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