net/sunrpc/xprtsock.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/net/sunrpc/xprtsock.c
   4  *
   5  * Client-side transport implementation for sockets.
   6  *
   7  * TCP callback races fixes (C) 1998 Red Hat
   8  * TCP send fixes (C) 1998 Red Hat
   9  * TCP NFS related read + write fixes
  10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  11  *
  12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  13  * Fix behaviour when socket buffer is full.
  14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  15  *
  16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  17  *
  18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  19  *   <gilles.quillard@bull.net>
  20  */
  21
  22 #include <linux/types.h>
  23 #include <linux/string.h>
  24 #include <linux/slab.h>
  25 #include <linux/module.h>
  26 #include <linux/capability.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/errno.h>
  29 #include <linux/socket.h>
  30 #include <linux/in.h>
  31 #include <linux/net.h>
  32 #include <linux/mm.h>
  33 #include <linux/un.h>
  34 #include <linux/udp.h>
  35 #include <linux/tcp.h>
  36 #include <linux/sunrpc/clnt.h>
  37 #include <linux/sunrpc/addr.h>
  38 #include <linux/sunrpc/sched.h>
  39 #include <linux/sunrpc/svcsock.h>
  40 #include <linux/sunrpc/xprtsock.h>
  41 #include <linux/file.h>
  42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  43 #include <linux/sunrpc/bc_xprt.h>
  44 #endif
  45
  46 #include <net/sock.h>
  47 #include <net/checksum.h>
  48 #include <net/udp.h>
  49 #include <net/tcp.h>
  50 #include <linux/bvec.h>
  51 #include <linux/highmem.h>
  52 #include <linux/uio.h>
  53 #include <linux/sched/mm.h>
  54
  55 #include <trace/events/sunrpc.h>
  56
  57 #include "socklib.h"
  58 #include "sunrpc.h"
  59
  60 static void xs_close(struct rpc_xprt *xprt);
  61 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  62                 struct socket *sock);
  63
  64 /*
  65  * xprtsock tunables
  66  */
  67 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  68 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  69 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  70
  71 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  72 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  73
  74 #define XS_TCP_LINGER_TO        (15U * HZ)
  75 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  76
  77 /*
  78  * We can register our own files under /proc/sys/sunrpc by
  79  * calling register_sysctl_table() again.  The files in that
  80  * directory become the union of all files registered there.
  81  *
  82  * We simply need to make sure that we don't collide with
  83  * someone else's file names!
  84  */
  85
  86 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  87 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  88 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  89 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  90 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  91
  92 static struct ctl_table_header *sunrpc_table_header;
  93
  94 static struct xprt_class xs_local_transport;
  95 static struct xprt_class xs_udp_transport;
  96 static struct xprt_class xs_tcp_transport;
  97 static struct xprt_class xs_bc_tcp_transport;
  98
  99 /*
 100  * FIXME: changing the UDP slot table size should also resize the UDP
 101  *        socket buffers for existing UDP transports
 102  */
 103 static struct ctl_table xs_tunables_table[] = {
 104         {
 105                 .procname       = "udp_slot_table_entries",
 106                 .data           = &xprt_udp_slot_table_entries,
 107                 .maxlen         = sizeof(unsigned int),
 108                 .mode           = 0644,
 109                 .proc_handler   = proc_dointvec_minmax,
 110                 .extra1         = &min_slot_table_size,
 111                 .extra2         = &max_slot_table_size
 112         },
 113         {
 114                 .procname       = "tcp_slot_table_entries",
 115                 .data           = &xprt_tcp_slot_table_entries,
 116                 .maxlen         = sizeof(unsigned int),
 117                 .mode           = 0644,
 118                 .proc_handler   = proc_dointvec_minmax,
 119                 .extra1         = &min_slot_table_size,
 120                 .extra2         = &max_slot_table_size
 121         },
 122         {
 123                 .procname       = "tcp_max_slot_table_entries",
 124                 .data           = &xprt_max_tcp_slot_table_entries,
 125                 .maxlen         = sizeof(unsigned int),
 126                 .mode           = 0644,
 127                 .proc_handler   = proc_dointvec_minmax,
 128                 .extra1         = &min_slot_table_size,
 129                 .extra2         = &max_tcp_slot_table_limit
 130         },
 131         {
 132                 .procname       = "min_resvport",
 133                 .data           = &xprt_min_resvport,
 134                 .maxlen         = sizeof(unsigned int),
 135                 .mode           = 0644,
 136                 .proc_handler   = proc_dointvec_minmax,
 137                 .extra1         = &xprt_min_resvport_limit,
 138                 .extra2         = &xprt_max_resvport_limit
 139         },
 140         {
 141                 .procname       = "max_resvport",
 142                 .data           = &xprt_max_resvport,
 143                 .maxlen         = sizeof(unsigned int),
 144                 .mode           = 0644,
 145                 .proc_handler   = proc_dointvec_minmax,
 146                 .extra1         = &xprt_min_resvport_limit,
 147                 .extra2         = &xprt_max_resvport_limit
 148         },
 149         {
 150                 .procname       = "tcp_fin_timeout",
 151                 .data           = &xs_tcp_fin_timeout,
 152                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 153                 .mode           = 0644,
 154                 .proc_handler   = proc_dointvec_jiffies,
 155         },
 156         { },
 157 };
 158
 159 static struct ctl_table sunrpc_table[] = {
 160         {
 161                 .procname       = "sunrpc",
 162                 .mode           = 0555,
 163                 .child          = xs_tunables_table
 164         },
 165         { },
 166 };
 167
 168 /*
 169  * Wait duration for a reply from the RPC portmapper.
 170  */
 171 #define XS_BIND_TO              (60U * HZ)
 172
 173 /*
 174  * Delay if a UDP socket connect error occurs.  This is most likely some
 175  * kind of resource problem on the local host.
 176  */
 177 #define XS_UDP_REEST_TO         (2U * HZ)
 178
 179 /*
 180  * The reestablish timeout allows clients to delay for a bit before attempting
 181  * to reconnect to a server that just dropped our connection.
 182  *
 183  * We implement an exponential backoff when trying to reestablish a TCP
 184  * transport connection with the server.  Some servers like to drop a TCP
 185  * connection when they are overworked, so we start with a short timeout and
 186  * increase over time if the server is down or not responding.
 187  */
 188 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 189
 190 /*
 191  * TCP idle timeout; client drops the transport socket if it is idle
 192  * for this long.  Note that we also timeout UDP sockets to prevent
 193  * holding port numbers when there is no RPC traffic.
 194  */
 195 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 196
 197 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 198 # undef  RPC_DEBUG_DATA
 199 # define RPCDBG_FACILITY        RPCDBG_TRANS
 200 #endif
 201
 202 #ifdef RPC_DEBUG_DATA
 203 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 204 {
 205         u8 *buf = (u8 *) packet;
 206         int j;
 207
 208         dprintk("RPC:       %s\n", msg);
 209         for (j = 0; j < count && j < 128; j += 4) {
 210                 if (!(j & 31)) {
 211                         if (j)
 212                                 dprintk("\n");
 213                         dprintk("0x%04x ", j);
 214                 }
 215                 dprintk("%02x%02x%02x%02x ",
 216                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 217         }
 218         dprintk("\n");
 219 }
 220 #else
 221 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 222 {
 223         /* NOP */
 224 }
 225 #endif
 226
 227 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 228 {
 229         return (struct rpc_xprt *) sk->sk_user_data;
 230 }
 231
 232 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 233 {
 234         return (struct sockaddr *) &xprt->addr;
 235 }
 236
 237 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 238 {
 239         return (struct sockaddr_un *) &xprt->addr;
 240 }
 241
 242 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 243 {
 244         return (struct sockaddr_in *) &xprt->addr;
 245 }
 246
 247 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 248 {
 249         return (struct sockaddr_in6 *) &xprt->addr;
 250 }
 251
 252 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 253 {
 254         struct sockaddr *sap = xs_addr(xprt);
 255         struct sockaddr_in6 *sin6;
 256         struct sockaddr_in *sin;
 257         struct sockaddr_un *sun;
 258         char buf[128];
 259
 260         switch (sap->sa_family) {
 261         case AF_LOCAL:
 262                 sun = xs_addr_un(xprt);
 263                 strlcpy(buf, sun->sun_path, sizeof(buf));
 264                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 265                                                 kstrdup(buf, GFP_KERNEL);
 266                 break;
 267         case AF_INET:
 268                 (void)rpc_ntop(sap, buf, sizeof(buf));
 269                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 270                                                 kstrdup(buf, GFP_KERNEL);
 271                 sin = xs_addr_in(xprt);
 272                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 273                 break;
 274         case AF_INET6:
 275                 (void)rpc_ntop(sap, buf, sizeof(buf));
 276                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 277                                                 kstrdup(buf, GFP_KERNEL);
 278                 sin6 = xs_addr_in6(xprt);
 279                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 280                 break;
 281         default:
 282                 BUG();
 283         }
 284
 285         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 286 }
 287
 288 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 289 {
 290         struct sockaddr *sap = xs_addr(xprt);
 291         char buf[128];
 292
 293         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 294         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 295
 296         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 297         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 298 }
 299
 300 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 301                                      const char *protocol,
 302                                      const char *netid)
 303 {
 304         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 305         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 306         xs_format_common_peer_addresses(xprt);
 307         xs_format_common_peer_ports(xprt);
 308 }
 309
 310 static void xs_update_peer_port(struct rpc_xprt *xprt)
 311 {
 312         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 313         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 314
 315         xs_format_common_peer_ports(xprt);
 316 }
 317
 318 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 319 {
 320         unsigned int i;
 321
 322         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 323                 switch (i) {
 324                 case RPC_DISPLAY_PROTO:
 325                 case RPC_DISPLAY_NETID:
 326                         continue;
 327                 default:
 328                         kfree(xprt->address_strings[i]);
 329                 }
 330 }
 331
 332 static size_t
 333 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 334 {
 335         size_t i,n;
 336
 337         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 338                 return want;
 339         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 340         for (i = 0; i < n; i++) {
 341                 if (buf->pages[i])
 342                         continue;
 343                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 344                 if (!buf->pages[i]) {
 345                         i *= PAGE_SIZE;
 346                         return i > buf->page_base ? i - buf->page_base : 0;
 347                 }
 348         }
 349         return want;
 350 }
 351
 352 static ssize_t
 353 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 354 {
 355         ssize_t ret;
 356         if (seek != 0)
 357                 iov_iter_advance(&msg->msg_iter, seek);
 358         ret = sock_recvmsg(sock, msg, flags);
 359         return ret > 0 ? ret + seek : ret;
 360 }
 361
 362 static ssize_t
 363 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 364                 struct kvec *kvec, size_t count, size_t seek)
 365 {
 366         iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
 367         return xs_sock_recvmsg(sock, msg, flags, seek);
 368 }
 369
 370 static ssize_t
 371 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 372                 struct bio_vec *bvec, unsigned long nr, size_t count,
 373                 size_t seek)
 374 {
 375         iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
 376         return xs_sock_recvmsg(sock, msg, flags, seek);
 377 }
 378
 379 static ssize_t
 380 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 381                 size_t count)
 382 {
 383         iov_iter_discard(&msg->msg_iter, READ, count);
 384         return sock_recvmsg(sock, msg, flags);
 385 }
 386
 387 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 388 static void
 389 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 390 {
 391         struct bvec_iter bi = {
 392                 .bi_size = count,
 393         };
 394         struct bio_vec bv;
 395
 396         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 397         for_each_bvec(bv, bvec, bi, bi)
 398                 flush_dcache_page(bv.bv_page);
 399 }
 400 #else
 401 static inline void
 402 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 403 {
 404 }
 405 #endif
 406
 407 static ssize_t
 408 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 409                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 410 {
 411         size_t want, seek_init = seek, offset = 0;
 412         ssize_t ret;
 413
 414         want = min_t(size_t, count, buf->head[0].iov_len);
 415         if (seek < want) {
 416                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 417                 if (ret <= 0)
 418                         goto sock_err;
 419                 offset += ret;
 420                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 421                         goto out;
 422                 if (ret != want)
 423                         goto out;
 424                 seek = 0;
 425         } else {
 426                 seek -= want;
 427                 offset += want;
 428         }
 429
 430         want = xs_alloc_sparse_pages(buf,
 431                         min_t(size_t, count - offset, buf->page_len),
 432                         GFP_KERNEL);
 433         if (seek < want) {
 434                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 435                                 xdr_buf_pagecount(buf),
 436                                 want + buf->page_base,
 437                                 seek + buf->page_base);
 438                 if (ret <= 0)
 439                         goto sock_err;
 440                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 441                 ret -= buf->page_base;
 442                 offset += ret;
 443                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 444                         goto out;
 445                 if (ret != want)
 446                         goto out;
 447                 seek = 0;
 448         } else {
 449                 seek -= want;
 450                 offset += want;
 451         }
 452
 453         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 454         if (seek < want) {
 455                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 456                 if (ret <= 0)
 457                         goto sock_err;
 458                 offset += ret;
 459                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 460                         goto out;
 461                 if (ret != want)
 462                         goto out;
 463         } else if (offset < seek_init)
 464                 offset = seek_init;
 465         ret = -EMSGSIZE;
 466 out:
 467         *read = offset - seek_init;
 468         return ret;
 469 sock_err:
 470         offset += seek;
 471         goto out;
 472 }
 473
 474 static void
 475 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 476 {
 477         if (!transport->recv.copied) {
 478                 if (buf->head[0].iov_len >= transport->recv.offset)
 479                         memcpy(buf->head[0].iov_base,
 480                                         &transport->recv.xid,
 481                                         transport->recv.offset);
 482                 transport->recv.copied = transport->recv.offset;
 483         }
 484 }
 485
 486 static bool
 487 xs_read_stream_request_done(struct sock_xprt *transport)
 488 {
 489         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 490 }
 491
 492 static void
 493 xs_read_stream_check_eor(struct sock_xprt *transport,
 494                 struct msghdr *msg)
 495 {
 496         if (xs_read_stream_request_done(transport))
 497                 msg->msg_flags |= MSG_EOR;
 498 }
 499
 500 static ssize_t
 501 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 502                 int flags, struct rpc_rqst *req)
 503 {
 504         struct xdr_buf *buf = &req->rq_private_buf;
 505         size_t want, read;
 506         ssize_t ret;
 507
 508         xs_read_header(transport, buf);
 509
 510         want = transport->recv.len - transport->recv.offset;
 511         if (want != 0) {
 512                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 513                                 transport->recv.copied + want,
 514                                 transport->recv.copied,
 515                                 &read);
 516                 transport->recv.offset += read;
 517                 transport->recv.copied += read;
 518         }
 519
 520         if (transport->recv.offset == transport->recv.len)
 521                 xs_read_stream_check_eor(transport, msg);
 522
 523         if (want == 0)
 524                 return 0;
 525
 526         switch (ret) {
 527         default:
 528                 break;
 529         case -EFAULT:
 530         case -EMSGSIZE:
 531                 msg->msg_flags |= MSG_TRUNC;
 532                 return read;
 533         case 0:
 534                 return -ESHUTDOWN;
 535         }
 536         return ret < 0 ? ret : read;
 537 }
 538
 539 static size_t
 540 xs_read_stream_headersize(bool isfrag)
 541 {
 542         if (isfrag)
 543                 return sizeof(__be32);
 544         return 3 * sizeof(__be32);
 545 }
 546
 547 static ssize_t
 548 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 549                 int flags, size_t want, size_t seek)
 550 {
 551         struct kvec kvec = {
 552                 .iov_base = &transport->recv.fraghdr,
 553                 .iov_len = want,
 554         };
 555         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 556 }
 557
 558 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 559 static ssize_t
 560 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 561 {
 562         struct rpc_xprt *xprt = &transport->xprt;
 563         struct rpc_rqst *req;
 564         ssize_t ret;
 565
 566         /* Is this transport associated with the backchannel? */
 567         if (!xprt->bc_serv)
 568                 return -ESHUTDOWN;
 569
 570         /* Look up and lock the request corresponding to the given XID */
 571         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 572         if (!req) {
 573                 printk(KERN_WARNING "Callback slot table overflowed\n");
 574                 return -ESHUTDOWN;
 575         }
 576         if (transport->recv.copied && !req->rq_private_buf.len)
 577                 return -ESHUTDOWN;
 578
 579         ret = xs_read_stream_request(transport, msg, flags, req);
 580         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 581                 xprt_complete_bc_request(req, transport->recv.copied);
 582         else
 583                 req->rq_private_buf.len = transport->recv.copied;
 584
 585         return ret;
 586 }
 587 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 588 static ssize_t
 589 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 590 {
 591         return -ESHUTDOWN;
 592 }
 593 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 594
 595 static ssize_t
 596 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 597 {
 598         struct rpc_xprt *xprt = &transport->xprt;
 599         struct rpc_rqst *req;
 600         ssize_t ret = 0;
 601
 602         /* Look up and lock the request corresponding to the given XID */
 603         spin_lock(&xprt->queue_lock);
 604         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 605         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 606                 msg->msg_flags |= MSG_TRUNC;
 607                 goto out;
 608         }
 609         xprt_pin_rqst(req);
 610         spin_unlock(&xprt->queue_lock);
 611
 612         ret = xs_read_stream_request(transport, msg, flags, req);
 613
 614         spin_lock(&xprt->queue_lock);
 615         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 616                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 617         else
 618                 req->rq_private_buf.len = transport->recv.copied;
 619         xprt_unpin_rqst(req);
 620 out:
 621         spin_unlock(&xprt->queue_lock);
 622         return ret;
 623 }
 624
 625 static ssize_t
 626 xs_read_stream(struct sock_xprt *transport, int flags)
 627 {
 628         struct msghdr msg = { 0 };
 629         size_t want, read = 0;
 630         ssize_t ret = 0;
 631
 632         if (transport->recv.len == 0) {
 633                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 634                 ret = xs_read_stream_header(transport, &msg, flags, want,
 635                                 transport->recv.offset);
 636                 if (ret <= 0)
 637                         goto out_err;
 638                 transport->recv.offset = ret;
 639                 if (transport->recv.offset != want)
 640                         return transport->recv.offset;
 641                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 642                         RPC_FRAGMENT_SIZE_MASK;
 643                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 644                 read = ret;
 645         }
 646
 647         switch (be32_to_cpu(transport->recv.calldir)) {
 648         default:
 649                 msg.msg_flags |= MSG_TRUNC;
 650                 break;
 651         case RPC_CALL:
 652                 ret = xs_read_stream_call(transport, &msg, flags);
 653                 break;
 654         case RPC_REPLY:
 655                 ret = xs_read_stream_reply(transport, &msg, flags);
 656         }
 657         if (msg.msg_flags & MSG_TRUNC) {
 658                 transport->recv.calldir = cpu_to_be32(-1);
 659                 transport->recv.copied = -1;
 660         }
 661         if (ret < 0)
 662                 goto out_err;
 663         read += ret;
 664         if (transport->recv.offset < transport->recv.len) {
 665                 if (!(msg.msg_flags & MSG_TRUNC))
 666                         return read;
 667                 msg.msg_flags = 0;
 668                 ret = xs_read_discard(transport->sock, &msg, flags,
 669                                 transport->recv.len - transport->recv.offset);
 670                 if (ret <= 0)
 671                         goto out_err;
 672                 transport->recv.offset += ret;
 673                 read += ret;
 674                 if (transport->recv.offset != transport->recv.len)
 675                         return read;
 676         }
 677         if (xs_read_stream_request_done(transport)) {
 678                 trace_xs_stream_read_request(transport);
 679                 transport->recv.copied = 0;
 680         }
 681         transport->recv.offset = 0;
 682         transport->recv.len = 0;
 683         return read;
 684 out_err:
 685         return ret != 0 ? ret : -ESHUTDOWN;
 686 }
 687
 688 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 689 {
 690         return transport->sock->ops->poll(transport->file, transport->sock,
 691                         NULL);
 692 }
 693
 694 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 695 {
 696         __poll_t events = xs_poll_socket(transport);
 697
 698         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 699 }
 700
 701 static void xs_poll_check_readable(struct sock_xprt *transport)
 702 {
 703
 704         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 705         if (!xs_poll_socket_readable(transport))
 706                 return;
 707         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 708                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 709 }
 710
 711 static void xs_stream_data_receive(struct sock_xprt *transport)
 712 {
 713         size_t read = 0;
 714         ssize_t ret = 0;
 715
 716         mutex_lock(&transport->recv_mutex);
 717         if (transport->sock == NULL)
 718                 goto out;
 719         for (;;) {
 720                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 721                 if (ret < 0)
 722                         break;
 723                 read += ret;
 724                 cond_resched();
 725         }
 726         if (ret == -ESHUTDOWN)
 727                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 728         else
 729                 xs_poll_check_readable(transport);
 730 out:
 731         mutex_unlock(&transport->recv_mutex);
 732         trace_xs_stream_read_data(&transport->xprt, ret, read);
 733 }
 734
 735 static void xs_stream_data_receive_workfn(struct work_struct *work)
 736 {
 737         struct sock_xprt *transport =
 738                 container_of(work, struct sock_xprt, recv_worker);
 739         unsigned int pflags = memalloc_nofs_save();
 740
 741         xs_stream_data_receive(transport);
 742         memalloc_nofs_restore(pflags);
 743 }
 744
 745 static void
 746 xs_stream_reset_connect(struct sock_xprt *transport)
 747 {
 748         transport->recv.offset = 0;
 749         transport->recv.len = 0;
 750         transport->recv.copied = 0;
 751         transport->xmit.offset = 0;
 752 }
 753
 754 static void
 755 xs_stream_start_connect(struct sock_xprt *transport)
 756 {
 757         transport->xprt.stat.connect_count++;
 758         transport->xprt.stat.connect_start = jiffies;
 759 }
 760
 761 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 762
 763 /**
 764  * xs_nospace - handle transmit was incomplete
 765  * @req: pointer to RPC request
 766  * @transport: pointer to struct sock_xprt
 767  *
 768  */
 769 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 770 {
 771         struct rpc_xprt *xprt = &transport->xprt;
 772         struct sock *sk = transport->inet;
 773         int ret = -EAGAIN;
 774
 775         trace_rpc_socket_nospace(req, transport);
 776
 777         /* Protect against races with write_space */
 778         spin_lock(&xprt->transport_lock);
 779
 780         /* Don't race with disconnect */
 781         if (xprt_connected(xprt)) {
 782                 struct socket_wq *wq;
 783
 784                 rcu_read_lock();
 785                 wq = rcu_dereference(sk->sk_wq);
 786                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 787                 rcu_read_unlock();
 788
 789                 /* wait for more buffer space */
 790                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 791                 sk->sk_write_pending++;
 792                 xprt_wait_for_buffer_space(xprt);
 793         } else
 794                 ret = -ENOTCONN;
 795
 796         spin_unlock(&xprt->transport_lock);
 797         return ret;
 798 }
 799
 800 static int xs_sock_nospace(struct rpc_rqst *req)
 801 {
 802         struct sock_xprt *transport =
 803                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 804         struct sock *sk = transport->inet;
 805         int ret = -EAGAIN;
 806
 807         lock_sock(sk);
 808         if (!sock_writeable(sk))
 809                 ret = xs_nospace(req, transport);
 810         release_sock(sk);
 811         return ret;
 812 }
 813
 814 static int xs_stream_nospace(struct rpc_rqst *req)
 815 {
 816         struct sock_xprt *transport =
 817                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 818         struct sock *sk = transport->inet;
 819         int ret = -EAGAIN;
 820
 821         lock_sock(sk);
 822         if (!sk_stream_memory_free(sk))
 823                 ret = xs_nospace(req, transport);
 824         release_sock(sk);
 825         return ret;
 826 }
 827
 828 static void
 829 xs_stream_prepare_request(struct rpc_rqst *req)
 830 {
 831         xdr_free_bvec(&req->rq_rcv_buf);
 832         req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
 833 }
 834
 835 /*
 836  * Determine if the previous message in the stream was aborted before it
 837  * could complete transmission.
 838  */
 839 static bool
 840 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 841 {
 842         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 843 }
 844
 845 /*
 846  * Return the stream record marker field for a record of length < 2^31-1
 847  */
 848 static rpc_fraghdr
 849 xs_stream_record_marker(struct xdr_buf *xdr)
 850 {
 851         if (!xdr->len)
 852                 return 0;
 853         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 854 }
 855
 856 /**
 857  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 858  * @req: pointer to RPC request
 859  *
 860  * Return values:
 861  *        0:    The request has been sent
 862  *   EAGAIN:    The socket was blocked, please call again later to
 863  *              complete the request
 864  * ENOTCONN:    Caller needs to invoke connect logic then call again
 865  *    other:    Some other error occurred, the request was not sent
 866  */
 867 static int xs_local_send_request(struct rpc_rqst *req)
 868 {
 869         struct rpc_xprt *xprt = req->rq_xprt;
 870         struct sock_xprt *transport =
 871                                 container_of(xprt, struct sock_xprt, xprt);
 872         struct xdr_buf *xdr = &req->rq_snd_buf;
 873         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 874         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 875         struct msghdr msg = {
 876                 .msg_flags      = XS_SENDMSG_FLAGS,
 877         };
 878         unsigned int sent;
 879         int status;
 880
 881         /* Close the stream if the previous transmission was incomplete */
 882         if (xs_send_request_was_aborted(transport, req)) {
 883                 xs_close(xprt);
 884                 return -ENOTCONN;
 885         }
 886
 887         xs_pktdump("packet data:",
 888                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 889
 890         req->rq_xtime = ktime_get();
 891         status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 892                                    transport->xmit.offset, rm, &sent);
 893         dprintk("RPC:       %s(%u) = %d\n",
 894                         __func__, xdr->len - transport->xmit.offset, status);
 895
 896         if (status == -EAGAIN && sock_writeable(transport->inet))
 897                 status = -ENOBUFS;
 898
 899         if (likely(sent > 0) || status == 0) {
 900                 transport->xmit.offset += sent;
 901                 req->rq_bytes_sent = transport->xmit.offset;
 902                 if (likely(req->rq_bytes_sent >= msglen)) {
 903                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 904                         transport->xmit.offset = 0;
 905                         return 0;
 906                 }
 907                 status = -EAGAIN;
 908         }
 909
 910         switch (status) {
 911         case -ENOBUFS:
 912                 break;
 913         case -EAGAIN:
 914                 status = xs_stream_nospace(req);
 915                 break;
 916         default:
 917                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 918                         -status);
 919                 fallthrough;
 920         case -EPIPE:
 921                 xs_close(xprt);
 922                 status = -ENOTCONN;
 923         }
 924
 925         return status;
 926 }
 927
 928 /**
 929  * xs_udp_send_request - write an RPC request to a UDP socket
 930  * @req: pointer to RPC request
 931  *
 932  * Return values:
 933  *        0:    The request has been sent
 934  *   EAGAIN:    The socket was blocked, please call again later to
 935  *              complete the request
 936  * ENOTCONN:    Caller needs to invoke connect logic then call again
 937  *    other:    Some other error occurred, the request was not sent
 938  */
 939 static int xs_udp_send_request(struct rpc_rqst *req)
 940 {
 941         struct rpc_xprt *xprt = req->rq_xprt;
 942         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 943         struct xdr_buf *xdr = &req->rq_snd_buf;
 944         struct msghdr msg = {
 945                 .msg_name       = xs_addr(xprt),
 946                 .msg_namelen    = xprt->addrlen,
 947                 .msg_flags      = XS_SENDMSG_FLAGS,
 948         };
 949         unsigned int sent;
 950         int status;
 951
 952         xs_pktdump("packet data:",
 953                                 req->rq_svec->iov_base,
 954                                 req->rq_svec->iov_len);
 955
 956         if (!xprt_bound(xprt))
 957                 return -ENOTCONN;
 958
 959         if (!xprt_request_get_cong(xprt, req))
 960                 return -EBADSLT;
 961
 962         req->rq_xtime = ktime_get();
 963         status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
 964
 965         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 966                         xdr->len, status);
 967
 968         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 969         if (status == -EPERM)
 970                 goto process_status;
 971
 972         if (status == -EAGAIN && sock_writeable(transport->inet))
 973                 status = -ENOBUFS;
 974
 975         if (sent > 0 || status == 0) {
 976                 req->rq_xmit_bytes_sent += sent;
 977                 if (sent >= req->rq_slen)
 978                         return 0;
 979                 /* Still some bytes left; set up for a retry later. */
 980                 status = -EAGAIN;
 981         }
 982
 983 process_status:
 984         switch (status) {
 985         case -ENOTSOCK:
 986                 status = -ENOTCONN;
 987                 /* Should we call xs_close() here? */
 988                 break;
 989         case -EAGAIN:
 990                 status = xs_sock_nospace(req);
 991                 break;
 992         case -ENETUNREACH:
 993         case -ENOBUFS:
 994         case -EPIPE:
 995         case -ECONNREFUSED:
 996         case -EPERM:
 997                 /* When the server has died, an ICMP port unreachable message
 998                  * prompts ECONNREFUSED. */
 999                 break;
1000         default:
1001                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1002                         -status);
1003         }
1004
1005         return status;
1006 }
1007
1008 /**
1009  * xs_tcp_send_request - write an RPC request to a TCP socket
1010  * @req: pointer to RPC request
1011  *
1012  * Return values:
1013  *        0:    The request has been sent
1014  *   EAGAIN:    The socket was blocked, please call again later to
1015  *              complete the request
1016  * ENOTCONN:    Caller needs to invoke connect logic then call again
1017  *    other:    Some other error occurred, the request was not sent
1018  *
1019  * XXX: In the case of soft timeouts, should we eventually give up
1020  *      if sendmsg is not able to make progress?
1021  */
1022 static int xs_tcp_send_request(struct rpc_rqst *req)
1023 {
1024         struct rpc_xprt *xprt = req->rq_xprt;
1025         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1026         struct xdr_buf *xdr = &req->rq_snd_buf;
1027         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1028         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1029         struct msghdr msg = {
1030                 .msg_flags      = XS_SENDMSG_FLAGS,
1031         };
1032         bool vm_wait = false;
1033         unsigned int sent;
1034         int status;
1035
1036         /* Close the stream if the previous transmission was incomplete */
1037         if (xs_send_request_was_aborted(transport, req)) {
1038                 if (transport->sock != NULL)
1039                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1040                 return -ENOTCONN;
1041         }
1042         if (!transport->inet)
1043                 return -ENOTCONN;
1044
1045         xs_pktdump("packet data:",
1046                                 req->rq_svec->iov_base,
1047                                 req->rq_svec->iov_len);
1048
1049         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1050                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1051
1052         /* Continue transmitting the packet/record. We must be careful
1053          * to cope with writespace callbacks arriving _after_ we have
1054          * called sendmsg(). */
1055         req->rq_xtime = ktime_get();
1056         tcp_sock_set_cork(transport->inet, true);
1057         while (1) {
1058                 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1059                                            transport->xmit.offset, rm, &sent);
1060
1061                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1062                                 xdr->len - transport->xmit.offset, status);
1063
1064                 /* If we've sent the entire packet, immediately
1065                  * reset the count of bytes sent. */
1066                 transport->xmit.offset += sent;
1067                 req->rq_bytes_sent = transport->xmit.offset;
1068                 if (likely(req->rq_bytes_sent >= msglen)) {
1069                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1070                         transport->xmit.offset = 0;
1071                         if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1072                                 tcp_sock_set_cork(transport->inet, false);
1073                         return 0;
1074                 }
1075
1076                 WARN_ON_ONCE(sent == 0 && status == 0);
1077
1078                 if (status == -EAGAIN ) {
1079                         /*
1080                          * Return EAGAIN if we're sure we're hitting the
1081                          * socket send buffer limits.
1082                          */
1083                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1084                                 break;
1085                         /*
1086                          * Did we hit a memory allocation failure?
1087                          */
1088                         if (sent == 0) {
1089                                 status = -ENOBUFS;
1090                                 if (vm_wait)
1091                                         break;
1092                                 /* Retry, knowing now that we're below the
1093                                  * socket send buffer limit
1094                                  */
1095                                 vm_wait = true;
1096                         }
1097                         continue;
1098                 }
1099                 if (status < 0)
1100                         break;
1101                 vm_wait = false;
1102         }
1103
1104         switch (status) {
1105         case -ENOTSOCK:
1106                 status = -ENOTCONN;
1107                 /* Should we call xs_close() here? */
1108                 break;
1109         case -EAGAIN:
1110                 status = xs_stream_nospace(req);
1111                 break;
1112         case -ECONNRESET:
1113         case -ECONNREFUSED:
1114         case -ENOTCONN:
1115         case -EADDRINUSE:
1116         case -ENOBUFS:
1117         case -EPIPE:
1118                 break;
1119         default:
1120                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1121                         -status);
1122         }
1123
1124         return status;
1125 }
1126
1127 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1128 {
1129         transport->old_data_ready = sk->sk_data_ready;
1130         transport->old_state_change = sk->sk_state_change;
1131         transport->old_write_space = sk->sk_write_space;
1132         transport->old_error_report = sk->sk_error_report;
1133 }
1134
1135 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1136 {
1137         sk->sk_data_ready = transport->old_data_ready;
1138         sk->sk_state_change = transport->old_state_change;
1139         sk->sk_write_space = transport->old_write_space;
1140         sk->sk_error_report = transport->old_error_report;
1141 }
1142
1143 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1144 {
1145         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1146
1147         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1148         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1149         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1150         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1151 }
1152
1153 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1154 {
1155         set_bit(nr, &transport->sock_state);
1156         queue_work(xprtiod_workqueue, &transport->error_worker);
1157 }
1158
1159 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1160 {
1161         smp_mb__before_atomic();
1162         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1163         clear_bit(XPRT_CLOSING, &xprt->state);
1164         xs_sock_reset_state_flags(xprt);
1165         smp_mb__after_atomic();
1166 }
1167
1168 /**
1169  * xs_error_report - callback to handle TCP socket state errors
1170  * @sk: socket
1171  *
1172  * Note: we don't call sock_error() since there may be a rpc_task
1173  * using the socket, and so we don't want to clear sk->sk_err.
1174  */
1175 static void xs_error_report(struct sock *sk)
1176 {
1177         struct sock_xprt *transport;
1178         struct rpc_xprt *xprt;
1179
1180         read_lock_bh(&sk->sk_callback_lock);
1181         if (!(xprt = xprt_from_sock(sk)))
1182                 goto out;
1183
1184         transport = container_of(xprt, struct sock_xprt, xprt);
1185         transport->xprt_err = -sk->sk_err;
1186         if (transport->xprt_err == 0)
1187                 goto out;
1188         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1189                         xprt, -transport->xprt_err);
1190         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1191
1192         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1193         smp_mb__before_atomic();
1194         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1195  out:
1196         read_unlock_bh(&sk->sk_callback_lock);
1197 }
1198
1199 static void xs_reset_transport(struct sock_xprt *transport)
1200 {
1201         struct socket *sock = transport->sock;
1202         struct sock *sk = transport->inet;
1203         struct rpc_xprt *xprt = &transport->xprt;
1204         struct file *filp = transport->file;
1205
1206         if (sk == NULL)
1207                 return;
1208
1209         if (atomic_read(&transport->xprt.swapper))
1210                 sk_clear_memalloc(sk);
1211
1212         kernel_sock_shutdown(sock, SHUT_RDWR);
1213
1214         mutex_lock(&transport->recv_mutex);
1215         write_lock_bh(&sk->sk_callback_lock);
1216         transport->inet = NULL;
1217         transport->sock = NULL;
1218         transport->file = NULL;
1219
1220         sk->sk_user_data = NULL;
1221
1222         xs_restore_old_callbacks(transport, sk);
1223         xprt_clear_connected(xprt);
1224         write_unlock_bh(&sk->sk_callback_lock);
1225         xs_sock_reset_connection_flags(xprt);
1226         /* Reset stream record info */
1227         xs_stream_reset_connect(transport);
1228         mutex_unlock(&transport->recv_mutex);
1229
1230         trace_rpc_socket_close(xprt, sock);
1231         fput(filp);
1232
1233         xprt_disconnect_done(xprt);
1234 }
1235
1236 /**
1237  * xs_close - close a socket
1238  * @xprt: transport
1239  *
1240  * This is used when all requests are complete; ie, no DRC state remains
1241  * on the server we want to save.
1242  *
1243  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1244  * xs_reset_transport() zeroing the socket from underneath a writer.
1245  */
1246 static void xs_close(struct rpc_xprt *xprt)
1247 {
1248         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1249
1250         dprintk("RPC:       xs_close xprt %p\n", xprt);
1251
1252         xs_reset_transport(transport);
1253         xprt->reestablish_timeout = 0;
1254 }
1255
1256 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1257 {
1258         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1259                 xprt);
1260         xprt_disconnect_done(xprt);
1261 }
1262
1263 static void xs_xprt_free(struct rpc_xprt *xprt)
1264 {
1265         xs_free_peer_addresses(xprt);
1266         xprt_free(xprt);
1267 }
1268
1269 /**
1270  * xs_destroy - prepare to shutdown a transport
1271  * @xprt: doomed transport
1272  *
1273  */
1274 static void xs_destroy(struct rpc_xprt *xprt)
1275 {
1276         struct sock_xprt *transport = container_of(xprt,
1277                         struct sock_xprt, xprt);
1278         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1279
1280         cancel_delayed_work_sync(&transport->connect_worker);
1281         xs_close(xprt);
1282         cancel_work_sync(&transport->recv_worker);
1283         cancel_work_sync(&transport->error_worker);
1284         xs_xprt_free(xprt);
1285         module_put(THIS_MODULE);
1286 }
1287
1288 /**
1289  * xs_udp_data_read_skb - receive callback for UDP sockets
1290  * @xprt: transport
1291  * @sk: socket
1292  * @skb: skbuff
1293  *
1294  */
1295 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1296                 struct sock *sk,
1297                 struct sk_buff *skb)
1298 {
1299         struct rpc_task *task;
1300         struct rpc_rqst *rovr;
1301         int repsize, copied;
1302         u32 _xid;
1303         __be32 *xp;
1304
1305         repsize = skb->len;
1306         if (repsize < 4) {
1307                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1308                 return;
1309         }
1310
1311         /* Copy the XID from the skb... */
1312         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1313         if (xp == NULL)
1314                 return;
1315
1316         /* Look up and lock the request corresponding to the given XID */
1317         spin_lock(&xprt->queue_lock);
1318         rovr = xprt_lookup_rqst(xprt, *xp);
1319         if (!rovr)
1320                 goto out_unlock;
1321         xprt_pin_rqst(rovr);
1322         xprt_update_rtt(rovr->rq_task);
1323         spin_unlock(&xprt->queue_lock);
1324         task = rovr->rq_task;
1325
1326         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1327                 copied = repsize;
1328
1329         /* Suck it into the iovec, verify checksum if not done by hw. */
1330         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1331                 spin_lock(&xprt->queue_lock);
1332                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1333                 goto out_unpin;
1334         }
1335
1336
1337         spin_lock(&xprt->transport_lock);
1338         xprt_adjust_cwnd(xprt, task, copied);
1339         spin_unlock(&xprt->transport_lock);
1340         spin_lock(&xprt->queue_lock);
1341         xprt_complete_rqst(task, copied);
1342         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1343 out_unpin:
1344         xprt_unpin_rqst(rovr);
1345  out_unlock:
1346         spin_unlock(&xprt->queue_lock);
1347 }
1348
1349 static void xs_udp_data_receive(struct sock_xprt *transport)
1350 {
1351         struct sk_buff *skb;
1352         struct sock *sk;
1353         int err;
1354
1355         mutex_lock(&transport->recv_mutex);
1356         sk = transport->inet;
1357         if (sk == NULL)
1358                 goto out;
1359         for (;;) {
1360                 skb = skb_recv_udp(sk, 0, 1, &err);
1361                 if (skb == NULL)
1362                         break;
1363                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1364                 consume_skb(skb);
1365                 cond_resched();
1366         }
1367         xs_poll_check_readable(transport);
1368 out:
1369         mutex_unlock(&transport->recv_mutex);
1370 }
1371
1372 static void xs_udp_data_receive_workfn(struct work_struct *work)
1373 {
1374         struct sock_xprt *transport =
1375                 container_of(work, struct sock_xprt, recv_worker);
1376         unsigned int pflags = memalloc_nofs_save();
1377
1378         xs_udp_data_receive(transport);
1379         memalloc_nofs_restore(pflags);
1380 }
1381
1382 /**
1383  * xs_data_ready - "data ready" callback for UDP sockets
1384  * @sk: socket with data to read
1385  *
1386  */
1387 static void xs_data_ready(struct sock *sk)
1388 {
1389         struct rpc_xprt *xprt;
1390
1391         read_lock_bh(&sk->sk_callback_lock);
1392         dprintk("RPC:       xs_data_ready...\n");
1393         xprt = xprt_from_sock(sk);
1394         if (xprt != NULL) {
1395                 struct sock_xprt *transport = container_of(xprt,
1396                                 struct sock_xprt, xprt);
1397                 transport->old_data_ready(sk);
1398                 /* Any data means we had a useful conversation, so
1399                  * then we don't need to delay the next reconnect
1400                  */
1401                 if (xprt->reestablish_timeout)
1402                         xprt->reestablish_timeout = 0;
1403                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1404                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1405         }
1406         read_unlock_bh(&sk->sk_callback_lock);
1407 }
1408
1409 /*
1410  * Helper function to force a TCP close if the server is sending
1411  * junk and/or it has put us in CLOSE_WAIT
1412  */
1413 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1414 {
1415         xprt_force_disconnect(xprt);
1416 }
1417
1418 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1419 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1420 {
1421         return PAGE_SIZE;
1422 }
1423 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1424
1425 /**
1426  * xs_tcp_state_change - callback to handle TCP socket state changes
1427  * @sk: socket whose state has changed
1428  *
1429  */
1430 static void xs_tcp_state_change(struct sock *sk)
1431 {
1432         struct rpc_xprt *xprt;
1433         struct sock_xprt *transport;
1434
1435         read_lock_bh(&sk->sk_callback_lock);
1436         if (!(xprt = xprt_from_sock(sk)))
1437                 goto out;
1438         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1439         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1440                         sk->sk_state, xprt_connected(xprt),
1441                         sock_flag(sk, SOCK_DEAD),
1442                         sock_flag(sk, SOCK_ZAPPED),
1443                         sk->sk_shutdown);
1444
1445         transport = container_of(xprt, struct sock_xprt, xprt);
1446         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1447         switch (sk->sk_state) {
1448         case TCP_ESTABLISHED:
1449                 if (!xprt_test_and_set_connected(xprt)) {
1450                         xprt->connect_cookie++;
1451                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1452                         xprt_clear_connecting(xprt);
1453
1454                         xprt->stat.connect_count++;
1455                         xprt->stat.connect_time += (long)jiffies -
1456                                                    xprt->stat.connect_start;
1457                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1458                 }
1459                 break;
1460         case TCP_FIN_WAIT1:
1461                 /* The client initiated a shutdown of the socket */
1462                 xprt->connect_cookie++;
1463                 xprt->reestablish_timeout = 0;
1464                 set_bit(XPRT_CLOSING, &xprt->state);
1465                 smp_mb__before_atomic();
1466                 clear_bit(XPRT_CONNECTED, &xprt->state);
1467                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1468                 smp_mb__after_atomic();
1469                 break;
1470         case TCP_CLOSE_WAIT:
1471                 /* The server initiated a shutdown of the socket */
1472                 xprt->connect_cookie++;
1473                 clear_bit(XPRT_CONNECTED, &xprt->state);
1474                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1475                 fallthrough;
1476         case TCP_CLOSING:
1477                 /*
1478                  * If the server closed down the connection, make sure that
1479                  * we back off before reconnecting
1480                  */
1481                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1482                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1483                 break;
1484         case TCP_LAST_ACK:
1485                 set_bit(XPRT_CLOSING, &xprt->state);
1486                 smp_mb__before_atomic();
1487                 clear_bit(XPRT_CONNECTED, &xprt->state);
1488                 smp_mb__after_atomic();
1489                 break;
1490         case TCP_CLOSE:
1491                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1492                                         &transport->sock_state))
1493                         xprt_clear_connecting(xprt);
1494                 clear_bit(XPRT_CLOSING, &xprt->state);
1495                 /* Trigger the socket release */
1496                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1497         }
1498  out:
1499         read_unlock_bh(&sk->sk_callback_lock);
1500 }
1501
1502 static void xs_write_space(struct sock *sk)
1503 {
1504         struct socket_wq *wq;
1505         struct sock_xprt *transport;
1506         struct rpc_xprt *xprt;
1507
1508         if (!sk->sk_socket)
1509                 return;
1510         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1511
1512         if (unlikely(!(xprt = xprt_from_sock(sk))))
1513                 return;
1514         transport = container_of(xprt, struct sock_xprt, xprt);
1515         rcu_read_lock();
1516         wq = rcu_dereference(sk->sk_wq);
1517         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1518                 goto out;
1519
1520         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1521         sk->sk_write_pending--;
1522 out:
1523         rcu_read_unlock();
1524 }
1525
1526 /**
1527  * xs_udp_write_space - callback invoked when socket buffer space
1528  *                             becomes available
1529  * @sk: socket whose state has changed
1530  *
1531  * Called when more output buffer space is available for this socket.
1532  * We try not to wake our writers until they can make "significant"
1533  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1534  * with a bunch of small requests.
1535  */
1536 static void xs_udp_write_space(struct sock *sk)
1537 {
1538         read_lock_bh(&sk->sk_callback_lock);
1539
1540         /* from net/core/sock.c:sock_def_write_space */
1541         if (sock_writeable(sk))
1542                 xs_write_space(sk);
1543
1544         read_unlock_bh(&sk->sk_callback_lock);
1545 }
1546
1547 /**
1548  * xs_tcp_write_space - callback invoked when socket buffer space
1549  *                             becomes available
1550  * @sk: socket whose state has changed
1551  *
1552  * Called when more output buffer space is available for this socket.
1553  * We try not to wake our writers until they can make "significant"
1554  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1555  * with a bunch of small requests.
1556  */
1557 static void xs_tcp_write_space(struct sock *sk)
1558 {
1559         read_lock_bh(&sk->sk_callback_lock);
1560
1561         /* from net/core/stream.c:sk_stream_write_space */
1562         if (sk_stream_is_writeable(sk))
1563                 xs_write_space(sk);
1564
1565         read_unlock_bh(&sk->sk_callback_lock);
1566 }
1567
1568 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1569 {
1570         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1571         struct sock *sk = transport->inet;
1572
1573         if (transport->rcvsize) {
1574                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1575                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1576         }
1577         if (transport->sndsize) {
1578                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1579                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1580                 sk->sk_write_space(sk);
1581         }
1582 }
1583
1584 /**
1585  * xs_udp_set_buffer_size - set send and receive limits
1586  * @xprt: generic transport
1587  * @sndsize: requested size of send buffer, in bytes
1588  * @rcvsize: requested size of receive buffer, in bytes
1589  *
1590  * Set socket send and receive buffer size limits.
1591  */
1592 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1593 {
1594         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1595
1596         transport->sndsize = 0;
1597         if (sndsize)
1598                 transport->sndsize = sndsize + 1024;
1599         transport->rcvsize = 0;
1600         if (rcvsize)
1601                 transport->rcvsize = rcvsize + 1024;
1602
1603         xs_udp_do_set_buffer_size(xprt);
1604 }
1605
1606 /**
1607  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1608  * @xprt: controlling transport
1609  * @task: task that timed out
1610  *
1611  * Adjust the congestion window after a retransmit timeout has occurred.
1612  */
1613 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1614 {
1615         spin_lock(&xprt->transport_lock);
1616         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1617         spin_unlock(&xprt->transport_lock);
1618 }
1619
1620 static int xs_get_random_port(void)
1621 {
1622         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1623         unsigned short range;
1624         unsigned short rand;
1625
1626         if (max < min)
1627                 return -EADDRINUSE;
1628         range = max - min + 1;
1629         rand = (unsigned short) prandom_u32() % range;
1630         return rand + min;
1631 }
1632
1633 static unsigned short xs_sock_getport(struct socket *sock)
1634 {
1635         struct sockaddr_storage buf;
1636         unsigned short port = 0;
1637
1638         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1639                 goto out;
1640         switch (buf.ss_family) {
1641         case AF_INET6:
1642                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1643                 break;
1644         case AF_INET:
1645                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1646         }
1647 out:
1648         return port;
1649 }
1650
1651 /**
1652  * xs_set_port - reset the port number in the remote endpoint address
1653  * @xprt: generic transport
1654  * @port: new port number
1655  *
1656  */
1657 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1658 {
1659         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1660
1661         rpc_set_port(xs_addr(xprt), port);
1662         xs_update_peer_port(xprt);
1663 }
1664
1665 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1666 {
1667         if (transport->srcport == 0 && transport->xprt.reuseport)
1668                 transport->srcport = xs_sock_getport(sock);
1669 }
1670
1671 static int xs_get_srcport(struct sock_xprt *transport)
1672 {
1673         int port = transport->srcport;
1674
1675         if (port == 0 && transport->xprt.resvport)
1676                 port = xs_get_random_port();
1677         return port;
1678 }
1679
1680 unsigned short get_srcport(struct rpc_xprt *xprt)
1681 {
1682         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1683         return xs_sock_getport(sock->sock);
1684 }
1685 EXPORT_SYMBOL(get_srcport);
1686
1687 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1688 {
1689         if (transport->srcport != 0)
1690                 transport->srcport = 0;
1691         if (!transport->xprt.resvport)
1692                 return 0;
1693         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1694                 return xprt_max_resvport;
1695         return --port;
1696 }
1697 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1698 {
1699         struct sockaddr_storage myaddr;
1700         int err, nloop = 0;
1701         int port = xs_get_srcport(transport);
1702         unsigned short last;
1703
1704         /*
1705          * If we are asking for any ephemeral port (i.e. port == 0 &&
1706          * transport->xprt.resvport == 0), don't bind.  Let the local
1707          * port selection happen implicitly when the socket is used
1708          * (for example at connect time).
1709          *
1710          * This ensures that we can continue to establish TCP
1711          * connections even when all local ephemeral ports are already
1712          * a part of some TCP connection.  This makes no difference
1713          * for UDP sockets, but also doesn't harm them.
1714          *
1715          * If we're asking for any reserved port (i.e. port == 0 &&
1716          * transport->xprt.resvport == 1) xs_get_srcport above will
1717          * ensure that port is non-zero and we will bind as needed.
1718          */
1719         if (port <= 0)
1720                 return port;
1721
1722         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1723         do {
1724                 rpc_set_port((struct sockaddr *)&myaddr, port);
1725                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1726                                 transport->xprt.addrlen);
1727                 if (err == 0) {
1728                         if (transport->xprt.reuseport)
1729                                 transport->srcport = port;
1730                         break;
1731                 }
1732                 last = port;
1733                 port = xs_next_srcport(transport, port);
1734                 if (port > last)
1735                         nloop++;
1736         } while (err == -EADDRINUSE && nloop != 2);
1737
1738         if (myaddr.ss_family == AF_INET)
1739                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1740                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1741                                 port, err ? "failed" : "ok", err);
1742         else
1743                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1744                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1745                                 port, err ? "failed" : "ok", err);
1746         return err;
1747 }
1748
1749 /*
1750  * We don't support autobind on AF_LOCAL sockets
1751  */
1752 static void xs_local_rpcbind(struct rpc_task *task)
1753 {
1754         xprt_set_bound(task->tk_xprt);
1755 }
1756
1757 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1758 {
1759 }
1760
1761 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1762 static struct lock_class_key xs_key[2];
1763 static struct lock_class_key xs_slock_key[2];
1764
1765 static inline void xs_reclassify_socketu(struct socket *sock)
1766 {
1767         struct sock *sk = sock->sk;
1768
1769         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1770                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1771 }
1772
1773 static inline void xs_reclassify_socket4(struct socket *sock)
1774 {
1775         struct sock *sk = sock->sk;
1776
1777         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1778                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1779 }
1780
1781 static inline void xs_reclassify_socket6(struct socket *sock)
1782 {
1783         struct sock *sk = sock->sk;
1784
1785         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1786                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1787 }
1788
1789 static inline void xs_reclassify_socket(int family, struct socket *sock)
1790 {
1791         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1792                 return;
1793
1794         switch (family) {
1795         case AF_LOCAL:
1796                 xs_reclassify_socketu(sock);
1797                 break;
1798         case AF_INET:
1799                 xs_reclassify_socket4(sock);
1800                 break;
1801         case AF_INET6:
1802                 xs_reclassify_socket6(sock);
1803                 break;
1804         }
1805 }
1806 #else
1807 static inline void xs_reclassify_socket(int family, struct socket *sock)
1808 {
1809 }
1810 #endif
1811
1812 static void xs_dummy_setup_socket(struct work_struct *work)
1813 {
1814 }
1815
1816 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1817                 struct sock_xprt *transport, int family, int type,
1818                 int protocol, bool reuseport)
1819 {
1820         struct file *filp;
1821         struct socket *sock;
1822         int err;
1823
1824         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1825         if (err < 0) {
1826                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1827                                 protocol, -err);
1828                 goto out;
1829         }
1830         xs_reclassify_socket(family, sock);
1831
1832         if (reuseport)
1833                 sock_set_reuseport(sock->sk);
1834
1835         err = xs_bind(transport, sock);
1836         if (err) {
1837                 sock_release(sock);
1838                 goto out;
1839         }
1840
1841         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1842         if (IS_ERR(filp))
1843                 return ERR_CAST(filp);
1844         transport->file = filp;
1845
1846         return sock;
1847 out:
1848         return ERR_PTR(err);
1849 }
1850
1851 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1852                                       struct socket *sock)
1853 {
1854         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1855                                                                         xprt);
1856
1857         if (!transport->inet) {
1858                 struct sock *sk = sock->sk;
1859
1860                 write_lock_bh(&sk->sk_callback_lock);
1861
1862                 xs_save_old_callbacks(transport, sk);
1863
1864                 sk->sk_user_data = xprt;
1865                 sk->sk_data_ready = xs_data_ready;
1866                 sk->sk_write_space = xs_udp_write_space;
1867                 sock_set_flag(sk, SOCK_FASYNC);
1868                 sk->sk_error_report = xs_error_report;
1869
1870                 xprt_clear_connected(xprt);
1871
1872                 /* Reset to new socket */
1873                 transport->sock = sock;
1874                 transport->inet = sk;
1875
1876                 write_unlock_bh(&sk->sk_callback_lock);
1877         }
1878
1879         xs_stream_start_connect(transport);
1880
1881         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1882 }
1883
1884 /**
1885  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1886  * @transport: socket transport to connect
1887  */
1888 static int xs_local_setup_socket(struct sock_xprt *transport)
1889 {
1890         struct rpc_xprt *xprt = &transport->xprt;
1891         struct file *filp;
1892         struct socket *sock;
1893         int status;
1894
1895         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1896                                         SOCK_STREAM, 0, &sock, 1);
1897         if (status < 0) {
1898                 dprintk("RPC:       can't create AF_LOCAL "
1899                         "transport socket (%d).\n", -status);
1900                 goto out;
1901         }
1902         xs_reclassify_socket(AF_LOCAL, sock);
1903
1904         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1905         if (IS_ERR(filp)) {
1906                 status = PTR_ERR(filp);
1907                 goto out;
1908         }
1909         transport->file = filp;
1910
1911         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1912                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1913
1914         status = xs_local_finish_connecting(xprt, sock);
1915         trace_rpc_socket_connect(xprt, sock, status);
1916         switch (status) {
1917         case 0:
1918                 dprintk("RPC:       xprt %p connected to %s\n",
1919                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1920                 xprt->stat.connect_count++;
1921                 xprt->stat.connect_time += (long)jiffies -
1922                                            xprt->stat.connect_start;
1923                 xprt_set_connected(xprt);
1924                 break;
1925         case -ENOBUFS:
1926                 break;
1927         case -ENOENT:
1928                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1929                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1930                 break;
1931         case -ECONNREFUSED:
1932                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1933                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1934                 break;
1935         default:
1936                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1937                                 __func__, -status,
1938                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1939         }
1940
1941 out:
1942         xprt_clear_connecting(xprt);
1943         xprt_wake_pending_tasks(xprt, status);
1944         return status;
1945 }
1946
1947 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1948 {
1949         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1950         int ret;
1951
1952          if (RPC_IS_ASYNC(task)) {
1953                 /*
1954                  * We want the AF_LOCAL connect to be resolved in the
1955                  * filesystem namespace of the process making the rpc
1956                  * call.  Thus we connect synchronously.
1957                  *
1958                  * If we want to support asynchronous AF_LOCAL calls,
1959                  * we'll need to figure out how to pass a namespace to
1960                  * connect.
1961                  */
1962                 task->tk_rpc_status = -ENOTCONN;
1963                 rpc_exit(task, -ENOTCONN);
1964                 return;
1965         }
1966         ret = xs_local_setup_socket(transport);
1967         if (ret && !RPC_IS_SOFTCONN(task))
1968                 msleep_interruptible(15000);
1969 }
1970
1971 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1972 /*
1973  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1974  * know that we have exclusive access to the socket), to guard against
1975  * races with xs_reset_transport.
1976  */
1977 static void xs_set_memalloc(struct rpc_xprt *xprt)
1978 {
1979         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1980                         xprt);
1981
1982         /*
1983          * If there's no sock, then we have nothing to set. The
1984          * reconnecting process will get it for us.
1985          */
1986         if (!transport->inet)
1987                 return;
1988         if (atomic_read(&xprt->swapper))
1989                 sk_set_memalloc(transport->inet);
1990 }
1991
1992 /**
1993  * xs_enable_swap - Tag this transport as being used for swap.
1994  * @xprt: transport to tag
1995  *
1996  * Take a reference to this transport on behalf of the rpc_clnt, and
1997  * optionally mark it for swapping if it wasn't already.
1998  */
1999 static int
2000 xs_enable_swap(struct rpc_xprt *xprt)
2001 {
2002         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2003
2004         if (atomic_inc_return(&xprt->swapper) != 1)
2005                 return 0;
2006         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2007                 return -ERESTARTSYS;
2008         if (xs->inet)
2009                 sk_set_memalloc(xs->inet);
2010         xprt_release_xprt(xprt, NULL);
2011         return 0;
2012 }
2013
2014 /**
2015  * xs_disable_swap - Untag this transport as being used for swap.
2016  * @xprt: transport to tag
2017  *
2018  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2019  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2020  */
2021 static void
2022 xs_disable_swap(struct rpc_xprt *xprt)
2023 {
2024         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2025
2026         if (!atomic_dec_and_test(&xprt->swapper))
2027                 return;
2028         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2029                 return;
2030         if (xs->inet)
2031                 sk_clear_memalloc(xs->inet);
2032         xprt_release_xprt(xprt, NULL);
2033 }
2034 #else
2035 static void xs_set_memalloc(struct rpc_xprt *xprt)
2036 {
2037 }
2038
2039 static int
2040 xs_enable_swap(struct rpc_xprt *xprt)
2041 {
2042         return -EINVAL;
2043 }
2044
2045 static void
2046 xs_disable_swap(struct rpc_xprt *xprt)
2047 {
2048 }
2049 #endif
2050
2051 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2052 {
2053         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2054
2055         if (!transport->inet) {
2056                 struct sock *sk = sock->sk;
2057
2058                 write_lock_bh(&sk->sk_callback_lock);
2059
2060                 xs_save_old_callbacks(transport, sk);
2061
2062                 sk->sk_user_data = xprt;
2063                 sk->sk_data_ready = xs_data_ready;
2064                 sk->sk_write_space = xs_udp_write_space;
2065                 sock_set_flag(sk, SOCK_FASYNC);
2066
2067                 xprt_set_connected(xprt);
2068
2069                 /* Reset to new socket */
2070                 transport->sock = sock;
2071                 transport->inet = sk;
2072
2073                 xs_set_memalloc(xprt);
2074
2075                 write_unlock_bh(&sk->sk_callback_lock);
2076         }
2077         xs_udp_do_set_buffer_size(xprt);
2078
2079         xprt->stat.connect_start = jiffies;
2080 }
2081
2082 static void xs_udp_setup_socket(struct work_struct *work)
2083 {
2084         struct sock_xprt *transport =
2085                 container_of(work, struct sock_xprt, connect_worker.work);
2086         struct rpc_xprt *xprt = &transport->xprt;
2087         struct socket *sock;
2088         int status = -EIO;
2089
2090         sock = xs_create_sock(xprt, transport,
2091                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2092                         IPPROTO_UDP, false);
2093         if (IS_ERR(sock))
2094                 goto out;
2095
2096         dprintk("RPC:       worker connecting xprt %p via %s to "
2097                                 "%s (port %s)\n", xprt,
2098                         xprt->address_strings[RPC_DISPLAY_PROTO],
2099                         xprt->address_strings[RPC_DISPLAY_ADDR],
2100                         xprt->address_strings[RPC_DISPLAY_PORT]);
2101
2102         xs_udp_finish_connecting(xprt, sock);
2103         trace_rpc_socket_connect(xprt, sock, 0);
2104         status = 0;
2105 out:
2106         xprt_clear_connecting(xprt);
2107         xprt_unlock_connect(xprt, transport);
2108         xprt_wake_pending_tasks(xprt, status);
2109 }
2110
2111 /**
2112  * xs_tcp_shutdown - gracefully shut down a TCP socket
2113  * @xprt: transport
2114  *
2115  * Initiates a graceful shutdown of the TCP socket by calling the
2116  * equivalent of shutdown(SHUT_RDWR);
2117  */
2118 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2119 {
2120         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2121         struct socket *sock = transport->sock;
2122         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2123
2124         if (sock == NULL)
2125                 return;
2126         if (!xprt->reuseport) {
2127                 xs_close(xprt);
2128                 return;
2129         }
2130         switch (skst) {
2131         case TCP_FIN_WAIT1:
2132         case TCP_FIN_WAIT2:
2133                 break;
2134         case TCP_ESTABLISHED:
2135         case TCP_CLOSE_WAIT:
2136                 kernel_sock_shutdown(sock, SHUT_RDWR);
2137                 trace_rpc_socket_shutdown(xprt, sock);
2138                 break;
2139         default:
2140                 xs_reset_transport(transport);
2141         }
2142 }
2143
2144 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2145                 struct socket *sock)
2146 {
2147         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2148         unsigned int keepidle;
2149         unsigned int keepcnt;
2150         unsigned int timeo;
2151
2152         spin_lock(&xprt->transport_lock);
2153         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2154         keepcnt = xprt->timeout->to_retries + 1;
2155         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2156                 (xprt->timeout->to_retries + 1);
2157         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2158         spin_unlock(&xprt->transport_lock);
2159
2160         /* TCP Keepalive options */
2161         sock_set_keepalive(sock->sk);
2162         tcp_sock_set_keepidle(sock->sk, keepidle);
2163         tcp_sock_set_keepintvl(sock->sk, keepidle);
2164         tcp_sock_set_keepcnt(sock->sk, keepcnt);
2165
2166         /* TCP user timeout (see RFC5482) */
2167         tcp_sock_set_user_timeout(sock->sk, timeo);
2168 }
2169
2170 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2171                 unsigned long connect_timeout,
2172                 unsigned long reconnect_timeout)
2173 {
2174         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2175         struct rpc_timeout to;
2176         unsigned long initval;
2177
2178         spin_lock(&xprt->transport_lock);
2179         if (reconnect_timeout < xprt->max_reconnect_timeout)
2180                 xprt->max_reconnect_timeout = reconnect_timeout;
2181         if (connect_timeout < xprt->connect_timeout) {
2182                 memcpy(&to, xprt->timeout, sizeof(to));
2183                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2184                 /* Arbitrary lower limit */
2185                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2186                         initval = XS_TCP_INIT_REEST_TO << 1;
2187                 to.to_initval = initval;
2188                 to.to_maxval = initval;
2189                 memcpy(&transport->tcp_timeout, &to,
2190                                 sizeof(transport->tcp_timeout));
2191                 xprt->timeout = &transport->tcp_timeout;
2192                 xprt->connect_timeout = connect_timeout;
2193         }
2194         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2195         spin_unlock(&xprt->transport_lock);
2196 }
2197
2198 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2199 {
2200         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2201         int ret = -ENOTCONN;
2202
2203         if (!transport->inet) {
2204                 struct sock *sk = sock->sk;
2205
2206                 /* Avoid temporary address, they are bad for long-lived
2207                  * connections such as NFS mounts.
2208                  * RFC4941, section 3.6 suggests that:
2209                  *    Individual applications, which have specific
2210                  *    knowledge about the normal duration of connections,
2211                  *    MAY override this as appropriate.
2212                  */
2213                 if (xs_addr(xprt)->sa_family == PF_INET6) {
2214                         ip6_sock_set_addr_preferences(sk,
2215                                 IPV6_PREFER_SRC_PUBLIC);
2216                 }
2217
2218                 xs_tcp_set_socket_timeouts(xprt, sock);
2219                 tcp_sock_set_nodelay(sk);
2220
2221                 write_lock_bh(&sk->sk_callback_lock);
2222
2223                 xs_save_old_callbacks(transport, sk);
2224
2225                 sk->sk_user_data = xprt;
2226                 sk->sk_data_ready = xs_data_ready;
2227                 sk->sk_state_change = xs_tcp_state_change;
2228                 sk->sk_write_space = xs_tcp_write_space;
2229                 sock_set_flag(sk, SOCK_FASYNC);
2230                 sk->sk_error_report = xs_error_report;
2231
2232                 /* socket options */
2233                 sock_reset_flag(sk, SOCK_LINGER);
2234
2235                 xprt_clear_connected(xprt);
2236
2237                 /* Reset to new socket */
2238                 transport->sock = sock;
2239                 transport->inet = sk;
2240
2241                 write_unlock_bh(&sk->sk_callback_lock);
2242         }
2243
2244         if (!xprt_bound(xprt))
2245                 goto out;
2246
2247         xs_set_memalloc(xprt);
2248
2249         xs_stream_start_connect(transport);
2250
2251         /* Tell the socket layer to start connecting... */
2252         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2253         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2254         switch (ret) {
2255         case 0:
2256                 xs_set_srcport(transport, sock);
2257                 fallthrough;
2258         case -EINPROGRESS:
2259                 /* SYN_SENT! */
2260                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2261                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2262                 break;
2263         case -EADDRNOTAVAIL:
2264                 /* Source port number is unavailable. Try a new one! */
2265                 transport->srcport = 0;
2266         }
2267 out:
2268         return ret;
2269 }
2270
2271 /**
2272  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2273  * @work: queued work item
2274  *
2275  * Invoked by a work queue tasklet.
2276  */
2277 static void xs_tcp_setup_socket(struct work_struct *work)
2278 {
2279         struct sock_xprt *transport =
2280                 container_of(work, struct sock_xprt, connect_worker.work);
2281         struct socket *sock = transport->sock;
2282         struct rpc_xprt *xprt = &transport->xprt;
2283         int status = -EIO;
2284
2285         if (!sock) {
2286                 sock = xs_create_sock(xprt, transport,
2287                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2288                                 IPPROTO_TCP, true);
2289                 if (IS_ERR(sock)) {
2290                         status = PTR_ERR(sock);
2291                         goto out;
2292                 }
2293         }
2294
2295         dprintk("RPC:       worker connecting xprt %p via %s to "
2296                                 "%s (port %s)\n", xprt,
2297                         xprt->address_strings[RPC_DISPLAY_PROTO],
2298                         xprt->address_strings[RPC_DISPLAY_ADDR],
2299                         xprt->address_strings[RPC_DISPLAY_PORT]);
2300
2301         status = xs_tcp_finish_connecting(xprt, sock);
2302         trace_rpc_socket_connect(xprt, sock, status);
2303         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2304                         xprt, -status, xprt_connected(xprt),
2305                         sock->sk->sk_state);
2306         switch (status) {
2307         default:
2308                 printk("%s: connect returned unhandled error %d\n",
2309                         __func__, status);
2310                 fallthrough;
2311         case -EADDRNOTAVAIL:
2312                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2313                  * and retry
2314                  */
2315                 xs_tcp_force_close(xprt);
2316                 break;
2317         case 0:
2318         case -EINPROGRESS:
2319         case -EALREADY:
2320                 xprt_unlock_connect(xprt, transport);
2321                 return;
2322         case -EINVAL:
2323                 /* Happens, for instance, if the user specified a link
2324                  * local IPv6 address without a scope-id.
2325                  */
2326         case -ECONNREFUSED:
2327         case -ECONNRESET:
2328         case -ENETDOWN:
2329         case -ENETUNREACH:
2330         case -EHOSTUNREACH:
2331         case -EADDRINUSE:
2332         case -ENOBUFS:
2333                 /* xs_tcp_force_close() wakes tasks with a fixed error code.
2334                  * We need to wake them first to ensure the correct error code.
2335                  */
2336                 xprt_wake_pending_tasks(xprt, status);
2337                 xs_tcp_force_close(xprt);
2338                 goto out;
2339         }
2340         status = -EAGAIN;
2341 out:
2342         xprt_clear_connecting(xprt);
2343         xprt_unlock_connect(xprt, transport);
2344         xprt_wake_pending_tasks(xprt, status);
2345 }
2346
2347 /**
2348  * xs_connect - connect a socket to a remote endpoint
2349  * @xprt: pointer to transport structure
2350  * @task: address of RPC task that manages state of connect request
2351  *
2352  * TCP: If the remote end dropped the connection, delay reconnecting.
2353  *
2354  * UDP socket connects are synchronous, but we use a work queue anyway
2355  * to guarantee that even unprivileged user processes can set up a
2356  * socket on a privileged port.
2357  *
2358  * If a UDP socket connect fails, the delay behavior here prevents
2359  * retry floods (hard mounts).
2360  */
2361 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2362 {
2363         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2364         unsigned long delay = 0;
2365
2366         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2367
2368         if (transport->sock != NULL && !xprt_connecting(xprt)) {
2369                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2370                                 "seconds\n",
2371                                 xprt, xprt->reestablish_timeout / HZ);
2372
2373                 /* Start by resetting any existing state */
2374                 xs_reset_transport(transport);
2375
2376                 delay = xprt_reconnect_delay(xprt);
2377                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2378
2379         } else
2380                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2381
2382         queue_delayed_work(xprtiod_workqueue,
2383                         &transport->connect_worker,
2384                         delay);
2385 }
2386
2387 static void xs_wake_disconnect(struct sock_xprt *transport)
2388 {
2389         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2390                 xs_tcp_force_close(&transport->xprt);
2391 }
2392
2393 static void xs_wake_write(struct sock_xprt *transport)
2394 {
2395         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2396                 xprt_write_space(&transport->xprt);
2397 }
2398
2399 static void xs_wake_error(struct sock_xprt *transport)
2400 {
2401         int sockerr;
2402
2403         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2404                 return;
2405         mutex_lock(&transport->recv_mutex);
2406         if (transport->sock == NULL)
2407                 goto out;
2408         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2409                 goto out;
2410         sockerr = xchg(&transport->xprt_err, 0);
2411         if (sockerr < 0)
2412                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2413 out:
2414         mutex_unlock(&transport->recv_mutex);
2415 }
2416
2417 static void xs_wake_pending(struct sock_xprt *transport)
2418 {
2419         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2420                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2421 }
2422
2423 static void xs_error_handle(struct work_struct *work)
2424 {
2425         struct sock_xprt *transport = container_of(work,
2426                         struct sock_xprt, error_worker);
2427
2428         xs_wake_disconnect(transport);
2429         xs_wake_write(transport);
2430         xs_wake_error(transport);
2431         xs_wake_pending(transport);
2432 }
2433
2434 /**
2435  * xs_local_print_stats - display AF_LOCAL socket-specific stats
2436  * @xprt: rpc_xprt struct containing statistics
2437  * @seq: output file
2438  *
2439  */
2440 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2441 {
2442         long idle_time = 0;
2443
2444         if (xprt_connected(xprt))
2445                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2446
2447         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2448                         "%llu %llu %lu %llu %llu\n",
2449                         xprt->stat.bind_count,
2450                         xprt->stat.connect_count,
2451                         xprt->stat.connect_time / HZ,
2452                         idle_time,
2453                         xprt->stat.sends,
2454                         xprt->stat.recvs,
2455                         xprt->stat.bad_xids,
2456                         xprt->stat.req_u,
2457                         xprt->stat.bklog_u,
2458                         xprt->stat.max_slots,
2459                         xprt->stat.sending_u,
2460                         xprt->stat.pending_u);
2461 }
2462
2463 /**
2464  * xs_udp_print_stats - display UDP socket-specific stats
2465  * @xprt: rpc_xprt struct containing statistics
2466  * @seq: output file
2467  *
2468  */
2469 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2470 {
2471         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2472
2473         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2474                         "%lu %llu %llu\n",
2475                         transport->srcport,
2476                         xprt->stat.bind_count,
2477                         xprt->stat.sends,
2478                         xprt->stat.recvs,
2479                         xprt->stat.bad_xids,
2480                         xprt->stat.req_u,
2481                         xprt->stat.bklog_u,
2482                         xprt->stat.max_slots,
2483                         xprt->stat.sending_u,
2484                         xprt->stat.pending_u);
2485 }
2486
2487 /**
2488  * xs_tcp_print_stats - display TCP socket-specific stats
2489  * @xprt: rpc_xprt struct containing statistics
2490  * @seq: output file
2491  *
2492  */
2493 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2494 {
2495         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2496         long idle_time = 0;
2497
2498         if (xprt_connected(xprt))
2499                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2500
2501         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2502                         "%llu %llu %lu %llu %llu\n",
2503                         transport->srcport,
2504                         xprt->stat.bind_count,
2505                         xprt->stat.connect_count,
2506                         xprt->stat.connect_time / HZ,
2507                         idle_time,
2508                         xprt->stat.sends,
2509                         xprt->stat.recvs,
2510                         xprt->stat.bad_xids,
2511                         xprt->stat.req_u,
2512                         xprt->stat.bklog_u,
2513                         xprt->stat.max_slots,
2514                         xprt->stat.sending_u,
2515                         xprt->stat.pending_u);
2516 }
2517
2518 /*
2519  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2520  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2521  * to use the server side send routines.
2522  */
2523 static int bc_malloc(struct rpc_task *task)
2524 {
2525         struct rpc_rqst *rqst = task->tk_rqstp;
2526         size_t size = rqst->rq_callsize;
2527         struct page *page;
2528         struct rpc_buffer *buf;
2529
2530         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2531                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2532                           size);
2533                 return -EINVAL;
2534         }
2535
2536         page = alloc_page(GFP_KERNEL);
2537         if (!page)
2538                 return -ENOMEM;
2539
2540         buf = page_address(page);
2541         buf->len = PAGE_SIZE;
2542
2543         rqst->rq_buffer = buf->data;
2544         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2545         return 0;
2546 }
2547
2548 /*
2549  * Free the space allocated in the bc_alloc routine
2550  */
2551 static void bc_free(struct rpc_task *task)
2552 {
2553         void *buffer = task->tk_rqstp->rq_buffer;
2554         struct rpc_buffer *buf;
2555
2556         buf = container_of(buffer, struct rpc_buffer, data);
2557         free_page((unsigned long)buf);
2558 }
2559
2560 static int bc_sendto(struct rpc_rqst *req)
2561 {
2562         struct xdr_buf *xdr = &req->rq_snd_buf;
2563         struct sock_xprt *transport =
2564                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2565         struct msghdr msg = {
2566                 .msg_flags      = 0,
2567         };
2568         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2569                                          (u32)xdr->len);
2570         unsigned int sent = 0;
2571         int err;
2572
2573         req->rq_xtime = ktime_get();
2574         err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2575         xdr_free_bvec(xdr);
2576         if (err < 0 || sent != (xdr->len + sizeof(marker)))
2577                 return -EAGAIN;
2578         return sent;
2579 }
2580
2581 /**
2582  * bc_send_request - Send a backchannel Call on a TCP socket
2583  * @req: rpc_rqst containing Call message to be sent
2584  *
2585  * xpt_mutex ensures @rqstp's whole message is written to the socket
2586  * without interruption.
2587  *
2588  * Return values:
2589  *   %0 if the message was sent successfully
2590  *   %ENOTCONN if the message was not sent
2591  */
2592 static int bc_send_request(struct rpc_rqst *req)
2593 {
2594         struct svc_xprt *xprt;
2595         int len;
2596
2597         /*
2598          * Get the server socket associated with this callback xprt
2599          */
2600         xprt = req->rq_xprt->bc_xprt;
2601
2602         /*
2603          * Grab the mutex to serialize data as the connection is shared
2604          * with the fore channel
2605          */
2606         mutex_lock(&xprt->xpt_mutex);
2607         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2608                 len = -ENOTCONN;
2609         else
2610                 len = bc_sendto(req);
2611         mutex_unlock(&xprt->xpt_mutex);
2612
2613         if (len > 0)
2614                 len = 0;
2615
2616         return len;
2617 }
2618
2619 /*
2620  * The close routine. Since this is client initiated, we do nothing
2621  */
2622
2623 static void bc_close(struct rpc_xprt *xprt)
2624 {
2625         xprt_disconnect_done(xprt);
2626 }
2627
2628 /*
2629  * The xprt destroy routine. Again, because this connection is client
2630  * initiated, we do nothing
2631  */
2632
2633 static void bc_destroy(struct rpc_xprt *xprt)
2634 {
2635         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2636
2637         xs_xprt_free(xprt);
2638         module_put(THIS_MODULE);
2639 }
2640
2641 static const struct rpc_xprt_ops xs_local_ops = {
2642         .reserve_xprt           = xprt_reserve_xprt,
2643         .release_xprt           = xprt_release_xprt,
2644         .alloc_slot             = xprt_alloc_slot,
2645         .free_slot              = xprt_free_slot,
2646         .rpcbind                = xs_local_rpcbind,
2647         .set_port               = xs_local_set_port,
2648         .connect                = xs_local_connect,
2649         .buf_alloc              = rpc_malloc,
2650         .buf_free               = rpc_free,
2651         .prepare_request        = xs_stream_prepare_request,
2652         .send_request           = xs_local_send_request,
2653         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2654         .close                  = xs_close,
2655         .destroy                = xs_destroy,
2656         .print_stats            = xs_local_print_stats,
2657         .enable_swap            = xs_enable_swap,
2658         .disable_swap           = xs_disable_swap,
2659 };
2660
2661 static const struct rpc_xprt_ops xs_udp_ops = {
2662         .set_buffer_size        = xs_udp_set_buffer_size,
2663         .reserve_xprt           = xprt_reserve_xprt_cong,
2664         .release_xprt           = xprt_release_xprt_cong,
2665         .alloc_slot             = xprt_alloc_slot,
2666         .free_slot              = xprt_free_slot,
2667         .rpcbind                = rpcb_getport_async,
2668         .set_port               = xs_set_port,
2669         .connect                = xs_connect,
2670         .buf_alloc              = rpc_malloc,
2671         .buf_free               = rpc_free,
2672         .send_request           = xs_udp_send_request,
2673         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2674         .timer                  = xs_udp_timer,
2675         .release_request        = xprt_release_rqst_cong,
2676         .close                  = xs_close,
2677         .destroy                = xs_destroy,
2678         .print_stats            = xs_udp_print_stats,
2679         .enable_swap            = xs_enable_swap,
2680         .disable_swap           = xs_disable_swap,
2681         .inject_disconnect      = xs_inject_disconnect,
2682 };
2683
2684 static const struct rpc_xprt_ops xs_tcp_ops = {
2685         .reserve_xprt           = xprt_reserve_xprt,
2686         .release_xprt           = xprt_release_xprt,
2687         .alloc_slot             = xprt_alloc_slot,
2688         .free_slot              = xprt_free_slot,
2689         .rpcbind                = rpcb_getport_async,
2690         .set_port               = xs_set_port,
2691         .connect                = xs_connect,
2692         .buf_alloc              = rpc_malloc,
2693         .buf_free               = rpc_free,
2694         .prepare_request        = xs_stream_prepare_request,
2695         .send_request           = xs_tcp_send_request,
2696         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2697         .close                  = xs_tcp_shutdown,
2698         .destroy                = xs_destroy,
2699         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2700         .print_stats            = xs_tcp_print_stats,
2701         .enable_swap            = xs_enable_swap,
2702         .disable_swap           = xs_disable_swap,
2703         .inject_disconnect      = xs_inject_disconnect,
2704 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2705         .bc_setup               = xprt_setup_bc,
2706         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2707         .bc_num_slots           = xprt_bc_max_slots,
2708         .bc_free_rqst           = xprt_free_bc_rqst,
2709         .bc_destroy             = xprt_destroy_bc,
2710 #endif
2711 };
2712
2713 /*
2714  * The rpc_xprt_ops for the server backchannel
2715  */
2716
2717 static const struct rpc_xprt_ops bc_tcp_ops = {
2718         .reserve_xprt           = xprt_reserve_xprt,
2719         .release_xprt           = xprt_release_xprt,
2720         .alloc_slot             = xprt_alloc_slot,
2721         .free_slot              = xprt_free_slot,
2722         .buf_alloc              = bc_malloc,
2723         .buf_free               = bc_free,
2724         .send_request           = bc_send_request,
2725         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2726         .close                  = bc_close,
2727         .destroy                = bc_destroy,
2728         .print_stats            = xs_tcp_print_stats,
2729         .enable_swap            = xs_enable_swap,
2730         .disable_swap           = xs_disable_swap,
2731         .inject_disconnect      = xs_inject_disconnect,
2732 };
2733
2734 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2735 {
2736         static const struct sockaddr_in sin = {
2737                 .sin_family             = AF_INET,
2738                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2739         };
2740         static const struct sockaddr_in6 sin6 = {
2741                 .sin6_family            = AF_INET6,
2742                 .sin6_addr              = IN6ADDR_ANY_INIT,
2743         };
2744
2745         switch (family) {
2746         case AF_LOCAL:
2747                 break;
2748         case AF_INET:
2749                 memcpy(sap, &sin, sizeof(sin));
2750                 break;
2751         case AF_INET6:
2752                 memcpy(sap, &sin6, sizeof(sin6));
2753                 break;
2754         default:
2755                 dprintk("RPC:       %s: Bad address family\n", __func__);
2756                 return -EAFNOSUPPORT;
2757         }
2758         return 0;
2759 }
2760
2761 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2762                                       unsigned int slot_table_size,
2763                                       unsigned int max_slot_table_size)
2764 {
2765         struct rpc_xprt *xprt;
2766         struct sock_xprt *new;
2767
2768         if (args->addrlen > sizeof(xprt->addr)) {
2769                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2770                 return ERR_PTR(-EBADF);
2771         }
2772
2773         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2774                         max_slot_table_size);
2775         if (xprt == NULL) {
2776                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2777                                 "rpc_xprt\n");
2778                 return ERR_PTR(-ENOMEM);
2779         }
2780
2781         new = container_of(xprt, struct sock_xprt, xprt);
2782         mutex_init(&new->recv_mutex);
2783         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2784         xprt->addrlen = args->addrlen;
2785         if (args->srcaddr)
2786                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2787         else {
2788                 int err;
2789                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2790                                         (struct sockaddr *)&new->srcaddr);
2791                 if (err != 0) {
2792                         xprt_free(xprt);
2793                         return ERR_PTR(err);
2794                 }
2795         }
2796
2797         return xprt;
2798 }
2799
2800 static const struct rpc_timeout xs_local_default_timeout = {
2801         .to_initval = 10 * HZ,
2802         .to_maxval = 10 * HZ,
2803         .to_retries = 2,
2804 };
2805
2806 /**
2807  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2808  * @args: rpc transport creation arguments
2809  *
2810  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2811  */
2812 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2813 {
2814         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2815         struct sock_xprt *transport;
2816         struct rpc_xprt *xprt;
2817         struct rpc_xprt *ret;
2818
2819         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2820                         xprt_max_tcp_slot_table_entries);
2821         if (IS_ERR(xprt))
2822                 return xprt;
2823         transport = container_of(xprt, struct sock_xprt, xprt);
2824
2825         xprt->prot = 0;
2826         xprt->xprt_class = &xs_local_transport;
2827         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2828
2829         xprt->bind_timeout = XS_BIND_TO;
2830         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2831         xprt->idle_timeout = XS_IDLE_DISC_TO;
2832
2833         xprt->ops = &xs_local_ops;
2834         xprt->timeout = &xs_local_default_timeout;
2835
2836         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2837         INIT_WORK(&transport->error_worker, xs_error_handle);
2838         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2839
2840         switch (sun->sun_family) {
2841         case AF_LOCAL:
2842                 if (sun->sun_path[0] != '/') {
2843                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2844                                         sun->sun_path);
2845                         ret = ERR_PTR(-EINVAL);
2846                         goto out_err;
2847                 }
2848                 xprt_set_bound(xprt);
2849                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2850                 ret = ERR_PTR(xs_local_setup_socket(transport));
2851                 if (ret)
2852                         goto out_err;
2853                 break;
2854         default:
2855                 ret = ERR_PTR(-EAFNOSUPPORT);
2856                 goto out_err;
2857         }
2858
2859         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2860                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2861
2862         if (try_module_get(THIS_MODULE))
2863                 return xprt;
2864         ret = ERR_PTR(-EINVAL);
2865 out_err:
2866         xs_xprt_free(xprt);
2867         return ret;
2868 }
2869
2870 static const struct rpc_timeout xs_udp_default_timeout = {
2871         .to_initval = 5 * HZ,
2872         .to_maxval = 30 * HZ,
2873         .to_increment = 5 * HZ,
2874         .to_retries = 5,
2875 };
2876
2877 /**
2878  * xs_setup_udp - Set up transport to use a UDP socket
2879  * @args: rpc transport creation arguments
2880  *
2881  */
2882 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2883 {
2884         struct sockaddr *addr = args->dstaddr;
2885         struct rpc_xprt *xprt;
2886         struct sock_xprt *transport;
2887         struct rpc_xprt *ret;
2888
2889         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2890                         xprt_udp_slot_table_entries);
2891         if (IS_ERR(xprt))
2892                 return xprt;
2893         transport = container_of(xprt, struct sock_xprt, xprt);
2894
2895         xprt->prot = IPPROTO_UDP;
2896         xprt->xprt_class = &xs_udp_transport;
2897         /* XXX: header size can vary due to auth type, IPv6, etc. */
2898         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2899
2900         xprt->bind_timeout = XS_BIND_TO;
2901         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2902         xprt->idle_timeout = XS_IDLE_DISC_TO;
2903
2904         xprt->ops = &xs_udp_ops;
2905
2906         xprt->timeout = &xs_udp_default_timeout;
2907
2908         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2909         INIT_WORK(&transport->error_worker, xs_error_handle);
2910         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2911
2912         switch (addr->sa_family) {
2913         case AF_INET:
2914                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2915                         xprt_set_bound(xprt);
2916
2917                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2918                 break;
2919         case AF_INET6:
2920                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2921                         xprt_set_bound(xprt);
2922
2923                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2924                 break;
2925         default:
2926                 ret = ERR_PTR(-EAFNOSUPPORT);
2927                 goto out_err;
2928         }
2929
2930         if (xprt_bound(xprt))
2931                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2932                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2933                                 xprt->address_strings[RPC_DISPLAY_PORT],
2934                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2935         else
2936                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2937                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2938                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2939
2940         if (try_module_get(THIS_MODULE))
2941                 return xprt;
2942         ret = ERR_PTR(-EINVAL);
2943 out_err:
2944         xs_xprt_free(xprt);
2945         return ret;
2946 }
2947
2948 static const struct rpc_timeout xs_tcp_default_timeout = {
2949         .to_initval = 60 * HZ,
2950         .to_maxval = 60 * HZ,
2951         .to_retries = 2,
2952 };
2953
2954 /**
2955  * xs_setup_tcp - Set up transport to use a TCP socket
2956  * @args: rpc transport creation arguments
2957  *
2958  */
2959 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2960 {
2961         struct sockaddr *addr = args->dstaddr;
2962         struct rpc_xprt *xprt;
2963         struct sock_xprt *transport;
2964         struct rpc_xprt *ret;
2965         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2966
2967         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2968                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2969
2970         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2971                         max_slot_table_size);
2972         if (IS_ERR(xprt))
2973                 return xprt;
2974         transport = container_of(xprt, struct sock_xprt, xprt);
2975
2976         xprt->prot = IPPROTO_TCP;
2977         xprt->xprt_class = &xs_tcp_transport;
2978         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2979
2980         xprt->bind_timeout = XS_BIND_TO;
2981         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2982         xprt->idle_timeout = XS_IDLE_DISC_TO;
2983
2984         xprt->ops = &xs_tcp_ops;
2985         xprt->timeout = &xs_tcp_default_timeout;
2986
2987         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2988         xprt->connect_timeout = xprt->timeout->to_initval *
2989                 (xprt->timeout->to_retries + 1);
2990
2991         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2992         INIT_WORK(&transport->error_worker, xs_error_handle);
2993         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2994
2995         switch (addr->sa_family) {
2996         case AF_INET:
2997                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2998                         xprt_set_bound(xprt);
2999
3000                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3001                 break;
3002         case AF_INET6:
3003                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3004                         xprt_set_bound(xprt);
3005
3006                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3007                 break;
3008         default:
3009                 ret = ERR_PTR(-EAFNOSUPPORT);
3010                 goto out_err;
3011         }
3012
3013         if (xprt_bound(xprt))
3014                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3015                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3016                                 xprt->address_strings[RPC_DISPLAY_PORT],
3017                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3018         else
3019                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3020                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3021                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3022
3023         if (try_module_get(THIS_MODULE))
3024                 return xprt;
3025         ret = ERR_PTR(-EINVAL);
3026 out_err:
3027         xs_xprt_free(xprt);
3028         return ret;
3029 }
3030
3031 /**
3032  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3033  * @args: rpc transport creation arguments
3034  *
3035  */
3036 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3037 {
3038         struct sockaddr *addr = args->dstaddr;
3039         struct rpc_xprt *xprt;
3040         struct sock_xprt *transport;
3041         struct svc_sock *bc_sock;
3042         struct rpc_xprt *ret;
3043
3044         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3045                         xprt_tcp_slot_table_entries);
3046         if (IS_ERR(xprt))
3047                 return xprt;
3048         transport = container_of(xprt, struct sock_xprt, xprt);
3049
3050         xprt->prot = IPPROTO_TCP;
3051         xprt->xprt_class = &xs_bc_tcp_transport;
3052         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3053         xprt->timeout = &xs_tcp_default_timeout;
3054
3055         /* backchannel */
3056         xprt_set_bound(xprt);
3057         xprt->bind_timeout = 0;
3058         xprt->reestablish_timeout = 0;
3059         xprt->idle_timeout = 0;
3060
3061         xprt->ops = &bc_tcp_ops;
3062
3063         switch (addr->sa_family) {
3064         case AF_INET:
3065                 xs_format_peer_addresses(xprt, "tcp",
3066                                          RPCBIND_NETID_TCP);
3067                 break;
3068         case AF_INET6:
3069                 xs_format_peer_addresses(xprt, "tcp",
3070                                    RPCBIND_NETID_TCP6);
3071                 break;
3072         default:
3073                 ret = ERR_PTR(-EAFNOSUPPORT);
3074                 goto out_err;
3075         }
3076
3077         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3078                         xprt->address_strings[RPC_DISPLAY_ADDR],
3079                         xprt->address_strings[RPC_DISPLAY_PORT],
3080                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3081
3082         /*
3083          * Once we've associated a backchannel xprt with a connection,
3084          * we want to keep it around as long as the connection lasts,
3085          * in case we need to start using it for a backchannel again;
3086          * this reference won't be dropped until bc_xprt is destroyed.
3087          */
3088         xprt_get(xprt);
3089         args->bc_xprt->xpt_bc_xprt = xprt;
3090         xprt->bc_xprt = args->bc_xprt;
3091         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3092         transport->sock = bc_sock->sk_sock;
3093         transport->inet = bc_sock->sk_sk;
3094
3095         /*
3096          * Since we don't want connections for the backchannel, we set
3097          * the xprt status to connected
3098          */
3099         xprt_set_connected(xprt);
3100
3101         if (try_module_get(THIS_MODULE))
3102                 return xprt;
3103
3104         args->bc_xprt->xpt_bc_xprt = NULL;
3105         args->bc_xprt->xpt_bc_xps = NULL;
3106         xprt_put(xprt);
3107         ret = ERR_PTR(-EINVAL);
3108 out_err:
3109         xs_xprt_free(xprt);
3110         return ret;
3111 }
3112
3113 static struct xprt_class        xs_local_transport = {
3114         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3115         .name           = "named UNIX socket",
3116         .owner          = THIS_MODULE,
3117         .ident          = XPRT_TRANSPORT_LOCAL,
3118         .setup          = xs_setup_local,
3119         .netid          = { "" },
3120 };
3121
3122 static struct xprt_class        xs_udp_transport = {
3123         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3124         .name           = "udp",
3125         .owner          = THIS_MODULE,
3126         .ident          = XPRT_TRANSPORT_UDP,
3127         .setup          = xs_setup_udp,
3128         .netid          = { "udp", "udp6", "" },
3129 };
3130
3131 static struct xprt_class        xs_tcp_transport = {
3132         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3133         .name           = "tcp",
3134         .owner          = THIS_MODULE,
3135         .ident          = XPRT_TRANSPORT_TCP,
3136         .setup          = xs_setup_tcp,
3137         .netid          = { "tcp", "tcp6", "" },
3138 };
3139
3140 static struct xprt_class        xs_bc_tcp_transport = {
3141         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3142         .name           = "tcp NFSv4.1 backchannel",
3143         .owner          = THIS_MODULE,
3144         .ident          = XPRT_TRANSPORT_BC_TCP,
3145         .setup          = xs_setup_bc_tcp,
3146         .netid          = { "" },
3147 };
3148
3149 /**
3150  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3151  *
3152  */
3153 int init_socket_xprt(void)
3154 {
3155         if (!sunrpc_table_header)
3156                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3157
3158         xprt_register_transport(&xs_local_transport);
3159         xprt_register_transport(&xs_udp_transport);
3160         xprt_register_transport(&xs_tcp_transport);
3161         xprt_register_transport(&xs_bc_tcp_transport);
3162
3163         return 0;
3164 }
3165
3166 /**
3167  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3168  *
3169  */
3170 void cleanup_socket_xprt(void)
3171 {
3172         if (sunrpc_table_header) {
3173                 unregister_sysctl_table(sunrpc_table_header);
3174                 sunrpc_table_header = NULL;
3175         }
3176
3177         xprt_unregister_transport(&xs_local_transport);
3178         xprt_unregister_transport(&xs_udp_transport);
3179         xprt_unregister_transport(&xs_tcp_transport);
3180         xprt_unregister_transport(&xs_bc_tcp_transport);
3181 }
3182
3183 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3184 {
3185         return param_set_uint_minmax(val, kp,
3186                         RPC_MIN_RESVPORT,
3187                         RPC_MAX_RESVPORT);
3188 }
3189
3190 static const struct kernel_param_ops param_ops_portnr = {
3191         .set = param_set_portnr,
3192         .get = param_get_uint,
3193 };
3194
3195 #define param_check_portnr(name, p) \
3196         __param_check(name, p, unsigned int);
3197
3198 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3199 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3200
3201 static int param_set_slot_table_size(const char *val,
3202                                      const struct kernel_param *kp)
3203 {
3204         return param_set_uint_minmax(val, kp,
3205                         RPC_MIN_SLOT_TABLE,
3206                         RPC_MAX_SLOT_TABLE);
3207 }
3208
3209 static const struct kernel_param_ops param_ops_slot_table_size = {
3210         .set = param_set_slot_table_size,
3211         .get = param_get_uint,
3212 };
3213
3214 #define param_check_slot_table_size(name, p) \
3215         __param_check(name, p, unsigned int);
3216
3217 static int param_set_max_slot_table_size(const char *val,
3218                                      const struct kernel_param *kp)
3219 {
3220         return param_set_uint_minmax(val, kp,
3221                         RPC_MIN_SLOT_TABLE,
3222                         RPC_MAX_SLOT_TABLE_LIMIT);
3223 }
3224
3225 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3226         .set = param_set_max_slot_table_size,
3227         .get = param_get_uint,
3228 };
3229
3230 #define param_check_max_slot_table_size(name, p) \
3231         __param_check(name, p, unsigned int);
3232
3233 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3234                    slot_table_size, 0644);
3235 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3236                    max_slot_table_size, 0644);
3237 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3238                    slot_table_size, 0644);