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_set_srcport(struct sock_xprt *transport, struct socket *sock);
  62 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  63                 struct socket *sock);
  64
  65 /*
  66  * xprtsock tunables
  67  */
  68 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  69 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  70 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  71
  72 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  73 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  74
  75 #define XS_TCP_LINGER_TO        (15U * HZ)
  76 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  77
  78 /*
  79  * We can register our own files under /proc/sys/sunrpc by
  80  * calling register_sysctl_table() again.  The files in that
  81  * directory become the union of all files registered there.
  82  *
  83  * We simply need to make sure that we don't collide with
  84  * someone else's file names!
  85  */
  86
  87 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  88 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  89 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  90 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  91 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  92
  93 static struct ctl_table_header *sunrpc_table_header;
  94
  95 static struct xprt_class xs_local_transport;
  96 static struct xprt_class xs_udp_transport;
  97 static struct xprt_class xs_tcp_transport;
  98 static struct xprt_class xs_bc_tcp_transport;
  99
 100 /*
 101  * FIXME: changing the UDP slot table size should also resize the UDP
 102  *        socket buffers for existing UDP transports
 103  */
 104 static struct ctl_table xs_tunables_table[] = {
 105         {
 106                 .procname       = "udp_slot_table_entries",
 107                 .data           = &xprt_udp_slot_table_entries,
 108                 .maxlen         = sizeof(unsigned int),
 109                 .mode           = 0644,
 110                 .proc_handler   = proc_dointvec_minmax,
 111                 .extra1         = &min_slot_table_size,
 112                 .extra2         = &max_slot_table_size
 113         },
 114         {
 115                 .procname       = "tcp_slot_table_entries",
 116                 .data           = &xprt_tcp_slot_table_entries,
 117                 .maxlen         = sizeof(unsigned int),
 118                 .mode           = 0644,
 119                 .proc_handler   = proc_dointvec_minmax,
 120                 .extra1         = &min_slot_table_size,
 121                 .extra2         = &max_slot_table_size
 122         },
 123         {
 124                 .procname       = "tcp_max_slot_table_entries",
 125                 .data           = &xprt_max_tcp_slot_table_entries,
 126                 .maxlen         = sizeof(unsigned int),
 127                 .mode           = 0644,
 128                 .proc_handler   = proc_dointvec_minmax,
 129                 .extra1         = &min_slot_table_size,
 130                 .extra2         = &max_tcp_slot_table_limit
 131         },
 132         {
 133                 .procname       = "min_resvport",
 134                 .data           = &xprt_min_resvport,
 135                 .maxlen         = sizeof(unsigned int),
 136                 .mode           = 0644,
 137                 .proc_handler   = proc_dointvec_minmax,
 138                 .extra1         = &xprt_min_resvport_limit,
 139                 .extra2         = &xprt_max_resvport_limit
 140         },
 141         {
 142                 .procname       = "max_resvport",
 143                 .data           = &xprt_max_resvport,
 144                 .maxlen         = sizeof(unsigned int),
 145                 .mode           = 0644,
 146                 .proc_handler   = proc_dointvec_minmax,
 147                 .extra1         = &xprt_min_resvport_limit,
 148                 .extra2         = &xprt_max_resvport_limit
 149         },
 150         {
 151                 .procname       = "tcp_fin_timeout",
 152                 .data           = &xs_tcp_fin_timeout,
 153                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 154                 .mode           = 0644,
 155                 .proc_handler   = proc_dointvec_jiffies,
 156         },
 157         { },
 158 };
 159
 160 static struct ctl_table sunrpc_table[] = {
 161         {
 162                 .procname       = "sunrpc",
 163                 .mode           = 0555,
 164                 .child          = xs_tunables_table
 165         },
 166         { },
 167 };
 168
 169 /*
 170  * Wait duration for a reply from the RPC portmapper.
 171  */
 172 #define XS_BIND_TO              (60U * HZ)
 173
 174 /*
 175  * Delay if a UDP socket connect error occurs.  This is most likely some
 176  * kind of resource problem on the local host.
 177  */
 178 #define XS_UDP_REEST_TO         (2U * HZ)
 179
 180 /*
 181  * The reestablish timeout allows clients to delay for a bit before attempting
 182  * to reconnect to a server that just dropped our connection.
 183  *
 184  * We implement an exponential backoff when trying to reestablish a TCP
 185  * transport connection with the server.  Some servers like to drop a TCP
 186  * connection when they are overworked, so we start with a short timeout and
 187  * increase over time if the server is down or not responding.
 188  */
 189 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 190
 191 /*
 192  * TCP idle timeout; client drops the transport socket if it is idle
 193  * for this long.  Note that we also timeout UDP sockets to prevent
 194  * holding port numbers when there is no RPC traffic.
 195  */
 196 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 197
 198 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 199 # undef  RPC_DEBUG_DATA
 200 # define RPCDBG_FACILITY        RPCDBG_TRANS
 201 #endif
 202
 203 #ifdef RPC_DEBUG_DATA
 204 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 205 {
 206         u8 *buf = (u8 *) packet;
 207         int j;
 208
 209         dprintk("RPC:       %s\n", msg);
 210         for (j = 0; j < count && j < 128; j += 4) {
 211                 if (!(j & 31)) {
 212                         if (j)
 213                                 dprintk("\n");
 214                         dprintk("0x%04x ", j);
 215                 }
 216                 dprintk("%02x%02x%02x%02x ",
 217                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 218         }
 219         dprintk("\n");
 220 }
 221 #else
 222 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 223 {
 224         /* NOP */
 225 }
 226 #endif
 227
 228 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 229 {
 230         return (struct rpc_xprt *) sk->sk_user_data;
 231 }
 232
 233 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 234 {
 235         return (struct sockaddr *) &xprt->addr;
 236 }
 237
 238 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 239 {
 240         return (struct sockaddr_un *) &xprt->addr;
 241 }
 242
 243 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 244 {
 245         return (struct sockaddr_in *) &xprt->addr;
 246 }
 247
 248 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 249 {
 250         return (struct sockaddr_in6 *) &xprt->addr;
 251 }
 252
 253 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 254 {
 255         struct sockaddr *sap = xs_addr(xprt);
 256         struct sockaddr_in6 *sin6;
 257         struct sockaddr_in *sin;
 258         struct sockaddr_un *sun;
 259         char buf[128];
 260
 261         switch (sap->sa_family) {
 262         case AF_LOCAL:
 263                 sun = xs_addr_un(xprt);
 264                 strlcpy(buf, sun->sun_path, sizeof(buf));
 265                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 266                                                 kstrdup(buf, GFP_KERNEL);
 267                 break;
 268         case AF_INET:
 269                 (void)rpc_ntop(sap, buf, sizeof(buf));
 270                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 271                                                 kstrdup(buf, GFP_KERNEL);
 272                 sin = xs_addr_in(xprt);
 273                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 274                 break;
 275         case AF_INET6:
 276                 (void)rpc_ntop(sap, buf, sizeof(buf));
 277                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 278                                                 kstrdup(buf, GFP_KERNEL);
 279                 sin6 = xs_addr_in6(xprt);
 280                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 281                 break;
 282         default:
 283                 BUG();
 284         }
 285
 286         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 287 }
 288
 289 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 290 {
 291         struct sockaddr *sap = xs_addr(xprt);
 292         char buf[128];
 293
 294         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 295         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 296
 297         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 298         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 299 }
 300
 301 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 302                                      const char *protocol,
 303                                      const char *netid)
 304 {
 305         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 306         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 307         xs_format_common_peer_addresses(xprt);
 308         xs_format_common_peer_ports(xprt);
 309 }
 310
 311 static void xs_update_peer_port(struct rpc_xprt *xprt)
 312 {
 313         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 314         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 315
 316         xs_format_common_peer_ports(xprt);
 317 }
 318
 319 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 320 {
 321         unsigned int i;
 322
 323         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 324                 switch (i) {
 325                 case RPC_DISPLAY_PROTO:
 326                 case RPC_DISPLAY_NETID:
 327                         continue;
 328                 default:
 329                         kfree(xprt->address_strings[i]);
 330                 }
 331 }
 332
 333 static size_t
 334 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 335 {
 336         size_t i,n;
 337
 338         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 339                 return want;
 340         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 341         for (i = 0; i < n; i++) {
 342                 if (buf->pages[i])
 343                         continue;
 344                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 345                 if (!buf->pages[i]) {
 346                         i *= PAGE_SIZE;
 347                         return i > buf->page_base ? i - buf->page_base : 0;
 348                 }
 349         }
 350         return want;
 351 }
 352
 353 static ssize_t
 354 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 355 {
 356         ssize_t ret;
 357         if (seek != 0)
 358                 iov_iter_advance(&msg->msg_iter, seek);
 359         ret = sock_recvmsg(sock, msg, flags);
 360         return ret > 0 ? ret + seek : ret;
 361 }
 362
 363 static ssize_t
 364 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 365                 struct kvec *kvec, size_t count, size_t seek)
 366 {
 367         iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
 368         return xs_sock_recvmsg(sock, msg, flags, seek);
 369 }
 370
 371 static ssize_t
 372 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 373                 struct bio_vec *bvec, unsigned long nr, size_t count,
 374                 size_t seek)
 375 {
 376         iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
 377         return xs_sock_recvmsg(sock, msg, flags, seek);
 378 }
 379
 380 static ssize_t
 381 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 382                 size_t count)
 383 {
 384         iov_iter_discard(&msg->msg_iter, READ, count);
 385         return sock_recvmsg(sock, msg, flags);
 386 }
 387
 388 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 389 static void
 390 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 391 {
 392         struct bvec_iter bi = {
 393                 .bi_size = count,
 394         };
 395         struct bio_vec bv;
 396
 397         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 398         for_each_bvec(bv, bvec, bi, bi)
 399                 flush_dcache_page(bv.bv_page);
 400 }
 401 #else
 402 static inline void
 403 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 404 {
 405 }
 406 #endif
 407
 408 static ssize_t
 409 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 410                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 411 {
 412         size_t want, seek_init = seek, offset = 0;
 413         ssize_t ret;
 414
 415         want = min_t(size_t, count, buf->head[0].iov_len);
 416         if (seek < want) {
 417                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 418                 if (ret <= 0)
 419                         goto sock_err;
 420                 offset += ret;
 421                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 422                         goto out;
 423                 if (ret != want)
 424                         goto out;
 425                 seek = 0;
 426         } else {
 427                 seek -= want;
 428                 offset += want;
 429         }
 430
 431         want = xs_alloc_sparse_pages(
 432                 buf, min_t(size_t, count - offset, buf->page_len),
 433                 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
 434         if (seek < want) {
 435                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 436                                 xdr_buf_pagecount(buf),
 437                                 want + buf->page_base,
 438                                 seek + buf->page_base);
 439                 if (ret <= 0)
 440                         goto sock_err;
 441                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 442                 ret -= buf->page_base;
 443                 offset += ret;
 444                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 445                         goto out;
 446                 if (ret != want)
 447                         goto out;
 448                 seek = 0;
 449         } else {
 450                 seek -= want;
 451                 offset += want;
 452         }
 453
 454         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 455         if (seek < want) {
 456                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 457                 if (ret <= 0)
 458                         goto sock_err;
 459                 offset += ret;
 460                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 461                         goto out;
 462                 if (ret != want)
 463                         goto out;
 464         } else if (offset < seek_init)
 465                 offset = seek_init;
 466         ret = -EMSGSIZE;
 467 out:
 468         *read = offset - seek_init;
 469         return ret;
 470 sock_err:
 471         offset += seek;
 472         goto out;
 473 }
 474
 475 static void
 476 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 477 {
 478         if (!transport->recv.copied) {
 479                 if (buf->head[0].iov_len >= transport->recv.offset)
 480                         memcpy(buf->head[0].iov_base,
 481                                         &transport->recv.xid,
 482                                         transport->recv.offset);
 483                 transport->recv.copied = transport->recv.offset;
 484         }
 485 }
 486
 487 static bool
 488 xs_read_stream_request_done(struct sock_xprt *transport)
 489 {
 490         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 491 }
 492
 493 static void
 494 xs_read_stream_check_eor(struct sock_xprt *transport,
 495                 struct msghdr *msg)
 496 {
 497         if (xs_read_stream_request_done(transport))
 498                 msg->msg_flags |= MSG_EOR;
 499 }
 500
 501 static ssize_t
 502 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 503                 int flags, struct rpc_rqst *req)
 504 {
 505         struct xdr_buf *buf = &req->rq_private_buf;
 506         size_t want, read;
 507         ssize_t ret;
 508
 509         xs_read_header(transport, buf);
 510
 511         want = transport->recv.len - transport->recv.offset;
 512         if (want != 0) {
 513                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 514                                 transport->recv.copied + want,
 515                                 transport->recv.copied,
 516                                 &read);
 517                 transport->recv.offset += read;
 518                 transport->recv.copied += read;
 519         }
 520
 521         if (transport->recv.offset == transport->recv.len)
 522                 xs_read_stream_check_eor(transport, msg);
 523
 524         if (want == 0)
 525                 return 0;
 526
 527         switch (ret) {
 528         default:
 529                 break;
 530         case -EFAULT:
 531         case -EMSGSIZE:
 532                 msg->msg_flags |= MSG_TRUNC;
 533                 return read;
 534         case 0:
 535                 return -ESHUTDOWN;
 536         }
 537         return ret < 0 ? ret : read;
 538 }
 539
 540 static size_t
 541 xs_read_stream_headersize(bool isfrag)
 542 {
 543         if (isfrag)
 544                 return sizeof(__be32);
 545         return 3 * sizeof(__be32);
 546 }
 547
 548 static ssize_t
 549 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 550                 int flags, size_t want, size_t seek)
 551 {
 552         struct kvec kvec = {
 553                 .iov_base = &transport->recv.fraghdr,
 554                 .iov_len = want,
 555         };
 556         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 557 }
 558
 559 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 560 static ssize_t
 561 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 562 {
 563         struct rpc_xprt *xprt = &transport->xprt;
 564         struct rpc_rqst *req;
 565         ssize_t ret;
 566
 567         /* Is this transport associated with the backchannel? */
 568         if (!xprt->bc_serv)
 569                 return -ESHUTDOWN;
 570
 571         /* Look up and lock the request corresponding to the given XID */
 572         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 573         if (!req) {
 574                 printk(KERN_WARNING "Callback slot table overflowed\n");
 575                 return -ESHUTDOWN;
 576         }
 577         if (transport->recv.copied && !req->rq_private_buf.len)
 578                 return -ESHUTDOWN;
 579
 580         ret = xs_read_stream_request(transport, msg, flags, req);
 581         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 582                 xprt_complete_bc_request(req, transport->recv.copied);
 583         else
 584                 req->rq_private_buf.len = transport->recv.copied;
 585
 586         return ret;
 587 }
 588 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 589 static ssize_t
 590 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 591 {
 592         return -ESHUTDOWN;
 593 }
 594 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 595
 596 static ssize_t
 597 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 598 {
 599         struct rpc_xprt *xprt = &transport->xprt;
 600         struct rpc_rqst *req;
 601         ssize_t ret = 0;
 602
 603         /* Look up and lock the request corresponding to the given XID */
 604         spin_lock(&xprt->queue_lock);
 605         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 606         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 607                 msg->msg_flags |= MSG_TRUNC;
 608                 goto out;
 609         }
 610         xprt_pin_rqst(req);
 611         spin_unlock(&xprt->queue_lock);
 612
 613         ret = xs_read_stream_request(transport, msg, flags, req);
 614
 615         spin_lock(&xprt->queue_lock);
 616         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 617                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 618         else
 619                 req->rq_private_buf.len = transport->recv.copied;
 620         xprt_unpin_rqst(req);
 621 out:
 622         spin_unlock(&xprt->queue_lock);
 623         return ret;
 624 }
 625
 626 static ssize_t
 627 xs_read_stream(struct sock_xprt *transport, int flags)
 628 {
 629         struct msghdr msg = { 0 };
 630         size_t want, read = 0;
 631         ssize_t ret = 0;
 632
 633         if (transport->recv.len == 0) {
 634                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 635                 ret = xs_read_stream_header(transport, &msg, flags, want,
 636                                 transport->recv.offset);
 637                 if (ret <= 0)
 638                         goto out_err;
 639                 transport->recv.offset = ret;
 640                 if (transport->recv.offset != want)
 641                         return transport->recv.offset;
 642                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 643                         RPC_FRAGMENT_SIZE_MASK;
 644                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 645                 read = ret;
 646         }
 647
 648         switch (be32_to_cpu(transport->recv.calldir)) {
 649         default:
 650                 msg.msg_flags |= MSG_TRUNC;
 651                 break;
 652         case RPC_CALL:
 653                 ret = xs_read_stream_call(transport, &msg, flags);
 654                 break;
 655         case RPC_REPLY:
 656                 ret = xs_read_stream_reply(transport, &msg, flags);
 657         }
 658         if (msg.msg_flags & MSG_TRUNC) {
 659                 transport->recv.calldir = cpu_to_be32(-1);
 660                 transport->recv.copied = -1;
 661         }
 662         if (ret < 0)
 663                 goto out_err;
 664         read += ret;
 665         if (transport->recv.offset < transport->recv.len) {
 666                 if (!(msg.msg_flags & MSG_TRUNC))
 667                         return read;
 668                 msg.msg_flags = 0;
 669                 ret = xs_read_discard(transport->sock, &msg, flags,
 670                                 transport->recv.len - transport->recv.offset);
 671                 if (ret <= 0)
 672                         goto out_err;
 673                 transport->recv.offset += ret;
 674                 read += ret;
 675                 if (transport->recv.offset != transport->recv.len)
 676                         return read;
 677         }
 678         if (xs_read_stream_request_done(transport)) {
 679                 trace_xs_stream_read_request(transport);
 680                 transport->recv.copied = 0;
 681         }
 682         transport->recv.offset = 0;
 683         transport->recv.len = 0;
 684         return read;
 685 out_err:
 686         return ret != 0 ? ret : -ESHUTDOWN;
 687 }
 688
 689 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 690 {
 691         return transport->sock->ops->poll(transport->file, transport->sock,
 692                         NULL);
 693 }
 694
 695 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 696 {
 697         __poll_t events = xs_poll_socket(transport);
 698
 699         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 700 }
 701
 702 static void xs_poll_check_readable(struct sock_xprt *transport)
 703 {
 704
 705         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 706         if (!xs_poll_socket_readable(transport))
 707                 return;
 708         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 709                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 710 }
 711
 712 static void xs_stream_data_receive(struct sock_xprt *transport)
 713 {
 714         size_t read = 0;
 715         ssize_t ret = 0;
 716
 717         mutex_lock(&transport->recv_mutex);
 718         if (transport->sock == NULL)
 719                 goto out;
 720         for (;;) {
 721                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 722                 if (ret < 0)
 723                         break;
 724                 read += ret;
 725                 cond_resched();
 726         }
 727         if (ret == -ESHUTDOWN)
 728                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 729         else
 730                 xs_poll_check_readable(transport);
 731 out:
 732         mutex_unlock(&transport->recv_mutex);
 733         trace_xs_stream_read_data(&transport->xprt, ret, read);
 734 }
 735
 736 static void xs_stream_data_receive_workfn(struct work_struct *work)
 737 {
 738         struct sock_xprt *transport =
 739                 container_of(work, struct sock_xprt, recv_worker);
 740         unsigned int pflags = memalloc_nofs_save();
 741
 742         xs_stream_data_receive(transport);
 743         memalloc_nofs_restore(pflags);
 744 }
 745
 746 static void
 747 xs_stream_reset_connect(struct sock_xprt *transport)
 748 {
 749         transport->recv.offset = 0;
 750         transport->recv.len = 0;
 751         transport->recv.copied = 0;
 752         transport->xmit.offset = 0;
 753 }
 754
 755 static void
 756 xs_stream_start_connect(struct sock_xprt *transport)
 757 {
 758         transport->xprt.stat.connect_count++;
 759         transport->xprt.stat.connect_start = jiffies;
 760 }
 761
 762 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 763
 764 /**
 765  * xs_nospace - handle transmit was incomplete
 766  * @req: pointer to RPC request
 767  * @transport: pointer to struct sock_xprt
 768  *
 769  */
 770 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 771 {
 772         struct rpc_xprt *xprt = &transport->xprt;
 773         struct sock *sk = transport->inet;
 774         int ret = -EAGAIN;
 775
 776         trace_rpc_socket_nospace(req, transport);
 777
 778         /* Protect against races with write_space */
 779         spin_lock(&xprt->transport_lock);
 780
 781         /* Don't race with disconnect */
 782         if (xprt_connected(xprt)) {
 783                 /* wait for more buffer space */
 784                 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
 785                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 786                 sk->sk_write_pending++;
 787                 xprt_wait_for_buffer_space(xprt);
 788         } else
 789                 ret = -ENOTCONN;
 790
 791         spin_unlock(&xprt->transport_lock);
 792         return ret;
 793 }
 794
 795 static int xs_sock_nospace(struct rpc_rqst *req)
 796 {
 797         struct sock_xprt *transport =
 798                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 799         struct sock *sk = transport->inet;
 800         int ret = -EAGAIN;
 801
 802         lock_sock(sk);
 803         if (!sock_writeable(sk))
 804                 ret = xs_nospace(req, transport);
 805         release_sock(sk);
 806         return ret;
 807 }
 808
 809 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
 810 {
 811         struct sock_xprt *transport =
 812                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 813         struct sock *sk = transport->inet;
 814         int ret = -EAGAIN;
 815
 816         if (vm_wait)
 817                 return -ENOBUFS;
 818         lock_sock(sk);
 819         if (!sk_stream_memory_free(sk))
 820                 ret = xs_nospace(req, transport);
 821         release_sock(sk);
 822         return ret;
 823 }
 824
 825 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
 826 {
 827         return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
 828 }
 829
 830 /*
 831  * Determine if the previous message in the stream was aborted before it
 832  * could complete transmission.
 833  */
 834 static bool
 835 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 836 {
 837         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 838 }
 839
 840 /*
 841  * Return the stream record marker field for a record of length < 2^31-1
 842  */
 843 static rpc_fraghdr
 844 xs_stream_record_marker(struct xdr_buf *xdr)
 845 {
 846         if (!xdr->len)
 847                 return 0;
 848         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 849 }
 850
 851 /**
 852  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 853  * @req: pointer to RPC request
 854  *
 855  * Return values:
 856  *        0:    The request has been sent
 857  *   EAGAIN:    The socket was blocked, please call again later to
 858  *              complete the request
 859  * ENOTCONN:    Caller needs to invoke connect logic then call again
 860  *    other:    Some other error occurred, the request was not sent
 861  */
 862 static int xs_local_send_request(struct rpc_rqst *req)
 863 {
 864         struct rpc_xprt *xprt = req->rq_xprt;
 865         struct sock_xprt *transport =
 866                                 container_of(xprt, struct sock_xprt, xprt);
 867         struct xdr_buf *xdr = &req->rq_snd_buf;
 868         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 869         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 870         struct msghdr msg = {
 871                 .msg_flags      = XS_SENDMSG_FLAGS,
 872         };
 873         bool vm_wait;
 874         unsigned int sent;
 875         int status;
 876
 877         /* Close the stream if the previous transmission was incomplete */
 878         if (xs_send_request_was_aborted(transport, req)) {
 879                 xprt_force_disconnect(xprt);
 880                 return -ENOTCONN;
 881         }
 882
 883         xs_pktdump("packet data:",
 884                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 885
 886         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
 887
 888         req->rq_xtime = ktime_get();
 889         status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 890                                    transport->xmit.offset, rm, &sent);
 891         dprintk("RPC:       %s(%u) = %d\n",
 892                         __func__, xdr->len - transport->xmit.offset, status);
 893
 894         if (likely(sent > 0) || status == 0) {
 895                 transport->xmit.offset += sent;
 896                 req->rq_bytes_sent = transport->xmit.offset;
 897                 if (likely(req->rq_bytes_sent >= msglen)) {
 898                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 899                         transport->xmit.offset = 0;
 900                         return 0;
 901                 }
 902                 status = -EAGAIN;
 903                 vm_wait = false;
 904         }
 905
 906         switch (status) {
 907         case -EAGAIN:
 908                 status = xs_stream_nospace(req, vm_wait);
 909                 break;
 910         default:
 911                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 912                         -status);
 913                 fallthrough;
 914         case -EPIPE:
 915                 xprt_force_disconnect(xprt);
 916                 status = -ENOTCONN;
 917         }
 918
 919         return status;
 920 }
 921
 922 /**
 923  * xs_udp_send_request - write an RPC request to a UDP socket
 924  * @req: pointer to RPC request
 925  *
 926  * Return values:
 927  *        0:    The request has been sent
 928  *   EAGAIN:    The socket was blocked, please call again later to
 929  *              complete the request
 930  * ENOTCONN:    Caller needs to invoke connect logic then call again
 931  *    other:    Some other error occurred, the request was not sent
 932  */
 933 static int xs_udp_send_request(struct rpc_rqst *req)
 934 {
 935         struct rpc_xprt *xprt = req->rq_xprt;
 936         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 937         struct xdr_buf *xdr = &req->rq_snd_buf;
 938         struct msghdr msg = {
 939                 .msg_name       = xs_addr(xprt),
 940                 .msg_namelen    = xprt->addrlen,
 941                 .msg_flags      = XS_SENDMSG_FLAGS,
 942         };
 943         unsigned int sent;
 944         int status;
 945
 946         xs_pktdump("packet data:",
 947                                 req->rq_svec->iov_base,
 948                                 req->rq_svec->iov_len);
 949
 950         if (!xprt_bound(xprt))
 951                 return -ENOTCONN;
 952
 953         if (!xprt_request_get_cong(xprt, req))
 954                 return -EBADSLT;
 955
 956         status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
 957         if (status < 0)
 958                 return status;
 959         req->rq_xtime = ktime_get();
 960         status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
 961
 962         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 963                         xdr->len, status);
 964
 965         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 966         if (status == -EPERM)
 967                 goto process_status;
 968
 969         if (status == -EAGAIN && sock_writeable(transport->inet))
 970                 status = -ENOBUFS;
 971
 972         if (sent > 0 || status == 0) {
 973                 req->rq_xmit_bytes_sent += sent;
 974                 if (sent >= req->rq_slen)
 975                         return 0;
 976                 /* Still some bytes left; set up for a retry later. */
 977                 status = -EAGAIN;
 978         }
 979
 980 process_status:
 981         switch (status) {
 982         case -ENOTSOCK:
 983                 status = -ENOTCONN;
 984                 /* Should we call xs_close() here? */
 985                 break;
 986         case -EAGAIN:
 987                 status = xs_sock_nospace(req);
 988                 break;
 989         case -ENETUNREACH:
 990         case -ENOBUFS:
 991         case -EPIPE:
 992         case -ECONNREFUSED:
 993         case -EPERM:
 994                 /* When the server has died, an ICMP port unreachable message
 995                  * prompts ECONNREFUSED. */
 996                 break;
 997         default:
 998                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 999                         -status);
1000         }
1001
1002         return status;
1003 }
1004
1005 /**
1006  * xs_tcp_send_request - write an RPC request to a TCP socket
1007  * @req: pointer to RPC request
1008  *
1009  * Return values:
1010  *        0:    The request has been sent
1011  *   EAGAIN:    The socket was blocked, please call again later to
1012  *              complete the request
1013  * ENOTCONN:    Caller needs to invoke connect logic then call again
1014  *    other:    Some other error occurred, the request was not sent
1015  *
1016  * XXX: In the case of soft timeouts, should we eventually give up
1017  *      if sendmsg is not able to make progress?
1018  */
1019 static int xs_tcp_send_request(struct rpc_rqst *req)
1020 {
1021         struct rpc_xprt *xprt = req->rq_xprt;
1022         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1023         struct xdr_buf *xdr = &req->rq_snd_buf;
1024         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1025         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1026         struct msghdr msg = {
1027                 .msg_flags      = XS_SENDMSG_FLAGS,
1028         };
1029         bool vm_wait;
1030         unsigned int sent;
1031         int status;
1032
1033         /* Close the stream if the previous transmission was incomplete */
1034         if (xs_send_request_was_aborted(transport, req)) {
1035                 if (transport->sock != NULL)
1036                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1037                 return -ENOTCONN;
1038         }
1039         if (!transport->inet)
1040                 return -ENOTCONN;
1041
1042         xs_pktdump("packet data:",
1043                                 req->rq_svec->iov_base,
1044                                 req->rq_svec->iov_len);
1045
1046         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1047                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1048
1049         xs_set_srcport(transport, transport->sock);
1050
1051         /* Continue transmitting the packet/record. We must be careful
1052          * to cope with writespace callbacks arriving _after_ we have
1053          * called sendmsg(). */
1054         req->rq_xtime = ktime_get();
1055         tcp_sock_set_cork(transport->inet, true);
1056
1057         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1058
1059         do {
1060                 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1061                                            transport->xmit.offset, rm, &sent);
1062
1063                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1064                                 xdr->len - transport->xmit.offset, status);
1065
1066                 /* If we've sent the entire packet, immediately
1067                  * reset the count of bytes sent. */
1068                 transport->xmit.offset += sent;
1069                 req->rq_bytes_sent = transport->xmit.offset;
1070                 if (likely(req->rq_bytes_sent >= msglen)) {
1071                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1072                         transport->xmit.offset = 0;
1073                         if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1074                                 tcp_sock_set_cork(transport->inet, false);
1075                         return 0;
1076                 }
1077
1078                 WARN_ON_ONCE(sent == 0 && status == 0);
1079
1080                 if (sent > 0)
1081                         vm_wait = false;
1082
1083         } while (status == 0);
1084
1085         switch (status) {
1086         case -ENOTSOCK:
1087                 status = -ENOTCONN;
1088                 /* Should we call xs_close() here? */
1089                 break;
1090         case -EAGAIN:
1091                 status = xs_stream_nospace(req, vm_wait);
1092                 break;
1093         case -ECONNRESET:
1094         case -ECONNREFUSED:
1095         case -ENOTCONN:
1096         case -EADDRINUSE:
1097         case -ENOBUFS:
1098         case -EPIPE:
1099                 break;
1100         default:
1101                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1102                         -status);
1103         }
1104
1105         return status;
1106 }
1107
1108 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1109 {
1110         transport->old_data_ready = sk->sk_data_ready;
1111         transport->old_state_change = sk->sk_state_change;
1112         transport->old_write_space = sk->sk_write_space;
1113         transport->old_error_report = sk->sk_error_report;
1114 }
1115
1116 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1117 {
1118         sk->sk_data_ready = transport->old_data_ready;
1119         sk->sk_state_change = transport->old_state_change;
1120         sk->sk_write_space = transport->old_write_space;
1121         sk->sk_error_report = transport->old_error_report;
1122 }
1123
1124 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1125 {
1126         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1127
1128         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1129         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1130         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1131         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1132         clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1133 }
1134
1135 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1136 {
1137         set_bit(nr, &transport->sock_state);
1138         queue_work(xprtiod_workqueue, &transport->error_worker);
1139 }
1140
1141 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1142 {
1143         xprt->connect_cookie++;
1144         smp_mb__before_atomic();
1145         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1146         clear_bit(XPRT_CLOSING, &xprt->state);
1147         xs_sock_reset_state_flags(xprt);
1148         smp_mb__after_atomic();
1149 }
1150
1151 /**
1152  * xs_error_report - callback to handle TCP socket state errors
1153  * @sk: socket
1154  *
1155  * Note: we don't call sock_error() since there may be a rpc_task
1156  * using the socket, and so we don't want to clear sk->sk_err.
1157  */
1158 static void xs_error_report(struct sock *sk)
1159 {
1160         struct sock_xprt *transport;
1161         struct rpc_xprt *xprt;
1162
1163         if (!(xprt = xprt_from_sock(sk)))
1164                 return;
1165
1166         transport = container_of(xprt, struct sock_xprt, xprt);
1167         transport->xprt_err = -sk->sk_err;
1168         if (transport->xprt_err == 0)
1169                 return;
1170         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1171                         xprt, -transport->xprt_err);
1172         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1173
1174         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1175         smp_mb__before_atomic();
1176         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1177 }
1178
1179 static void xs_reset_transport(struct sock_xprt *transport)
1180 {
1181         struct socket *sock = transport->sock;
1182         struct sock *sk = transport->inet;
1183         struct rpc_xprt *xprt = &transport->xprt;
1184         struct file *filp = transport->file;
1185
1186         if (sk == NULL)
1187                 return;
1188         /*
1189          * Make sure we're calling this in a context from which it is safe
1190          * to call __fput_sync(). In practice that means rpciod and the
1191          * system workqueue.
1192          */
1193         if (!(current->flags & PF_WQ_WORKER)) {
1194                 WARN_ON_ONCE(1);
1195                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1196                 return;
1197         }
1198
1199         if (atomic_read(&transport->xprt.swapper))
1200                 sk_clear_memalloc(sk);
1201
1202         kernel_sock_shutdown(sock, SHUT_RDWR);
1203
1204         mutex_lock(&transport->recv_mutex);
1205         lock_sock(sk);
1206         transport->inet = NULL;
1207         transport->sock = NULL;
1208         transport->file = NULL;
1209
1210         sk->sk_user_data = NULL;
1211
1212         xs_restore_old_callbacks(transport, sk);
1213         xprt_clear_connected(xprt);
1214         xs_sock_reset_connection_flags(xprt);
1215         /* Reset stream record info */
1216         xs_stream_reset_connect(transport);
1217         release_sock(sk);
1218         mutex_unlock(&transport->recv_mutex);
1219
1220         trace_rpc_socket_close(xprt, sock);
1221         __fput_sync(filp);
1222
1223         xprt_disconnect_done(xprt);
1224 }
1225
1226 /**
1227  * xs_close - close a socket
1228  * @xprt: transport
1229  *
1230  * This is used when all requests are complete; ie, no DRC state remains
1231  * on the server we want to save.
1232  *
1233  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1234  * xs_reset_transport() zeroing the socket from underneath a writer.
1235  */
1236 static void xs_close(struct rpc_xprt *xprt)
1237 {
1238         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1239
1240         dprintk("RPC:       xs_close xprt %p\n", xprt);
1241
1242         xs_reset_transport(transport);
1243         xprt->reestablish_timeout = 0;
1244 }
1245
1246 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1247 {
1248         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1249                 xprt);
1250         xprt_disconnect_done(xprt);
1251 }
1252
1253 static void xs_xprt_free(struct rpc_xprt *xprt)
1254 {
1255         xs_free_peer_addresses(xprt);
1256         xprt_free(xprt);
1257 }
1258
1259 /**
1260  * xs_destroy - prepare to shutdown a transport
1261  * @xprt: doomed transport
1262  *
1263  */
1264 static void xs_destroy(struct rpc_xprt *xprt)
1265 {
1266         struct sock_xprt *transport = container_of(xprt,
1267                         struct sock_xprt, xprt);
1268         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1269
1270         cancel_delayed_work_sync(&transport->connect_worker);
1271         xs_close(xprt);
1272         cancel_work_sync(&transport->recv_worker);
1273         cancel_work_sync(&transport->error_worker);
1274         xs_xprt_free(xprt);
1275         module_put(THIS_MODULE);
1276 }
1277
1278 /**
1279  * xs_udp_data_read_skb - receive callback for UDP sockets
1280  * @xprt: transport
1281  * @sk: socket
1282  * @skb: skbuff
1283  *
1284  */
1285 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1286                 struct sock *sk,
1287                 struct sk_buff *skb)
1288 {
1289         struct rpc_task *task;
1290         struct rpc_rqst *rovr;
1291         int repsize, copied;
1292         u32 _xid;
1293         __be32 *xp;
1294
1295         repsize = skb->len;
1296         if (repsize < 4) {
1297                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1298                 return;
1299         }
1300
1301         /* Copy the XID from the skb... */
1302         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1303         if (xp == NULL)
1304                 return;
1305
1306         /* Look up and lock the request corresponding to the given XID */
1307         spin_lock(&xprt->queue_lock);
1308         rovr = xprt_lookup_rqst(xprt, *xp);
1309         if (!rovr)
1310                 goto out_unlock;
1311         xprt_pin_rqst(rovr);
1312         xprt_update_rtt(rovr->rq_task);
1313         spin_unlock(&xprt->queue_lock);
1314         task = rovr->rq_task;
1315
1316         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1317                 copied = repsize;
1318
1319         /* Suck it into the iovec, verify checksum if not done by hw. */
1320         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1321                 spin_lock(&xprt->queue_lock);
1322                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1323                 goto out_unpin;
1324         }
1325
1326
1327         spin_lock(&xprt->transport_lock);
1328         xprt_adjust_cwnd(xprt, task, copied);
1329         spin_unlock(&xprt->transport_lock);
1330         spin_lock(&xprt->queue_lock);
1331         xprt_complete_rqst(task, copied);
1332         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1333 out_unpin:
1334         xprt_unpin_rqst(rovr);
1335  out_unlock:
1336         spin_unlock(&xprt->queue_lock);
1337 }
1338
1339 static void xs_udp_data_receive(struct sock_xprt *transport)
1340 {
1341         struct sk_buff *skb;
1342         struct sock *sk;
1343         int err;
1344
1345         mutex_lock(&transport->recv_mutex);
1346         sk = transport->inet;
1347         if (sk == NULL)
1348                 goto out;
1349         for (;;) {
1350                 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1351                 if (skb == NULL)
1352                         break;
1353                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1354                 consume_skb(skb);
1355                 cond_resched();
1356         }
1357         xs_poll_check_readable(transport);
1358 out:
1359         mutex_unlock(&transport->recv_mutex);
1360 }
1361
1362 static void xs_udp_data_receive_workfn(struct work_struct *work)
1363 {
1364         struct sock_xprt *transport =
1365                 container_of(work, struct sock_xprt, recv_worker);
1366         unsigned int pflags = memalloc_nofs_save();
1367
1368         xs_udp_data_receive(transport);
1369         memalloc_nofs_restore(pflags);
1370 }
1371
1372 /**
1373  * xs_data_ready - "data ready" callback for sockets
1374  * @sk: socket with data to read
1375  *
1376  */
1377 static void xs_data_ready(struct sock *sk)
1378 {
1379         struct rpc_xprt *xprt;
1380
1381         xprt = xprt_from_sock(sk);
1382         if (xprt != NULL) {
1383                 struct sock_xprt *transport = container_of(xprt,
1384                                 struct sock_xprt, xprt);
1385
1386                 trace_xs_data_ready(xprt);
1387
1388                 transport->old_data_ready(sk);
1389                 /* Any data means we had a useful conversation, so
1390                  * then we don't need to delay the next reconnect
1391                  */
1392                 if (xprt->reestablish_timeout)
1393                         xprt->reestablish_timeout = 0;
1394                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1395                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1396         }
1397 }
1398
1399 /*
1400  * Helper function to force a TCP close if the server is sending
1401  * junk and/or it has put us in CLOSE_WAIT
1402  */
1403 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1404 {
1405         xprt_force_disconnect(xprt);
1406 }
1407
1408 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1409 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1410 {
1411         return PAGE_SIZE;
1412 }
1413 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1414
1415 /**
1416  * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1417  * @sk: socket whose state has changed
1418  *
1419  */
1420 static void xs_local_state_change(struct sock *sk)
1421 {
1422         struct rpc_xprt *xprt;
1423         struct sock_xprt *transport;
1424
1425         if (!(xprt = xprt_from_sock(sk)))
1426                 return;
1427         transport = container_of(xprt, struct sock_xprt, xprt);
1428         if (sk->sk_shutdown & SHUTDOWN_MASK) {
1429                 clear_bit(XPRT_CONNECTED, &xprt->state);
1430                 /* Trigger the socket release */
1431                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1432         }
1433 }
1434
1435 /**
1436  * xs_tcp_state_change - callback to handle TCP socket state changes
1437  * @sk: socket whose state has changed
1438  *
1439  */
1440 static void xs_tcp_state_change(struct sock *sk)
1441 {
1442         struct rpc_xprt *xprt;
1443         struct sock_xprt *transport;
1444
1445         if (!(xprt = xprt_from_sock(sk)))
1446                 return;
1447         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1448         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1449                         sk->sk_state, xprt_connected(xprt),
1450                         sock_flag(sk, SOCK_DEAD),
1451                         sock_flag(sk, SOCK_ZAPPED),
1452                         sk->sk_shutdown);
1453
1454         transport = container_of(xprt, struct sock_xprt, xprt);
1455         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1456         switch (sk->sk_state) {
1457         case TCP_ESTABLISHED:
1458                 if (!xprt_test_and_set_connected(xprt)) {
1459                         xprt->connect_cookie++;
1460                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1461                         xprt_clear_connecting(xprt);
1462
1463                         xprt->stat.connect_count++;
1464                         xprt->stat.connect_time += (long)jiffies -
1465                                                    xprt->stat.connect_start;
1466                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1467                 }
1468                 break;
1469         case TCP_FIN_WAIT1:
1470                 /* The client initiated a shutdown of the socket */
1471                 xprt->connect_cookie++;
1472                 xprt->reestablish_timeout = 0;
1473                 set_bit(XPRT_CLOSING, &xprt->state);
1474                 smp_mb__before_atomic();
1475                 clear_bit(XPRT_CONNECTED, &xprt->state);
1476                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1477                 smp_mb__after_atomic();
1478                 break;
1479         case TCP_CLOSE_WAIT:
1480                 /* The server initiated a shutdown of the socket */
1481                 xprt->connect_cookie++;
1482                 clear_bit(XPRT_CONNECTED, &xprt->state);
1483                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1484                 fallthrough;
1485         case TCP_CLOSING:
1486                 /*
1487                  * If the server closed down the connection, make sure that
1488                  * we back off before reconnecting
1489                  */
1490                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1491                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1492                 break;
1493         case TCP_LAST_ACK:
1494                 set_bit(XPRT_CLOSING, &xprt->state);
1495                 smp_mb__before_atomic();
1496                 clear_bit(XPRT_CONNECTED, &xprt->state);
1497                 smp_mb__after_atomic();
1498                 break;
1499         case TCP_CLOSE:
1500                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1501                                         &transport->sock_state))
1502                         xprt_clear_connecting(xprt);
1503                 clear_bit(XPRT_CLOSING, &xprt->state);
1504                 /* Trigger the socket release */
1505                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1506         }
1507 }
1508
1509 static void xs_write_space(struct sock *sk)
1510 {
1511         struct sock_xprt *transport;
1512         struct rpc_xprt *xprt;
1513
1514         if (!sk->sk_socket)
1515                 return;
1516         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1517
1518         if (unlikely(!(xprt = xprt_from_sock(sk))))
1519                 return;
1520         transport = container_of(xprt, struct sock_xprt, xprt);
1521         if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1522                 return;
1523         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1524         sk->sk_write_pending--;
1525 }
1526
1527 /**
1528  * xs_udp_write_space - callback invoked when socket buffer space
1529  *                             becomes available
1530  * @sk: socket whose state has changed
1531  *
1532  * Called when more output buffer space is available for this socket.
1533  * We try not to wake our writers until they can make "significant"
1534  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1535  * with a bunch of small requests.
1536  */
1537 static void xs_udp_write_space(struct sock *sk)
1538 {
1539         /* from net/core/sock.c:sock_def_write_space */
1540         if (sock_writeable(sk))
1541                 xs_write_space(sk);
1542 }
1543
1544 /**
1545  * xs_tcp_write_space - callback invoked when socket buffer space
1546  *                             becomes available
1547  * @sk: socket whose state has changed
1548  *
1549  * Called when more output buffer space is available for this socket.
1550  * We try not to wake our writers until they can make "significant"
1551  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1552  * with a bunch of small requests.
1553  */
1554 static void xs_tcp_write_space(struct sock *sk)
1555 {
1556         /* from net/core/stream.c:sk_stream_write_space */
1557         if (sk_stream_is_writeable(sk))
1558                 xs_write_space(sk);
1559 }
1560
1561 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1562 {
1563         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1564         struct sock *sk = transport->inet;
1565
1566         if (transport->rcvsize) {
1567                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1568                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1569         }
1570         if (transport->sndsize) {
1571                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1572                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1573                 sk->sk_write_space(sk);
1574         }
1575 }
1576
1577 /**
1578  * xs_udp_set_buffer_size - set send and receive limits
1579  * @xprt: generic transport
1580  * @sndsize: requested size of send buffer, in bytes
1581  * @rcvsize: requested size of receive buffer, in bytes
1582  *
1583  * Set socket send and receive buffer size limits.
1584  */
1585 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1586 {
1587         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1588
1589         transport->sndsize = 0;
1590         if (sndsize)
1591                 transport->sndsize = sndsize + 1024;
1592         transport->rcvsize = 0;
1593         if (rcvsize)
1594                 transport->rcvsize = rcvsize + 1024;
1595
1596         xs_udp_do_set_buffer_size(xprt);
1597 }
1598
1599 /**
1600  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1601  * @xprt: controlling transport
1602  * @task: task that timed out
1603  *
1604  * Adjust the congestion window after a retransmit timeout has occurred.
1605  */
1606 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1607 {
1608         spin_lock(&xprt->transport_lock);
1609         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1610         spin_unlock(&xprt->transport_lock);
1611 }
1612
1613 static int xs_get_random_port(void)
1614 {
1615         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1616         unsigned short range;
1617         unsigned short rand;
1618
1619         if (max < min)
1620                 return -EADDRINUSE;
1621         range = max - min + 1;
1622         rand = (unsigned short) prandom_u32() % range;
1623         return rand + min;
1624 }
1625
1626 static unsigned short xs_sock_getport(struct socket *sock)
1627 {
1628         struct sockaddr_storage buf;
1629         unsigned short port = 0;
1630
1631         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1632                 goto out;
1633         switch (buf.ss_family) {
1634         case AF_INET6:
1635                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1636                 break;
1637         case AF_INET:
1638                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1639         }
1640 out:
1641         return port;
1642 }
1643
1644 /**
1645  * xs_set_port - reset the port number in the remote endpoint address
1646  * @xprt: generic transport
1647  * @port: new port number
1648  *
1649  */
1650 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1651 {
1652         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1653
1654         rpc_set_port(xs_addr(xprt), port);
1655         xs_update_peer_port(xprt);
1656 }
1657
1658 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1659 {
1660         if (transport->srcport == 0 && transport->xprt.reuseport)
1661                 transport->srcport = xs_sock_getport(sock);
1662 }
1663
1664 static int xs_get_srcport(struct sock_xprt *transport)
1665 {
1666         int port = transport->srcport;
1667
1668         if (port == 0 && transport->xprt.resvport)
1669                 port = xs_get_random_port();
1670         return port;
1671 }
1672
1673 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1674 {
1675         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1676         unsigned short ret = 0;
1677         mutex_lock(&sock->recv_mutex);
1678         if (sock->sock)
1679                 ret = xs_sock_getport(sock->sock);
1680         mutex_unlock(&sock->recv_mutex);
1681         return ret;
1682 }
1683
1684 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1685 {
1686         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1687         union {
1688                 struct sockaddr sa;
1689                 struct sockaddr_storage st;
1690         } saddr;
1691         int ret = -ENOTCONN;
1692
1693         mutex_lock(&sock->recv_mutex);
1694         if (sock->sock) {
1695                 ret = kernel_getsockname(sock->sock, &saddr.sa);
1696                 if (ret >= 0)
1697                         ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1698         }
1699         mutex_unlock(&sock->recv_mutex);
1700         return ret;
1701 }
1702
1703 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1704 {
1705         if (transport->srcport != 0)
1706                 transport->srcport = 0;
1707         if (!transport->xprt.resvport)
1708                 return 0;
1709         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1710                 return xprt_max_resvport;
1711         return --port;
1712 }
1713 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1714 {
1715         struct sockaddr_storage myaddr;
1716         int err, nloop = 0;
1717         int port = xs_get_srcport(transport);
1718         unsigned short last;
1719
1720         /*
1721          * If we are asking for any ephemeral port (i.e. port == 0 &&
1722          * transport->xprt.resvport == 0), don't bind.  Let the local
1723          * port selection happen implicitly when the socket is used
1724          * (for example at connect time).
1725          *
1726          * This ensures that we can continue to establish TCP
1727          * connections even when all local ephemeral ports are already
1728          * a part of some TCP connection.  This makes no difference
1729          * for UDP sockets, but also doesn't harm them.
1730          *
1731          * If we're asking for any reserved port (i.e. port == 0 &&
1732          * transport->xprt.resvport == 1) xs_get_srcport above will
1733          * ensure that port is non-zero and we will bind as needed.
1734          */
1735         if (port <= 0)
1736                 return port;
1737
1738         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1739         do {
1740                 rpc_set_port((struct sockaddr *)&myaddr, port);
1741                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1742                                 transport->xprt.addrlen);
1743                 if (err == 0) {
1744                         if (transport->xprt.reuseport)
1745                                 transport->srcport = port;
1746                         break;
1747                 }
1748                 last = port;
1749                 port = xs_next_srcport(transport, port);
1750                 if (port > last)
1751                         nloop++;
1752         } while (err == -EADDRINUSE && nloop != 2);
1753
1754         if (myaddr.ss_family == AF_INET)
1755                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1756                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1757                                 port, err ? "failed" : "ok", err);
1758         else
1759                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1760                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1761                                 port, err ? "failed" : "ok", err);
1762         return err;
1763 }
1764
1765 /*
1766  * We don't support autobind on AF_LOCAL sockets
1767  */
1768 static void xs_local_rpcbind(struct rpc_task *task)
1769 {
1770         xprt_set_bound(task->tk_xprt);
1771 }
1772
1773 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1774 {
1775 }
1776
1777 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1778 static struct lock_class_key xs_key[3];
1779 static struct lock_class_key xs_slock_key[3];
1780
1781 static inline void xs_reclassify_socketu(struct socket *sock)
1782 {
1783         struct sock *sk = sock->sk;
1784
1785         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1786                 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1787 }
1788
1789 static inline void xs_reclassify_socket4(struct socket *sock)
1790 {
1791         struct sock *sk = sock->sk;
1792
1793         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1794                 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1795 }
1796
1797 static inline void xs_reclassify_socket6(struct socket *sock)
1798 {
1799         struct sock *sk = sock->sk;
1800
1801         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1802                 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1803 }
1804
1805 static inline void xs_reclassify_socket(int family, struct socket *sock)
1806 {
1807         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1808                 return;
1809
1810         switch (family) {
1811         case AF_LOCAL:
1812                 xs_reclassify_socketu(sock);
1813                 break;
1814         case AF_INET:
1815                 xs_reclassify_socket4(sock);
1816                 break;
1817         case AF_INET6:
1818                 xs_reclassify_socket6(sock);
1819                 break;
1820         }
1821 }
1822 #else
1823 static inline void xs_reclassify_socket(int family, struct socket *sock)
1824 {
1825 }
1826 #endif
1827
1828 static void xs_dummy_setup_socket(struct work_struct *work)
1829 {
1830 }
1831
1832 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1833                 struct sock_xprt *transport, int family, int type,
1834                 int protocol, bool reuseport)
1835 {
1836         struct file *filp;
1837         struct socket *sock;
1838         int err;
1839
1840         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1841         if (err < 0) {
1842                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1843                                 protocol, -err);
1844                 goto out;
1845         }
1846         xs_reclassify_socket(family, sock);
1847
1848         if (reuseport)
1849                 sock_set_reuseport(sock->sk);
1850
1851         err = xs_bind(transport, sock);
1852         if (err) {
1853                 sock_release(sock);
1854                 goto out;
1855         }
1856
1857         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1858         if (IS_ERR(filp))
1859                 return ERR_CAST(filp);
1860         transport->file = filp;
1861
1862         return sock;
1863 out:
1864         return ERR_PTR(err);
1865 }
1866
1867 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1868                                       struct socket *sock)
1869 {
1870         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1871                                                                         xprt);
1872
1873         if (!transport->inet) {
1874                 struct sock *sk = sock->sk;
1875
1876                 lock_sock(sk);
1877
1878                 xs_save_old_callbacks(transport, sk);
1879
1880                 sk->sk_user_data = xprt;
1881                 sk->sk_data_ready = xs_data_ready;
1882                 sk->sk_write_space = xs_udp_write_space;
1883                 sk->sk_state_change = xs_local_state_change;
1884                 sk->sk_error_report = xs_error_report;
1885
1886                 xprt_clear_connected(xprt);
1887
1888                 /* Reset to new socket */
1889                 transport->sock = sock;
1890                 transport->inet = sk;
1891
1892                 release_sock(sk);
1893         }
1894
1895         xs_stream_start_connect(transport);
1896
1897         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1898 }
1899
1900 /**
1901  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1902  * @transport: socket transport to connect
1903  */
1904 static int xs_local_setup_socket(struct sock_xprt *transport)
1905 {
1906         struct rpc_xprt *xprt = &transport->xprt;
1907         struct file *filp;
1908         struct socket *sock;
1909         int status;
1910
1911         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1912                                         SOCK_STREAM, 0, &sock, 1);
1913         if (status < 0) {
1914                 dprintk("RPC:       can't create AF_LOCAL "
1915                         "transport socket (%d).\n", -status);
1916                 goto out;
1917         }
1918         xs_reclassify_socket(AF_LOCAL, sock);
1919
1920         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1921         if (IS_ERR(filp)) {
1922                 status = PTR_ERR(filp);
1923                 goto out;
1924         }
1925         transport->file = filp;
1926
1927         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1928                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1929
1930         status = xs_local_finish_connecting(xprt, sock);
1931         trace_rpc_socket_connect(xprt, sock, status);
1932         switch (status) {
1933         case 0:
1934                 dprintk("RPC:       xprt %p connected to %s\n",
1935                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1936                 xprt->stat.connect_count++;
1937                 xprt->stat.connect_time += (long)jiffies -
1938                                            xprt->stat.connect_start;
1939                 xprt_set_connected(xprt);
1940                 break;
1941         case -ENOBUFS:
1942                 break;
1943         case -ENOENT:
1944                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1945                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1946                 break;
1947         case -ECONNREFUSED:
1948                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1949                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1950                 break;
1951         default:
1952                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1953                                 __func__, -status,
1954                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1955         }
1956
1957 out:
1958         xprt_clear_connecting(xprt);
1959         xprt_wake_pending_tasks(xprt, status);
1960         return status;
1961 }
1962
1963 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1964 {
1965         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1966         int ret;
1967
1968         if (transport->file)
1969                 goto force_disconnect;
1970
1971         if (RPC_IS_ASYNC(task)) {
1972                 /*
1973                  * We want the AF_LOCAL connect to be resolved in the
1974                  * filesystem namespace of the process making the rpc
1975                  * call.  Thus we connect synchronously.
1976                  *
1977                  * If we want to support asynchronous AF_LOCAL calls,
1978                  * we'll need to figure out how to pass a namespace to
1979                  * connect.
1980                  */
1981                 task->tk_rpc_status = -ENOTCONN;
1982                 rpc_exit(task, -ENOTCONN);
1983                 goto out_wake;
1984         }
1985         ret = xs_local_setup_socket(transport);
1986         if (ret && !RPC_IS_SOFTCONN(task))
1987                 msleep_interruptible(15000);
1988         return;
1989 force_disconnect:
1990         xprt_force_disconnect(xprt);
1991 out_wake:
1992         xprt_clear_connecting(xprt);
1993         xprt_wake_pending_tasks(xprt, -ENOTCONN);
1994 }
1995
1996 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1997 /*
1998  * Note that this should be called with XPRT_LOCKED held, or recv_mutex
1999  * held, or when we otherwise know that we have exclusive access to the
2000  * socket, to guard against races with xs_reset_transport.
2001  */
2002 static void xs_set_memalloc(struct rpc_xprt *xprt)
2003 {
2004         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2005                         xprt);
2006
2007         /*
2008          * If there's no sock, then we have nothing to set. The
2009          * reconnecting process will get it for us.
2010          */
2011         if (!transport->inet)
2012                 return;
2013         if (atomic_read(&xprt->swapper))
2014                 sk_set_memalloc(transport->inet);
2015 }
2016
2017 /**
2018  * xs_enable_swap - Tag this transport as being used for swap.
2019  * @xprt: transport to tag
2020  *
2021  * Take a reference to this transport on behalf of the rpc_clnt, and
2022  * optionally mark it for swapping if it wasn't already.
2023  */
2024 static int
2025 xs_enable_swap(struct rpc_xprt *xprt)
2026 {
2027         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2028
2029         mutex_lock(&xs->recv_mutex);
2030         if (atomic_inc_return(&xprt->swapper) == 1 &&
2031             xs->inet)
2032                 sk_set_memalloc(xs->inet);
2033         mutex_unlock(&xs->recv_mutex);
2034         return 0;
2035 }
2036
2037 /**
2038  * xs_disable_swap - Untag this transport as being used for swap.
2039  * @xprt: transport to tag
2040  *
2041  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2042  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2043  */
2044 static void
2045 xs_disable_swap(struct rpc_xprt *xprt)
2046 {
2047         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2048
2049         mutex_lock(&xs->recv_mutex);
2050         if (atomic_dec_and_test(&xprt->swapper) &&
2051             xs->inet)
2052                 sk_clear_memalloc(xs->inet);
2053         mutex_unlock(&xs->recv_mutex);
2054 }
2055 #else
2056 static void xs_set_memalloc(struct rpc_xprt *xprt)
2057 {
2058 }
2059
2060 static int
2061 xs_enable_swap(struct rpc_xprt *xprt)
2062 {
2063         return -EINVAL;
2064 }
2065
2066 static void
2067 xs_disable_swap(struct rpc_xprt *xprt)
2068 {
2069 }
2070 #endif
2071
2072 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2073 {
2074         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2075
2076         if (!transport->inet) {
2077                 struct sock *sk = sock->sk;
2078
2079                 lock_sock(sk);
2080
2081                 xs_save_old_callbacks(transport, sk);
2082
2083                 sk->sk_user_data = xprt;
2084                 sk->sk_data_ready = xs_data_ready;
2085                 sk->sk_write_space = xs_udp_write_space;
2086
2087                 xprt_set_connected(xprt);
2088
2089                 /* Reset to new socket */
2090                 transport->sock = sock;
2091                 transport->inet = sk;
2092
2093                 xs_set_memalloc(xprt);
2094
2095                 release_sock(sk);
2096         }
2097         xs_udp_do_set_buffer_size(xprt);
2098
2099         xprt->stat.connect_start = jiffies;
2100 }
2101
2102 static void xs_udp_setup_socket(struct work_struct *work)
2103 {
2104         struct sock_xprt *transport =
2105                 container_of(work, struct sock_xprt, connect_worker.work);
2106         struct rpc_xprt *xprt = &transport->xprt;
2107         struct socket *sock;
2108         int status = -EIO;
2109         unsigned int pflags = current->flags;
2110
2111         if (atomic_read(&xprt->swapper))
2112                 current->flags |= PF_MEMALLOC;
2113         sock = xs_create_sock(xprt, transport,
2114                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2115                         IPPROTO_UDP, false);
2116         if (IS_ERR(sock))
2117                 goto out;
2118
2119         dprintk("RPC:       worker connecting xprt %p via %s to "
2120                                 "%s (port %s)\n", xprt,
2121                         xprt->address_strings[RPC_DISPLAY_PROTO],
2122                         xprt->address_strings[RPC_DISPLAY_ADDR],
2123                         xprt->address_strings[RPC_DISPLAY_PORT]);
2124
2125         xs_udp_finish_connecting(xprt, sock);
2126         trace_rpc_socket_connect(xprt, sock, 0);
2127         status = 0;
2128 out:
2129         xprt_clear_connecting(xprt);
2130         xprt_unlock_connect(xprt, transport);
2131         xprt_wake_pending_tasks(xprt, status);
2132         current_restore_flags(pflags, PF_MEMALLOC);
2133 }
2134
2135 /**
2136  * xs_tcp_shutdown - gracefully shut down a TCP socket
2137  * @xprt: transport
2138  *
2139  * Initiates a graceful shutdown of the TCP socket by calling the
2140  * equivalent of shutdown(SHUT_RDWR);
2141  */
2142 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2143 {
2144         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2145         struct socket *sock = transport->sock;
2146         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2147
2148         if (sock == NULL)
2149                 return;
2150         if (!xprt->reuseport) {
2151                 xs_close(xprt);
2152                 return;
2153         }
2154         switch (skst) {
2155         case TCP_FIN_WAIT1:
2156         case TCP_FIN_WAIT2:
2157                 break;
2158         case TCP_ESTABLISHED:
2159         case TCP_CLOSE_WAIT:
2160                 kernel_sock_shutdown(sock, SHUT_RDWR);
2161                 trace_rpc_socket_shutdown(xprt, sock);
2162                 break;
2163         default:
2164                 xs_reset_transport(transport);
2165         }
2166 }
2167
2168 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2169                 struct socket *sock)
2170 {
2171         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2172         unsigned int keepidle;
2173         unsigned int keepcnt;
2174         unsigned int timeo;
2175
2176         spin_lock(&xprt->transport_lock);
2177         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2178         keepcnt = xprt->timeout->to_retries + 1;
2179         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2180                 (xprt->timeout->to_retries + 1);
2181         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2182         spin_unlock(&xprt->transport_lock);
2183
2184         /* TCP Keepalive options */
2185         sock_set_keepalive(sock->sk);
2186         tcp_sock_set_keepidle(sock->sk, keepidle);
2187         tcp_sock_set_keepintvl(sock->sk, keepidle);
2188         tcp_sock_set_keepcnt(sock->sk, keepcnt);
2189
2190         /* TCP user timeout (see RFC5482) */
2191         tcp_sock_set_user_timeout(sock->sk, timeo);
2192 }
2193
2194 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2195                 unsigned long connect_timeout,
2196                 unsigned long reconnect_timeout)
2197 {
2198         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2199         struct rpc_timeout to;
2200         unsigned long initval;
2201
2202         spin_lock(&xprt->transport_lock);
2203         if (reconnect_timeout < xprt->max_reconnect_timeout)
2204                 xprt->max_reconnect_timeout = reconnect_timeout;
2205         if (connect_timeout < xprt->connect_timeout) {
2206                 memcpy(&to, xprt->timeout, sizeof(to));
2207                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2208                 /* Arbitrary lower limit */
2209                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2210                         initval = XS_TCP_INIT_REEST_TO << 1;
2211                 to.to_initval = initval;
2212                 to.to_maxval = initval;
2213                 memcpy(&transport->tcp_timeout, &to,
2214                                 sizeof(transport->tcp_timeout));
2215                 xprt->timeout = &transport->tcp_timeout;
2216                 xprt->connect_timeout = connect_timeout;
2217         }
2218         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2219         spin_unlock(&xprt->transport_lock);
2220 }
2221
2222 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2223 {
2224         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2225
2226         if (!transport->inet) {
2227                 struct sock *sk = sock->sk;
2228
2229                 /* Avoid temporary address, they are bad for long-lived
2230                  * connections such as NFS mounts.
2231                  * RFC4941, section 3.6 suggests that:
2232                  *    Individual applications, which have specific
2233                  *    knowledge about the normal duration of connections,
2234                  *    MAY override this as appropriate.
2235                  */
2236                 if (xs_addr(xprt)->sa_family == PF_INET6) {
2237                         ip6_sock_set_addr_preferences(sk,
2238                                 IPV6_PREFER_SRC_PUBLIC);
2239                 }
2240
2241                 xs_tcp_set_socket_timeouts(xprt, sock);
2242                 tcp_sock_set_nodelay(sk);
2243
2244                 lock_sock(sk);
2245
2246                 xs_save_old_callbacks(transport, sk);
2247
2248                 sk->sk_user_data = xprt;
2249                 sk->sk_data_ready = xs_data_ready;
2250                 sk->sk_state_change = xs_tcp_state_change;
2251                 sk->sk_write_space = xs_tcp_write_space;
2252                 sk->sk_error_report = xs_error_report;
2253
2254                 /* socket options */
2255                 sock_reset_flag(sk, SOCK_LINGER);
2256
2257                 xprt_clear_connected(xprt);
2258
2259                 /* Reset to new socket */
2260                 transport->sock = sock;
2261                 transport->inet = sk;
2262
2263                 release_sock(sk);
2264         }
2265
2266         if (!xprt_bound(xprt))
2267                 return -ENOTCONN;
2268
2269         xs_set_memalloc(xprt);
2270
2271         xs_stream_start_connect(transport);
2272
2273         /* Tell the socket layer to start connecting... */
2274         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2275         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2276 }
2277
2278 /**
2279  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2280  * @work: queued work item
2281  *
2282  * Invoked by a work queue tasklet.
2283  */
2284 static void xs_tcp_setup_socket(struct work_struct *work)
2285 {
2286         struct sock_xprt *transport =
2287                 container_of(work, struct sock_xprt, connect_worker.work);
2288         struct socket *sock = transport->sock;
2289         struct rpc_xprt *xprt = &transport->xprt;
2290         int status;
2291         unsigned int pflags = current->flags;
2292
2293         if (atomic_read(&xprt->swapper))
2294                 current->flags |= PF_MEMALLOC;
2295
2296         if (xprt_connected(xprt))
2297                 goto out;
2298         if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2299                                &transport->sock_state) ||
2300             !sock) {
2301                 xs_reset_transport(transport);
2302                 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2303                                       SOCK_STREAM, IPPROTO_TCP, true);
2304                 if (IS_ERR(sock)) {
2305                         xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2306                         goto out;
2307                 }
2308         }
2309
2310         dprintk("RPC:       worker connecting xprt %p via %s to "
2311                                 "%s (port %s)\n", xprt,
2312                         xprt->address_strings[RPC_DISPLAY_PROTO],
2313                         xprt->address_strings[RPC_DISPLAY_ADDR],
2314                         xprt->address_strings[RPC_DISPLAY_PORT]);
2315
2316         status = xs_tcp_finish_connecting(xprt, sock);
2317         trace_rpc_socket_connect(xprt, sock, status);
2318         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2319                         xprt, -status, xprt_connected(xprt),
2320                         sock->sk->sk_state);
2321         switch (status) {
2322         case 0:
2323         case -EINPROGRESS:
2324                 /* SYN_SENT! */
2325                 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2326                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2327                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2328                 fallthrough;
2329         case -EALREADY:
2330                 goto out_unlock;
2331         case -EADDRNOTAVAIL:
2332                 /* Source port number is unavailable. Try a new one! */
2333                 transport->srcport = 0;
2334                 status = -EAGAIN;
2335                 break;
2336         case -EINVAL:
2337                 /* Happens, for instance, if the user specified a link
2338                  * local IPv6 address without a scope-id.
2339                  */
2340         case -ECONNREFUSED:
2341         case -ECONNRESET:
2342         case -ENETDOWN:
2343         case -ENETUNREACH:
2344         case -EHOSTUNREACH:
2345         case -EADDRINUSE:
2346         case -ENOBUFS:
2347                 break;
2348         default:
2349                 printk("%s: connect returned unhandled error %d\n",
2350                         __func__, status);
2351                 status = -EAGAIN;
2352         }
2353
2354         /* xs_tcp_force_close() wakes tasks with a fixed error code.
2355          * We need to wake them first to ensure the correct error code.
2356          */
2357         xprt_wake_pending_tasks(xprt, status);
2358         xs_tcp_force_close(xprt);
2359 out:
2360         xprt_clear_connecting(xprt);
2361 out_unlock:
2362         xprt_unlock_connect(xprt, transport);
2363         current_restore_flags(pflags, PF_MEMALLOC);
2364 }
2365
2366 /**
2367  * xs_connect - connect a socket to a remote endpoint
2368  * @xprt: pointer to transport structure
2369  * @task: address of RPC task that manages state of connect request
2370  *
2371  * TCP: If the remote end dropped the connection, delay reconnecting.
2372  *
2373  * UDP socket connects are synchronous, but we use a work queue anyway
2374  * to guarantee that even unprivileged user processes can set up a
2375  * socket on a privileged port.
2376  *
2377  * If a UDP socket connect fails, the delay behavior here prevents
2378  * retry floods (hard mounts).
2379  */
2380 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2381 {
2382         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2383         unsigned long delay = 0;
2384
2385         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2386
2387         if (transport->sock != NULL) {
2388                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2389                         "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2390
2391                 delay = xprt_reconnect_delay(xprt);
2392                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2393
2394         } else
2395                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2396
2397         queue_delayed_work(xprtiod_workqueue,
2398                         &transport->connect_worker,
2399                         delay);
2400 }
2401
2402 static void xs_wake_disconnect(struct sock_xprt *transport)
2403 {
2404         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2405                 xs_tcp_force_close(&transport->xprt);
2406 }
2407
2408 static void xs_wake_write(struct sock_xprt *transport)
2409 {
2410         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2411                 xprt_write_space(&transport->xprt);
2412 }
2413
2414 static void xs_wake_error(struct sock_xprt *transport)
2415 {
2416         int sockerr;
2417
2418         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2419                 return;
2420         mutex_lock(&transport->recv_mutex);
2421         if (transport->sock == NULL)
2422                 goto out;
2423         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2424                 goto out;
2425         sockerr = xchg(&transport->xprt_err, 0);
2426         if (sockerr < 0)
2427                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2428 out:
2429         mutex_unlock(&transport->recv_mutex);
2430 }
2431
2432 static void xs_wake_pending(struct sock_xprt *transport)
2433 {
2434         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2435                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2436 }
2437
2438 static void xs_error_handle(struct work_struct *work)
2439 {
2440         struct sock_xprt *transport = container_of(work,
2441                         struct sock_xprt, error_worker);
2442
2443         xs_wake_disconnect(transport);
2444         xs_wake_write(transport);
2445         xs_wake_error(transport);
2446         xs_wake_pending(transport);
2447 }
2448
2449 /**
2450  * xs_local_print_stats - display AF_LOCAL socket-specific stats
2451  * @xprt: rpc_xprt struct containing statistics
2452  * @seq: output file
2453  *
2454  */
2455 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2456 {
2457         long idle_time = 0;
2458
2459         if (xprt_connected(xprt))
2460                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2461
2462         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2463                         "%llu %llu %lu %llu %llu\n",
2464                         xprt->stat.bind_count,
2465                         xprt->stat.connect_count,
2466                         xprt->stat.connect_time / HZ,
2467                         idle_time,
2468                         xprt->stat.sends,
2469                         xprt->stat.recvs,
2470                         xprt->stat.bad_xids,
2471                         xprt->stat.req_u,
2472                         xprt->stat.bklog_u,
2473                         xprt->stat.max_slots,
2474                         xprt->stat.sending_u,
2475                         xprt->stat.pending_u);
2476 }
2477
2478 /**
2479  * xs_udp_print_stats - display UDP socket-specific stats
2480  * @xprt: rpc_xprt struct containing statistics
2481  * @seq: output file
2482  *
2483  */
2484 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2485 {
2486         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2487
2488         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2489                         "%lu %llu %llu\n",
2490                         transport->srcport,
2491                         xprt->stat.bind_count,
2492                         xprt->stat.sends,
2493                         xprt->stat.recvs,
2494                         xprt->stat.bad_xids,
2495                         xprt->stat.req_u,
2496                         xprt->stat.bklog_u,
2497                         xprt->stat.max_slots,
2498                         xprt->stat.sending_u,
2499                         xprt->stat.pending_u);
2500 }
2501
2502 /**
2503  * xs_tcp_print_stats - display TCP socket-specific stats
2504  * @xprt: rpc_xprt struct containing statistics
2505  * @seq: output file
2506  *
2507  */
2508 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2509 {
2510         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2511         long idle_time = 0;
2512
2513         if (xprt_connected(xprt))
2514                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2515
2516         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2517                         "%llu %llu %lu %llu %llu\n",
2518                         transport->srcport,
2519                         xprt->stat.bind_count,
2520                         xprt->stat.connect_count,
2521                         xprt->stat.connect_time / HZ,
2522                         idle_time,
2523                         xprt->stat.sends,
2524                         xprt->stat.recvs,
2525                         xprt->stat.bad_xids,
2526                         xprt->stat.req_u,
2527                         xprt->stat.bklog_u,
2528                         xprt->stat.max_slots,
2529                         xprt->stat.sending_u,
2530                         xprt->stat.pending_u);
2531 }
2532
2533 /*
2534  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2535  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2536  * to use the server side send routines.
2537  */
2538 static int bc_malloc(struct rpc_task *task)
2539 {
2540         struct rpc_rqst *rqst = task->tk_rqstp;
2541         size_t size = rqst->rq_callsize;
2542         struct page *page;
2543         struct rpc_buffer *buf;
2544
2545         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2546                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2547                           size);
2548                 return -EINVAL;
2549         }
2550
2551         page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2552         if (!page)
2553                 return -ENOMEM;
2554
2555         buf = page_address(page);
2556         buf->len = PAGE_SIZE;
2557
2558         rqst->rq_buffer = buf->data;
2559         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2560         return 0;
2561 }
2562
2563 /*
2564  * Free the space allocated in the bc_alloc routine
2565  */
2566 static void bc_free(struct rpc_task *task)
2567 {
2568         void *buffer = task->tk_rqstp->rq_buffer;
2569         struct rpc_buffer *buf;
2570
2571         buf = container_of(buffer, struct rpc_buffer, data);
2572         free_page((unsigned long)buf);
2573 }
2574
2575 static int bc_sendto(struct rpc_rqst *req)
2576 {
2577         struct xdr_buf *xdr = &req->rq_snd_buf;
2578         struct sock_xprt *transport =
2579                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2580         struct msghdr msg = {
2581                 .msg_flags      = 0,
2582         };
2583         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2584                                          (u32)xdr->len);
2585         unsigned int sent = 0;
2586         int err;
2587
2588         req->rq_xtime = ktime_get();
2589         err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2590         if (err < 0)
2591                 return err;
2592         err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2593         xdr_free_bvec(xdr);
2594         if (err < 0 || sent != (xdr->len + sizeof(marker)))
2595                 return -EAGAIN;
2596         return sent;
2597 }
2598
2599 /**
2600  * bc_send_request - Send a backchannel Call on a TCP socket
2601  * @req: rpc_rqst containing Call message to be sent
2602  *
2603  * xpt_mutex ensures @rqstp's whole message is written to the socket
2604  * without interruption.
2605  *
2606  * Return values:
2607  *   %0 if the message was sent successfully
2608  *   %ENOTCONN if the message was not sent
2609  */
2610 static int bc_send_request(struct rpc_rqst *req)
2611 {
2612         struct svc_xprt *xprt;
2613         int len;
2614
2615         /*
2616          * Get the server socket associated with this callback xprt
2617          */
2618         xprt = req->rq_xprt->bc_xprt;
2619
2620         /*
2621          * Grab the mutex to serialize data as the connection is shared
2622          * with the fore channel
2623          */
2624         mutex_lock(&xprt->xpt_mutex);
2625         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2626                 len = -ENOTCONN;
2627         else
2628                 len = bc_sendto(req);
2629         mutex_unlock(&xprt->xpt_mutex);
2630
2631         if (len > 0)
2632                 len = 0;
2633
2634         return len;
2635 }
2636
2637 /*
2638  * The close routine. Since this is client initiated, we do nothing
2639  */
2640
2641 static void bc_close(struct rpc_xprt *xprt)
2642 {
2643         xprt_disconnect_done(xprt);
2644 }
2645
2646 /*
2647  * The xprt destroy routine. Again, because this connection is client
2648  * initiated, we do nothing
2649  */
2650
2651 static void bc_destroy(struct rpc_xprt *xprt)
2652 {
2653         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2654
2655         xs_xprt_free(xprt);
2656         module_put(THIS_MODULE);
2657 }
2658
2659 static const struct rpc_xprt_ops xs_local_ops = {
2660         .reserve_xprt           = xprt_reserve_xprt,
2661         .release_xprt           = xprt_release_xprt,
2662         .alloc_slot             = xprt_alloc_slot,
2663         .free_slot              = xprt_free_slot,
2664         .rpcbind                = xs_local_rpcbind,
2665         .set_port               = xs_local_set_port,
2666         .connect                = xs_local_connect,
2667         .buf_alloc              = rpc_malloc,
2668         .buf_free               = rpc_free,
2669         .prepare_request        = xs_stream_prepare_request,
2670         .send_request           = xs_local_send_request,
2671         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2672         .close                  = xs_close,
2673         .destroy                = xs_destroy,
2674         .print_stats            = xs_local_print_stats,
2675         .enable_swap            = xs_enable_swap,
2676         .disable_swap           = xs_disable_swap,
2677 };
2678
2679 static const struct rpc_xprt_ops xs_udp_ops = {
2680         .set_buffer_size        = xs_udp_set_buffer_size,
2681         .reserve_xprt           = xprt_reserve_xprt_cong,
2682         .release_xprt           = xprt_release_xprt_cong,
2683         .alloc_slot             = xprt_alloc_slot,
2684         .free_slot              = xprt_free_slot,
2685         .rpcbind                = rpcb_getport_async,
2686         .set_port               = xs_set_port,
2687         .connect                = xs_connect,
2688         .get_srcaddr            = xs_sock_srcaddr,
2689         .get_srcport            = xs_sock_srcport,
2690         .buf_alloc              = rpc_malloc,
2691         .buf_free               = rpc_free,
2692         .send_request           = xs_udp_send_request,
2693         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2694         .timer                  = xs_udp_timer,
2695         .release_request        = xprt_release_rqst_cong,
2696         .close                  = xs_close,
2697         .destroy                = xs_destroy,
2698         .print_stats            = xs_udp_print_stats,
2699         .enable_swap            = xs_enable_swap,
2700         .disable_swap           = xs_disable_swap,
2701         .inject_disconnect      = xs_inject_disconnect,
2702 };
2703
2704 static const struct rpc_xprt_ops xs_tcp_ops = {
2705         .reserve_xprt           = xprt_reserve_xprt,
2706         .release_xprt           = xprt_release_xprt,
2707         .alloc_slot             = xprt_alloc_slot,
2708         .free_slot              = xprt_free_slot,
2709         .rpcbind                = rpcb_getport_async,
2710         .set_port               = xs_set_port,
2711         .connect                = xs_connect,
2712         .get_srcaddr            = xs_sock_srcaddr,
2713         .get_srcport            = xs_sock_srcport,
2714         .buf_alloc              = rpc_malloc,
2715         .buf_free               = rpc_free,
2716         .prepare_request        = xs_stream_prepare_request,
2717         .send_request           = xs_tcp_send_request,
2718         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2719         .close                  = xs_tcp_shutdown,
2720         .destroy                = xs_destroy,
2721         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2722         .print_stats            = xs_tcp_print_stats,
2723         .enable_swap            = xs_enable_swap,
2724         .disable_swap           = xs_disable_swap,
2725         .inject_disconnect      = xs_inject_disconnect,
2726 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2727         .bc_setup               = xprt_setup_bc,
2728         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2729         .bc_num_slots           = xprt_bc_max_slots,
2730         .bc_free_rqst           = xprt_free_bc_rqst,
2731         .bc_destroy             = xprt_destroy_bc,
2732 #endif
2733 };
2734
2735 /*
2736  * The rpc_xprt_ops for the server backchannel
2737  */
2738
2739 static const struct rpc_xprt_ops bc_tcp_ops = {
2740         .reserve_xprt           = xprt_reserve_xprt,
2741         .release_xprt           = xprt_release_xprt,
2742         .alloc_slot             = xprt_alloc_slot,
2743         .free_slot              = xprt_free_slot,
2744         .buf_alloc              = bc_malloc,
2745         .buf_free               = bc_free,
2746         .send_request           = bc_send_request,
2747         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2748         .close                  = bc_close,
2749         .destroy                = bc_destroy,
2750         .print_stats            = xs_tcp_print_stats,
2751         .enable_swap            = xs_enable_swap,
2752         .disable_swap           = xs_disable_swap,
2753         .inject_disconnect      = xs_inject_disconnect,
2754 };
2755
2756 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2757 {
2758         static const struct sockaddr_in sin = {
2759                 .sin_family             = AF_INET,
2760                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2761         };
2762         static const struct sockaddr_in6 sin6 = {
2763                 .sin6_family            = AF_INET6,
2764                 .sin6_addr              = IN6ADDR_ANY_INIT,
2765         };
2766
2767         switch (family) {
2768         case AF_LOCAL:
2769                 break;
2770         case AF_INET:
2771                 memcpy(sap, &sin, sizeof(sin));
2772                 break;
2773         case AF_INET6:
2774                 memcpy(sap, &sin6, sizeof(sin6));
2775                 break;
2776         default:
2777                 dprintk("RPC:       %s: Bad address family\n", __func__);
2778                 return -EAFNOSUPPORT;
2779         }
2780         return 0;
2781 }
2782
2783 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2784                                       unsigned int slot_table_size,
2785                                       unsigned int max_slot_table_size)
2786 {
2787         struct rpc_xprt *xprt;
2788         struct sock_xprt *new;
2789
2790         if (args->addrlen > sizeof(xprt->addr)) {
2791                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2792                 return ERR_PTR(-EBADF);
2793         }
2794
2795         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2796                         max_slot_table_size);
2797         if (xprt == NULL) {
2798                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2799                                 "rpc_xprt\n");
2800                 return ERR_PTR(-ENOMEM);
2801         }
2802
2803         new = container_of(xprt, struct sock_xprt, xprt);
2804         mutex_init(&new->recv_mutex);
2805         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2806         xprt->addrlen = args->addrlen;
2807         if (args->srcaddr)
2808                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2809         else {
2810                 int err;
2811                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2812                                         (struct sockaddr *)&new->srcaddr);
2813                 if (err != 0) {
2814                         xprt_free(xprt);
2815                         return ERR_PTR(err);
2816                 }
2817         }
2818
2819         return xprt;
2820 }
2821
2822 static const struct rpc_timeout xs_local_default_timeout = {
2823         .to_initval = 10 * HZ,
2824         .to_maxval = 10 * HZ,
2825         .to_retries = 2,
2826 };
2827
2828 /**
2829  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2830  * @args: rpc transport creation arguments
2831  *
2832  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2833  */
2834 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2835 {
2836         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2837         struct sock_xprt *transport;
2838         struct rpc_xprt *xprt;
2839         struct rpc_xprt *ret;
2840
2841         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2842                         xprt_max_tcp_slot_table_entries);
2843         if (IS_ERR(xprt))
2844                 return xprt;
2845         transport = container_of(xprt, struct sock_xprt, xprt);
2846
2847         xprt->prot = 0;
2848         xprt->xprt_class = &xs_local_transport;
2849         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2850
2851         xprt->bind_timeout = XS_BIND_TO;
2852         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2853         xprt->idle_timeout = XS_IDLE_DISC_TO;
2854
2855         xprt->ops = &xs_local_ops;
2856         xprt->timeout = &xs_local_default_timeout;
2857
2858         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2859         INIT_WORK(&transport->error_worker, xs_error_handle);
2860         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2861
2862         switch (sun->sun_family) {
2863         case AF_LOCAL:
2864                 if (sun->sun_path[0] != '/') {
2865                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2866                                         sun->sun_path);
2867                         ret = ERR_PTR(-EINVAL);
2868                         goto out_err;
2869                 }
2870                 xprt_set_bound(xprt);
2871                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2872                 break;
2873         default:
2874                 ret = ERR_PTR(-EAFNOSUPPORT);
2875                 goto out_err;
2876         }
2877
2878         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2879                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2880
2881         if (try_module_get(THIS_MODULE))
2882                 return xprt;
2883         ret = ERR_PTR(-EINVAL);
2884 out_err:
2885         xs_xprt_free(xprt);
2886         return ret;
2887 }
2888
2889 static const struct rpc_timeout xs_udp_default_timeout = {
2890         .to_initval = 5 * HZ,
2891         .to_maxval = 30 * HZ,
2892         .to_increment = 5 * HZ,
2893         .to_retries = 5,
2894 };
2895
2896 /**
2897  * xs_setup_udp - Set up transport to use a UDP socket
2898  * @args: rpc transport creation arguments
2899  *
2900  */
2901 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2902 {
2903         struct sockaddr *addr = args->dstaddr;
2904         struct rpc_xprt *xprt;
2905         struct sock_xprt *transport;
2906         struct rpc_xprt *ret;
2907
2908         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2909                         xprt_udp_slot_table_entries);
2910         if (IS_ERR(xprt))
2911                 return xprt;
2912         transport = container_of(xprt, struct sock_xprt, xprt);
2913
2914         xprt->prot = IPPROTO_UDP;
2915         xprt->xprt_class = &xs_udp_transport;
2916         /* XXX: header size can vary due to auth type, IPv6, etc. */
2917         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2918
2919         xprt->bind_timeout = XS_BIND_TO;
2920         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2921         xprt->idle_timeout = XS_IDLE_DISC_TO;
2922
2923         xprt->ops = &xs_udp_ops;
2924
2925         xprt->timeout = &xs_udp_default_timeout;
2926
2927         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2928         INIT_WORK(&transport->error_worker, xs_error_handle);
2929         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2930
2931         switch (addr->sa_family) {
2932         case AF_INET:
2933                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2934                         xprt_set_bound(xprt);
2935
2936                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2937                 break;
2938         case AF_INET6:
2939                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2940                         xprt_set_bound(xprt);
2941
2942                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2943                 break;
2944         default:
2945                 ret = ERR_PTR(-EAFNOSUPPORT);
2946                 goto out_err;
2947         }
2948
2949         if (xprt_bound(xprt))
2950                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2951                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2952                                 xprt->address_strings[RPC_DISPLAY_PORT],
2953                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2954         else
2955                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2956                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2957                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2958
2959         if (try_module_get(THIS_MODULE))
2960                 return xprt;
2961         ret = ERR_PTR(-EINVAL);
2962 out_err:
2963         xs_xprt_free(xprt);
2964         return ret;
2965 }
2966
2967 static const struct rpc_timeout xs_tcp_default_timeout = {
2968         .to_initval = 60 * HZ,
2969         .to_maxval = 60 * HZ,
2970         .to_retries = 2,
2971 };
2972
2973 /**
2974  * xs_setup_tcp - Set up transport to use a TCP socket
2975  * @args: rpc transport creation arguments
2976  *
2977  */
2978 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2979 {
2980         struct sockaddr *addr = args->dstaddr;
2981         struct rpc_xprt *xprt;
2982         struct sock_xprt *transport;
2983         struct rpc_xprt *ret;
2984         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2985
2986         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2987                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2988
2989         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2990                         max_slot_table_size);
2991         if (IS_ERR(xprt))
2992                 return xprt;
2993         transport = container_of(xprt, struct sock_xprt, xprt);
2994
2995         xprt->prot = IPPROTO_TCP;
2996         xprt->xprt_class = &xs_tcp_transport;
2997         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2998
2999         xprt->bind_timeout = XS_BIND_TO;
3000         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3001         xprt->idle_timeout = XS_IDLE_DISC_TO;
3002
3003         xprt->ops = &xs_tcp_ops;
3004         xprt->timeout = &xs_tcp_default_timeout;
3005
3006         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3007         xprt->connect_timeout = xprt->timeout->to_initval *
3008                 (xprt->timeout->to_retries + 1);
3009
3010         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3011         INIT_WORK(&transport->error_worker, xs_error_handle);
3012         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3013
3014         switch (addr->sa_family) {
3015         case AF_INET:
3016                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3017                         xprt_set_bound(xprt);
3018
3019                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3020                 break;
3021         case AF_INET6:
3022                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3023                         xprt_set_bound(xprt);
3024
3025                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3026                 break;
3027         default:
3028                 ret = ERR_PTR(-EAFNOSUPPORT);
3029                 goto out_err;
3030         }
3031
3032         if (xprt_bound(xprt))
3033                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3034                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3035                                 xprt->address_strings[RPC_DISPLAY_PORT],
3036                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3037         else
3038                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3039                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3040                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3041
3042         if (try_module_get(THIS_MODULE))
3043                 return xprt;
3044         ret = ERR_PTR(-EINVAL);
3045 out_err:
3046         xs_xprt_free(xprt);
3047         return ret;
3048 }
3049
3050 /**
3051  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3052  * @args: rpc transport creation arguments
3053  *
3054  */
3055 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3056 {
3057         struct sockaddr *addr = args->dstaddr;
3058         struct rpc_xprt *xprt;
3059         struct sock_xprt *transport;
3060         struct svc_sock *bc_sock;
3061         struct rpc_xprt *ret;
3062
3063         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3064                         xprt_tcp_slot_table_entries);
3065         if (IS_ERR(xprt))
3066                 return xprt;
3067         transport = container_of(xprt, struct sock_xprt, xprt);
3068
3069         xprt->prot = IPPROTO_TCP;
3070         xprt->xprt_class = &xs_bc_tcp_transport;
3071         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3072         xprt->timeout = &xs_tcp_default_timeout;
3073
3074         /* backchannel */
3075         xprt_set_bound(xprt);
3076         xprt->bind_timeout = 0;
3077         xprt->reestablish_timeout = 0;
3078         xprt->idle_timeout = 0;
3079
3080         xprt->ops = &bc_tcp_ops;
3081
3082         switch (addr->sa_family) {
3083         case AF_INET:
3084                 xs_format_peer_addresses(xprt, "tcp",
3085                                          RPCBIND_NETID_TCP);
3086                 break;
3087         case AF_INET6:
3088                 xs_format_peer_addresses(xprt, "tcp",
3089                                    RPCBIND_NETID_TCP6);
3090                 break;
3091         default:
3092                 ret = ERR_PTR(-EAFNOSUPPORT);
3093                 goto out_err;
3094         }
3095
3096         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3097                         xprt->address_strings[RPC_DISPLAY_ADDR],
3098                         xprt->address_strings[RPC_DISPLAY_PORT],
3099                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3100
3101         /*
3102          * Once we've associated a backchannel xprt with a connection,
3103          * we want to keep it around as long as the connection lasts,
3104          * in case we need to start using it for a backchannel again;
3105          * this reference won't be dropped until bc_xprt is destroyed.
3106          */
3107         xprt_get(xprt);
3108         args->bc_xprt->xpt_bc_xprt = xprt;
3109         xprt->bc_xprt = args->bc_xprt;
3110         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3111         transport->sock = bc_sock->sk_sock;
3112         transport->inet = bc_sock->sk_sk;
3113
3114         /*
3115          * Since we don't want connections for the backchannel, we set
3116          * the xprt status to connected
3117          */
3118         xprt_set_connected(xprt);
3119
3120         if (try_module_get(THIS_MODULE))
3121                 return xprt;
3122
3123         args->bc_xprt->xpt_bc_xprt = NULL;
3124         args->bc_xprt->xpt_bc_xps = NULL;
3125         xprt_put(xprt);
3126         ret = ERR_PTR(-EINVAL);
3127 out_err:
3128         xs_xprt_free(xprt);
3129         return ret;
3130 }
3131
3132 static struct xprt_class        xs_local_transport = {
3133         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3134         .name           = "named UNIX socket",
3135         .owner          = THIS_MODULE,
3136         .ident          = XPRT_TRANSPORT_LOCAL,
3137         .setup          = xs_setup_local,
3138         .netid          = { "" },
3139 };
3140
3141 static struct xprt_class        xs_udp_transport = {
3142         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3143         .name           = "udp",
3144         .owner          = THIS_MODULE,
3145         .ident          = XPRT_TRANSPORT_UDP,
3146         .setup          = xs_setup_udp,
3147         .netid          = { "udp", "udp6", "" },
3148 };
3149
3150 static struct xprt_class        xs_tcp_transport = {
3151         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3152         .name           = "tcp",
3153         .owner          = THIS_MODULE,
3154         .ident          = XPRT_TRANSPORT_TCP,
3155         .setup          = xs_setup_tcp,
3156         .netid          = { "tcp", "tcp6", "" },
3157 };
3158
3159 static struct xprt_class        xs_bc_tcp_transport = {
3160         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3161         .name           = "tcp NFSv4.1 backchannel",
3162         .owner          = THIS_MODULE,
3163         .ident          = XPRT_TRANSPORT_BC_TCP,
3164         .setup          = xs_setup_bc_tcp,
3165         .netid          = { "" },
3166 };
3167
3168 /**
3169  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3170  *
3171  */
3172 int init_socket_xprt(void)
3173 {
3174         if (!sunrpc_table_header)
3175                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3176
3177         xprt_register_transport(&xs_local_transport);
3178         xprt_register_transport(&xs_udp_transport);
3179         xprt_register_transport(&xs_tcp_transport);
3180         xprt_register_transport(&xs_bc_tcp_transport);
3181
3182         return 0;
3183 }
3184
3185 /**
3186  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3187  *
3188  */
3189 void cleanup_socket_xprt(void)
3190 {
3191         if (sunrpc_table_header) {
3192                 unregister_sysctl_table(sunrpc_table_header);
3193                 sunrpc_table_header = NULL;
3194         }
3195
3196         xprt_unregister_transport(&xs_local_transport);
3197         xprt_unregister_transport(&xs_udp_transport);
3198         xprt_unregister_transport(&xs_tcp_transport);
3199         xprt_unregister_transport(&xs_bc_tcp_transport);
3200 }
3201
3202 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3203 {
3204         return param_set_uint_minmax(val, kp,
3205                         RPC_MIN_RESVPORT,
3206                         RPC_MAX_RESVPORT);
3207 }
3208
3209 static const struct kernel_param_ops param_ops_portnr = {
3210         .set = param_set_portnr,
3211         .get = param_get_uint,
3212 };
3213
3214 #define param_check_portnr(name, p) \
3215         __param_check(name, p, unsigned int);
3216
3217 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3218 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3219
3220 static int param_set_slot_table_size(const char *val,
3221                                      const struct kernel_param *kp)
3222 {
3223         return param_set_uint_minmax(val, kp,
3224                         RPC_MIN_SLOT_TABLE,
3225                         RPC_MAX_SLOT_TABLE);
3226 }
3227
3228 static const struct kernel_param_ops param_ops_slot_table_size = {
3229         .set = param_set_slot_table_size,
3230         .get = param_get_uint,
3231 };
3232
3233 #define param_check_slot_table_size(name, p) \
3234         __param_check(name, p, unsigned int);
3235
3236 static int param_set_max_slot_table_size(const char *val,
3237                                      const struct kernel_param *kp)
3238 {
3239         return param_set_uint_minmax(val, kp,
3240                         RPC_MIN_SLOT_TABLE,
3241                         RPC_MAX_SLOT_TABLE_LIMIT);
3242 }
3243
3244 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3245         .set = param_set_max_slot_table_size,
3246         .get = param_get_uint,
3247 };
3248
3249 #define param_check_max_slot_table_size(name, p) \
3250         __param_check(name, p, unsigned int);
3251
3252 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3253                    slot_table_size, 0644);
3254 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3255                    max_slot_table_size, 0644);
3256 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3257                    slot_table_size, 0644);