net/sunrpc/xprtsock.c

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