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 "sunrpc.h"
  58
  59 static void xs_close(struct rpc_xprt *xprt);
  60 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  61                 struct socket *sock);
  62
  63 /*
  64  * xprtsock tunables
  65  */
  66 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  67 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  68 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  69
  70 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  71 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  72
  73 #define XS_TCP_LINGER_TO        (15U * HZ)
  74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  75
  76 /*
  77  * We can register our own files under /proc/sys/sunrpc by
  78  * calling register_sysctl_table() again.  The files in that
  79  * directory become the union of all files registered there.
  80  *
  81  * We simply need to make sure that we don't collide with
  82  * someone else's file names!
  83  */
  84
  85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  90
  91 static struct ctl_table_header *sunrpc_table_header;
  92
  93 /*
  94  * FIXME: changing the UDP slot table size should also resize the UDP
  95  *        socket buffers for existing UDP transports
  96  */
  97 static struct ctl_table xs_tunables_table[] = {
  98         {
  99                 .procname       = "udp_slot_table_entries",
 100                 .data           = &xprt_udp_slot_table_entries,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = proc_dointvec_minmax,
 104                 .extra1         = &min_slot_table_size,
 105                 .extra2         = &max_slot_table_size
 106         },
 107         {
 108                 .procname       = "tcp_slot_table_entries",
 109                 .data           = &xprt_tcp_slot_table_entries,
 110                 .maxlen         = sizeof(unsigned int),
 111                 .mode           = 0644,
 112                 .proc_handler   = proc_dointvec_minmax,
 113                 .extra1         = &min_slot_table_size,
 114                 .extra2         = &max_slot_table_size
 115         },
 116         {
 117                 .procname       = "tcp_max_slot_table_entries",
 118                 .data           = &xprt_max_tcp_slot_table_entries,
 119                 .maxlen         = sizeof(unsigned int),
 120                 .mode           = 0644,
 121                 .proc_handler   = proc_dointvec_minmax,
 122                 .extra1         = &min_slot_table_size,
 123                 .extra2         = &max_tcp_slot_table_limit
 124         },
 125         {
 126                 .procname       = "min_resvport",
 127                 .data           = &xprt_min_resvport,
 128                 .maxlen         = sizeof(unsigned int),
 129                 .mode           = 0644,
 130                 .proc_handler   = proc_dointvec_minmax,
 131                 .extra1         = &xprt_min_resvport_limit,
 132                 .extra2         = &xprt_max_resvport_limit
 133         },
 134         {
 135                 .procname       = "max_resvport",
 136                 .data           = &xprt_max_resvport,
 137                 .maxlen         = sizeof(unsigned int),
 138                 .mode           = 0644,
 139                 .proc_handler   = proc_dointvec_minmax,
 140                 .extra1         = &xprt_min_resvport_limit,
 141                 .extra2         = &xprt_max_resvport_limit
 142         },
 143         {
 144                 .procname       = "tcp_fin_timeout",
 145                 .data           = &xs_tcp_fin_timeout,
 146                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 147                 .mode           = 0644,
 148                 .proc_handler   = proc_dointvec_jiffies,
 149         },
 150         { },
 151 };
 152
 153 static struct ctl_table sunrpc_table[] = {
 154         {
 155                 .procname       = "sunrpc",
 156                 .mode           = 0555,
 157                 .child          = xs_tunables_table
 158         },
 159         { },
 160 };
 161
 162 /*
 163  * Wait duration for a reply from the RPC portmapper.
 164  */
 165 #define XS_BIND_TO              (60U * HZ)
 166
 167 /*
 168  * Delay if a UDP socket connect error occurs.  This is most likely some
 169  * kind of resource problem on the local host.
 170  */
 171 #define XS_UDP_REEST_TO         (2U * HZ)
 172
 173 /*
 174  * The reestablish timeout allows clients to delay for a bit before attempting
 175  * to reconnect to a server that just dropped our connection.
 176  *
 177  * We implement an exponential backoff when trying to reestablish a TCP
 178  * transport connection with the server.  Some servers like to drop a TCP
 179  * connection when they are overworked, so we start with a short timeout and
 180  * increase over time if the server is down or not responding.
 181  */
 182 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 183
 184 /*
 185  * TCP idle timeout; client drops the transport socket if it is idle
 186  * for this long.  Note that we also timeout UDP sockets to prevent
 187  * holding port numbers when there is no RPC traffic.
 188  */
 189 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 190
 191 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 192 # undef  RPC_DEBUG_DATA
 193 # define RPCDBG_FACILITY        RPCDBG_TRANS
 194 #endif
 195
 196 #ifdef RPC_DEBUG_DATA
 197 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 198 {
 199         u8 *buf = (u8 *) packet;
 200         int j;
 201
 202         dprintk("RPC:       %s\n", msg);
 203         for (j = 0; j < count && j < 128; j += 4) {
 204                 if (!(j & 31)) {
 205                         if (j)
 206                                 dprintk("\n");
 207                         dprintk("0x%04x ", j);
 208                 }
 209                 dprintk("%02x%02x%02x%02x ",
 210                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 211         }
 212         dprintk("\n");
 213 }
 214 #else
 215 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 216 {
 217         /* NOP */
 218 }
 219 #endif
 220
 221 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 222 {
 223         return (struct rpc_xprt *) sk->sk_user_data;
 224 }
 225
 226 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 227 {
 228         return (struct sockaddr *) &xprt->addr;
 229 }
 230
 231 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 232 {
 233         return (struct sockaddr_un *) &xprt->addr;
 234 }
 235
 236 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 237 {
 238         return (struct sockaddr_in *) &xprt->addr;
 239 }
 240
 241 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 242 {
 243         return (struct sockaddr_in6 *) &xprt->addr;
 244 }
 245
 246 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 247 {
 248         struct sockaddr *sap = xs_addr(xprt);
 249         struct sockaddr_in6 *sin6;
 250         struct sockaddr_in *sin;
 251         struct sockaddr_un *sun;
 252         char buf[128];
 253
 254         switch (sap->sa_family) {
 255         case AF_LOCAL:
 256                 sun = xs_addr_un(xprt);
 257                 strlcpy(buf, sun->sun_path, sizeof(buf));
 258                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 259                                                 kstrdup(buf, GFP_KERNEL);
 260                 break;
 261         case AF_INET:
 262                 (void)rpc_ntop(sap, buf, sizeof(buf));
 263                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 264                                                 kstrdup(buf, GFP_KERNEL);
 265                 sin = xs_addr_in(xprt);
 266                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 267                 break;
 268         case AF_INET6:
 269                 (void)rpc_ntop(sap, buf, sizeof(buf));
 270                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 271                                                 kstrdup(buf, GFP_KERNEL);
 272                 sin6 = xs_addr_in6(xprt);
 273                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 274                 break;
 275         default:
 276                 BUG();
 277         }
 278
 279         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 280 }
 281
 282 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 283 {
 284         struct sockaddr *sap = xs_addr(xprt);
 285         char buf[128];
 286
 287         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 288         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 289
 290         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 291         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 292 }
 293
 294 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 295                                      const char *protocol,
 296                                      const char *netid)
 297 {
 298         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 299         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 300         xs_format_common_peer_addresses(xprt);
 301         xs_format_common_peer_ports(xprt);
 302 }
 303
 304 static void xs_update_peer_port(struct rpc_xprt *xprt)
 305 {
 306         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 307         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 308
 309         xs_format_common_peer_ports(xprt);
 310 }
 311
 312 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 313 {
 314         unsigned int i;
 315
 316         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 317                 switch (i) {
 318                 case RPC_DISPLAY_PROTO:
 319                 case RPC_DISPLAY_NETID:
 320                         continue;
 321                 default:
 322                         kfree(xprt->address_strings[i]);
 323                 }
 324 }
 325
 326 static size_t
 327 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 328 {
 329         size_t i,n;
 330
 331         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 332                 return want;
 333         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 334         for (i = 0; i < n; i++) {
 335                 if (buf->pages[i])
 336                         continue;
 337                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 338                 if (!buf->pages[i]) {
 339                         i *= PAGE_SIZE;
 340                         return i > buf->page_base ? i - buf->page_base : 0;
 341                 }
 342         }
 343         return want;
 344 }
 345
 346 static ssize_t
 347 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 348 {
 349         ssize_t ret;
 350         if (seek != 0)
 351                 iov_iter_advance(&msg->msg_iter, seek);
 352         ret = sock_recvmsg(sock, msg, flags);
 353         return ret > 0 ? ret + seek : ret;
 354 }
 355
 356 static ssize_t
 357 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 358                 struct kvec *kvec, size_t count, size_t seek)
 359 {
 360         iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
 361         return xs_sock_recvmsg(sock, msg, flags, seek);
 362 }
 363
 364 static ssize_t
 365 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 366                 struct bio_vec *bvec, unsigned long nr, size_t count,
 367                 size_t seek)
 368 {
 369         iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
 370         return xs_sock_recvmsg(sock, msg, flags, seek);
 371 }
 372
 373 static ssize_t
 374 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 375                 size_t count)
 376 {
 377         iov_iter_discard(&msg->msg_iter, READ, count);
 378         return sock_recvmsg(sock, msg, flags);
 379 }
 380
 381 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 382 static void
 383 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 384 {
 385         struct bvec_iter bi = {
 386                 .bi_size = count,
 387         };
 388         struct bio_vec bv;
 389
 390         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 391         for_each_bvec(bv, bvec, bi, bi)
 392                 flush_dcache_page(bv.bv_page);
 393 }
 394 #else
 395 static inline void
 396 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 397 {
 398 }
 399 #endif
 400
 401 static ssize_t
 402 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 403                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 404 {
 405         size_t want, seek_init = seek, offset = 0;
 406         ssize_t ret;
 407
 408         want = min_t(size_t, count, buf->head[0].iov_len);
 409         if (seek < want) {
 410                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 411                 if (ret <= 0)
 412                         goto sock_err;
 413                 offset += ret;
 414                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 415                         goto out;
 416                 if (ret != want)
 417                         goto out;
 418                 seek = 0;
 419         } else {
 420                 seek -= want;
 421                 offset += want;
 422         }
 423
 424         want = xs_alloc_sparse_pages(buf,
 425                         min_t(size_t, count - offset, buf->page_len),
 426                         GFP_KERNEL);
 427         if (seek < want) {
 428                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 429                                 xdr_buf_pagecount(buf),
 430                                 want + buf->page_base,
 431                                 seek + buf->page_base);
 432                 if (ret <= 0)
 433                         goto sock_err;
 434                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 435                 ret -= buf->page_base;
 436                 offset += ret;
 437                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 438                         goto out;
 439                 if (ret != want)
 440                         goto out;
 441                 seek = 0;
 442         } else {
 443                 seek -= want;
 444                 offset += want;
 445         }
 446
 447         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 448         if (seek < want) {
 449                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 450                 if (ret <= 0)
 451                         goto sock_err;
 452                 offset += ret;
 453                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 454                         goto out;
 455                 if (ret != want)
 456                         goto out;
 457         } else if (offset < seek_init)
 458                 offset = seek_init;
 459         ret = -EMSGSIZE;
 460 out:
 461         *read = offset - seek_init;
 462         return ret;
 463 sock_err:
 464         offset += seek;
 465         goto out;
 466 }
 467
 468 static void
 469 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 470 {
 471         if (!transport->recv.copied) {
 472                 if (buf->head[0].iov_len >= transport->recv.offset)
 473                         memcpy(buf->head[0].iov_base,
 474                                         &transport->recv.xid,
 475                                         transport->recv.offset);
 476                 transport->recv.copied = transport->recv.offset;
 477         }
 478 }
 479
 480 static bool
 481 xs_read_stream_request_done(struct sock_xprt *transport)
 482 {
 483         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 484 }
 485
 486 static void
 487 xs_read_stream_check_eor(struct sock_xprt *transport,
 488                 struct msghdr *msg)
 489 {
 490         if (xs_read_stream_request_done(transport))
 491                 msg->msg_flags |= MSG_EOR;
 492 }
 493
 494 static ssize_t
 495 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 496                 int flags, struct rpc_rqst *req)
 497 {
 498         struct xdr_buf *buf = &req->rq_private_buf;
 499         size_t want, uninitialized_var(read);
 500         ssize_t uninitialized_var(ret);
 501
 502         xs_read_header(transport, buf);
 503
 504         want = transport->recv.len - transport->recv.offset;
 505         if (want != 0) {
 506                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 507                                 transport->recv.copied + want,
 508                                 transport->recv.copied,
 509                                 &read);
 510                 transport->recv.offset += read;
 511                 transport->recv.copied += read;
 512         }
 513
 514         if (transport->recv.offset == transport->recv.len)
 515                 xs_read_stream_check_eor(transport, msg);
 516
 517         if (want == 0)
 518                 return 0;
 519
 520         switch (ret) {
 521         default:
 522                 break;
 523         case -EFAULT:
 524         case -EMSGSIZE:
 525                 msg->msg_flags |= MSG_TRUNC;
 526                 return read;
 527         case 0:
 528                 return -ESHUTDOWN;
 529         }
 530         return ret < 0 ? ret : read;
 531 }
 532
 533 static size_t
 534 xs_read_stream_headersize(bool isfrag)
 535 {
 536         if (isfrag)
 537                 return sizeof(__be32);
 538         return 3 * sizeof(__be32);
 539 }
 540
 541 static ssize_t
 542 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 543                 int flags, size_t want, size_t seek)
 544 {
 545         struct kvec kvec = {
 546                 .iov_base = &transport->recv.fraghdr,
 547                 .iov_len = want,
 548         };
 549         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 550 }
 551
 552 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 553 static ssize_t
 554 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 555 {
 556         struct rpc_xprt *xprt = &transport->xprt;
 557         struct rpc_rqst *req;
 558         ssize_t ret;
 559
 560         /* Look up and lock the request corresponding to the given XID */
 561         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 562         if (!req) {
 563                 printk(KERN_WARNING "Callback slot table overflowed\n");
 564                 return -ESHUTDOWN;
 565         }
 566         if (transport->recv.copied && !req->rq_private_buf.len)
 567                 return -ESHUTDOWN;
 568
 569         ret = xs_read_stream_request(transport, msg, flags, req);
 570         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 571                 xprt_complete_bc_request(req, transport->recv.copied);
 572         else
 573                 req->rq_private_buf.len = transport->recv.copied;
 574
 575         return ret;
 576 }
 577 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 578 static ssize_t
 579 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 580 {
 581         return -ESHUTDOWN;
 582 }
 583 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 584
 585 static ssize_t
 586 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 587 {
 588         struct rpc_xprt *xprt = &transport->xprt;
 589         struct rpc_rqst *req;
 590         ssize_t ret = 0;
 591
 592         /* Look up and lock the request corresponding to the given XID */
 593         spin_lock(&xprt->queue_lock);
 594         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 595         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 596                 msg->msg_flags |= MSG_TRUNC;
 597                 goto out;
 598         }
 599         xprt_pin_rqst(req);
 600         spin_unlock(&xprt->queue_lock);
 601
 602         ret = xs_read_stream_request(transport, msg, flags, req);
 603
 604         spin_lock(&xprt->queue_lock);
 605         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 606                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 607         else
 608                 req->rq_private_buf.len = transport->recv.copied;
 609         xprt_unpin_rqst(req);
 610 out:
 611         spin_unlock(&xprt->queue_lock);
 612         return ret;
 613 }
 614
 615 static ssize_t
 616 xs_read_stream(struct sock_xprt *transport, int flags)
 617 {
 618         struct msghdr msg = { 0 };
 619         size_t want, read = 0;
 620         ssize_t ret = 0;
 621
 622         if (transport->recv.len == 0) {
 623                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 624                 ret = xs_read_stream_header(transport, &msg, flags, want,
 625                                 transport->recv.offset);
 626                 if (ret <= 0)
 627                         goto out_err;
 628                 transport->recv.offset = ret;
 629                 if (transport->recv.offset != want)
 630                         return transport->recv.offset;
 631                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 632                         RPC_FRAGMENT_SIZE_MASK;
 633                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 634                 read = ret;
 635         }
 636
 637         switch (be32_to_cpu(transport->recv.calldir)) {
 638         default:
 639                 msg.msg_flags |= MSG_TRUNC;
 640                 break;
 641         case RPC_CALL:
 642                 ret = xs_read_stream_call(transport, &msg, flags);
 643                 break;
 644         case RPC_REPLY:
 645                 ret = xs_read_stream_reply(transport, &msg, flags);
 646         }
 647         if (msg.msg_flags & MSG_TRUNC) {
 648                 transport->recv.calldir = cpu_to_be32(-1);
 649                 transport->recv.copied = -1;
 650         }
 651         if (ret < 0)
 652                 goto out_err;
 653         read += ret;
 654         if (transport->recv.offset < transport->recv.len) {
 655                 if (!(msg.msg_flags & MSG_TRUNC))
 656                         return read;
 657                 msg.msg_flags = 0;
 658                 ret = xs_read_discard(transport->sock, &msg, flags,
 659                                 transport->recv.len - transport->recv.offset);
 660                 if (ret <= 0)
 661                         goto out_err;
 662                 transport->recv.offset += ret;
 663                 read += ret;
 664                 if (transport->recv.offset != transport->recv.len)
 665                         return read;
 666         }
 667         if (xs_read_stream_request_done(transport)) {
 668                 trace_xs_stream_read_request(transport);
 669                 transport->recv.copied = 0;
 670         }
 671         transport->recv.offset = 0;
 672         transport->recv.len = 0;
 673         return read;
 674 out_err:
 675         return ret != 0 ? ret : -ESHUTDOWN;
 676 }
 677
 678 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 679 {
 680         return transport->sock->ops->poll(transport->file, transport->sock,
 681                         NULL);
 682 }
 683
 684 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 685 {
 686         __poll_t events = xs_poll_socket(transport);
 687
 688         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 689 }
 690
 691 static void xs_poll_check_readable(struct sock_xprt *transport)
 692 {
 693
 694         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 695         if (!xs_poll_socket_readable(transport))
 696                 return;
 697         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 698                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 699 }
 700
 701 static void xs_stream_data_receive(struct sock_xprt *transport)
 702 {
 703         size_t read = 0;
 704         ssize_t ret = 0;
 705
 706         mutex_lock(&transport->recv_mutex);
 707         if (transport->sock == NULL)
 708                 goto out;
 709         for (;;) {
 710                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 711                 if (ret < 0)
 712                         break;
 713                 read += ret;
 714                 cond_resched();
 715         }
 716         if (ret == -ESHUTDOWN)
 717                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 718         else
 719                 xs_poll_check_readable(transport);
 720 out:
 721         mutex_unlock(&transport->recv_mutex);
 722         trace_xs_stream_read_data(&transport->xprt, ret, read);
 723 }
 724
 725 static void xs_stream_data_receive_workfn(struct work_struct *work)
 726 {
 727         struct sock_xprt *transport =
 728                 container_of(work, struct sock_xprt, recv_worker);
 729         unsigned int pflags = memalloc_nofs_save();
 730
 731         xs_stream_data_receive(transport);
 732         memalloc_nofs_restore(pflags);
 733 }
 734
 735 static void
 736 xs_stream_reset_connect(struct sock_xprt *transport)
 737 {
 738         transport->recv.offset = 0;
 739         transport->recv.len = 0;
 740         transport->recv.copied = 0;
 741         transport->xmit.offset = 0;
 742 }
 743
 744 static void
 745 xs_stream_start_connect(struct sock_xprt *transport)
 746 {
 747         transport->xprt.stat.connect_count++;
 748         transport->xprt.stat.connect_start = jiffies;
 749 }
 750
 751 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 752
 753 static int xs_sendmsg(struct socket *sock, struct msghdr *msg, size_t seek)
 754 {
 755         if (seek)
 756                 iov_iter_advance(&msg->msg_iter, seek);
 757         return sock_sendmsg(sock, msg);
 758 }
 759
 760 static int xs_send_kvec(struct socket *sock, struct msghdr *msg, struct kvec *vec, size_t seek)
 761 {
 762         iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
 763         return xs_sendmsg(sock, msg, seek);
 764 }
 765
 766 static int xs_send_pagedata(struct socket *sock, struct msghdr *msg, struct xdr_buf *xdr, size_t base)
 767 {
 768         int err;
 769
 770         err = xdr_alloc_bvec(xdr, GFP_KERNEL);
 771         if (err < 0)
 772                 return err;
 773
 774         iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec,
 775                         xdr_buf_pagecount(xdr),
 776                         xdr->page_len + xdr->page_base);
 777         return xs_sendmsg(sock, msg, base + xdr->page_base);
 778 }
 779
 780 #define xs_record_marker_len() sizeof(rpc_fraghdr)
 781
 782 /* Common case:
 783  *  - stream transport
 784  *  - sending from byte 0 of the message
 785  *  - the message is wholly contained in @xdr's head iovec
 786  */
 787 static int xs_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
 788                 rpc_fraghdr marker, struct kvec *vec, size_t base)
 789 {
 790         struct kvec iov[2] = {
 791                 [0] = {
 792                         .iov_base       = &marker,
 793                         .iov_len        = sizeof(marker)
 794                 },
 795                 [1] = *vec,
 796         };
 797         size_t len = iov[0].iov_len + iov[1].iov_len;
 798
 799         iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
 800         return xs_sendmsg(sock, msg, base);
 801 }
 802
 803 /**
 804  * xs_sendpages - write pages directly to a socket
 805  * @sock: socket to send on
 806  * @addr: UDP only -- address of destination
 807  * @addrlen: UDP only -- length of destination address
 808  * @xdr: buffer containing this request
 809  * @base: starting position in the buffer
 810  * @rm: stream record marker field
 811  * @sent_p: return the total number of bytes successfully queued for sending
 812  *
 813  */
 814 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, rpc_fraghdr rm, int *sent_p)
 815 {
 816         struct msghdr msg = {
 817                 .msg_name = addr,
 818                 .msg_namelen = addrlen,
 819                 .msg_flags = XS_SENDMSG_FLAGS | MSG_MORE,
 820         };
 821         unsigned int rmsize = rm ? sizeof(rm) : 0;
 822         unsigned int remainder = rmsize + xdr->len - base;
 823         unsigned int want;
 824         int err = 0;
 825
 826         if (unlikely(!sock))
 827                 return -ENOTSOCK;
 828
 829         want = xdr->head[0].iov_len + rmsize;
 830         if (base < want) {
 831                 unsigned int len = want - base;
 832                 remainder -= len;
 833                 if (remainder == 0)
 834                         msg.msg_flags &= ~MSG_MORE;
 835                 if (rmsize)
 836                         err = xs_send_rm_and_kvec(sock, &msg, rm,
 837                                         &xdr->head[0], base);
 838                 else
 839                         err = xs_send_kvec(sock, &msg, &xdr->head[0], base);
 840                 if (remainder == 0 || err != len)
 841                         goto out;
 842                 *sent_p += err;
 843                 base = 0;
 844         } else
 845                 base -= want;
 846
 847         if (base < xdr->page_len) {
 848                 unsigned int len = xdr->page_len - base;
 849                 remainder -= len;
 850                 if (remainder == 0)
 851                         msg.msg_flags &= ~MSG_MORE;
 852                 err = xs_send_pagedata(sock, &msg, xdr, base);
 853                 if (remainder == 0 || err != len)
 854                         goto out;
 855                 *sent_p += err;
 856                 base = 0;
 857         } else
 858                 base -= xdr->page_len;
 859
 860         if (base >= xdr->tail[0].iov_len)
 861                 return 0;
 862         msg.msg_flags &= ~MSG_MORE;
 863         err = xs_send_kvec(sock, &msg, &xdr->tail[0], base);
 864 out:
 865         if (err > 0) {
 866                 *sent_p += err;
 867                 err = 0;
 868         }
 869         return err;
 870 }
 871
 872 /**
 873  * xs_nospace - handle transmit was incomplete
 874  * @req: pointer to RPC request
 875  * @transport: pointer to struct sock_xprt
 876  *
 877  */
 878 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 879 {
 880         struct rpc_xprt *xprt = &transport->xprt;
 881         struct sock *sk = transport->inet;
 882         int ret = -EAGAIN;
 883
 884         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 885                         req->rq_task->tk_pid,
 886                         req->rq_slen - transport->xmit.offset,
 887                         req->rq_slen);
 888
 889         /* Protect against races with write_space */
 890         spin_lock(&xprt->transport_lock);
 891
 892         /* Don't race with disconnect */
 893         if (xprt_connected(xprt)) {
 894                 struct socket_wq *wq;
 895
 896                 rcu_read_lock();
 897                 wq = rcu_dereference(sk->sk_wq);
 898                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 899                 rcu_read_unlock();
 900
 901                 /* wait for more buffer space */
 902                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 903                 sk->sk_write_pending++;
 904                 xprt_wait_for_buffer_space(xprt);
 905         } else
 906                 ret = -ENOTCONN;
 907
 908         spin_unlock(&xprt->transport_lock);
 909         return ret;
 910 }
 911
 912 static int xs_sock_nospace(struct rpc_rqst *req)
 913 {
 914         struct sock_xprt *transport =
 915                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 916         struct sock *sk = transport->inet;
 917         int ret = -EAGAIN;
 918
 919         lock_sock(sk);
 920         if (!sock_writeable(sk))
 921                 ret = xs_nospace(req, transport);
 922         release_sock(sk);
 923         return ret;
 924 }
 925
 926 static int xs_stream_nospace(struct rpc_rqst *req)
 927 {
 928         struct sock_xprt *transport =
 929                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 930         struct sock *sk = transport->inet;
 931         int ret = -EAGAIN;
 932
 933         lock_sock(sk);
 934         if (!sk_stream_memory_free(sk))
 935                 ret = xs_nospace(req, transport);
 936         release_sock(sk);
 937         return ret;
 938 }
 939
 940 static void
 941 xs_stream_prepare_request(struct rpc_rqst *req)
 942 {
 943         xdr_free_bvec(&req->rq_rcv_buf);
 944         req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
 945 }
 946
 947 /*
 948  * Determine if the previous message in the stream was aborted before it
 949  * could complete transmission.
 950  */
 951 static bool
 952 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 953 {
 954         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 955 }
 956
 957 /*
 958  * Return the stream record marker field for a record of length < 2^31-1
 959  */
 960 static rpc_fraghdr
 961 xs_stream_record_marker(struct xdr_buf *xdr)
 962 {
 963         if (!xdr->len)
 964                 return 0;
 965         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 966 }
 967
 968 /**
 969  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 970  * @req: pointer to RPC request
 971  *
 972  * Return values:
 973  *        0:    The request has been sent
 974  *   EAGAIN:    The socket was blocked, please call again later to
 975  *              complete the request
 976  * ENOTCONN:    Caller needs to invoke connect logic then call again
 977  *    other:    Some other error occured, the request was not sent
 978  */
 979 static int xs_local_send_request(struct rpc_rqst *req)
 980 {
 981         struct rpc_xprt *xprt = req->rq_xprt;
 982         struct sock_xprt *transport =
 983                                 container_of(xprt, struct sock_xprt, xprt);
 984         struct xdr_buf *xdr = &req->rq_snd_buf;
 985         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 986         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 987         int status;
 988         int sent = 0;
 989
 990         /* Close the stream if the previous transmission was incomplete */
 991         if (xs_send_request_was_aborted(transport, req)) {
 992                 xprt_force_disconnect(xprt);
 993                 return -ENOTCONN;
 994         }
 995
 996         xs_pktdump("packet data:",
 997                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 998
 999         req->rq_xtime = ktime_get();
1000         status = xs_sendpages(transport->sock, NULL, 0, xdr,
1001                               transport->xmit.offset, rm, &sent);
1002         dprintk("RPC:       %s(%u) = %d\n",
1003                         __func__, xdr->len - transport->xmit.offset, status);
1004
1005         if (status == -EAGAIN && sock_writeable(transport->inet))
1006                 status = -ENOBUFS;
1007
1008         if (likely(sent > 0) || status == 0) {
1009                 transport->xmit.offset += sent;
1010                 req->rq_bytes_sent = transport->xmit.offset;
1011                 if (likely(req->rq_bytes_sent >= msglen)) {
1012                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1013                         transport->xmit.offset = 0;
1014                         return 0;
1015                 }
1016                 status = -EAGAIN;
1017         }
1018
1019         switch (status) {
1020         case -ENOBUFS:
1021                 break;
1022         case -EAGAIN:
1023                 status = xs_stream_nospace(req);
1024                 break;
1025         default:
1026                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1027                         -status);
1028                 /* fall through */
1029         case -EPIPE:
1030                 xprt_force_disconnect(xprt);
1031                 status = -ENOTCONN;
1032         }
1033
1034         return status;
1035 }
1036
1037 /**
1038  * xs_udp_send_request - write an RPC request to a UDP socket
1039  * @req: pointer to RPC request
1040  *
1041  * Return values:
1042  *        0:    The request has been sent
1043  *   EAGAIN:    The socket was blocked, please call again later to
1044  *              complete the request
1045  * ENOTCONN:    Caller needs to invoke connect logic then call again
1046  *    other:    Some other error occurred, the request was not sent
1047  */
1048 static int xs_udp_send_request(struct rpc_rqst *req)
1049 {
1050         struct rpc_xprt *xprt = req->rq_xprt;
1051         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1052         struct xdr_buf *xdr = &req->rq_snd_buf;
1053         int sent = 0;
1054         int status;
1055
1056         xs_pktdump("packet data:",
1057                                 req->rq_svec->iov_base,
1058                                 req->rq_svec->iov_len);
1059
1060         if (!xprt_bound(xprt))
1061                 return -ENOTCONN;
1062
1063         if (!xprt_request_get_cong(xprt, req))
1064                 return -EBADSLT;
1065
1066         req->rq_xtime = ktime_get();
1067         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
1068                               xdr, 0, 0, &sent);
1069
1070         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
1071                         xdr->len, status);
1072
1073         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1074         if (status == -EPERM)
1075                 goto process_status;
1076
1077         if (status == -EAGAIN && sock_writeable(transport->inet))
1078                 status = -ENOBUFS;
1079
1080         if (sent > 0 || status == 0) {
1081                 req->rq_xmit_bytes_sent += sent;
1082                 if (sent >= req->rq_slen)
1083                         return 0;
1084                 /* Still some bytes left; set up for a retry later. */
1085                 status = -EAGAIN;
1086         }
1087
1088 process_status:
1089         switch (status) {
1090         case -ENOTSOCK:
1091                 status = -ENOTCONN;
1092                 /* Should we call xs_close() here? */
1093                 break;
1094         case -EAGAIN:
1095                 status = xs_sock_nospace(req);
1096                 break;
1097         case -ENETUNREACH:
1098         case -ENOBUFS:
1099         case -EPIPE:
1100         case -ECONNREFUSED:
1101         case -EPERM:
1102                 /* When the server has died, an ICMP port unreachable message
1103                  * prompts ECONNREFUSED. */
1104                 break;
1105         default:
1106                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1107                         -status);
1108         }
1109
1110         return status;
1111 }
1112
1113 /**
1114  * xs_tcp_send_request - write an RPC request to a TCP socket
1115  * @req: pointer to RPC request
1116  *
1117  * Return values:
1118  *        0:    The request has been sent
1119  *   EAGAIN:    The socket was blocked, please call again later to
1120  *              complete the request
1121  * ENOTCONN:    Caller needs to invoke connect logic then call again
1122  *    other:    Some other error occurred, the request was not sent
1123  *
1124  * XXX: In the case of soft timeouts, should we eventually give up
1125  *      if sendmsg is not able to make progress?
1126  */
1127 static int xs_tcp_send_request(struct rpc_rqst *req)
1128 {
1129         struct rpc_xprt *xprt = req->rq_xprt;
1130         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1131         struct xdr_buf *xdr = &req->rq_snd_buf;
1132         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1133         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1134         bool vm_wait = false;
1135         int status;
1136         int sent;
1137
1138         /* Close the stream if the previous transmission was incomplete */
1139         if (xs_send_request_was_aborted(transport, req)) {
1140                 if (transport->sock != NULL)
1141                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1142                 return -ENOTCONN;
1143         }
1144
1145         xs_pktdump("packet data:",
1146                                 req->rq_svec->iov_base,
1147                                 req->rq_svec->iov_len);
1148
1149         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1150                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1151
1152         /* Continue transmitting the packet/record. We must be careful
1153          * to cope with writespace callbacks arriving _after_ we have
1154          * called sendmsg(). */
1155         req->rq_xtime = ktime_get();
1156         while (1) {
1157                 sent = 0;
1158                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
1159                                       transport->xmit.offset, rm, &sent);
1160
1161                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1162                                 xdr->len - transport->xmit.offset, status);
1163
1164                 /* If we've sent the entire packet, immediately
1165                  * reset the count of bytes sent. */
1166                 transport->xmit.offset += sent;
1167                 req->rq_bytes_sent = transport->xmit.offset;
1168                 if (likely(req->rq_bytes_sent >= msglen)) {
1169                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1170                         transport->xmit.offset = 0;
1171                         return 0;
1172                 }
1173
1174                 WARN_ON_ONCE(sent == 0 && status == 0);
1175
1176                 if (status == -EAGAIN ) {
1177                         /*
1178                          * Return EAGAIN if we're sure we're hitting the
1179                          * socket send buffer limits.
1180                          */
1181                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1182                                 break;
1183                         /*
1184                          * Did we hit a memory allocation failure?
1185                          */
1186                         if (sent == 0) {
1187                                 status = -ENOBUFS;
1188                                 if (vm_wait)
1189                                         break;
1190                                 /* Retry, knowing now that we're below the
1191                                  * socket send buffer limit
1192                                  */
1193                                 vm_wait = true;
1194                         }
1195                         continue;
1196                 }
1197                 if (status < 0)
1198                         break;
1199                 vm_wait = false;
1200         }
1201
1202         switch (status) {
1203         case -ENOTSOCK:
1204                 status = -ENOTCONN;
1205                 /* Should we call xs_close() here? */
1206                 break;
1207         case -EAGAIN:
1208                 status = xs_stream_nospace(req);
1209                 break;
1210         case -ECONNRESET:
1211         case -ECONNREFUSED:
1212         case -ENOTCONN:
1213         case -EADDRINUSE:
1214         case -ENOBUFS:
1215         case -EPIPE:
1216                 break;
1217         default:
1218                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1219                         -status);
1220         }
1221
1222         return status;
1223 }
1224
1225 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1226 {
1227         transport->old_data_ready = sk->sk_data_ready;
1228         transport->old_state_change = sk->sk_state_change;
1229         transport->old_write_space = sk->sk_write_space;
1230         transport->old_error_report = sk->sk_error_report;
1231 }
1232
1233 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1234 {
1235         sk->sk_data_ready = transport->old_data_ready;
1236         sk->sk_state_change = transport->old_state_change;
1237         sk->sk_write_space = transport->old_write_space;
1238         sk->sk_error_report = transport->old_error_report;
1239 }
1240
1241 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1242 {
1243         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1244
1245         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1246         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1247         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1248         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1249 }
1250
1251 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1252 {
1253         set_bit(nr, &transport->sock_state);
1254         queue_work(xprtiod_workqueue, &transport->error_worker);
1255 }
1256
1257 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1258 {
1259         smp_mb__before_atomic();
1260         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1261         clear_bit(XPRT_CLOSING, &xprt->state);
1262         xs_sock_reset_state_flags(xprt);
1263         smp_mb__after_atomic();
1264 }
1265
1266 /**
1267  * xs_error_report - callback to handle TCP socket state errors
1268  * @sk: socket
1269  *
1270  * Note: we don't call sock_error() since there may be a rpc_task
1271  * using the socket, and so we don't want to clear sk->sk_err.
1272  */
1273 static void xs_error_report(struct sock *sk)
1274 {
1275         struct sock_xprt *transport;
1276         struct rpc_xprt *xprt;
1277
1278         read_lock_bh(&sk->sk_callback_lock);
1279         if (!(xprt = xprt_from_sock(sk)))
1280                 goto out;
1281
1282         transport = container_of(xprt, struct sock_xprt, xprt);
1283         transport->xprt_err = -sk->sk_err;
1284         if (transport->xprt_err == 0)
1285                 goto out;
1286         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1287                         xprt, -transport->xprt_err);
1288         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1289
1290         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1291         smp_mb__before_atomic();
1292         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1293  out:
1294         read_unlock_bh(&sk->sk_callback_lock);
1295 }
1296
1297 static void xs_reset_transport(struct sock_xprt *transport)
1298 {
1299         struct socket *sock = transport->sock;
1300         struct sock *sk = transport->inet;
1301         struct rpc_xprt *xprt = &transport->xprt;
1302         struct file *filp = transport->file;
1303
1304         if (sk == NULL)
1305                 return;
1306         /*
1307          * Make sure we're calling this in a context from which it is safe
1308          * to call __fput_sync(). In practice that means rpciod and the
1309          * system workqueue.
1310          */
1311         if (!(current->flags & PF_WQ_WORKER)) {
1312                 WARN_ON_ONCE(1);
1313                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1314                 return;
1315         }
1316
1317         if (atomic_read(&transport->xprt.swapper))
1318                 sk_clear_memalloc(sk);
1319
1320         kernel_sock_shutdown(sock, SHUT_RDWR);
1321
1322         mutex_lock(&transport->recv_mutex);
1323         write_lock_bh(&sk->sk_callback_lock);
1324         transport->inet = NULL;
1325         transport->sock = NULL;
1326         transport->file = NULL;
1327
1328         sk->sk_user_data = NULL;
1329
1330         xs_restore_old_callbacks(transport, sk);
1331         xprt_clear_connected(xprt);
1332         write_unlock_bh(&sk->sk_callback_lock);
1333         xs_sock_reset_connection_flags(xprt);
1334         /* Reset stream record info */
1335         xs_stream_reset_connect(transport);
1336         mutex_unlock(&transport->recv_mutex);
1337
1338         trace_rpc_socket_close(xprt, sock);
1339         __fput_sync(filp);
1340
1341         xprt_disconnect_done(xprt);
1342 }
1343
1344 /**
1345  * xs_close - close a socket
1346  * @xprt: transport
1347  *
1348  * This is used when all requests are complete; ie, no DRC state remains
1349  * on the server we want to save.
1350  *
1351  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1352  * xs_reset_transport() zeroing the socket from underneath a writer.
1353  */
1354 static void xs_close(struct rpc_xprt *xprt)
1355 {
1356         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1357
1358         dprintk("RPC:       xs_close xprt %p\n", xprt);
1359
1360         xs_reset_transport(transport);
1361         xprt->reestablish_timeout = 0;
1362 }
1363
1364 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1365 {
1366         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1367                 xprt);
1368         xprt_disconnect_done(xprt);
1369 }
1370
1371 static void xs_xprt_free(struct rpc_xprt *xprt)
1372 {
1373         xs_free_peer_addresses(xprt);
1374         xprt_free(xprt);
1375 }
1376
1377 /**
1378  * xs_destroy - prepare to shutdown a transport
1379  * @xprt: doomed transport
1380  *
1381  */
1382 static void xs_destroy(struct rpc_xprt *xprt)
1383 {
1384         struct sock_xprt *transport = container_of(xprt,
1385                         struct sock_xprt, xprt);
1386         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1387
1388         cancel_delayed_work_sync(&transport->connect_worker);
1389         xs_close(xprt);
1390         cancel_work_sync(&transport->recv_worker);
1391         cancel_work_sync(&transport->error_worker);
1392         xs_xprt_free(xprt);
1393         module_put(THIS_MODULE);
1394 }
1395
1396 /**
1397  * xs_udp_data_read_skb - receive callback for UDP sockets
1398  * @xprt: transport
1399  * @sk: socket
1400  * @skb: skbuff
1401  *
1402  */
1403 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1404                 struct sock *sk,
1405                 struct sk_buff *skb)
1406 {
1407         struct rpc_task *task;
1408         struct rpc_rqst *rovr;
1409         int repsize, copied;
1410         u32 _xid;
1411         __be32 *xp;
1412
1413         repsize = skb->len;
1414         if (repsize < 4) {
1415                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1416                 return;
1417         }
1418
1419         /* Copy the XID from the skb... */
1420         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1421         if (xp == NULL)
1422                 return;
1423
1424         /* Look up and lock the request corresponding to the given XID */
1425         spin_lock(&xprt->queue_lock);
1426         rovr = xprt_lookup_rqst(xprt, *xp);
1427         if (!rovr)
1428                 goto out_unlock;
1429         xprt_pin_rqst(rovr);
1430         xprt_update_rtt(rovr->rq_task);
1431         spin_unlock(&xprt->queue_lock);
1432         task = rovr->rq_task;
1433
1434         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1435                 copied = repsize;
1436
1437         /* Suck it into the iovec, verify checksum if not done by hw. */
1438         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1439                 spin_lock(&xprt->queue_lock);
1440                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1441                 goto out_unpin;
1442         }
1443
1444
1445         spin_lock(&xprt->transport_lock);
1446         xprt_adjust_cwnd(xprt, task, copied);
1447         spin_unlock(&xprt->transport_lock);
1448         spin_lock(&xprt->queue_lock);
1449         xprt_complete_rqst(task, copied);
1450         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1451 out_unpin:
1452         xprt_unpin_rqst(rovr);
1453  out_unlock:
1454         spin_unlock(&xprt->queue_lock);
1455 }
1456
1457 static void xs_udp_data_receive(struct sock_xprt *transport)
1458 {
1459         struct sk_buff *skb;
1460         struct sock *sk;
1461         int err;
1462
1463         mutex_lock(&transport->recv_mutex);
1464         sk = transport->inet;
1465         if (sk == NULL)
1466                 goto out;
1467         for (;;) {
1468                 skb = skb_recv_udp(sk, 0, 1, &err);
1469                 if (skb == NULL)
1470                         break;
1471                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1472                 consume_skb(skb);
1473                 cond_resched();
1474         }
1475         xs_poll_check_readable(transport);
1476 out:
1477         mutex_unlock(&transport->recv_mutex);
1478 }
1479
1480 static void xs_udp_data_receive_workfn(struct work_struct *work)
1481 {
1482         struct sock_xprt *transport =
1483                 container_of(work, struct sock_xprt, recv_worker);
1484         unsigned int pflags = memalloc_nofs_save();
1485
1486         xs_udp_data_receive(transport);
1487         memalloc_nofs_restore(pflags);
1488 }
1489
1490 /**
1491  * xs_data_ready - "data ready" callback for UDP sockets
1492  * @sk: socket with data to read
1493  *
1494  */
1495 static void xs_data_ready(struct sock *sk)
1496 {
1497         struct rpc_xprt *xprt;
1498
1499         read_lock_bh(&sk->sk_callback_lock);
1500         dprintk("RPC:       xs_data_ready...\n");
1501         xprt = xprt_from_sock(sk);
1502         if (xprt != NULL) {
1503                 struct sock_xprt *transport = container_of(xprt,
1504                                 struct sock_xprt, xprt);
1505                 transport->old_data_ready(sk);
1506                 /* Any data means we had a useful conversation, so
1507                  * then we don't need to delay the next reconnect
1508                  */
1509                 if (xprt->reestablish_timeout)
1510                         xprt->reestablish_timeout = 0;
1511                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1512                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1513         }
1514         read_unlock_bh(&sk->sk_callback_lock);
1515 }
1516
1517 /*
1518  * Helper function to force a TCP close if the server is sending
1519  * junk and/or it has put us in CLOSE_WAIT
1520  */
1521 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1522 {
1523         xprt_force_disconnect(xprt);
1524 }
1525
1526 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1527 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1528 {
1529         return PAGE_SIZE;
1530 }
1531 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1532
1533 /**
1534  * xs_tcp_state_change - callback to handle TCP socket state changes
1535  * @sk: socket whose state has changed
1536  *
1537  */
1538 static void xs_tcp_state_change(struct sock *sk)
1539 {
1540         struct rpc_xprt *xprt;
1541         struct sock_xprt *transport;
1542
1543         read_lock_bh(&sk->sk_callback_lock);
1544         if (!(xprt = xprt_from_sock(sk)))
1545                 goto out;
1546         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1547         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1548                         sk->sk_state, xprt_connected(xprt),
1549                         sock_flag(sk, SOCK_DEAD),
1550                         sock_flag(sk, SOCK_ZAPPED),
1551                         sk->sk_shutdown);
1552
1553         transport = container_of(xprt, struct sock_xprt, xprt);
1554         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1555         switch (sk->sk_state) {
1556         case TCP_ESTABLISHED:
1557                 if (!xprt_test_and_set_connected(xprt)) {
1558                         xprt->connect_cookie++;
1559                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1560                         xprt_clear_connecting(xprt);
1561
1562                         xprt->stat.connect_count++;
1563                         xprt->stat.connect_time += (long)jiffies -
1564                                                    xprt->stat.connect_start;
1565                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1566                 }
1567                 break;
1568         case TCP_FIN_WAIT1:
1569                 /* The client initiated a shutdown of the socket */
1570                 xprt->connect_cookie++;
1571                 xprt->reestablish_timeout = 0;
1572                 set_bit(XPRT_CLOSING, &xprt->state);
1573                 smp_mb__before_atomic();
1574                 clear_bit(XPRT_CONNECTED, &xprt->state);
1575                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1576                 smp_mb__after_atomic();
1577                 break;
1578         case TCP_CLOSE_WAIT:
1579                 /* The server initiated a shutdown of the socket */
1580                 xprt->connect_cookie++;
1581                 clear_bit(XPRT_CONNECTED, &xprt->state);
1582                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1583                 /* fall through */
1584         case TCP_CLOSING:
1585                 /*
1586                  * If the server closed down the connection, make sure that
1587                  * we back off before reconnecting
1588                  */
1589                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1590                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1591                 break;
1592         case TCP_LAST_ACK:
1593                 set_bit(XPRT_CLOSING, &xprt->state);
1594                 smp_mb__before_atomic();
1595                 clear_bit(XPRT_CONNECTED, &xprt->state);
1596                 smp_mb__after_atomic();
1597                 break;
1598         case TCP_CLOSE:
1599                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1600                                         &transport->sock_state))
1601                         xprt_clear_connecting(xprt);
1602                 clear_bit(XPRT_CLOSING, &xprt->state);
1603                 /* Trigger the socket release */
1604                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1605         }
1606  out:
1607         read_unlock_bh(&sk->sk_callback_lock);
1608 }
1609
1610 static void xs_write_space(struct sock *sk)
1611 {
1612         struct socket_wq *wq;
1613         struct sock_xprt *transport;
1614         struct rpc_xprt *xprt;
1615
1616         if (!sk->sk_socket)
1617                 return;
1618         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1619
1620         if (unlikely(!(xprt = xprt_from_sock(sk))))
1621                 return;
1622         transport = container_of(xprt, struct sock_xprt, xprt);
1623         rcu_read_lock();
1624         wq = rcu_dereference(sk->sk_wq);
1625         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1626                 goto out;
1627
1628         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1629         sk->sk_write_pending--;
1630 out:
1631         rcu_read_unlock();
1632 }
1633
1634 /**
1635  * xs_udp_write_space - callback invoked when socket buffer space
1636  *                             becomes available
1637  * @sk: socket whose state has changed
1638  *
1639  * Called when more output buffer space is available for this socket.
1640  * We try not to wake our writers until they can make "significant"
1641  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1642  * with a bunch of small requests.
1643  */
1644 static void xs_udp_write_space(struct sock *sk)
1645 {
1646         read_lock_bh(&sk->sk_callback_lock);
1647
1648         /* from net/core/sock.c:sock_def_write_space */
1649         if (sock_writeable(sk))
1650                 xs_write_space(sk);
1651
1652         read_unlock_bh(&sk->sk_callback_lock);
1653 }
1654
1655 /**
1656  * xs_tcp_write_space - callback invoked when socket buffer space
1657  *                             becomes available
1658  * @sk: socket whose state has changed
1659  *
1660  * Called when more output buffer space is available for this socket.
1661  * We try not to wake our writers until they can make "significant"
1662  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1663  * with a bunch of small requests.
1664  */
1665 static void xs_tcp_write_space(struct sock *sk)
1666 {
1667         read_lock_bh(&sk->sk_callback_lock);
1668
1669         /* from net/core/stream.c:sk_stream_write_space */
1670         if (sk_stream_is_writeable(sk))
1671                 xs_write_space(sk);
1672
1673         read_unlock_bh(&sk->sk_callback_lock);
1674 }
1675
1676 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1677 {
1678         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1679         struct sock *sk = transport->inet;
1680
1681         if (transport->rcvsize) {
1682                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1683                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1684         }
1685         if (transport->sndsize) {
1686                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1687                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1688                 sk->sk_write_space(sk);
1689         }
1690 }
1691
1692 /**
1693  * xs_udp_set_buffer_size - set send and receive limits
1694  * @xprt: generic transport
1695  * @sndsize: requested size of send buffer, in bytes
1696  * @rcvsize: requested size of receive buffer, in bytes
1697  *
1698  * Set socket send and receive buffer size limits.
1699  */
1700 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1701 {
1702         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1703
1704         transport->sndsize = 0;
1705         if (sndsize)
1706                 transport->sndsize = sndsize + 1024;
1707         transport->rcvsize = 0;
1708         if (rcvsize)
1709                 transport->rcvsize = rcvsize + 1024;
1710
1711         xs_udp_do_set_buffer_size(xprt);
1712 }
1713
1714 /**
1715  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1716  * @xprt: controlling transport
1717  * @task: task that timed out
1718  *
1719  * Adjust the congestion window after a retransmit timeout has occurred.
1720  */
1721 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1722 {
1723         spin_lock(&xprt->transport_lock);
1724         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1725         spin_unlock(&xprt->transport_lock);
1726 }
1727
1728 static int xs_get_random_port(void)
1729 {
1730         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1731         unsigned short range;
1732         unsigned short rand;
1733
1734         if (max < min)
1735                 return -EADDRINUSE;
1736         range = max - min + 1;
1737         rand = (unsigned short) prandom_u32() % range;
1738         return rand + min;
1739 }
1740
1741 /**
1742  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1743  * @sock: socket
1744  *
1745  * Note that this function has to be called on all sockets that share the
1746  * same port, and it must be called before binding.
1747  */
1748 static void xs_sock_set_reuseport(struct socket *sock)
1749 {
1750         int opt = 1;
1751
1752         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1753                         (char *)&opt, sizeof(opt));
1754 }
1755
1756 static unsigned short xs_sock_getport(struct socket *sock)
1757 {
1758         struct sockaddr_storage buf;
1759         unsigned short port = 0;
1760
1761         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1762                 goto out;
1763         switch (buf.ss_family) {
1764         case AF_INET6:
1765                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1766                 break;
1767         case AF_INET:
1768                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1769         }
1770 out:
1771         return port;
1772 }
1773
1774 /**
1775  * xs_set_port - reset the port number in the remote endpoint address
1776  * @xprt: generic transport
1777  * @port: new port number
1778  *
1779  */
1780 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1781 {
1782         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1783
1784         rpc_set_port(xs_addr(xprt), port);
1785         xs_update_peer_port(xprt);
1786 }
1787
1788 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1789 {
1790         if (transport->srcport == 0)
1791                 transport->srcport = xs_sock_getport(sock);
1792 }
1793
1794 static int xs_get_srcport(struct sock_xprt *transport)
1795 {
1796         int port = transport->srcport;
1797
1798         if (port == 0 && transport->xprt.resvport)
1799                 port = xs_get_random_port();
1800         return port;
1801 }
1802
1803 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1804 {
1805         if (transport->srcport != 0)
1806                 transport->srcport = 0;
1807         if (!transport->xprt.resvport)
1808                 return 0;
1809         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1810                 return xprt_max_resvport;
1811         return --port;
1812 }
1813 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1814 {
1815         struct sockaddr_storage myaddr;
1816         int err, nloop = 0;
1817         int port = xs_get_srcport(transport);
1818         unsigned short last;
1819
1820         /*
1821          * If we are asking for any ephemeral port (i.e. port == 0 &&
1822          * transport->xprt.resvport == 0), don't bind.  Let the local
1823          * port selection happen implicitly when the socket is used
1824          * (for example at connect time).
1825          *
1826          * This ensures that we can continue to establish TCP
1827          * connections even when all local ephemeral ports are already
1828          * a part of some TCP connection.  This makes no difference
1829          * for UDP sockets, but also doens't harm them.
1830          *
1831          * If we're asking for any reserved port (i.e. port == 0 &&
1832          * transport->xprt.resvport == 1) xs_get_srcport above will
1833          * ensure that port is non-zero and we will bind as needed.
1834          */
1835         if (port <= 0)
1836                 return port;
1837
1838         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1839         do {
1840                 rpc_set_port((struct sockaddr *)&myaddr, port);
1841                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1842                                 transport->xprt.addrlen);
1843                 if (err == 0) {
1844                         transport->srcport = port;
1845                         break;
1846                 }
1847                 last = port;
1848                 port = xs_next_srcport(transport, port);
1849                 if (port > last)
1850                         nloop++;
1851         } while (err == -EADDRINUSE && nloop != 2);
1852
1853         if (myaddr.ss_family == AF_INET)
1854                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1855                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1856                                 port, err ? "failed" : "ok", err);
1857         else
1858                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1859                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1860                                 port, err ? "failed" : "ok", err);
1861         return err;
1862 }
1863
1864 /*
1865  * We don't support autobind on AF_LOCAL sockets
1866  */
1867 static void xs_local_rpcbind(struct rpc_task *task)
1868 {
1869         xprt_set_bound(task->tk_xprt);
1870 }
1871
1872 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1873 {
1874 }
1875
1876 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1877 static struct lock_class_key xs_key[2];
1878 static struct lock_class_key xs_slock_key[2];
1879
1880 static inline void xs_reclassify_socketu(struct socket *sock)
1881 {
1882         struct sock *sk = sock->sk;
1883
1884         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1885                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1886 }
1887
1888 static inline void xs_reclassify_socket4(struct socket *sock)
1889 {
1890         struct sock *sk = sock->sk;
1891
1892         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1893                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1894 }
1895
1896 static inline void xs_reclassify_socket6(struct socket *sock)
1897 {
1898         struct sock *sk = sock->sk;
1899
1900         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1901                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1902 }
1903
1904 static inline void xs_reclassify_socket(int family, struct socket *sock)
1905 {
1906         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1907                 return;
1908
1909         switch (family) {
1910         case AF_LOCAL:
1911                 xs_reclassify_socketu(sock);
1912                 break;
1913         case AF_INET:
1914                 xs_reclassify_socket4(sock);
1915                 break;
1916         case AF_INET6:
1917                 xs_reclassify_socket6(sock);
1918                 break;
1919         }
1920 }
1921 #else
1922 static inline void xs_reclassify_socket(int family, struct socket *sock)
1923 {
1924 }
1925 #endif
1926
1927 static void xs_dummy_setup_socket(struct work_struct *work)
1928 {
1929 }
1930
1931 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1932                 struct sock_xprt *transport, int family, int type,
1933                 int protocol, bool reuseport)
1934 {
1935         struct file *filp;
1936         struct socket *sock;
1937         int err;
1938
1939         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1940         if (err < 0) {
1941                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1942                                 protocol, -err);
1943                 goto out;
1944         }
1945         xs_reclassify_socket(family, sock);
1946
1947         if (reuseport)
1948                 xs_sock_set_reuseport(sock);
1949
1950         err = xs_bind(transport, sock);
1951         if (err) {
1952                 sock_release(sock);
1953                 goto out;
1954         }
1955
1956         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1957         if (IS_ERR(filp))
1958                 return ERR_CAST(filp);
1959         transport->file = filp;
1960
1961         return sock;
1962 out:
1963         return ERR_PTR(err);
1964 }
1965
1966 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1967                                       struct socket *sock)
1968 {
1969         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1970                                                                         xprt);
1971
1972         if (!transport->inet) {
1973                 struct sock *sk = sock->sk;
1974
1975                 write_lock_bh(&sk->sk_callback_lock);
1976
1977                 xs_save_old_callbacks(transport, sk);
1978
1979                 sk->sk_user_data = xprt;
1980                 sk->sk_data_ready = xs_data_ready;
1981                 sk->sk_write_space = xs_udp_write_space;
1982                 sock_set_flag(sk, SOCK_FASYNC);
1983                 sk->sk_error_report = xs_error_report;
1984
1985                 xprt_clear_connected(xprt);
1986
1987                 /* Reset to new socket */
1988                 transport->sock = sock;
1989                 transport->inet = sk;
1990
1991                 write_unlock_bh(&sk->sk_callback_lock);
1992         }
1993
1994         xs_stream_start_connect(transport);
1995
1996         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1997 }
1998
1999 /**
2000  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2001  * @transport: socket transport to connect
2002  */
2003 static int xs_local_setup_socket(struct sock_xprt *transport)
2004 {
2005         struct rpc_xprt *xprt = &transport->xprt;
2006         struct file *filp;
2007         struct socket *sock;
2008         int status = -EIO;
2009
2010         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2011                                         SOCK_STREAM, 0, &sock, 1);
2012         if (status < 0) {
2013                 dprintk("RPC:       can't create AF_LOCAL "
2014                         "transport socket (%d).\n", -status);
2015                 goto out;
2016         }
2017         xs_reclassify_socket(AF_LOCAL, sock);
2018
2019         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
2020         if (IS_ERR(filp)) {
2021                 status = PTR_ERR(filp);
2022                 goto out;
2023         }
2024         transport->file = filp;
2025
2026         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2027                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2028
2029         status = xs_local_finish_connecting(xprt, sock);
2030         trace_rpc_socket_connect(xprt, sock, status);
2031         switch (status) {
2032         case 0:
2033                 dprintk("RPC:       xprt %p connected to %s\n",
2034                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2035                 xprt->stat.connect_count++;
2036                 xprt->stat.connect_time += (long)jiffies -
2037                                            xprt->stat.connect_start;
2038                 xprt_set_connected(xprt);
2039         case -ENOBUFS:
2040                 break;
2041         case -ENOENT:
2042                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2043                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2044                 break;
2045         case -ECONNREFUSED:
2046                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2047                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2048                 break;
2049         default:
2050                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2051                                 __func__, -status,
2052                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2053         }
2054
2055 out:
2056         xprt_clear_connecting(xprt);
2057         xprt_wake_pending_tasks(xprt, status);
2058         return status;
2059 }
2060
2061 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2062 {
2063         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2064         int ret;
2065
2066          if (RPC_IS_ASYNC(task)) {
2067                 /*
2068                  * We want the AF_LOCAL connect to be resolved in the
2069                  * filesystem namespace of the process making the rpc
2070                  * call.  Thus we connect synchronously.
2071                  *
2072                  * If we want to support asynchronous AF_LOCAL calls,
2073                  * we'll need to figure out how to pass a namespace to
2074                  * connect.
2075                  */
2076                 task->tk_rpc_status = -ENOTCONN;
2077                 rpc_exit(task, -ENOTCONN);
2078                 return;
2079         }
2080         ret = xs_local_setup_socket(transport);
2081         if (ret && !RPC_IS_SOFTCONN(task))
2082                 msleep_interruptible(15000);
2083 }
2084
2085 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2086 /*
2087  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2088  * know that we have exclusive access to the socket), to guard against
2089  * races with xs_reset_transport.
2090  */
2091 static void xs_set_memalloc(struct rpc_xprt *xprt)
2092 {
2093         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2094                         xprt);
2095
2096         /*
2097          * If there's no sock, then we have nothing to set. The
2098          * reconnecting process will get it for us.
2099          */
2100         if (!transport->inet)
2101                 return;
2102         if (atomic_read(&xprt->swapper))
2103                 sk_set_memalloc(transport->inet);
2104 }
2105
2106 /**
2107  * xs_enable_swap - Tag this transport as being used for swap.
2108  * @xprt: transport to tag
2109  *
2110  * Take a reference to this transport on behalf of the rpc_clnt, and
2111  * optionally mark it for swapping if it wasn't already.
2112  */
2113 static int
2114 xs_enable_swap(struct rpc_xprt *xprt)
2115 {
2116         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2117
2118         if (atomic_inc_return(&xprt->swapper) != 1)
2119                 return 0;
2120         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2121                 return -ERESTARTSYS;
2122         if (xs->inet)
2123                 sk_set_memalloc(xs->inet);
2124         xprt_release_xprt(xprt, NULL);
2125         return 0;
2126 }
2127
2128 /**
2129  * xs_disable_swap - Untag this transport as being used for swap.
2130  * @xprt: transport to tag
2131  *
2132  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2133  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2134  */
2135 static void
2136 xs_disable_swap(struct rpc_xprt *xprt)
2137 {
2138         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2139
2140         if (!atomic_dec_and_test(&xprt->swapper))
2141                 return;
2142         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2143                 return;
2144         if (xs->inet)
2145                 sk_clear_memalloc(xs->inet);
2146         xprt_release_xprt(xprt, NULL);
2147 }
2148 #else
2149 static void xs_set_memalloc(struct rpc_xprt *xprt)
2150 {
2151 }
2152
2153 static int
2154 xs_enable_swap(struct rpc_xprt *xprt)
2155 {
2156         return -EINVAL;
2157 }
2158
2159 static void
2160 xs_disable_swap(struct rpc_xprt *xprt)
2161 {
2162 }
2163 #endif
2164
2165 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2166 {
2167         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2168
2169         if (!transport->inet) {
2170                 struct sock *sk = sock->sk;
2171
2172                 write_lock_bh(&sk->sk_callback_lock);
2173
2174                 xs_save_old_callbacks(transport, sk);
2175
2176                 sk->sk_user_data = xprt;
2177                 sk->sk_data_ready = xs_data_ready;
2178                 sk->sk_write_space = xs_udp_write_space;
2179                 sock_set_flag(sk, SOCK_FASYNC);
2180
2181                 xprt_set_connected(xprt);
2182
2183                 /* Reset to new socket */
2184                 transport->sock = sock;
2185                 transport->inet = sk;
2186
2187                 xs_set_memalloc(xprt);
2188
2189                 write_unlock_bh(&sk->sk_callback_lock);
2190         }
2191         xs_udp_do_set_buffer_size(xprt);
2192
2193         xprt->stat.connect_start = jiffies;
2194 }
2195
2196 static void xs_udp_setup_socket(struct work_struct *work)
2197 {
2198         struct sock_xprt *transport =
2199                 container_of(work, struct sock_xprt, connect_worker.work);
2200         struct rpc_xprt *xprt = &transport->xprt;
2201         struct socket *sock;
2202         int status = -EIO;
2203
2204         sock = xs_create_sock(xprt, transport,
2205                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2206                         IPPROTO_UDP, false);
2207         if (IS_ERR(sock))
2208                 goto out;
2209
2210         dprintk("RPC:       worker connecting xprt %p via %s to "
2211                                 "%s (port %s)\n", xprt,
2212                         xprt->address_strings[RPC_DISPLAY_PROTO],
2213                         xprt->address_strings[RPC_DISPLAY_ADDR],
2214                         xprt->address_strings[RPC_DISPLAY_PORT]);
2215
2216         xs_udp_finish_connecting(xprt, sock);
2217         trace_rpc_socket_connect(xprt, sock, 0);
2218         status = 0;
2219 out:
2220         xprt_clear_connecting(xprt);
2221         xprt_unlock_connect(xprt, transport);
2222         xprt_wake_pending_tasks(xprt, status);
2223 }
2224
2225 /**
2226  * xs_tcp_shutdown - gracefully shut down a TCP socket
2227  * @xprt: transport
2228  *
2229  * Initiates a graceful shutdown of the TCP socket by calling the
2230  * equivalent of shutdown(SHUT_RDWR);
2231  */
2232 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2233 {
2234         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2235         struct socket *sock = transport->sock;
2236         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2237
2238         if (sock == NULL)
2239                 return;
2240         switch (skst) {
2241         default:
2242                 kernel_sock_shutdown(sock, SHUT_RDWR);
2243                 trace_rpc_socket_shutdown(xprt, sock);
2244                 break;
2245         case TCP_CLOSE:
2246         case TCP_TIME_WAIT:
2247                 xs_reset_transport(transport);
2248         }
2249 }
2250
2251 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2252                 struct socket *sock)
2253 {
2254         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2255         unsigned int keepidle;
2256         unsigned int keepcnt;
2257         unsigned int opt_on = 1;
2258         unsigned int timeo;
2259
2260         spin_lock(&xprt->transport_lock);
2261         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2262         keepcnt = xprt->timeout->to_retries + 1;
2263         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2264                 (xprt->timeout->to_retries + 1);
2265         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2266         spin_unlock(&xprt->transport_lock);
2267
2268         /* TCP Keepalive options */
2269         kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2270                         (char *)&opt_on, sizeof(opt_on));
2271         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2272                         (char *)&keepidle, sizeof(keepidle));
2273         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2274                         (char *)&keepidle, sizeof(keepidle));
2275         kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2276                         (char *)&keepcnt, sizeof(keepcnt));
2277
2278         /* TCP user timeout (see RFC5482) */
2279         kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2280                         (char *)&timeo, sizeof(timeo));
2281 }
2282
2283 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2284                 unsigned long connect_timeout,
2285                 unsigned long reconnect_timeout)
2286 {
2287         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2288         struct rpc_timeout to;
2289         unsigned long initval;
2290
2291         spin_lock(&xprt->transport_lock);
2292         if (reconnect_timeout < xprt->max_reconnect_timeout)
2293                 xprt->max_reconnect_timeout = reconnect_timeout;
2294         if (connect_timeout < xprt->connect_timeout) {
2295                 memcpy(&to, xprt->timeout, sizeof(to));
2296                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2297                 /* Arbitrary lower limit */
2298                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2299                         initval = XS_TCP_INIT_REEST_TO << 1;
2300                 to.to_initval = initval;
2301                 to.to_maxval = initval;
2302                 memcpy(&transport->tcp_timeout, &to,
2303                                 sizeof(transport->tcp_timeout));
2304                 xprt->timeout = &transport->tcp_timeout;
2305                 xprt->connect_timeout = connect_timeout;
2306         }
2307         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2308         spin_unlock(&xprt->transport_lock);
2309 }
2310
2311 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2312 {
2313         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314         int ret = -ENOTCONN;
2315
2316         if (!transport->inet) {
2317                 struct sock *sk = sock->sk;
2318                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2319
2320                 /* Avoid temporary address, they are bad for long-lived
2321                  * connections such as NFS mounts.
2322                  * RFC4941, section 3.6 suggests that:
2323                  *    Individual applications, which have specific
2324                  *    knowledge about the normal duration of connections,
2325                  *    MAY override this as appropriate.
2326                  */
2327                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2328                                 (char *)&addr_pref, sizeof(addr_pref));
2329
2330                 xs_tcp_set_socket_timeouts(xprt, sock);
2331
2332                 write_lock_bh(&sk->sk_callback_lock);
2333
2334                 xs_save_old_callbacks(transport, sk);
2335
2336                 sk->sk_user_data = xprt;
2337                 sk->sk_data_ready = xs_data_ready;
2338                 sk->sk_state_change = xs_tcp_state_change;
2339                 sk->sk_write_space = xs_tcp_write_space;
2340                 sock_set_flag(sk, SOCK_FASYNC);
2341                 sk->sk_error_report = xs_error_report;
2342
2343                 /* socket options */
2344                 sock_reset_flag(sk, SOCK_LINGER);
2345                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2346
2347                 xprt_clear_connected(xprt);
2348
2349                 /* Reset to new socket */
2350                 transport->sock = sock;
2351                 transport->inet = sk;
2352
2353                 write_unlock_bh(&sk->sk_callback_lock);
2354         }
2355
2356         if (!xprt_bound(xprt))
2357                 goto out;
2358
2359         xs_set_memalloc(xprt);
2360
2361         xs_stream_start_connect(transport);
2362
2363         /* Tell the socket layer to start connecting... */
2364         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2365         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2366         switch (ret) {
2367         case 0:
2368                 xs_set_srcport(transport, sock);
2369                 /* fall through */
2370         case -EINPROGRESS:
2371                 /* SYN_SENT! */
2372                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2373                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2374                 break;
2375         case -EADDRNOTAVAIL:
2376                 /* Source port number is unavailable. Try a new one! */
2377                 transport->srcport = 0;
2378         }
2379 out:
2380         return ret;
2381 }
2382
2383 /**
2384  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2385  * @work: queued work item
2386  *
2387  * Invoked by a work queue tasklet.
2388  */
2389 static void xs_tcp_setup_socket(struct work_struct *work)
2390 {
2391         struct sock_xprt *transport =
2392                 container_of(work, struct sock_xprt, connect_worker.work);
2393         struct socket *sock = transport->sock;
2394         struct rpc_xprt *xprt = &transport->xprt;
2395         int status = -EIO;
2396
2397         if (xprt_connected(xprt))
2398                 goto out;
2399         if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2400                                &transport->sock_state) ||
2401             !sock) {
2402                 xs_reset_transport(transport);
2403                 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2404                                       SOCK_STREAM, IPPROTO_TCP, true);
2405                 if (IS_ERR(sock)) {
2406                         status = PTR_ERR(sock);
2407                         goto out;
2408                 }
2409         }
2410
2411         dprintk("RPC:       worker connecting xprt %p via %s to "
2412                                 "%s (port %s)\n", xprt,
2413                         xprt->address_strings[RPC_DISPLAY_PROTO],
2414                         xprt->address_strings[RPC_DISPLAY_ADDR],
2415                         xprt->address_strings[RPC_DISPLAY_PORT]);
2416
2417         status = xs_tcp_finish_connecting(xprt, sock);
2418         trace_rpc_socket_connect(xprt, sock, status);
2419         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2420                         xprt, -status, xprt_connected(xprt),
2421                         sock->sk->sk_state);
2422         switch (status) {
2423         default:
2424                 printk("%s: connect returned unhandled error %d\n",
2425                         __func__, status);
2426                 /* fall through */
2427         case -EADDRNOTAVAIL:
2428                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2429                  * and retry
2430                  */
2431                 xs_tcp_force_close(xprt);
2432                 break;
2433         case 0:
2434         case -EINPROGRESS:
2435                 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2436                 fallthrough;
2437         case -EALREADY:
2438                 xprt_unlock_connect(xprt, transport);
2439                 return;
2440         case -EINVAL:
2441                 /* Happens, for instance, if the user specified a link
2442                  * local IPv6 address without a scope-id.
2443                  */
2444         case -ECONNREFUSED:
2445         case -ECONNRESET:
2446         case -ENETDOWN:
2447         case -ENETUNREACH:
2448         case -EHOSTUNREACH:
2449         case -EADDRINUSE:
2450         case -ENOBUFS:
2451                 /*
2452                  * xs_tcp_force_close() wakes tasks with -EIO.
2453                  * We need to wake them first to ensure the
2454                  * correct error code.
2455                  */
2456                 xprt_wake_pending_tasks(xprt, status);
2457                 xs_tcp_force_close(xprt);
2458                 goto out;
2459         }
2460         status = -EAGAIN;
2461 out:
2462         xprt_clear_connecting(xprt);
2463         xprt_unlock_connect(xprt, transport);
2464         xprt_wake_pending_tasks(xprt, status);
2465 }
2466
2467 /**
2468  * xs_connect - connect a socket to a remote endpoint
2469  * @xprt: pointer to transport structure
2470  * @task: address of RPC task that manages state of connect request
2471  *
2472  * TCP: If the remote end dropped the connection, delay reconnecting.
2473  *
2474  * UDP socket connects are synchronous, but we use a work queue anyway
2475  * to guarantee that even unprivileged user processes can set up a
2476  * socket on a privileged port.
2477  *
2478  * If a UDP socket connect fails, the delay behavior here prevents
2479  * retry floods (hard mounts).
2480  */
2481 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2482 {
2483         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2484         unsigned long delay = 0;
2485
2486         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2487
2488         if (transport->sock != NULL) {
2489                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2490                         "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2491
2492                 delay = xprt_reconnect_delay(xprt);
2493                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2494
2495         } else
2496                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2497
2498         queue_delayed_work(xprtiod_workqueue,
2499                         &transport->connect_worker,
2500                         delay);
2501 }
2502
2503 static void xs_wake_disconnect(struct sock_xprt *transport)
2504 {
2505         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2506                 xs_tcp_force_close(&transport->xprt);
2507 }
2508
2509 static void xs_wake_write(struct sock_xprt *transport)
2510 {
2511         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2512                 xprt_write_space(&transport->xprt);
2513 }
2514
2515 static void xs_wake_error(struct sock_xprt *transport)
2516 {
2517         int sockerr;
2518
2519         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2520                 return;
2521         mutex_lock(&transport->recv_mutex);
2522         if (transport->sock == NULL)
2523                 goto out;
2524         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2525                 goto out;
2526         sockerr = xchg(&transport->xprt_err, 0);
2527         if (sockerr < 0)
2528                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2529 out:
2530         mutex_unlock(&transport->recv_mutex);
2531 }
2532
2533 static void xs_wake_pending(struct sock_xprt *transport)
2534 {
2535         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2536                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2537 }
2538
2539 static void xs_error_handle(struct work_struct *work)
2540 {
2541         struct sock_xprt *transport = container_of(work,
2542                         struct sock_xprt, error_worker);
2543
2544         xs_wake_disconnect(transport);
2545         xs_wake_write(transport);
2546         xs_wake_error(transport);
2547         xs_wake_pending(transport);
2548 }
2549
2550 /**
2551  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2552  * @xprt: rpc_xprt struct containing statistics
2553  * @seq: output file
2554  *
2555  */
2556 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2557 {
2558         long idle_time = 0;
2559
2560         if (xprt_connected(xprt))
2561                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2562
2563         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2564                         "%llu %llu %lu %llu %llu\n",
2565                         xprt->stat.bind_count,
2566                         xprt->stat.connect_count,
2567                         xprt->stat.connect_time / HZ,
2568                         idle_time,
2569                         xprt->stat.sends,
2570                         xprt->stat.recvs,
2571                         xprt->stat.bad_xids,
2572                         xprt->stat.req_u,
2573                         xprt->stat.bklog_u,
2574                         xprt->stat.max_slots,
2575                         xprt->stat.sending_u,
2576                         xprt->stat.pending_u);
2577 }
2578
2579 /**
2580  * xs_udp_print_stats - display UDP socket-specifc stats
2581  * @xprt: rpc_xprt struct containing statistics
2582  * @seq: output file
2583  *
2584  */
2585 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2586 {
2587         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2588
2589         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2590                         "%lu %llu %llu\n",
2591                         transport->srcport,
2592                         xprt->stat.bind_count,
2593                         xprt->stat.sends,
2594                         xprt->stat.recvs,
2595                         xprt->stat.bad_xids,
2596                         xprt->stat.req_u,
2597                         xprt->stat.bklog_u,
2598                         xprt->stat.max_slots,
2599                         xprt->stat.sending_u,
2600                         xprt->stat.pending_u);
2601 }
2602
2603 /**
2604  * xs_tcp_print_stats - display TCP socket-specifc stats
2605  * @xprt: rpc_xprt struct containing statistics
2606  * @seq: output file
2607  *
2608  */
2609 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2610 {
2611         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2612         long idle_time = 0;
2613
2614         if (xprt_connected(xprt))
2615                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2616
2617         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2618                         "%llu %llu %lu %llu %llu\n",
2619                         transport->srcport,
2620                         xprt->stat.bind_count,
2621                         xprt->stat.connect_count,
2622                         xprt->stat.connect_time / HZ,
2623                         idle_time,
2624                         xprt->stat.sends,
2625                         xprt->stat.recvs,
2626                         xprt->stat.bad_xids,
2627                         xprt->stat.req_u,
2628                         xprt->stat.bklog_u,
2629                         xprt->stat.max_slots,
2630                         xprt->stat.sending_u,
2631                         xprt->stat.pending_u);
2632 }
2633
2634 /*
2635  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2636  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2637  * to use the server side send routines.
2638  */
2639 static int bc_malloc(struct rpc_task *task)
2640 {
2641         struct rpc_rqst *rqst = task->tk_rqstp;
2642         size_t size = rqst->rq_callsize;
2643         struct page *page;
2644         struct rpc_buffer *buf;
2645
2646         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2647                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2648                           size);
2649                 return -EINVAL;
2650         }
2651
2652         page = alloc_page(GFP_KERNEL);
2653         if (!page)
2654                 return -ENOMEM;
2655
2656         buf = page_address(page);
2657         buf->len = PAGE_SIZE;
2658
2659         rqst->rq_buffer = buf->data;
2660         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2661         return 0;
2662 }
2663
2664 /*
2665  * Free the space allocated in the bc_alloc routine
2666  */
2667 static void bc_free(struct rpc_task *task)
2668 {
2669         void *buffer = task->tk_rqstp->rq_buffer;
2670         struct rpc_buffer *buf;
2671
2672         buf = container_of(buffer, struct rpc_buffer, data);
2673         free_page((unsigned long)buf);
2674 }
2675
2676 /*
2677  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2678  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2679  */
2680 static int bc_sendto(struct rpc_rqst *req)
2681 {
2682         int len;
2683         struct xdr_buf *xbufp = &req->rq_snd_buf;
2684         struct sock_xprt *transport =
2685                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2686         unsigned long headoff;
2687         unsigned long tailoff;
2688         struct page *tailpage;
2689         struct msghdr msg = {
2690                 .msg_flags      = MSG_MORE
2691         };
2692         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2693                                          (u32)xbufp->len);
2694         struct kvec iov = {
2695                 .iov_base       = &marker,
2696                 .iov_len        = sizeof(marker),
2697         };
2698
2699         req->rq_xtime = ktime_get();
2700
2701         len = kernel_sendmsg(transport->sock, &msg, &iov, 1, iov.iov_len);
2702         if (len != iov.iov_len)
2703                 return -EAGAIN;
2704
2705         tailpage = NULL;
2706         if (xbufp->tail[0].iov_len)
2707                 tailpage = virt_to_page(xbufp->tail[0].iov_base);
2708         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2709         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2710         len = svc_send_common(transport->sock, xbufp,
2711                               virt_to_page(xbufp->head[0].iov_base), headoff,
2712                               tailpage, tailoff);
2713         if (len != xbufp->len)
2714                 return -EAGAIN;
2715         return len;
2716 }
2717
2718 /*
2719  * The send routine. Borrows from svc_send
2720  */
2721 static int bc_send_request(struct rpc_rqst *req)
2722 {
2723         struct svc_xprt *xprt;
2724         int len;
2725
2726         /*
2727          * Get the server socket associated with this callback xprt
2728          */
2729         xprt = req->rq_xprt->bc_xprt;
2730
2731         /*
2732          * Grab the mutex to serialize data as the connection is shared
2733          * with the fore channel
2734          */
2735         mutex_lock(&xprt->xpt_mutex);
2736         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2737                 len = -ENOTCONN;
2738         else
2739                 len = bc_sendto(req);
2740         mutex_unlock(&xprt->xpt_mutex);
2741
2742         if (len > 0)
2743                 len = 0;
2744
2745         return len;
2746 }
2747
2748 /*
2749  * The close routine. Since this is client initiated, we do nothing
2750  */
2751
2752 static void bc_close(struct rpc_xprt *xprt)
2753 {
2754         xprt_disconnect_done(xprt);
2755 }
2756
2757 /*
2758  * The xprt destroy routine. Again, because this connection is client
2759  * initiated, we do nothing
2760  */
2761
2762 static void bc_destroy(struct rpc_xprt *xprt)
2763 {
2764         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2765
2766         xs_xprt_free(xprt);
2767         module_put(THIS_MODULE);
2768 }
2769
2770 static const struct rpc_xprt_ops xs_local_ops = {
2771         .reserve_xprt           = xprt_reserve_xprt,
2772         .release_xprt           = xprt_release_xprt,
2773         .alloc_slot             = xprt_alloc_slot,
2774         .free_slot              = xprt_free_slot,
2775         .rpcbind                = xs_local_rpcbind,
2776         .set_port               = xs_local_set_port,
2777         .connect                = xs_local_connect,
2778         .buf_alloc              = rpc_malloc,
2779         .buf_free               = rpc_free,
2780         .prepare_request        = xs_stream_prepare_request,
2781         .send_request           = xs_local_send_request,
2782         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2783         .close                  = xs_close,
2784         .destroy                = xs_destroy,
2785         .print_stats            = xs_local_print_stats,
2786         .enable_swap            = xs_enable_swap,
2787         .disable_swap           = xs_disable_swap,
2788 };
2789
2790 static const struct rpc_xprt_ops xs_udp_ops = {
2791         .set_buffer_size        = xs_udp_set_buffer_size,
2792         .reserve_xprt           = xprt_reserve_xprt_cong,
2793         .release_xprt           = xprt_release_xprt_cong,
2794         .alloc_slot             = xprt_alloc_slot,
2795         .free_slot              = xprt_free_slot,
2796         .rpcbind                = rpcb_getport_async,
2797         .set_port               = xs_set_port,
2798         .connect                = xs_connect,
2799         .buf_alloc              = rpc_malloc,
2800         .buf_free               = rpc_free,
2801         .send_request           = xs_udp_send_request,
2802         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2803         .timer                  = xs_udp_timer,
2804         .release_request        = xprt_release_rqst_cong,
2805         .close                  = xs_close,
2806         .destroy                = xs_destroy,
2807         .print_stats            = xs_udp_print_stats,
2808         .enable_swap            = xs_enable_swap,
2809         .disable_swap           = xs_disable_swap,
2810         .inject_disconnect      = xs_inject_disconnect,
2811 };
2812
2813 static const struct rpc_xprt_ops xs_tcp_ops = {
2814         .reserve_xprt           = xprt_reserve_xprt,
2815         .release_xprt           = xprt_release_xprt,
2816         .alloc_slot             = xprt_alloc_slot,
2817         .free_slot              = xprt_free_slot,
2818         .rpcbind                = rpcb_getport_async,
2819         .set_port               = xs_set_port,
2820         .connect                = xs_connect,
2821         .buf_alloc              = rpc_malloc,
2822         .buf_free               = rpc_free,
2823         .prepare_request        = xs_stream_prepare_request,
2824         .send_request           = xs_tcp_send_request,
2825         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2826         .close                  = xs_tcp_shutdown,
2827         .destroy                = xs_destroy,
2828         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2829         .print_stats            = xs_tcp_print_stats,
2830         .enable_swap            = xs_enable_swap,
2831         .disable_swap           = xs_disable_swap,
2832         .inject_disconnect      = xs_inject_disconnect,
2833 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2834         .bc_setup               = xprt_setup_bc,
2835         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2836         .bc_num_slots           = xprt_bc_max_slots,
2837         .bc_free_rqst           = xprt_free_bc_rqst,
2838         .bc_destroy             = xprt_destroy_bc,
2839 #endif
2840 };
2841
2842 /*
2843  * The rpc_xprt_ops for the server backchannel
2844  */
2845
2846 static const struct rpc_xprt_ops bc_tcp_ops = {
2847         .reserve_xprt           = xprt_reserve_xprt,
2848         .release_xprt           = xprt_release_xprt,
2849         .alloc_slot             = xprt_alloc_slot,
2850         .free_slot              = xprt_free_slot,
2851         .buf_alloc              = bc_malloc,
2852         .buf_free               = bc_free,
2853         .send_request           = bc_send_request,
2854         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2855         .close                  = bc_close,
2856         .destroy                = bc_destroy,
2857         .print_stats            = xs_tcp_print_stats,
2858         .enable_swap            = xs_enable_swap,
2859         .disable_swap           = xs_disable_swap,
2860         .inject_disconnect      = xs_inject_disconnect,
2861 };
2862
2863 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2864 {
2865         static const struct sockaddr_in sin = {
2866                 .sin_family             = AF_INET,
2867                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2868         };
2869         static const struct sockaddr_in6 sin6 = {
2870                 .sin6_family            = AF_INET6,
2871                 .sin6_addr              = IN6ADDR_ANY_INIT,
2872         };
2873
2874         switch (family) {
2875         case AF_LOCAL:
2876                 break;
2877         case AF_INET:
2878                 memcpy(sap, &sin, sizeof(sin));
2879                 break;
2880         case AF_INET6:
2881                 memcpy(sap, &sin6, sizeof(sin6));
2882                 break;
2883         default:
2884                 dprintk("RPC:       %s: Bad address family\n", __func__);
2885                 return -EAFNOSUPPORT;
2886         }
2887         return 0;
2888 }
2889
2890 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2891                                       unsigned int slot_table_size,
2892                                       unsigned int max_slot_table_size)
2893 {
2894         struct rpc_xprt *xprt;
2895         struct sock_xprt *new;
2896
2897         if (args->addrlen > sizeof(xprt->addr)) {
2898                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2899                 return ERR_PTR(-EBADF);
2900         }
2901
2902         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2903                         max_slot_table_size);
2904         if (xprt == NULL) {
2905                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2906                                 "rpc_xprt\n");
2907                 return ERR_PTR(-ENOMEM);
2908         }
2909
2910         new = container_of(xprt, struct sock_xprt, xprt);
2911         mutex_init(&new->recv_mutex);
2912         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2913         xprt->addrlen = args->addrlen;
2914         if (args->srcaddr)
2915                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2916         else {
2917                 int err;
2918                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2919                                         (struct sockaddr *)&new->srcaddr);
2920                 if (err != 0) {
2921                         xprt_free(xprt);
2922                         return ERR_PTR(err);
2923                 }
2924         }
2925
2926         return xprt;
2927 }
2928
2929 static const struct rpc_timeout xs_local_default_timeout = {
2930         .to_initval = 10 * HZ,
2931         .to_maxval = 10 * HZ,
2932         .to_retries = 2,
2933 };
2934
2935 /**
2936  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2937  * @args: rpc transport creation arguments
2938  *
2939  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2940  */
2941 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2942 {
2943         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2944         struct sock_xprt *transport;
2945         struct rpc_xprt *xprt;
2946         struct rpc_xprt *ret;
2947
2948         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2949                         xprt_max_tcp_slot_table_entries);
2950         if (IS_ERR(xprt))
2951                 return xprt;
2952         transport = container_of(xprt, struct sock_xprt, xprt);
2953
2954         xprt->prot = 0;
2955         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2956
2957         xprt->bind_timeout = XS_BIND_TO;
2958         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2959         xprt->idle_timeout = XS_IDLE_DISC_TO;
2960
2961         xprt->ops = &xs_local_ops;
2962         xprt->timeout = &xs_local_default_timeout;
2963
2964         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2965         INIT_WORK(&transport->error_worker, xs_error_handle);
2966         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2967
2968         switch (sun->sun_family) {
2969         case AF_LOCAL:
2970                 if (sun->sun_path[0] != '/') {
2971                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2972                                         sun->sun_path);
2973                         ret = ERR_PTR(-EINVAL);
2974                         goto out_err;
2975                 }
2976                 xprt_set_bound(xprt);
2977                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2978                 break;
2979         default:
2980                 ret = ERR_PTR(-EAFNOSUPPORT);
2981                 goto out_err;
2982         }
2983
2984         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2985                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2986
2987         if (try_module_get(THIS_MODULE))
2988                 return xprt;
2989         ret = ERR_PTR(-EINVAL);
2990 out_err:
2991         xs_xprt_free(xprt);
2992         return ret;
2993 }
2994
2995 static const struct rpc_timeout xs_udp_default_timeout = {
2996         .to_initval = 5 * HZ,
2997         .to_maxval = 30 * HZ,
2998         .to_increment = 5 * HZ,
2999         .to_retries = 5,
3000 };
3001
3002 /**
3003  * xs_setup_udp - Set up transport to use a UDP socket
3004  * @args: rpc transport creation arguments
3005  *
3006  */
3007 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3008 {
3009         struct sockaddr *addr = args->dstaddr;
3010         struct rpc_xprt *xprt;
3011         struct sock_xprt *transport;
3012         struct rpc_xprt *ret;
3013
3014         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3015                         xprt_udp_slot_table_entries);
3016         if (IS_ERR(xprt))
3017                 return xprt;
3018         transport = container_of(xprt, struct sock_xprt, xprt);
3019
3020         xprt->prot = IPPROTO_UDP;
3021         /* XXX: header size can vary due to auth type, IPv6, etc. */
3022         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3023
3024         xprt->bind_timeout = XS_BIND_TO;
3025         xprt->reestablish_timeout = XS_UDP_REEST_TO;
3026         xprt->idle_timeout = XS_IDLE_DISC_TO;
3027
3028         xprt->ops = &xs_udp_ops;
3029
3030         xprt->timeout = &xs_udp_default_timeout;
3031
3032         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3033         INIT_WORK(&transport->error_worker, xs_error_handle);
3034         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3035
3036         switch (addr->sa_family) {
3037         case AF_INET:
3038                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3039                         xprt_set_bound(xprt);
3040
3041                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3042                 break;
3043         case AF_INET6:
3044                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3045                         xprt_set_bound(xprt);
3046
3047                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3048                 break;
3049         default:
3050                 ret = ERR_PTR(-EAFNOSUPPORT);
3051                 goto out_err;
3052         }
3053
3054         if (xprt_bound(xprt))
3055                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3056                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3057                                 xprt->address_strings[RPC_DISPLAY_PORT],
3058                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3059         else
3060                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3061                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3062                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3063
3064         if (try_module_get(THIS_MODULE))
3065                 return xprt;
3066         ret = ERR_PTR(-EINVAL);
3067 out_err:
3068         xs_xprt_free(xprt);
3069         return ret;
3070 }
3071
3072 static const struct rpc_timeout xs_tcp_default_timeout = {
3073         .to_initval = 60 * HZ,
3074         .to_maxval = 60 * HZ,
3075         .to_retries = 2,
3076 };
3077
3078 /**
3079  * xs_setup_tcp - Set up transport to use a TCP socket
3080  * @args: rpc transport creation arguments
3081  *
3082  */
3083 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3084 {
3085         struct sockaddr *addr = args->dstaddr;
3086         struct rpc_xprt *xprt;
3087         struct sock_xprt *transport;
3088         struct rpc_xprt *ret;
3089         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3090
3091         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3092                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3093
3094         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3095                         max_slot_table_size);
3096         if (IS_ERR(xprt))
3097                 return xprt;
3098         transport = container_of(xprt, struct sock_xprt, xprt);
3099
3100         xprt->prot = IPPROTO_TCP;
3101         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3102
3103         xprt->bind_timeout = XS_BIND_TO;
3104         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3105         xprt->idle_timeout = XS_IDLE_DISC_TO;
3106
3107         xprt->ops = &xs_tcp_ops;
3108         xprt->timeout = &xs_tcp_default_timeout;
3109
3110         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3111         xprt->connect_timeout = xprt->timeout->to_initval *
3112                 (xprt->timeout->to_retries + 1);
3113
3114         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3115         INIT_WORK(&transport->error_worker, xs_error_handle);
3116         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3117
3118         switch (addr->sa_family) {
3119         case AF_INET:
3120                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3121                         xprt_set_bound(xprt);
3122
3123                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3124                 break;
3125         case AF_INET6:
3126                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3127                         xprt_set_bound(xprt);
3128
3129                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3130                 break;
3131         default:
3132                 ret = ERR_PTR(-EAFNOSUPPORT);
3133                 goto out_err;
3134         }
3135
3136         if (xprt_bound(xprt))
3137                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3138                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3139                                 xprt->address_strings[RPC_DISPLAY_PORT],
3140                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3141         else
3142                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3143                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3144                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3145
3146         if (try_module_get(THIS_MODULE))
3147                 return xprt;
3148         ret = ERR_PTR(-EINVAL);
3149 out_err:
3150         xs_xprt_free(xprt);
3151         return ret;
3152 }
3153
3154 /**
3155  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3156  * @args: rpc transport creation arguments
3157  *
3158  */
3159 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3160 {
3161         struct sockaddr *addr = args->dstaddr;
3162         struct rpc_xprt *xprt;
3163         struct sock_xprt *transport;
3164         struct svc_sock *bc_sock;
3165         struct rpc_xprt *ret;
3166
3167         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3168                         xprt_tcp_slot_table_entries);
3169         if (IS_ERR(xprt))
3170                 return xprt;
3171         transport = container_of(xprt, struct sock_xprt, xprt);
3172
3173         xprt->prot = IPPROTO_TCP;
3174         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3175         xprt->timeout = &xs_tcp_default_timeout;
3176
3177         /* backchannel */
3178         xprt_set_bound(xprt);
3179         xprt->bind_timeout = 0;
3180         xprt->reestablish_timeout = 0;
3181         xprt->idle_timeout = 0;
3182
3183         xprt->ops = &bc_tcp_ops;
3184
3185         switch (addr->sa_family) {
3186         case AF_INET:
3187                 xs_format_peer_addresses(xprt, "tcp",
3188                                          RPCBIND_NETID_TCP);
3189                 break;
3190         case AF_INET6:
3191                 xs_format_peer_addresses(xprt, "tcp",
3192                                    RPCBIND_NETID_TCP6);
3193                 break;
3194         default:
3195                 ret = ERR_PTR(-EAFNOSUPPORT);
3196                 goto out_err;
3197         }
3198
3199         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3200                         xprt->address_strings[RPC_DISPLAY_ADDR],
3201                         xprt->address_strings[RPC_DISPLAY_PORT],
3202                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3203
3204         /*
3205          * Once we've associated a backchannel xprt with a connection,
3206          * we want to keep it around as long as the connection lasts,
3207          * in case we need to start using it for a backchannel again;
3208          * this reference won't be dropped until bc_xprt is destroyed.
3209          */
3210         xprt_get(xprt);
3211         args->bc_xprt->xpt_bc_xprt = xprt;
3212         xprt->bc_xprt = args->bc_xprt;
3213         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3214         transport->sock = bc_sock->sk_sock;
3215         transport->inet = bc_sock->sk_sk;
3216
3217         /*
3218          * Since we don't want connections for the backchannel, we set
3219          * the xprt status to connected
3220          */
3221         xprt_set_connected(xprt);
3222
3223         if (try_module_get(THIS_MODULE))
3224                 return xprt;
3225
3226         args->bc_xprt->xpt_bc_xprt = NULL;
3227         args->bc_xprt->xpt_bc_xps = NULL;
3228         xprt_put(xprt);
3229         ret = ERR_PTR(-EINVAL);
3230 out_err:
3231         xs_xprt_free(xprt);
3232         return ret;
3233 }
3234
3235 static struct xprt_class        xs_local_transport = {
3236         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3237         .name           = "named UNIX socket",
3238         .owner          = THIS_MODULE,
3239         .ident          = XPRT_TRANSPORT_LOCAL,
3240         .setup          = xs_setup_local,
3241         .netid          = { "" },
3242 };
3243
3244 static struct xprt_class        xs_udp_transport = {
3245         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3246         .name           = "udp",
3247         .owner          = THIS_MODULE,
3248         .ident          = XPRT_TRANSPORT_UDP,
3249         .setup          = xs_setup_udp,
3250         .netid          = { "udp", "udp6", "" },
3251 };
3252
3253 static struct xprt_class        xs_tcp_transport = {
3254         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3255         .name           = "tcp",
3256         .owner          = THIS_MODULE,
3257         .ident          = XPRT_TRANSPORT_TCP,
3258         .setup          = xs_setup_tcp,
3259         .netid          = { "tcp", "tcp6", "" },
3260 };
3261
3262 static struct xprt_class        xs_bc_tcp_transport = {
3263         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3264         .name           = "tcp NFSv4.1 backchannel",
3265         .owner          = THIS_MODULE,
3266         .ident          = XPRT_TRANSPORT_BC_TCP,
3267         .setup          = xs_setup_bc_tcp,
3268         .netid          = { "" },
3269 };
3270
3271 /**
3272  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3273  *
3274  */
3275 int init_socket_xprt(void)
3276 {
3277         if (!sunrpc_table_header)
3278                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3279
3280         xprt_register_transport(&xs_local_transport);
3281         xprt_register_transport(&xs_udp_transport);
3282         xprt_register_transport(&xs_tcp_transport);
3283         xprt_register_transport(&xs_bc_tcp_transport);
3284
3285         return 0;
3286 }
3287
3288 /**
3289  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3290  *
3291  */
3292 void cleanup_socket_xprt(void)
3293 {
3294         if (sunrpc_table_header) {
3295                 unregister_sysctl_table(sunrpc_table_header);
3296                 sunrpc_table_header = NULL;
3297         }
3298
3299         xprt_unregister_transport(&xs_local_transport);
3300         xprt_unregister_transport(&xs_udp_transport);
3301         xprt_unregister_transport(&xs_tcp_transport);
3302         xprt_unregister_transport(&xs_bc_tcp_transport);
3303 }
3304
3305 static int param_set_uint_minmax(const char *val,
3306                 const struct kernel_param *kp,
3307                 unsigned int min, unsigned int max)
3308 {
3309         unsigned int num;
3310         int ret;
3311
3312         if (!val)
3313                 return -EINVAL;
3314         ret = kstrtouint(val, 0, &num);
3315         if (ret)
3316                 return ret;
3317         if (num < min || num > max)
3318                 return -EINVAL;
3319         *((unsigned int *)kp->arg) = num;
3320         return 0;
3321 }
3322
3323 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3324 {
3325         return param_set_uint_minmax(val, kp,
3326                         RPC_MIN_RESVPORT,
3327                         RPC_MAX_RESVPORT);
3328 }
3329
3330 static const struct kernel_param_ops param_ops_portnr = {
3331         .set = param_set_portnr,
3332         .get = param_get_uint,
3333 };
3334
3335 #define param_check_portnr(name, p) \
3336         __param_check(name, p, unsigned int);
3337
3338 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3339 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3340
3341 static int param_set_slot_table_size(const char *val,
3342                                      const struct kernel_param *kp)
3343 {
3344         return param_set_uint_minmax(val, kp,
3345                         RPC_MIN_SLOT_TABLE,
3346                         RPC_MAX_SLOT_TABLE);
3347 }
3348
3349 static const struct kernel_param_ops param_ops_slot_table_size = {
3350         .set = param_set_slot_table_size,
3351         .get = param_get_uint,
3352 };
3353
3354 #define param_check_slot_table_size(name, p) \
3355         __param_check(name, p, unsigned int);
3356
3357 static int param_set_max_slot_table_size(const char *val,
3358                                      const struct kernel_param *kp)
3359 {
3360         return param_set_uint_minmax(val, kp,
3361                         RPC_MIN_SLOT_TABLE,
3362                         RPC_MAX_SLOT_TABLE_LIMIT);
3363 }
3364
3365 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3366         .set = param_set_max_slot_table_size,
3367         .get = param_get_uint,
3368 };
3369
3370 #define param_check_max_slot_table_size(name, p) \
3371         __param_check(name, p, unsigned int);
3372
3373 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3374                    slot_table_size, 0644);
3375 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3376                    max_slot_table_size, 0644);
3377 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3378                    slot_table_size, 0644);