
2918 {
2919         void *buffer = task->tk_rqstp->rq_buffer;
2920         struct rpc_buffer *buf;
2921
2922         buf = container_of(buffer, struct rpc_buffer, data);
2923         free_page((unsigned long)buf);
2924 }
2925
2926 static int bc_sendto(struct rpc_rqst *req)
2927 {
2928         struct xdr_buf *xdr = &req->rq_snd_buf;
2929         struct sock_xprt *transport =
2930                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2931         struct msghdr msg = {
2932                 .msg_flags      = 0,
2933         };
2934         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2935                                          (u32)xdr->len);
2936         unsigned int sent = 0;
2937         int err;
2938
2939         req->rq_xtime = ktime_get();
2940         err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2941         if (err < 0)
2942                 return err;
2943         err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2944         xdr_free_bvec(xdr);
2945         if (err < 0 || sent != (xdr->len + sizeof(marker)))
2946                 return -EAGAIN;
2947         return sent;
2948 }
2949
2950 /**
2951  * bc_send_request - Send a backchannel Call on a TCP socket
2952  * @req: rpc_rqst containing Call message to be sent
2953  *
2954  * xpt_mutex ensures @rqstp's whole message is written to the socket
2955  * without interruption.
2956  *
2957  * Return values:
2958  *   %0 if the message was sent successfully
2959  *   %ENOTCONN if the message was not sent
2960  */
2961 static int bc_send_request(struct rpc_rqst *req)
2962 {
2963         struct svc_xprt *xprt;
2964         int len;
2965
2966         /*
2967          * Get the server socket associated with this callback xprt
2968          */
2969         xprt = req->rq_xprt->bc_xprt;
2970
2971         /*
2972          * Grab the mutex to serialize data as the connection is shared
2973          * with the fore channel
2974          */
2975         mutex_lock(&xprt->xpt_mutex);
2976         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2977                 len = -ENOTCONN;
2978         else
2979                 len = bc_sendto(req);
2980         mutex_unlock(&xprt->xpt_mutex);
2981
2982         if (len > 0)
2983                 len = 0;
2984
2985         return len;
2986 }
2987
2988 /*
2989  * The close routine. Since this is client initiated, we do nothing
2990  */
2991
2992 static void bc_close(struct rpc_xprt *xprt)
2993 {
2994         xprt_disconnect_done(xprt);
2995 }
2996
2997 /*
2998  * The xprt destroy routine. Again, because this connection is client
2999  * initiated, we do nothing
3000  */
3001
3002 static void bc_destroy(struct rpc_xprt *xprt)
3003 {
3004         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
3005
3006         xs_xprt_free(xprt);
3007         module_put(THIS_MODULE);
3008 }
3009
3010 static const struct rpc_xprt_ops xs_local_ops = {
3011         .reserve_xprt           = xprt_reserve_xprt,
3012         .release_xprt           = xprt_release_xprt,
3013         .alloc_slot             = xprt_alloc_slot,
3014         .free_slot              = xprt_free_slot,
3015         .rpcbind                = xs_local_rpcbind,
3016         .set_port               = xs_local_set_port,
3017         .connect                = xs_local_connect,
3018         .buf_alloc              = rpc_malloc,
3019         .buf_free               = rpc_free,
3020         .prepare_request        = xs_stream_prepare_request,
3021         .send_request           = xs_local_send_request,
3022         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3023         .close                  = xs_close,
3024         .destroy                = xs_destroy,
3025         .print_stats            = xs_local_print_stats,
3026         .enable_swap            = xs_enable_swap,
3027         .disable_swap           = xs_disable_swap,
3028 };
3029
3030 static const struct rpc_xprt_ops xs_udp_ops = {
3031         .set_buffer_size        = xs_udp_set_buffer_size,
3032         .reserve_xprt           = xprt_reserve_xprt_cong,
3033         .release_xprt           = xprt_release_xprt_cong,
3034         .alloc_slot             = xprt_alloc_slot,
3035         .free_slot              = xprt_free_slot,
3036         .rpcbind                = rpcb_getport_async,
3037         .set_port               = xs_set_port,
3038         .connect                = xs_connect,
3039         .get_srcaddr            = xs_sock_srcaddr,
3040         .get_srcport            = xs_sock_srcport,
3041         .buf_alloc              = rpc_malloc,
3042         .buf_free               = rpc_free,
3043         .send_request           = xs_udp_send_request,
3044         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3045         .timer                  = xs_udp_timer,
3046         .release_request        = xprt_release_rqst_cong,
3047         .close                  = xs_close,
3048         .destroy                = xs_destroy,
3049         .print_stats            = xs_udp_print_stats,
3050         .enable_swap            = xs_enable_swap,
3051         .disable_swap           = xs_disable_swap,
3052         .inject_disconnect      = xs_inject_disconnect,
3053 };
3054
3055 static const struct rpc_xprt_ops xs_tcp_ops = {
3056         .reserve_xprt           = xprt_reserve_xprt,
3057         .release_xprt           = xprt_release_xprt,
3058         .alloc_slot             = xprt_alloc_slot,
3059         .free_slot              = xprt_free_slot,
3060         .rpcbind                = rpcb_getport_async,
3061         .set_port               = xs_set_port,
3062         .connect                = xs_connect,
3063         .get_srcaddr            = xs_sock_srcaddr,
3064         .get_srcport            = xs_sock_srcport,
3065         .buf_alloc              = rpc_malloc,
3066         .buf_free               = rpc_free,
3067         .prepare_request        = xs_stream_prepare_request,
3068         .send_request           = xs_tcp_send_request,
3069         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3070         .close                  = xs_tcp_shutdown,
3071         .destroy                = xs_destroy,
3072         .set_connect_timeout    = xs_tcp_set_connect_timeout,
3073         .print_stats            = xs_tcp_print_stats,
3074         .enable_swap            = xs_enable_swap,
3075         .disable_swap           = xs_disable_swap,
3076         .inject_disconnect      = xs_inject_disconnect,
3077 #ifdef CONFIG_SUNRPC_BACKCHANNEL
3078         .bc_setup               = xprt_setup_bc,
3079         .bc_maxpayload          = xs_tcp_bc_maxpayload,
3080         .bc_num_slots           = xprt_bc_max_slots,
3081         .bc_free_rqst           = xprt_free_bc_rqst,
3082         .bc_destroy             = xprt_destroy_bc,
3083 #endif
3084 };
3085
3086 /*
3087  * The rpc_xprt_ops for the server backchannel
3088  */
3089
3090 static const struct rpc_xprt_ops bc_tcp_ops = {
3091         .reserve_xprt           = xprt_reserve_xprt,
3092         .release_xprt           = xprt_release_xprt,
3093         .alloc_slot             = xprt_alloc_slot,
3094         .free_slot              = xprt_free_slot,
3095         .buf_alloc              = bc_malloc,
3096         .buf_free               = bc_free,
3097         .send_request           = bc_send_request,
3098         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3099         .close                  = bc_close,
3100         .destroy                = bc_destroy,
3101         .print_stats            = xs_tcp_print_stats,
3102         .enable_swap            = xs_enable_swap,
3103         .disable_swap           = xs_disable_swap,
3104         .inject_disconnect      = xs_inject_disconnect,
3105 };
3106
3107 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3108 {
3109         static const struct sockaddr_in sin = {
3110                 .sin_family             = AF_INET,
3111                 .sin_addr.s_addr        = htonl(INADDR_ANY),
3112         };
3113         static const struct sockaddr_in6 sin6 = {
3114                 .sin6_family            = AF_INET6,
3115                 .sin6_addr              = IN6ADDR_ANY_INIT,
3116         };
3117
3118         switch (family) {
3119         case AF_LOCAL:
3120                 break;
3121         case AF_INET:
3122                 memcpy(sap, &sin, sizeof(sin));
3123                 break;
3124         case AF_INET6:
3125                 memcpy(sap, &sin6, sizeof(sin6));
3126                 break;
3127         default:
3128                 dprintk("RPC:       %s: Bad address family\n", __func__);
3129                 return -EAFNOSUPPORT;
3130         }
3131         return 0;
3132 }
3133
3134 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3135                                       unsigned int slot_table_size,
3136                                       unsigned int max_slot_table_size)
3137 {
3138         struct rpc_xprt *xprt;
3139         struct sock_xprt *new;
3140
3141         if (args->addrlen > sizeof(xprt->addr)) {
3142                 dprintk("RPC:       xs_setup_xprt: address too large\n");
3143                 return ERR_PTR(-EBADF);
3144         }
3145
3146         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3147                         max_slot_table_size);
3148         if (xprt == NULL) {
3149                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
3150                                 "rpc_xprt\n");
3151                 return ERR_PTR(-ENOMEM);
3152         }
3153
3154         new = container_of(xprt, struct sock_xprt, xprt);
3155         mutex_init(&new->recv_mutex);
3156         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3157         xprt->addrlen = args->addrlen;
3158         if (args->srcaddr)
3159                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3160         else {
3161                 int err;
3162                 err = xs_init_anyaddr(args->dstaddr->sa_family,
3163                                         (struct sockaddr *)&new->srcaddr);
3164                 if (err != 0) {
3165                         xprt_free(xprt);
3166                         return ERR_PTR(err);
3167                 }
3168         }
3169
3170         return xprt;
3171 }
3172
3173 static const struct rpc_timeout xs_local_default_timeout = {
3174         .to_initval = 10 * HZ,
3175         .to_maxval = 10 * HZ,
3176         .to_retries = 2,
3177 };
3178
3179 /**
3180  * xs_setup_local - Set up transport to use an AF_LOCAL socket
3181  * @args: rpc transport creation arguments
3182  *
3183  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3184  */
3185 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3186 {
3187         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3188         struct sock_xprt *transport;
3189         struct rpc_xprt *xprt;
3190         struct rpc_xprt *ret;
3191
3192         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3193                         xprt_max_tcp_slot_table_entries);
3194         if (IS_ERR(xprt))
3195                 return xprt;
3196         transport = container_of(xprt, struct sock_xprt, xprt);
3197
3198         xprt->prot = 0;
3199         xprt->xprt_class = &xs_local_transport;
3200         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3201
3202         xprt->bind_timeout = XS_BIND_TO;
3203         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3204         xprt->idle_timeout = XS_IDLE_DISC_TO;
3205
3206         xprt->ops = &xs_local_ops;
3207         xprt->timeout = &xs_local_default_timeout;
3208
3209         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3210         INIT_WORK(&transport->error_worker, xs_error_handle);
3211         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3212
3213         switch (sun->sun_family) {
3214         case AF_LOCAL:
3215                 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3216                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
3217                                         sun->sun_path);
3218                         ret = ERR_PTR(-EINVAL);
3219                         goto out_err;
3220                 }
3221                 xprt_set_bound(xprt);
3222                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3223                 break;
3224         default:
3225                 ret = ERR_PTR(-EAFNOSUPPORT);
3226                 goto out_err;
3227         }
3228
3229         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
3230                         xprt->address_strings[RPC_DISPLAY_ADDR]);
3231
3232         if (try_module_get(THIS_MODULE))
3233                 return xprt;
3234         ret = ERR_PTR(-EINVAL);
3235 out_err:
3236         xs_xprt_free(xprt);
3237         return ret;
3238 }
3239
3240 static const struct rpc_timeout xs_udp_default_timeout = {
3241         .to_initval = 5 * HZ,
3242         .to_maxval = 30 * HZ,
3243         .to_increment = 5 * HZ,
3244         .to_retries = 5,
3245 };
3246
3247 /**
3248  * xs_setup_udp - Set up transport to use a UDP socket
3249  * @args: rpc transport creation arguments
3250  *
3251  */
3252 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3253 {
3254         struct sockaddr *addr = args->dstaddr;
3255         struct rpc_xprt *xprt;
3256         struct sock_xprt *transport;
3257         struct rpc_xprt *ret;
3258
3259         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3260                         xprt_udp_slot_table_entries);
3261         if (IS_ERR(xprt))
3262                 return xprt;
3263         transport = container_of(xprt, struct sock_xprt, xprt);
3264
3265         xprt->prot = IPPROTO_UDP;
3266         xprt->xprt_class = &xs_udp_transport;
3267         /* XXX: header size can vary due to auth type, IPv6, etc. */
3268         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3269
3270         xprt->bind_timeout = XS_BIND_TO;
3271         xprt->reestablish_timeout = XS_UDP_REEST_TO;
3272         xprt->idle_timeout = XS_IDLE_DISC_TO;
3273
3274         xprt->ops = &xs_udp_ops;
3275
3276         xprt->timeout = &xs_udp_default_timeout;
3277
3278         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3279         INIT_WORK(&transport->error_worker, xs_error_handle);
3280         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3281
3282         switch (addr->sa_family) {
3283         case AF_INET:
3284                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3285                         xprt_set_bound(xprt);
3286
3287                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3288                 break;
3289         case AF_INET6:
3290                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3291                         xprt_set_bound(xprt);
3292
3293                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3294                 break;
3295         default:
3296                 ret = ERR_PTR(-EAFNOSUPPORT);
3297                 goto out_err;
3298         }
3299
3300         if (xprt_bound(xprt))
3301                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3302                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3303                                 xprt->address_strings[RPC_DISPLAY_PORT],
3304                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3305         else
3306                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3307                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3308                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3309
3310         if (try_module_get(THIS_MODULE))
3311                 return xprt;
3312         ret = ERR_PTR(-EINVAL);
3313 out_err:
3314         xs_xprt_free(xprt);
3315         return ret;
3316 }
3317
3318 static const struct rpc_timeout xs_tcp_default_timeout = {
3319         .to_initval = 60 * HZ,
3320         .to_maxval = 60 * HZ,
3321         .to_retries = 2,
3322 };
3323
3324 /**
3325  * xs_setup_tcp - Set up transport to use a TCP socket
3326  * @args: rpc transport creation arguments
3327  *
3328  */
3329 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3330 {
3331         struct sockaddr *addr = args->dstaddr;
3332         struct rpc_xprt *xprt;
3333         struct sock_xprt *transport;
3334         struct rpc_xprt *ret;
3335         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3336
3337         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3338                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3339
3340         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3341                         max_slot_table_size);
3342         if (IS_ERR(xprt))
3343                 return xprt;
3344         transport = container_of(xprt, struct sock_xprt, xprt);
3345
3346         xprt->prot = IPPROTO_TCP;
3347         xprt->xprt_class = &xs_tcp_transport;
3348         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3349
3350         xprt->bind_timeout = XS_BIND_TO;
3351         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3352         xprt->idle_timeout = XS_IDLE_DISC_TO;
3353
3354         xprt->ops = &xs_tcp_ops;
3355         xprt->timeout = &xs_tcp_default_timeout;
3356
3357         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3358         if (args->reconnect_timeout)
3359                 xprt->max_reconnect_timeout = args->reconnect_timeout;
3360
3361         xprt->connect_timeout = xprt->timeout->to_initval *
3362                 (xprt->timeout->to_retries + 1);
3363         if (args->connect_timeout)
3364                 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3365
3366         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3367         INIT_WORK(&transport->error_worker, xs_error_handle);
3368         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3369
3370         switch (addr->sa_family) {
3371         case AF_INET:
3372                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3373                         xprt_set_bound(xprt);
3374
3375                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3376                 break;
3377         case AF_INET6:
3378                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3379                         xprt_set_bound(xprt);
3380
3381                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3382                 break;
3383         default:
3384                 ret = ERR_PTR(-EAFNOSUPPORT);
3385                 goto out_err;
3386         }
3387
3388         if (xprt_bound(xprt))
3389                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3390                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3391                                 xprt->address_strings[RPC_DISPLAY_PORT],
3392                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3393         else
3394                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3395                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3396                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3397
3398         if (try_module_get(THIS_MODULE))
3399                 return xprt;
3400         ret = ERR_PTR(-EINVAL);
3401 out_err:
3402         xs_xprt_free(xprt);
3403         return ret;
3404 }
3405
3406 /**
3407  * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3408  * @args: rpc transport creation arguments
3409  *
3410  */
3411 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3412 {
3413         struct sockaddr *addr = args->dstaddr;
3414         struct rpc_xprt *xprt;
3415         struct sock_xprt *transport;
3416         struct rpc_xprt *ret;
3417         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3418
3419         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3420                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3421
3422         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3423                              max_slot_table_size);
3424         if (IS_ERR(xprt))
3425                 return xprt;
3426         transport = container_of(xprt, struct sock_xprt, xprt);
3427
3428         xprt->prot = IPPROTO_TCP;
3429         xprt->xprt_class = &xs_tcp_transport;
3430         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3431
3432         xprt->bind_timeout = XS_BIND_TO;
3433         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3434         xprt->idle_timeout = XS_IDLE_DISC_TO;
3435
3436         xprt->ops = &xs_tcp_ops;
3437         xprt->timeout = &xs_tcp_default_timeout;
3438
3439         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3440         xprt->connect_timeout = xprt->timeout->to_initval *
3441                 (xprt->timeout->to_retries + 1);
3442
3443         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3444         INIT_WORK(&transport->error_worker, xs_error_handle);
3445
3446         switch (args->xprtsec.policy) {
3447         case RPC_XPRTSEC_TLS_ANON:
3448         case RPC_XPRTSEC_TLS_X509:
3449                 xprt->xprtsec = args->xprtsec;
3450                 INIT_DELAYED_WORK(&transport->connect_worker,
3451                                   xs_tcp_tls_setup_socket);
3452                 break;
3453         default:
3454                 ret = ERR_PTR(-EACCES);
3455                 goto out_err;
3456         }
3457
3458         switch (addr->sa_family) {
3459         case AF_INET:
3460                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3461                         xprt_set_bound(xprt);
3462
3463                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3464                 break;
3465         case AF_INET6:
3466                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3467                         xprt_set_bound(xprt);
3468
3469                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3470                 break;
3471         default:
3472                 ret = ERR_PTR(-EAFNOSUPPORT);
3473                 goto out_err;
3474         }
3475
3476         if (xprt_bound(xprt))
3477                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3478                         xprt->address_strings[RPC_DISPLAY_ADDR],
3479                         xprt->address_strings[RPC_DISPLAY_PORT],
3480                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3481         else
3482                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3483                         xprt->address_strings[RPC_DISPLAY_ADDR],
3484                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3485
3486         if (try_module_get(THIS_MODULE))
3487                 return xprt;
3488         ret = ERR_PTR(-EINVAL);
3489 out_err:
3490         xs_xprt_free(xprt);
3491         return ret;
3492 }
3493
3494 /**
3495  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3496  * @args: rpc transport creation arguments
3497  *
3498  */
3499 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3500 {
3501         struct sockaddr *addr = args->dstaddr;
3502         struct rpc_xprt *xprt;
3503         struct sock_xprt *transport;
3504         struct svc_sock *bc_sock;
3505         struct rpc_xprt *ret;
3506
3507         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3508                         xprt_tcp_slot_table_entries);
3509         if (IS_ERR(xprt))
3510                 return xprt;
3511         transport = container_of(xprt, struct sock_xprt, xprt);
3512
3513         xprt->prot = IPPROTO_TCP;
3514         xprt->xprt_class = &xs_bc_tcp_transport;
3515         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3516         xprt->timeout = &xs_tcp_default_timeout;
3517
3518         /* backchannel */
3519         xprt_set_bound(xprt);
3520         xprt->bind_timeout = 0;
3521         xprt->reestablish_timeout = 0;
3522         xprt->idle_timeout = 0;
3523
3524         xprt->ops = &bc_tcp_ops;
3525
3526         switch (addr->sa_family) {
3527         case AF_INET:
3528                 xs_format_peer_addresses(xprt, "tcp",
3529                                          RPCBIND_NETID_TCP);
3530                 break;
3531         case AF_INET6:
3532                 xs_format_peer_addresses(xprt, "tcp",
3533                                    RPCBIND_NETID_TCP6);
3534                 break;
3535         default:
3536                 ret = ERR_PTR(-EAFNOSUPPORT);
3537                 goto out_err;
3538         }
3539
3540         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3541                         xprt->address_strings[RPC_DISPLAY_ADDR],
3542                         xprt->address_strings[RPC_DISPLAY_PORT],
3543                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3544
3545         /*
3546          * Once we've associated a backchannel xprt with a connection,
3547          * we want to keep it around as long as the connection lasts,
3548          * in case we need to start using it for a backchannel again;
3549          * this reference won't be dropped until bc_xprt is destroyed.
3550          */
3551         xprt_get(xprt);
3552         args->bc_xprt->xpt_bc_xprt = xprt;
3553         xprt->bc_xprt = args->bc_xprt;
3554         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3555         transport->sock = bc_sock->sk_sock;
3556         transport->inet = bc_sock->sk_sk;
3557
3558         /*
3559          * Since we don't want connections for the backchannel, we set
3560          * the xprt status to connected
3561          */
3562         xprt_set_connected(xprt);
3563
3564         if (try_module_get(THIS_MODULE))
3565                 return xprt;
3566
3567         args->bc_xprt->xpt_bc_xprt = NULL;
3568         args->bc_xprt->xpt_bc_xps = NULL;
3569         xprt_put(xprt);
3570         ret = ERR_PTR(-EINVAL);
3571 out_err:
3572         xs_xprt_free(xprt);
3573         return ret;
3574 }
3575
3576 static struct xprt_class        xs_local_transport = {
3577         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3578         .name           = "named UNIX socket",
3579         .owner          = THIS_MODULE,
3580         .ident          = XPRT_TRANSPORT_LOCAL,
3581         .setup          = xs_setup_local,
3582         .netid          = { "" },
3583 };
3584
3585 static struct xprt_class        xs_udp_transport = {
3586         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3587         .name           = "udp",
3588         .owner          = THIS_MODULE,
3589         .ident          = XPRT_TRANSPORT_UDP,
3590         .setup          = xs_setup_udp,
3591         .netid          = { "udp", "udp6", "" },
3592 };
3593
3594 static struct xprt_class        xs_tcp_transport = {
3595         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3596         .name           = "tcp",
3597         .owner          = THIS_MODULE,
3598         .ident          = XPRT_TRANSPORT_TCP,
3599         .setup          = xs_setup_tcp,
3600         .netid          = { "tcp", "tcp6", "" },
3601 };
3602
3603 static struct xprt_class        xs_tcp_tls_transport = {
3604         .list           = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3605         .name           = "tcp-with-tls",
3606         .owner          = THIS_MODULE,
3607         .ident          = XPRT_TRANSPORT_TCP_TLS,
3608         .setup          = xs_setup_tcp_tls,
3609         .netid          = { "tcp", "tcp6", "" },
3610 };
3611
3612 static struct xprt_class        xs_bc_tcp_transport = {
3613         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3614         .name           = "tcp NFSv4.1 backchannel",
3615         .owner          = THIS_MODULE,
3616         .ident          = XPRT_TRANSPORT_BC_TCP,
3617         .setup          = xs_setup_bc_tcp,
3618         .netid          = { "" },
3619 };
3620
3621 /**
3622  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3623  *
3624  */
3625 int init_socket_xprt(void)
3626 {
3627         if (!sunrpc_table_header)
3628                 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3629
3630         xprt_register_transport(&xs_local_transport);
3631         xprt_register_transport(&xs_udp_transport);
3632         xprt_register_transport(&xs_tcp_transport);
3633         xprt_register_transport(&xs_tcp_tls_transport);
3634         xprt_register_transport(&xs_bc_tcp_transport);
3635
3636         return 0;
3637 }
3638
3639 /**
3640  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3641  *
3642  */
3643 void cleanup_socket_xprt(void)
3644 {
3645         if (sunrpc_table_header) {
3646                 unregister_sysctl_table(sunrpc_table_header);
3647                 sunrpc_table_header = NULL;
3648         }
3649
3650         xprt_unregister_transport(&xs_local_transport);
3651         xprt_unregister_transport(&xs_udp_transport);
3652         xprt_unregister_transport(&xs_tcp_transport);
3653         xprt_unregister_transport(&xs_tcp_tls_transport);
3654         xprt_unregister_transport(&xs_bc_tcp_transport);
3655 }
3656
3657 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3658 {
3659         return param_set_uint_minmax(val, kp,
3660                         RPC_MIN_RESVPORT,
3661                         RPC_MAX_RESVPORT);
3662 }
3663
3664 static const struct kernel_param_ops param_ops_portnr = {
3665         .set = param_set_portnr,
3666         .get = param_get_uint,
3667 };
3668
3669 #define param_check_portnr(name, p) \
3670         __param_check(name, p, unsigned int);
3671
3672 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3673 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3674
3675 static int param_set_slot_table_size(const char *val,
3676                                      const struct kernel_param *kp)
3677 {
3678         return param_set_uint_minmax(val, kp,
3679                         RPC_MIN_SLOT_TABLE,
3680                         RPC_MAX_SLOT_TABLE);
3681 }
3682
3683 static const struct kernel_param_ops param_ops_slot_table_size = {
3684         .set = param_set_slot_table_size,
3685         .get = param_get_uint,
3686 };
3687
3688 #define param_check_slot_table_size(name, p) \
3689         __param_check(name, p, unsigned int);
3690
3691 static int param_set_max_slot_table_size(const char *val,
3692                                      const struct kernel_param *kp)
3693 {
3694         return param_set_uint_minmax(val, kp,
3695                         RPC_MIN_SLOT_TABLE,
3696                         RPC_MAX_SLOT_TABLE_LIMIT);
3697 }
3698
3699 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3700         .set = param_set_max_slot_table_size,
3701         .get = param_get_uint,
3702 };
3703
3704 #define param_check_max_slot_table_size(name, p) \
3705         __param_check(name, p, unsigned int);
3706
3707 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3708                    slot_table_size, 0644);
3709 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3710                    max_slot_table_size, 0644);
3711 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3712                    slot_table_size, 0644);