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 <net/tls_prot.h>
  51 #include <net/handshake.h>
  52
  53 #include <linux/bvec.h>
  54 #include <linux/highmem.h>
  55 #include <linux/uio.h>
  56 #include <linux/sched/mm.h>
  57
  58 #include <trace/events/sock.h>
  59 #include <trace/events/sunrpc.h>
  60
  61 #include "socklib.h"
  62 #include "sunrpc.h"
  63
  64 static void xs_close(struct rpc_xprt *xprt);
  65 static void xs_reset_srcport(struct sock_xprt *transport);
  66 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
  67 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  68                 struct socket *sock);
  69
  70 /*
  71  * xprtsock tunables
  72  */
  73 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  74 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  75 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  76
  77 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  78 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  79
  80 #define XS_TCP_LINGER_TO        (15U * HZ)
  81 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  82
  83 /*
  84  * We can register our own files under /proc/sys/sunrpc by
  85  * calling register_sysctl() again.  The files in that
  86  * directory become the union of all files registered there.
  87  *
  88  * We simply need to make sure that we don't collide with
  89  * someone else's file names!
  90  */
  91
  92 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  93 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  94 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  95 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  96 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  97
  98 static struct ctl_table_header *sunrpc_table_header;
  99
 100 static struct xprt_class xs_local_transport;
 101 static struct xprt_class xs_udp_transport;
 102 static struct xprt_class xs_tcp_transport;
 103 static struct xprt_class xs_tcp_tls_transport;
 104 static struct xprt_class xs_bc_tcp_transport;
 105
 106 /*
 107  * FIXME: changing the UDP slot table size should also resize the UDP
 108  *        socket buffers for existing UDP transports
 109  */
 110 static struct ctl_table xs_tunables_table[] = {
 111         {
 112                 .procname       = "udp_slot_table_entries",
 113                 .data           = &xprt_udp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_slot_table_size
 119         },
 120         {
 121                 .procname       = "tcp_slot_table_entries",
 122                 .data           = &xprt_tcp_slot_table_entries,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &min_slot_table_size,
 127                 .extra2         = &max_slot_table_size
 128         },
 129         {
 130                 .procname       = "tcp_max_slot_table_entries",
 131                 .data           = &xprt_max_tcp_slot_table_entries,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &min_slot_table_size,
 136                 .extra2         = &max_tcp_slot_table_limit
 137         },
 138         {
 139                 .procname       = "min_resvport",
 140                 .data           = &xprt_min_resvport,
 141                 .maxlen         = sizeof(unsigned int),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_minmax,
 144                 .extra1         = &xprt_min_resvport_limit,
 145                 .extra2         = &xprt_max_resvport_limit
 146         },
 147         {
 148                 .procname       = "max_resvport",
 149                 .data           = &xprt_max_resvport,
 150                 .maxlen         = sizeof(unsigned int),
 151                 .mode           = 0644,
 152                 .proc_handler   = proc_dointvec_minmax,
 153                 .extra1         = &xprt_min_resvport_limit,
 154                 .extra2         = &xprt_max_resvport_limit
 155         },
 156         {
 157                 .procname       = "tcp_fin_timeout",
 158                 .data           = &xs_tcp_fin_timeout,
 159                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 160                 .mode           = 0644,
 161                 .proc_handler   = proc_dointvec_jiffies,
 162         },
 163 };
 164
 165 /*
 166  * Wait duration for a reply from the RPC portmapper.
 167  */
 168 #define XS_BIND_TO              (60U * HZ)
 169
 170 /*
 171  * Delay if a UDP socket connect error occurs.  This is most likely some
 172  * kind of resource problem on the local host.
 173  */
 174 #define XS_UDP_REEST_TO         (2U * HZ)
 175
 176 /*
 177  * The reestablish timeout allows clients to delay for a bit before attempting
 178  * to reconnect to a server that just dropped our connection.
 179  *
 180  * We implement an exponential backoff when trying to reestablish a TCP
 181  * transport connection with the server.  Some servers like to drop a TCP
 182  * connection when they are overworked, so we start with a short timeout and
 183  * increase over time if the server is down or not responding.
 184  */
 185 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 186
 187 /*
 188  * TCP idle timeout; client drops the transport socket if it is idle
 189  * for this long.  Note that we also timeout UDP sockets to prevent
 190  * holding port numbers when there is no RPC traffic.
 191  */
 192 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 193
 194 /*
 195  * TLS handshake timeout.
 196  */
 197 #define XS_TLS_HANDSHAKE_TO     (10U * HZ)
 198
 199 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 200 # undef  RPC_DEBUG_DATA
 201 # define RPCDBG_FACILITY        RPCDBG_TRANS
 202 #endif
 203
 204 #ifdef RPC_DEBUG_DATA
 205 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 206 {
 207         u8 *buf = (u8 *) packet;
 208         int j;
 209
 210         dprintk("RPC:       %s\n", msg);
 211         for (j = 0; j < count && j < 128; j += 4) {
 212                 if (!(j & 31)) {
 213                         if (j)
 214                                 dprintk("\n");
 215                         dprintk("0x%04x ", j);
 216                 }
 217                 dprintk("%02x%02x%02x%02x ",
 218                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 219         }
 220         dprintk("\n");
 221 }
 222 #else
 223 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 224 {
 225         /* NOP */
 226 }
 227 #endif
 228
 229 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 230 {
 231         return (struct rpc_xprt *) sk->sk_user_data;
 232 }
 233
 234 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_un *) &xprt->addr;
 242 }
 243
 244 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 245 {
 246         return (struct sockaddr_in *) &xprt->addr;
 247 }
 248
 249 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 250 {
 251         return (struct sockaddr_in6 *) &xprt->addr;
 252 }
 253
 254 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 255 {
 256         struct sockaddr *sap = xs_addr(xprt);
 257         struct sockaddr_in6 *sin6;
 258         struct sockaddr_in *sin;
 259         struct sockaddr_un *sun;
 260         char buf[128];
 261
 262         switch (sap->sa_family) {
 263         case AF_LOCAL:
 264                 sun = xs_addr_un(xprt);
 265                 if (sun->sun_path[0]) {
 266                         strscpy(buf, sun->sun_path, sizeof(buf));
 267                 } else {
 268                         buf[0] = '@';
 269                         strscpy(buf+1, sun->sun_path+1, sizeof(buf)-1);
 270                 }
 271                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 272                                                 kstrdup(buf, GFP_KERNEL);
 273                 break;
 274         case AF_INET:
 275                 (void)rpc_ntop(sap, buf, sizeof(buf));
 276                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 277                                                 kstrdup(buf, GFP_KERNEL);
 278                 sin = xs_addr_in(xprt);
 279                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 280                 break;
 281         case AF_INET6:
 282                 (void)rpc_ntop(sap, buf, sizeof(buf));
 283                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 284                                                 kstrdup(buf, GFP_KERNEL);
 285                 sin6 = xs_addr_in6(xprt);
 286                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 287                 break;
 288         default:
 289                 BUG();
 290         }
 291
 292         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 293 }
 294
 295 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 296 {
 297         struct sockaddr *sap = xs_addr(xprt);
 298         char buf[128];
 299
 300         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 301         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 302
 303         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 304         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 305 }
 306
 307 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 308                                      const char *protocol,
 309                                      const char *netid)
 310 {
 311         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 312         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 313         xs_format_common_peer_addresses(xprt);
 314         xs_format_common_peer_ports(xprt);
 315 }
 316
 317 static void xs_update_peer_port(struct rpc_xprt *xprt)
 318 {
 319         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 320         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 321
 322         xs_format_common_peer_ports(xprt);
 323 }
 324
 325 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 326 {
 327         unsigned int i;
 328
 329         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 330                 switch (i) {
 331                 case RPC_DISPLAY_PROTO:
 332                 case RPC_DISPLAY_NETID:
 333                         continue;
 334                 default:
 335                         kfree(xprt->address_strings[i]);
 336                 }
 337 }
 338
 339 static size_t
 340 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 341 {
 342         size_t i,n;
 343
 344         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 345                 return want;
 346         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 347         for (i = 0; i < n; i++) {
 348                 if (buf->pages[i])
 349                         continue;
 350                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 351                 if (!buf->pages[i]) {
 352                         i *= PAGE_SIZE;
 353                         return i > buf->page_base ? i - buf->page_base : 0;
 354                 }
 355         }
 356         return want;
 357 }
 358
 359 static int
 360 xs_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
 361                      struct cmsghdr *cmsg, int ret)
 362 {
 363         u8 content_type = tls_get_record_type(sock->sk, cmsg);
 364         u8 level, description;
 365
 366         switch (content_type) {
 367         case 0:
 368                 break;
 369         case TLS_RECORD_TYPE_DATA:
 370                 /* TLS sets EOR at the end of each application data
 371                  * record, even though there might be more frames
 372                  * waiting to be decrypted.
 373                  */
 374                 msg->msg_flags &= ~MSG_EOR;
 375                 break;
 376         case TLS_RECORD_TYPE_ALERT:
 377                 tls_alert_recv(sock->sk, msg, &level, &description);
 378                 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
 379                         -EACCES : -EAGAIN;
 380                 break;
 381         default:
 382                 /* discard this record type */
 383                 ret = -EAGAIN;
 384         }
 385         return ret;
 386 }
 387
 388 static int
 389 xs_sock_recv_cmsg(struct socket *sock, struct msghdr *msg, int flags)
 390 {
 391         union {
 392                 struct cmsghdr  cmsg;
 393                 u8              buf[CMSG_SPACE(sizeof(u8))];
 394         } u;
 395         int ret;
 396
 397         msg->msg_control = &u;
 398         msg->msg_controllen = sizeof(u);
 399         ret = sock_recvmsg(sock, msg, flags);
 400         if (msg->msg_controllen != sizeof(u))
 401                 ret = xs_sock_process_cmsg(sock, msg, &u.cmsg, ret);
 402         return ret;
 403 }
 404
 405 static ssize_t
 406 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 407 {
 408         ssize_t ret;
 409         if (seek != 0)
 410                 iov_iter_advance(&msg->msg_iter, seek);
 411         ret = xs_sock_recv_cmsg(sock, msg, flags);
 412         return ret > 0 ? ret + seek : ret;
 413 }
 414
 415 static ssize_t
 416 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 417                 struct kvec *kvec, size_t count, size_t seek)
 418 {
 419         iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
 420         return xs_sock_recvmsg(sock, msg, flags, seek);
 421 }
 422
 423 static ssize_t
 424 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 425                 struct bio_vec *bvec, unsigned long nr, size_t count,
 426                 size_t seek)
 427 {
 428         iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
 429         return xs_sock_recvmsg(sock, msg, flags, seek);
 430 }
 431
 432 static ssize_t
 433 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 434                 size_t count)
 435 {
 436         iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
 437         return xs_sock_recv_cmsg(sock, msg, flags);
 438 }
 439
 440 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 441 static void
 442 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 443 {
 444         struct bvec_iter bi = {
 445                 .bi_size = count,
 446         };
 447         struct bio_vec bv;
 448
 449         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 450         for_each_bvec(bv, bvec, bi, bi)
 451                 flush_dcache_page(bv.bv_page);
 452 }
 453 #else
 454 static inline void
 455 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 456 {
 457 }
 458 #endif
 459
 460 static ssize_t
 461 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 462                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 463 {
 464         size_t want, seek_init = seek, offset = 0;
 465         ssize_t ret;
 466
 467         want = min_t(size_t, count, buf->head[0].iov_len);
 468         if (seek < want) {
 469                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 470                 if (ret <= 0)
 471                         goto sock_err;
 472                 offset += ret;
 473                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 474                         goto out;
 475                 if (ret != want)
 476                         goto out;
 477                 seek = 0;
 478         } else {
 479                 seek -= want;
 480                 offset += want;
 481         }
 482
 483         want = xs_alloc_sparse_pages(
 484                 buf, min_t(size_t, count - offset, buf->page_len),
 485                 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
 486         if (seek < want) {
 487                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 488                                 xdr_buf_pagecount(buf),
 489                                 want + buf->page_base,
 490                                 seek + buf->page_base);
 491                 if (ret <= 0)
 492                         goto sock_err;
 493                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 494                 ret -= buf->page_base;
 495                 offset += ret;
 496                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 497                         goto out;
 498                 if (ret != want)
 499                         goto out;
 500                 seek = 0;
 501         } else {
 502                 seek -= want;
 503                 offset += want;
 504         }
 505
 506         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 507         if (seek < want) {
 508                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 509                 if (ret <= 0)
 510                         goto sock_err;
 511                 offset += ret;
 512                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 513                         goto out;
 514                 if (ret != want)
 515                         goto out;
 516         } else if (offset < seek_init)
 517                 offset = seek_init;
 518         ret = -EMSGSIZE;
 519 out:
 520         *read = offset - seek_init;
 521         return ret;
 522 sock_err:
 523         offset += seek;
 524         goto out;
 525 }
 526
 527 static void
 528 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 529 {
 530         if (!transport->recv.copied) {
 531                 if (buf->head[0].iov_len >= transport->recv.offset)
 532                         memcpy(buf->head[0].iov_base,
 533                                         &transport->recv.xid,
 534                                         transport->recv.offset);
 535                 transport->recv.copied = transport->recv.offset;
 536         }
 537 }
 538
 539 static bool
 540 xs_read_stream_request_done(struct sock_xprt *transport)
 541 {
 542         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 543 }
 544
 545 static void
 546 xs_read_stream_check_eor(struct sock_xprt *transport,
 547                 struct msghdr *msg)
 548 {
 549         if (xs_read_stream_request_done(transport))
 550                 msg->msg_flags |= MSG_EOR;
 551 }
 552
 553 static ssize_t
 554 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 555                 int flags, struct rpc_rqst *req)
 556 {
 557         struct xdr_buf *buf = &req->rq_private_buf;
 558         size_t want, read;
 559         ssize_t ret;
 560
 561         xs_read_header(transport, buf);
 562
 563         want = transport->recv.len - transport->recv.offset;
 564         if (want != 0) {
 565                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 566                                 transport->recv.copied + want,
 567                                 transport->recv.copied,
 568                                 &read);
 569                 transport->recv.offset += read;
 570                 transport->recv.copied += read;
 571         }
 572
 573         if (transport->recv.offset == transport->recv.len)
 574                 xs_read_stream_check_eor(transport, msg);
 575
 576         if (want == 0)
 577                 return 0;
 578
 579         switch (ret) {
 580         default:
 581                 break;
 582         case -EFAULT:
 583         case -EMSGSIZE:
 584                 msg->msg_flags |= MSG_TRUNC;
 585                 return read;
 586         case 0:
 587                 return -ESHUTDOWN;
 588         }
 589         return ret < 0 ? ret : read;
 590 }
 591
 592 static size_t
 593 xs_read_stream_headersize(bool isfrag)
 594 {
 595         if (isfrag)
 596                 return sizeof(__be32);
 597         return 3 * sizeof(__be32);
 598 }
 599
 600 static ssize_t
 601 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 602                 int flags, size_t want, size_t seek)
 603 {
 604         struct kvec kvec = {
 605                 .iov_base = &transport->recv.fraghdr,
 606                 .iov_len = want,
 607         };
 608         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 609 }
 610
 611 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 612 static ssize_t
 613 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 614 {
 615         struct rpc_xprt *xprt = &transport->xprt;
 616         struct rpc_rqst *req;
 617         ssize_t ret;
 618
 619         /* Is this transport associated with the backchannel? */
 620         if (!xprt->bc_serv)
 621                 return -ESHUTDOWN;
 622
 623         /* Look up and lock the request corresponding to the given XID */
 624         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 625         if (!req) {
 626                 printk(KERN_WARNING "Callback slot table overflowed\n");
 627                 return -ESHUTDOWN;
 628         }
 629         if (transport->recv.copied && !req->rq_private_buf.len)
 630                 return -ESHUTDOWN;
 631
 632         ret = xs_read_stream_request(transport, msg, flags, req);
 633         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 634                 xprt_complete_bc_request(req, transport->recv.copied);
 635         else
 636                 req->rq_private_buf.len = transport->recv.copied;
 637
 638         return ret;
 639 }
 640 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 641 static ssize_t
 642 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 643 {
 644         return -ESHUTDOWN;
 645 }
 646 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 647
 648 static ssize_t
 649 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 650 {
 651         struct rpc_xprt *xprt = &transport->xprt;
 652         struct rpc_rqst *req;
 653         ssize_t ret = 0;
 654
 655         /* Look up and lock the request corresponding to the given XID */
 656         spin_lock(&xprt->queue_lock);
 657         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 658         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 659                 msg->msg_flags |= MSG_TRUNC;
 660                 goto out;
 661         }
 662         xprt_pin_rqst(req);
 663         spin_unlock(&xprt->queue_lock);
 664
 665         ret = xs_read_stream_request(transport, msg, flags, req);
 666
 667         spin_lock(&xprt->queue_lock);
 668         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 669                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 670         else
 671                 req->rq_private_buf.len = transport->recv.copied;
 672         xprt_unpin_rqst(req);
 673 out:
 674         spin_unlock(&xprt->queue_lock);
 675         return ret;
 676 }
 677
 678 static ssize_t
 679 xs_read_stream(struct sock_xprt *transport, int flags)
 680 {
 681         struct msghdr msg = { 0 };
 682         size_t want, read = 0;
 683         ssize_t ret = 0;
 684
 685         if (transport->recv.len == 0) {
 686                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 687                 ret = xs_read_stream_header(transport, &msg, flags, want,
 688                                 transport->recv.offset);
 689                 if (ret <= 0)
 690                         goto out_err;
 691                 transport->recv.offset = ret;
 692                 if (transport->recv.offset != want)
 693                         return transport->recv.offset;
 694                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 695                         RPC_FRAGMENT_SIZE_MASK;
 696                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 697                 read = ret;
 698         }
 699
 700         switch (be32_to_cpu(transport->recv.calldir)) {
 701         default:
 702                 msg.msg_flags |= MSG_TRUNC;
 703                 break;
 704         case RPC_CALL:
 705                 ret = xs_read_stream_call(transport, &msg, flags);
 706                 break;
 707         case RPC_REPLY:
 708                 ret = xs_read_stream_reply(transport, &msg, flags);
 709         }
 710         if (msg.msg_flags & MSG_TRUNC) {
 711                 transport->recv.calldir = cpu_to_be32(-1);
 712                 transport->recv.copied = -1;
 713         }
 714         if (ret < 0)
 715                 goto out_err;
 716         read += ret;
 717         if (transport->recv.offset < transport->recv.len) {
 718                 if (!(msg.msg_flags & MSG_TRUNC))
 719                         return read;
 720                 msg.msg_flags = 0;
 721                 ret = xs_read_discard(transport->sock, &msg, flags,
 722                                 transport->recv.len - transport->recv.offset);
 723                 if (ret <= 0)
 724                         goto out_err;
 725                 transport->recv.offset += ret;
 726                 read += ret;
 727                 if (transport->recv.offset != transport->recv.len)
 728                         return read;
 729         }
 730         if (xs_read_stream_request_done(transport)) {
 731                 trace_xs_stream_read_request(transport);
 732                 transport->recv.copied = 0;
 733         }
 734         transport->recv.offset = 0;
 735         transport->recv.len = 0;
 736         return read;
 737 out_err:
 738         return ret != 0 ? ret : -ESHUTDOWN;
 739 }
 740
 741 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 742 {
 743         return transport->sock->ops->poll(transport->file, transport->sock,
 744                         NULL);
 745 }
 746
 747 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 748 {
 749         __poll_t events = xs_poll_socket(transport);
 750
 751         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 752 }
 753
 754 static void xs_poll_check_readable(struct sock_xprt *transport)
 755 {
 756
 757         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 758         if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
 759                 return;
 760         if (!xs_poll_socket_readable(transport))
 761                 return;
 762         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 763                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 764 }
 765
 766 static void xs_stream_data_receive(struct sock_xprt *transport)
 767 {
 768         size_t read = 0;
 769         ssize_t ret = 0;
 770
 771         mutex_lock(&transport->recv_mutex);
 772         if (transport->sock == NULL)
 773                 goto out;
 774         for (;;) {
 775                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 776                 if (ret < 0)
 777                         break;
 778                 read += ret;
 779                 cond_resched();
 780         }
 781         if (ret == -ESHUTDOWN)
 782                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 783         else if (ret == -EACCES)
 784                 xprt_wake_pending_tasks(&transport->xprt, -EACCES);
 785         else
 786                 xs_poll_check_readable(transport);
 787 out:
 788         mutex_unlock(&transport->recv_mutex);
 789         trace_xs_stream_read_data(&transport->xprt, ret, read);
 790 }
 791
 792 static void xs_stream_data_receive_workfn(struct work_struct *work)
 793 {
 794         struct sock_xprt *transport =
 795                 container_of(work, struct sock_xprt, recv_worker);
 796         unsigned int pflags = memalloc_nofs_save();
 797
 798         xs_stream_data_receive(transport);
 799         memalloc_nofs_restore(pflags);
 800 }
 801
 802 static void
 803 xs_stream_reset_connect(struct sock_xprt *transport)
 804 {
 805         transport->recv.offset = 0;
 806         transport->recv.len = 0;
 807         transport->recv.copied = 0;
 808         transport->xmit.offset = 0;
 809 }
 810
 811 static void
 812 xs_stream_start_connect(struct sock_xprt *transport)
 813 {
 814         transport->xprt.stat.connect_count++;
 815         transport->xprt.stat.connect_start = jiffies;
 816 }
 817
 818 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 819
 820 /**
 821  * xs_nospace - handle transmit was incomplete
 822  * @req: pointer to RPC request
 823  * @transport: pointer to struct sock_xprt
 824  *
 825  */
 826 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 827 {
 828         struct rpc_xprt *xprt = &transport->xprt;
 829         struct sock *sk = transport->inet;
 830         int ret = -EAGAIN;
 831
 832         trace_rpc_socket_nospace(req, transport);
 833
 834         /* Protect against races with write_space */
 835         spin_lock(&xprt->transport_lock);
 836
 837         /* Don't race with disconnect */
 838         if (xprt_connected(xprt)) {
 839                 /* wait for more buffer space */
 840                 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
 841                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 842                 sk->sk_write_pending++;
 843                 xprt_wait_for_buffer_space(xprt);
 844         } else
 845                 ret = -ENOTCONN;
 846
 847         spin_unlock(&xprt->transport_lock);
 848         return ret;
 849 }
 850
 851 static int xs_sock_nospace(struct rpc_rqst *req)
 852 {
 853         struct sock_xprt *transport =
 854                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 855         struct sock *sk = transport->inet;
 856         int ret = -EAGAIN;
 857
 858         lock_sock(sk);
 859         if (!sock_writeable(sk))
 860                 ret = xs_nospace(req, transport);
 861         release_sock(sk);
 862         return ret;
 863 }
 864
 865 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
 866 {
 867         struct sock_xprt *transport =
 868                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 869         struct sock *sk = transport->inet;
 870         int ret = -EAGAIN;
 871
 872         if (vm_wait)
 873                 return -ENOBUFS;
 874         lock_sock(sk);
 875         if (!sk_stream_memory_free(sk))
 876                 ret = xs_nospace(req, transport);
 877         release_sock(sk);
 878         return ret;
 879 }
 880
 881 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
 882 {
 883         return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
 884 }
 885
 886 static void xs_stream_abort_send_request(struct rpc_rqst *req)
 887 {
 888         struct rpc_xprt *xprt = req->rq_xprt;
 889         struct sock_xprt *transport =
 890                 container_of(xprt, struct sock_xprt, xprt);
 891
 892         if (transport->xmit.offset != 0 &&
 893             !test_bit(XPRT_CLOSE_WAIT, &xprt->state))
 894                 xprt_force_disconnect(xprt);
 895 }
 896
 897 /*
 898  * Determine if the previous message in the stream was aborted before it
 899  * could complete transmission.
 900  */
 901 static bool
 902 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 903 {
 904         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 905 }
 906
 907 /*
 908  * Return the stream record marker field for a record of length < 2^31-1
 909  */
 910 static rpc_fraghdr
 911 xs_stream_record_marker(struct xdr_buf *xdr)
 912 {
 913         if (!xdr->len)
 914                 return 0;
 915         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 916 }
 917
 918 /**
 919  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 920  * @req: pointer to RPC request
 921  *
 922  * Return values:
 923  *        0:    The request has been sent
 924  *   EAGAIN:    The socket was blocked, please call again later to
 925  *              complete the request
 926  * ENOTCONN:    Caller needs to invoke connect logic then call again
 927  *    other:    Some other error occurred, the request was not sent
 928  */
 929 static int xs_local_send_request(struct rpc_rqst *req)
 930 {
 931         struct rpc_xprt *xprt = req->rq_xprt;
 932         struct sock_xprt *transport =
 933                                 container_of(xprt, struct sock_xprt, xprt);
 934         struct xdr_buf *xdr = &req->rq_snd_buf;
 935         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 936         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 937         struct msghdr msg = {
 938                 .msg_flags      = XS_SENDMSG_FLAGS,
 939         };
 940         bool vm_wait;
 941         unsigned int sent;
 942         int status;
 943
 944         /* Close the stream if the previous transmission was incomplete */
 945         if (xs_send_request_was_aborted(transport, req)) {
 946                 xprt_force_disconnect(xprt);
 947                 return -ENOTCONN;
 948         }
 949
 950         xs_pktdump("packet data:",
 951                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 952
 953         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
 954
 955         req->rq_xtime = ktime_get();
 956         status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 957                                    transport->xmit.offset, rm, &sent);
 958         dprintk("RPC:       %s(%u) = %d\n",
 959                         __func__, xdr->len - transport->xmit.offset, status);
 960
 961         if (likely(sent > 0) || status == 0) {
 962                 transport->xmit.offset += sent;
 963                 req->rq_bytes_sent = transport->xmit.offset;
 964                 if (likely(req->rq_bytes_sent >= msglen)) {
 965                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 966                         transport->xmit.offset = 0;
 967                         return 0;
 968                 }
 969                 status = -EAGAIN;
 970                 vm_wait = false;
 971         }
 972
 973         switch (status) {
 974         case -EAGAIN:
 975                 status = xs_stream_nospace(req, vm_wait);
 976                 break;
 977         default:
 978                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 979                         -status);
 980                 fallthrough;
 981         case -EPIPE:
 982                 xprt_force_disconnect(xprt);
 983                 status = -ENOTCONN;
 984         }
 985
 986         return status;
 987 }
 988
 989 /**
 990  * xs_udp_send_request - write an RPC request to a UDP socket
 991  * @req: pointer to RPC request
 992  *
 993  * Return values:
 994  *        0:    The request has been sent
 995  *   EAGAIN:    The socket was blocked, please call again later to
 996  *              complete the request
 997  * ENOTCONN:    Caller needs to invoke connect logic then call again
 998  *    other:    Some other error occurred, the request was not sent
 999  */
1000 static int xs_udp_send_request(struct rpc_rqst *req)
1001 {
1002         struct rpc_xprt *xprt = req->rq_xprt;
1003         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1004         struct xdr_buf *xdr = &req->rq_snd_buf;
1005         struct msghdr msg = {
1006                 .msg_name       = xs_addr(xprt),
1007                 .msg_namelen    = xprt->addrlen,
1008                 .msg_flags      = XS_SENDMSG_FLAGS,
1009         };
1010         unsigned int sent;
1011         int status;
1012
1013         xs_pktdump("packet data:",
1014                                 req->rq_svec->iov_base,
1015                                 req->rq_svec->iov_len);
1016
1017         if (!xprt_bound(xprt))
1018                 return -ENOTCONN;
1019
1020         if (!xprt_request_get_cong(xprt, req))
1021                 return -EBADSLT;
1022
1023         status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
1024         if (status < 0)
1025                 return status;
1026         req->rq_xtime = ktime_get();
1027         status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
1028
1029         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
1030                         xdr->len, status);
1031
1032         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
1033         if (status == -EPERM)
1034                 goto process_status;
1035
1036         if (status == -EAGAIN && sock_writeable(transport->inet))
1037                 status = -ENOBUFS;
1038
1039         if (sent > 0 || status == 0) {
1040                 req->rq_xmit_bytes_sent += sent;
1041                 if (sent >= req->rq_slen)
1042                         return 0;
1043                 /* Still some bytes left; set up for a retry later. */
1044                 status = -EAGAIN;
1045         }
1046
1047 process_status:
1048         switch (status) {
1049         case -ENOTSOCK:
1050                 status = -ENOTCONN;
1051                 /* Should we call xs_close() here? */
1052                 break;
1053         case -EAGAIN:
1054                 status = xs_sock_nospace(req);
1055                 break;
1056         case -ENETUNREACH:
1057         case -ENOBUFS:
1058         case -EPIPE:
1059         case -ECONNREFUSED:
1060         case -EPERM:
1061                 /* When the server has died, an ICMP port unreachable message
1062                  * prompts ECONNREFUSED. */
1063                 break;
1064         default:
1065                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1066                         -status);
1067         }
1068
1069         return status;
1070 }
1071
1072 /**
1073  * xs_tcp_send_request - write an RPC request to a TCP socket
1074  * @req: pointer to RPC request
1075  *
1076  * Return values:
1077  *        0:    The request has been sent
1078  *   EAGAIN:    The socket was blocked, please call again later to
1079  *              complete the request
1080  * ENOTCONN:    Caller needs to invoke connect logic then call again
1081  *    other:    Some other error occurred, the request was not sent
1082  *
1083  * XXX: In the case of soft timeouts, should we eventually give up
1084  *      if sendmsg is not able to make progress?
1085  */
1086 static int xs_tcp_send_request(struct rpc_rqst *req)
1087 {
1088         struct rpc_xprt *xprt = req->rq_xprt;
1089         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1090         struct xdr_buf *xdr = &req->rq_snd_buf;
1091         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1092         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1093         struct msghdr msg = {
1094                 .msg_flags      = XS_SENDMSG_FLAGS,
1095         };
1096         bool vm_wait;
1097         unsigned int sent;
1098         int status;
1099
1100         /* Close the stream if the previous transmission was incomplete */
1101         if (xs_send_request_was_aborted(transport, req)) {
1102                 if (transport->sock != NULL)
1103                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1104                 return -ENOTCONN;
1105         }
1106         if (!transport->inet)
1107                 return -ENOTCONN;
1108
1109         xs_pktdump("packet data:",
1110                                 req->rq_svec->iov_base,
1111                                 req->rq_svec->iov_len);
1112
1113         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1114                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1115
1116         xs_set_srcport(transport, transport->sock);
1117
1118         /* Continue transmitting the packet/record. We must be careful
1119          * to cope with writespace callbacks arriving _after_ we have
1120          * called sendmsg(). */
1121         req->rq_xtime = ktime_get();
1122         tcp_sock_set_cork(transport->inet, true);
1123
1124         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1125
1126         do {
1127                 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1128                                            transport->xmit.offset, rm, &sent);
1129
1130                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1131                                 xdr->len - transport->xmit.offset, status);
1132
1133                 /* If we've sent the entire packet, immediately
1134                  * reset the count of bytes sent. */
1135                 transport->xmit.offset += sent;
1136                 req->rq_bytes_sent = transport->xmit.offset;
1137                 if (likely(req->rq_bytes_sent >= msglen)) {
1138                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1139                         transport->xmit.offset = 0;
1140                         if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1141                                 tcp_sock_set_cork(transport->inet, false);
1142                         return 0;
1143                 }
1144
1145                 WARN_ON_ONCE(sent == 0 && status == 0);
1146
1147                 if (sent > 0)
1148                         vm_wait = false;
1149
1150         } while (status == 0);
1151
1152         switch (status) {
1153         case -ENOTSOCK:
1154                 status = -ENOTCONN;
1155                 /* Should we call xs_close() here? */
1156                 break;
1157         case -EAGAIN:
1158                 status = xs_stream_nospace(req, vm_wait);
1159                 break;
1160         case -ECONNRESET:
1161         case -ECONNREFUSED:
1162         case -ENOTCONN:
1163         case -EADDRINUSE:
1164         case -ENOBUFS:
1165         case -EPIPE:
1166                 break;
1167         default:
1168                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1169                         -status);
1170         }
1171
1172         return status;
1173 }
1174
1175 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1176 {
1177         transport->old_data_ready = sk->sk_data_ready;
1178         transport->old_state_change = sk->sk_state_change;
1179         transport->old_write_space = sk->sk_write_space;
1180         transport->old_error_report = sk->sk_error_report;
1181 }
1182
1183 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1184 {
1185         sk->sk_data_ready = transport->old_data_ready;
1186         sk->sk_state_change = transport->old_state_change;
1187         sk->sk_write_space = transport->old_write_space;
1188         sk->sk_error_report = transport->old_error_report;
1189 }
1190
1191 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1192 {
1193         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1194
1195         transport->xprt_err = 0;
1196         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1197         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1198         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1199         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1200         clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1201         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
1202 }
1203
1204 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1205 {
1206         set_bit(nr, &transport->sock_state);
1207         queue_work(xprtiod_workqueue, &transport->error_worker);
1208 }
1209
1210 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1211 {
1212         xprt->connect_cookie++;
1213         smp_mb__before_atomic();
1214         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1215         clear_bit(XPRT_CLOSING, &xprt->state);
1216         xs_sock_reset_state_flags(xprt);
1217         smp_mb__after_atomic();
1218 }
1219
1220 /**
1221  * xs_error_report - callback to handle TCP socket state errors
1222  * @sk: socket
1223  *
1224  * Note: we don't call sock_error() since there may be a rpc_task
1225  * using the socket, and so we don't want to clear sk->sk_err.
1226  */
1227 static void xs_error_report(struct sock *sk)
1228 {
1229         struct sock_xprt *transport;
1230         struct rpc_xprt *xprt;
1231
1232         if (!(xprt = xprt_from_sock(sk)))
1233                 return;
1234
1235         transport = container_of(xprt, struct sock_xprt, xprt);
1236         transport->xprt_err = -sk->sk_err;
1237         if (transport->xprt_err == 0)
1238                 return;
1239         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1240                         xprt, -transport->xprt_err);
1241         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1242
1243         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1244         smp_mb__before_atomic();
1245         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1246 }
1247
1248 static void xs_reset_transport(struct sock_xprt *transport)
1249 {
1250         struct socket *sock = transport->sock;
1251         struct sock *sk = transport->inet;
1252         struct rpc_xprt *xprt = &transport->xprt;
1253         struct file *filp = transport->file;
1254
1255         if (sk == NULL)
1256                 return;
1257         /*
1258          * Make sure we're calling this in a context from which it is safe
1259          * to call __fput_sync(). In practice that means rpciod and the
1260          * system workqueue.
1261          */
1262         if (!(current->flags & PF_WQ_WORKER)) {
1263                 WARN_ON_ONCE(1);
1264                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1265                 return;
1266         }
1267
1268         if (atomic_read(&transport->xprt.swapper))
1269                 sk_clear_memalloc(sk);
1270
1271         tls_handshake_cancel(sk);
1272
1273         kernel_sock_shutdown(sock, SHUT_RDWR);
1274
1275         mutex_lock(&transport->recv_mutex);
1276         lock_sock(sk);
1277         transport->inet = NULL;
1278         transport->sock = NULL;
1279         transport->file = NULL;
1280
1281         sk->sk_user_data = NULL;
1282         sk->sk_sndtimeo = 0;
1283
1284         xs_restore_old_callbacks(transport, sk);
1285         xprt_clear_connected(xprt);
1286         xs_sock_reset_connection_flags(xprt);
1287         /* Reset stream record info */
1288         xs_stream_reset_connect(transport);
1289         release_sock(sk);
1290         mutex_unlock(&transport->recv_mutex);
1291
1292         trace_rpc_socket_close(xprt, sock);
1293         __fput_sync(filp);
1294
1295         xprt_disconnect_done(xprt);
1296 }
1297
1298 /**
1299  * xs_close - close a socket
1300  * @xprt: transport
1301  *
1302  * This is used when all requests are complete; ie, no DRC state remains
1303  * on the server we want to save.
1304  *
1305  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1306  * xs_reset_transport() zeroing the socket from underneath a writer.
1307  */
1308 static void xs_close(struct rpc_xprt *xprt)
1309 {
1310         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1311
1312         dprintk("RPC:       xs_close xprt %p\n", xprt);
1313
1314         if (transport->sock)
1315                 tls_handshake_close(transport->sock);
1316         xs_reset_transport(transport);
1317         xprt->reestablish_timeout = 0;
1318 }
1319
1320 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1321 {
1322         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1323                 xprt);
1324         xprt_disconnect_done(xprt);
1325 }
1326
1327 static void xs_xprt_free(struct rpc_xprt *xprt)
1328 {
1329         xs_free_peer_addresses(xprt);
1330         xprt_free(xprt);
1331 }
1332
1333 /**
1334  * xs_destroy - prepare to shutdown a transport
1335  * @xprt: doomed transport
1336  *
1337  */
1338 static void xs_destroy(struct rpc_xprt *xprt)
1339 {
1340         struct sock_xprt *transport = container_of(xprt,
1341                         struct sock_xprt, xprt);
1342         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1343
1344         cancel_delayed_work_sync(&transport->connect_worker);
1345         xs_close(xprt);
1346         cancel_work_sync(&transport->recv_worker);
1347         cancel_work_sync(&transport->error_worker);
1348         xs_xprt_free(xprt);
1349         module_put(THIS_MODULE);
1350 }
1351
1352 /**
1353  * xs_udp_data_read_skb - receive callback for UDP sockets
1354  * @xprt: transport
1355  * @sk: socket
1356  * @skb: skbuff
1357  *
1358  */
1359 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1360                 struct sock *sk,
1361                 struct sk_buff *skb)
1362 {
1363         struct rpc_task *task;
1364         struct rpc_rqst *rovr;
1365         int repsize, copied;
1366         u32 _xid;
1367         __be32 *xp;
1368
1369         repsize = skb->len;
1370         if (repsize < 4) {
1371                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1372                 return;
1373         }
1374
1375         /* Copy the XID from the skb... */
1376         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1377         if (xp == NULL)
1378                 return;
1379
1380         /* Look up and lock the request corresponding to the given XID */
1381         spin_lock(&xprt->queue_lock);
1382         rovr = xprt_lookup_rqst(xprt, *xp);
1383         if (!rovr)
1384                 goto out_unlock;
1385         xprt_pin_rqst(rovr);
1386         xprt_update_rtt(rovr->rq_task);
1387         spin_unlock(&xprt->queue_lock);
1388         task = rovr->rq_task;
1389
1390         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1391                 copied = repsize;
1392
1393         /* Suck it into the iovec, verify checksum if not done by hw. */
1394         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1395                 spin_lock(&xprt->queue_lock);
1396                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1397                 goto out_unpin;
1398         }
1399
1400
1401         spin_lock(&xprt->transport_lock);
1402         xprt_adjust_cwnd(xprt, task, copied);
1403         spin_unlock(&xprt->transport_lock);
1404         spin_lock(&xprt->queue_lock);
1405         xprt_complete_rqst(task, copied);
1406         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1407 out_unpin:
1408         xprt_unpin_rqst(rovr);
1409  out_unlock:
1410         spin_unlock(&xprt->queue_lock);
1411 }
1412
1413 static void xs_udp_data_receive(struct sock_xprt *transport)
1414 {
1415         struct sk_buff *skb;
1416         struct sock *sk;
1417         int err;
1418
1419         mutex_lock(&transport->recv_mutex);
1420         sk = transport->inet;
1421         if (sk == NULL)
1422                 goto out;
1423         for (;;) {
1424                 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1425                 if (skb == NULL)
1426                         break;
1427                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1428                 consume_skb(skb);
1429                 cond_resched();
1430         }
1431         xs_poll_check_readable(transport);
1432 out:
1433         mutex_unlock(&transport->recv_mutex);
1434 }
1435
1436 static void xs_udp_data_receive_workfn(struct work_struct *work)
1437 {
1438         struct sock_xprt *transport =
1439                 container_of(work, struct sock_xprt, recv_worker);
1440         unsigned int pflags = memalloc_nofs_save();
1441
1442         xs_udp_data_receive(transport);
1443         memalloc_nofs_restore(pflags);
1444 }
1445
1446 /**
1447  * xs_data_ready - "data ready" callback for sockets
1448  * @sk: socket with data to read
1449  *
1450  */
1451 static void xs_data_ready(struct sock *sk)
1452 {
1453         struct rpc_xprt *xprt;
1454
1455         trace_sk_data_ready(sk);
1456
1457         xprt = xprt_from_sock(sk);
1458         if (xprt != NULL) {
1459                 struct sock_xprt *transport = container_of(xprt,
1460                                 struct sock_xprt, xprt);
1461
1462                 trace_xs_data_ready(xprt);
1463
1464                 transport->old_data_ready(sk);
1465
1466                 if (test_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state))
1467                         return;
1468
1469                 /* Any data means we had a useful conversation, so
1470                  * then we don't need to delay the next reconnect
1471                  */
1472                 if (xprt->reestablish_timeout)
1473                         xprt->reestablish_timeout = 0;
1474                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1475                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1476         }
1477 }
1478
1479 /*
1480  * Helper function to force a TCP close if the server is sending
1481  * junk and/or it has put us in CLOSE_WAIT
1482  */
1483 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1484 {
1485         xprt_force_disconnect(xprt);
1486 }
1487
1488 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1489 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1490 {
1491         return PAGE_SIZE;
1492 }
1493 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1494
1495 /**
1496  * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1497  * @sk: socket whose state has changed
1498  *
1499  */
1500 static void xs_local_state_change(struct sock *sk)
1501 {
1502         struct rpc_xprt *xprt;
1503         struct sock_xprt *transport;
1504
1505         if (!(xprt = xprt_from_sock(sk)))
1506                 return;
1507         transport = container_of(xprt, struct sock_xprt, xprt);
1508         if (sk->sk_shutdown & SHUTDOWN_MASK) {
1509                 clear_bit(XPRT_CONNECTED, &xprt->state);
1510                 /* Trigger the socket release */
1511                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1512         }
1513 }
1514
1515 /**
1516  * xs_tcp_state_change - callback to handle TCP socket state changes
1517  * @sk: socket whose state has changed
1518  *
1519  */
1520 static void xs_tcp_state_change(struct sock *sk)
1521 {
1522         struct rpc_xprt *xprt;
1523         struct sock_xprt *transport;
1524
1525         if (!(xprt = xprt_from_sock(sk)))
1526                 return;
1527         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1528         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1529                         sk->sk_state, xprt_connected(xprt),
1530                         sock_flag(sk, SOCK_DEAD),
1531                         sock_flag(sk, SOCK_ZAPPED),
1532                         sk->sk_shutdown);
1533
1534         transport = container_of(xprt, struct sock_xprt, xprt);
1535         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1536         switch (sk->sk_state) {
1537         case TCP_ESTABLISHED:
1538                 if (!xprt_test_and_set_connected(xprt)) {
1539                         xprt->connect_cookie++;
1540                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1541                         xprt_clear_connecting(xprt);
1542
1543                         xprt->stat.connect_count++;
1544                         xprt->stat.connect_time += (long)jiffies -
1545                                                    xprt->stat.connect_start;
1546                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1547                 }
1548                 break;
1549         case TCP_FIN_WAIT1:
1550                 /* The client initiated a shutdown of the socket */
1551                 xprt->connect_cookie++;
1552                 xprt->reestablish_timeout = 0;
1553                 set_bit(XPRT_CLOSING, &xprt->state);
1554                 smp_mb__before_atomic();
1555                 clear_bit(XPRT_CONNECTED, &xprt->state);
1556                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1557                 smp_mb__after_atomic();
1558                 break;
1559         case TCP_CLOSE_WAIT:
1560                 /* The server initiated a shutdown of the socket */
1561                 xprt->connect_cookie++;
1562                 clear_bit(XPRT_CONNECTED, &xprt->state);
1563                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1564                 fallthrough;
1565         case TCP_CLOSING:
1566                 /*
1567                  * If the server closed down the connection, make sure that
1568                  * we back off before reconnecting
1569                  */
1570                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1571                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1572                 break;
1573         case TCP_LAST_ACK:
1574                 set_bit(XPRT_CLOSING, &xprt->state);
1575                 smp_mb__before_atomic();
1576                 clear_bit(XPRT_CONNECTED, &xprt->state);
1577                 smp_mb__after_atomic();
1578                 break;
1579         case TCP_CLOSE:
1580                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1581                                        &transport->sock_state)) {
1582                         xs_reset_srcport(transport);
1583                         xprt_clear_connecting(xprt);
1584                 }
1585                 clear_bit(XPRT_CLOSING, &xprt->state);
1586                 /* Trigger the socket release */
1587                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1588         }
1589 }
1590
1591 static void xs_write_space(struct sock *sk)
1592 {
1593         struct sock_xprt *transport;
1594         struct rpc_xprt *xprt;
1595
1596         if (!sk->sk_socket)
1597                 return;
1598         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1599
1600         if (unlikely(!(xprt = xprt_from_sock(sk))))
1601                 return;
1602         transport = container_of(xprt, struct sock_xprt, xprt);
1603         if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1604                 return;
1605         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1606         sk->sk_write_pending--;
1607 }
1608
1609 /**
1610  * xs_udp_write_space - callback invoked when socket buffer space
1611  *                             becomes available
1612  * @sk: socket whose state has changed
1613  *
1614  * Called when more output buffer space is available for this socket.
1615  * We try not to wake our writers until they can make "significant"
1616  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1617  * with a bunch of small requests.
1618  */
1619 static void xs_udp_write_space(struct sock *sk)
1620 {
1621         /* from net/core/sock.c:sock_def_write_space */
1622         if (sock_writeable(sk))
1623                 xs_write_space(sk);
1624 }
1625
1626 /**
1627  * xs_tcp_write_space - callback invoked when socket buffer space
1628  *                             becomes available
1629  * @sk: socket whose state has changed
1630  *
1631  * Called when more output buffer space is available for this socket.
1632  * We try not to wake our writers until they can make "significant"
1633  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1634  * with a bunch of small requests.
1635  */
1636 static void xs_tcp_write_space(struct sock *sk)
1637 {
1638         /* from net/core/stream.c:sk_stream_write_space */
1639         if (sk_stream_is_writeable(sk))
1640                 xs_write_space(sk);
1641 }
1642
1643 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1644 {
1645         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1646         struct sock *sk = transport->inet;
1647
1648         if (transport->rcvsize) {
1649                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1650                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1651         }
1652         if (transport->sndsize) {
1653                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1654                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1655                 sk->sk_write_space(sk);
1656         }
1657 }
1658
1659 /**
1660  * xs_udp_set_buffer_size - set send and receive limits
1661  * @xprt: generic transport
1662  * @sndsize: requested size of send buffer, in bytes
1663  * @rcvsize: requested size of receive buffer, in bytes
1664  *
1665  * Set socket send and receive buffer size limits.
1666  */
1667 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1668 {
1669         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1670
1671         transport->sndsize = 0;
1672         if (sndsize)
1673                 transport->sndsize = sndsize + 1024;
1674         transport->rcvsize = 0;
1675         if (rcvsize)
1676                 transport->rcvsize = rcvsize + 1024;
1677
1678         xs_udp_do_set_buffer_size(xprt);
1679 }
1680
1681 /**
1682  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1683  * @xprt: controlling transport
1684  * @task: task that timed out
1685  *
1686  * Adjust the congestion window after a retransmit timeout has occurred.
1687  */
1688 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1689 {
1690         spin_lock(&xprt->transport_lock);
1691         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1692         spin_unlock(&xprt->transport_lock);
1693 }
1694
1695 static int xs_get_random_port(void)
1696 {
1697         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1698         unsigned short range;
1699         unsigned short rand;
1700
1701         if (max < min)
1702                 return -EADDRINUSE;
1703         range = max - min + 1;
1704         rand = get_random_u32_below(range);
1705         return rand + min;
1706 }
1707
1708 static unsigned short xs_sock_getport(struct socket *sock)
1709 {
1710         struct sockaddr_storage buf;
1711         unsigned short port = 0;
1712
1713         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1714                 goto out;
1715         switch (buf.ss_family) {
1716         case AF_INET6:
1717                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1718                 break;
1719         case AF_INET:
1720                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1721         }
1722 out:
1723         return port;
1724 }
1725
1726 /**
1727  * xs_set_port - reset the port number in the remote endpoint address
1728  * @xprt: generic transport
1729  * @port: new port number
1730  *
1731  */
1732 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1733 {
1734         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1735
1736         rpc_set_port(xs_addr(xprt), port);
1737         xs_update_peer_port(xprt);
1738 }
1739
1740 static void xs_reset_srcport(struct sock_xprt *transport)
1741 {
1742         transport->srcport = 0;
1743 }
1744
1745 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1746 {
1747         if (transport->srcport == 0 && transport->xprt.reuseport)
1748                 transport->srcport = xs_sock_getport(sock);
1749 }
1750
1751 static int xs_get_srcport(struct sock_xprt *transport)
1752 {
1753         int port = transport->srcport;
1754
1755         if (port == 0 && transport->xprt.resvport)
1756                 port = xs_get_random_port();
1757         return port;
1758 }
1759
1760 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1761 {
1762         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1763         unsigned short ret = 0;
1764         mutex_lock(&sock->recv_mutex);
1765         if (sock->sock)
1766                 ret = xs_sock_getport(sock->sock);
1767         mutex_unlock(&sock->recv_mutex);
1768         return ret;
1769 }
1770
1771 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1772 {
1773         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1774         union {
1775                 struct sockaddr sa;
1776                 struct sockaddr_storage st;
1777         } saddr;
1778         int ret = -ENOTCONN;
1779
1780         mutex_lock(&sock->recv_mutex);
1781         if (sock->sock) {
1782                 ret = kernel_getsockname(sock->sock, &saddr.sa);
1783                 if (ret >= 0)
1784                         ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1785         }
1786         mutex_unlock(&sock->recv_mutex);
1787         return ret;
1788 }
1789
1790 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1791 {
1792         if (transport->srcport != 0)
1793                 transport->srcport = 0;
1794         if (!transport->xprt.resvport)
1795                 return 0;
1796         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1797                 return xprt_max_resvport;
1798         return --port;
1799 }
1800 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1801 {
1802         struct sockaddr_storage myaddr;
1803         int err, nloop = 0;
1804         int port = xs_get_srcport(transport);
1805         unsigned short last;
1806
1807         /*
1808          * If we are asking for any ephemeral port (i.e. port == 0 &&
1809          * transport->xprt.resvport == 0), don't bind.  Let the local
1810          * port selection happen implicitly when the socket is used
1811          * (for example at connect time).
1812          *
1813          * This ensures that we can continue to establish TCP
1814          * connections even when all local ephemeral ports are already
1815          * a part of some TCP connection.  This makes no difference
1816          * for UDP sockets, but also doesn't harm them.
1817          *
1818          * If we're asking for any reserved port (i.e. port == 0 &&
1819          * transport->xprt.resvport == 1) xs_get_srcport above will
1820          * ensure that port is non-zero and we will bind as needed.
1821          */
1822         if (port <= 0)
1823                 return port;
1824
1825         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1826         do {
1827                 rpc_set_port((struct sockaddr *)&myaddr, port);
1828                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1829                                 transport->xprt.addrlen);
1830                 if (err == 0) {
1831                         if (transport->xprt.reuseport)
1832                                 transport->srcport = port;
1833                         break;
1834                 }
1835                 last = port;
1836                 port = xs_next_srcport(transport, port);
1837                 if (port > last)
1838                         nloop++;
1839         } while (err == -EADDRINUSE && nloop != 2);
1840
1841         if (myaddr.ss_family == AF_INET)
1842                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1843                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1844                                 port, err ? "failed" : "ok", err);
1845         else
1846                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1847                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1848                                 port, err ? "failed" : "ok", err);
1849         return err;
1850 }
1851
1852 /*
1853  * We don't support autobind on AF_LOCAL sockets
1854  */
1855 static void xs_local_rpcbind(struct rpc_task *task)
1856 {
1857         xprt_set_bound(task->tk_xprt);
1858 }
1859
1860 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1861 {
1862 }
1863
1864 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1865 static struct lock_class_key xs_key[3];
1866 static struct lock_class_key xs_slock_key[3];
1867
1868 static inline void xs_reclassify_socketu(struct socket *sock)
1869 {
1870         struct sock *sk = sock->sk;
1871
1872         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1873                 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1874 }
1875
1876 static inline void xs_reclassify_socket4(struct socket *sock)
1877 {
1878         struct sock *sk = sock->sk;
1879
1880         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1881                 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1882 }
1883
1884 static inline void xs_reclassify_socket6(struct socket *sock)
1885 {
1886         struct sock *sk = sock->sk;
1887
1888         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1889                 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1890 }
1891
1892 static inline void xs_reclassify_socket(int family, struct socket *sock)
1893 {
1894         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1895                 return;
1896
1897         switch (family) {
1898         case AF_LOCAL:
1899                 xs_reclassify_socketu(sock);
1900                 break;
1901         case AF_INET:
1902                 xs_reclassify_socket4(sock);
1903                 break;
1904         case AF_INET6:
1905                 xs_reclassify_socket6(sock);
1906                 break;
1907         }
1908 }
1909 #else
1910 static inline void xs_reclassify_socket(int family, struct socket *sock)
1911 {
1912 }
1913 #endif
1914
1915 static void xs_dummy_setup_socket(struct work_struct *work)
1916 {
1917 }
1918
1919 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1920                 struct sock_xprt *transport, int family, int type,
1921                 int protocol, bool reuseport)
1922 {
1923         struct file *filp;
1924         struct socket *sock;
1925         int err;
1926
1927         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1928         if (err < 0) {
1929                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1930                                 protocol, -err);
1931                 goto out;
1932         }
1933         xs_reclassify_socket(family, sock);
1934
1935         if (reuseport)
1936                 sock_set_reuseport(sock->sk);
1937
1938         err = xs_bind(transport, sock);
1939         if (err) {
1940                 sock_release(sock);
1941                 goto out;
1942         }
1943
1944         if (protocol == IPPROTO_TCP)
1945                 sk_net_refcnt_upgrade(sock->sk);
1946
1947         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1948         if (IS_ERR(filp))
1949                 return ERR_CAST(filp);
1950         transport->file = filp;
1951
1952         return sock;
1953 out:
1954         return ERR_PTR(err);
1955 }
1956
1957 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1958                                       struct socket *sock)
1959 {
1960         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1961                                                                         xprt);
1962
1963         if (!transport->inet) {
1964                 struct sock *sk = sock->sk;
1965
1966                 lock_sock(sk);
1967
1968                 xs_save_old_callbacks(transport, sk);
1969
1970                 sk->sk_user_data = xprt;
1971                 sk->sk_data_ready = xs_data_ready;
1972                 sk->sk_write_space = xs_udp_write_space;
1973                 sk->sk_state_change = xs_local_state_change;
1974                 sk->sk_error_report = xs_error_report;
1975                 sk->sk_use_task_frag = false;
1976
1977                 xprt_clear_connected(xprt);
1978
1979                 /* Reset to new socket */
1980                 transport->sock = sock;
1981                 transport->inet = sk;
1982
1983                 release_sock(sk);
1984         }
1985
1986         xs_stream_start_connect(transport);
1987
1988         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1989 }
1990
1991 /**
1992  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1993  * @transport: socket transport to connect
1994  */
1995 static int xs_local_setup_socket(struct sock_xprt *transport)
1996 {
1997         struct rpc_xprt *xprt = &transport->xprt;
1998         struct file *filp;
1999         struct socket *sock;
2000         int status;
2001
2002         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2003                                         SOCK_STREAM, 0, &sock, 1);
2004         if (status < 0) {
2005                 dprintk("RPC:       can't create AF_LOCAL "
2006                         "transport socket (%d).\n", -status);
2007                 goto out;
2008         }
2009         xs_reclassify_socket(AF_LOCAL, sock);
2010
2011         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
2012         if (IS_ERR(filp)) {
2013                 status = PTR_ERR(filp);
2014                 goto out;
2015         }
2016         transport->file = filp;
2017
2018         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2019                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2020
2021         status = xs_local_finish_connecting(xprt, sock);
2022         trace_rpc_socket_connect(xprt, sock, status);
2023         switch (status) {
2024         case 0:
2025                 dprintk("RPC:       xprt %p connected to %s\n",
2026                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2027                 xprt->stat.connect_count++;
2028                 xprt->stat.connect_time += (long)jiffies -
2029                                            xprt->stat.connect_start;
2030                 xprt_set_connected(xprt);
2031                 break;
2032         case -ENOBUFS:
2033                 break;
2034         case -ENOENT:
2035                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2036                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2037                 break;
2038         case -ECONNREFUSED:
2039                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2040                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2041                 break;
2042         default:
2043                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2044                                 __func__, -status,
2045                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2046         }
2047
2048 out:
2049         xprt_clear_connecting(xprt);
2050         xprt_wake_pending_tasks(xprt, status);
2051         return status;
2052 }
2053
2054 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2055 {
2056         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2057         int ret;
2058
2059         if (transport->file)
2060                 goto force_disconnect;
2061
2062         if (RPC_IS_ASYNC(task)) {
2063                 /*
2064                  * We want the AF_LOCAL connect to be resolved in the
2065                  * filesystem namespace of the process making the rpc
2066                  * call.  Thus we connect synchronously.
2067                  *
2068                  * If we want to support asynchronous AF_LOCAL calls,
2069                  * we'll need to figure out how to pass a namespace to
2070                  * connect.
2071                  */
2072                 rpc_task_set_rpc_status(task, -ENOTCONN);
2073                 goto out_wake;
2074         }
2075         ret = xs_local_setup_socket(transport);
2076         if (ret && !RPC_IS_SOFTCONN(task))
2077                 msleep_interruptible(15000);
2078         return;
2079 force_disconnect:
2080         xprt_force_disconnect(xprt);
2081 out_wake:
2082         xprt_clear_connecting(xprt);
2083         xprt_wake_pending_tasks(xprt, -ENOTCONN);
2084 }
2085
2086 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2087 /*
2088  * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2089  * held, or when we otherwise know that we have exclusive access to the
2090  * socket, to guard against races with xs_reset_transport.
2091  */
2092 static void xs_set_memalloc(struct rpc_xprt *xprt)
2093 {
2094         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2095                         xprt);
2096
2097         /*
2098          * If there's no sock, then we have nothing to set. The
2099          * reconnecting process will get it for us.
2100          */
2101         if (!transport->inet)
2102                 return;
2103         if (atomic_read(&xprt->swapper))
2104                 sk_set_memalloc(transport->inet);
2105 }
2106
2107 /**
2108  * xs_enable_swap - Tag this transport as being used for swap.
2109  * @xprt: transport to tag
2110  *
2111  * Take a reference to this transport on behalf of the rpc_clnt, and
2112  * optionally mark it for swapping if it wasn't already.
2113  */
2114 static int
2115 xs_enable_swap(struct rpc_xprt *xprt)
2116 {
2117         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2118
2119         mutex_lock(&xs->recv_mutex);
2120         if (atomic_inc_return(&xprt->swapper) == 1 &&
2121             xs->inet)
2122                 sk_set_memalloc(xs->inet);
2123         mutex_unlock(&xs->recv_mutex);
2124         return 0;
2125 }
2126
2127 /**
2128  * xs_disable_swap - Untag this transport as being used for swap.
2129  * @xprt: transport to tag
2130  *
2131  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2132  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2133  */
2134 static void
2135 xs_disable_swap(struct rpc_xprt *xprt)
2136 {
2137         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2138
2139         mutex_lock(&xs->recv_mutex);
2140         if (atomic_dec_and_test(&xprt->swapper) &&
2141             xs->inet)
2142                 sk_clear_memalloc(xs->inet);
2143         mutex_unlock(&xs->recv_mutex);
2144 }
2145 #else
2146 static void xs_set_memalloc(struct rpc_xprt *xprt)
2147 {
2148 }
2149
2150 static int
2151 xs_enable_swap(struct rpc_xprt *xprt)
2152 {
2153         return -EINVAL;
2154 }
2155
2156 static void
2157 xs_disable_swap(struct rpc_xprt *xprt)
2158 {
2159 }
2160 #endif
2161
2162 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2163 {
2164         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2165
2166         if (!transport->inet) {
2167                 struct sock *sk = sock->sk;
2168
2169                 lock_sock(sk);
2170
2171                 xs_save_old_callbacks(transport, sk);
2172
2173                 sk->sk_user_data = xprt;
2174                 sk->sk_data_ready = xs_data_ready;
2175                 sk->sk_write_space = xs_udp_write_space;
2176                 sk->sk_use_task_frag = false;
2177
2178                 xprt_set_connected(xprt);
2179
2180                 /* Reset to new socket */
2181                 transport->sock = sock;
2182                 transport->inet = sk;
2183
2184                 xs_set_memalloc(xprt);
2185
2186                 release_sock(sk);
2187         }
2188         xs_udp_do_set_buffer_size(xprt);
2189
2190         xprt->stat.connect_start = jiffies;
2191 }
2192
2193 static void xs_udp_setup_socket(struct work_struct *work)
2194 {
2195         struct sock_xprt *transport =
2196                 container_of(work, struct sock_xprt, connect_worker.work);
2197         struct rpc_xprt *xprt = &transport->xprt;
2198         struct socket *sock;
2199         int status = -EIO;
2200         unsigned int pflags = current->flags;
2201
2202         if (atomic_read(&xprt->swapper))
2203                 current->flags |= PF_MEMALLOC;
2204         sock = xs_create_sock(xprt, transport,
2205                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2206                         IPPROTO_UDP, false);
2207         if (IS_ERR(sock))
2208                 goto out;
2209
2210         dprintk("RPC:       worker connecting xprt %p via %s to "
2211                                 "%s (port %s)\n", xprt,
2212                         xprt->address_strings[RPC_DISPLAY_PROTO],
2213                         xprt->address_strings[RPC_DISPLAY_ADDR],
2214                         xprt->address_strings[RPC_DISPLAY_PORT]);
2215
2216         xs_udp_finish_connecting(xprt, sock);
2217         trace_rpc_socket_connect(xprt, sock, 0);
2218         status = 0;
2219 out:
2220         xprt_clear_connecting(xprt);
2221         xprt_unlock_connect(xprt, transport);
2222         xprt_wake_pending_tasks(xprt, status);
2223         current_restore_flags(pflags, PF_MEMALLOC);
2224 }
2225
2226 /**
2227  * xs_tcp_shutdown - gracefully shut down a TCP socket
2228  * @xprt: transport
2229  *
2230  * Initiates a graceful shutdown of the TCP socket by calling the
2231  * equivalent of shutdown(SHUT_RDWR);
2232  */
2233 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2234 {
2235         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2236         struct socket *sock = transport->sock;
2237         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2238
2239         if (sock == NULL)
2240                 return;
2241         if (!xprt->reuseport) {
2242                 xs_close(xprt);
2243                 return;
2244         }
2245         switch (skst) {
2246         case TCP_FIN_WAIT1:
2247         case TCP_FIN_WAIT2:
2248         case TCP_LAST_ACK:
2249                 break;
2250         case TCP_ESTABLISHED:
2251         case TCP_CLOSE_WAIT:
2252                 kernel_sock_shutdown(sock, SHUT_RDWR);
2253                 trace_rpc_socket_shutdown(xprt, sock);
2254                 break;
2255         default:
2256                 xs_reset_transport(transport);
2257         }
2258 }
2259
2260 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2261                 struct socket *sock)
2262 {
2263         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2264         struct net *net = sock_net(sock->sk);
2265         unsigned long connect_timeout;
2266         unsigned long syn_retries;
2267         unsigned int keepidle;
2268         unsigned int keepcnt;
2269         unsigned int timeo;
2270         unsigned long t;
2271
2272         spin_lock(&xprt->transport_lock);
2273         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2274         keepcnt = xprt->timeout->to_retries + 1;
2275         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2276                 (xprt->timeout->to_retries + 1);
2277         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2278         spin_unlock(&xprt->transport_lock);
2279
2280         /* TCP Keepalive options */
2281         sock_set_keepalive(sock->sk);
2282         tcp_sock_set_keepidle(sock->sk, keepidle);
2283         tcp_sock_set_keepintvl(sock->sk, keepidle);
2284         tcp_sock_set_keepcnt(sock->sk, keepcnt);
2285
2286         /* TCP user timeout (see RFC5482) */
2287         tcp_sock_set_user_timeout(sock->sk, timeo);
2288
2289         /* Connect timeout */
2290         connect_timeout = max_t(unsigned long,
2291                                 DIV_ROUND_UP(xprt->connect_timeout, HZ), 1);
2292         syn_retries = max_t(unsigned long,
2293                             READ_ONCE(net->ipv4.sysctl_tcp_syn_retries), 1);
2294         for (t = 0; t <= syn_retries && (1UL << t) < connect_timeout; t++)
2295                 ;
2296         if (t <= syn_retries)
2297                 tcp_sock_set_syncnt(sock->sk, t - 1);
2298 }
2299
2300 static void xs_tcp_do_set_connect_timeout(struct rpc_xprt *xprt,
2301                                           unsigned long connect_timeout)
2302 {
2303         struct sock_xprt *transport =
2304                 container_of(xprt, struct sock_xprt, xprt);
2305         struct rpc_timeout to;
2306         unsigned long initval;
2307
2308         memcpy(&to, xprt->timeout, sizeof(to));
2309         /* Arbitrary lower limit */
2310         initval = max_t(unsigned long, connect_timeout, XS_TCP_INIT_REEST_TO);
2311         to.to_initval = initval;
2312         to.to_maxval = initval;
2313         to.to_retries = 0;
2314         memcpy(&transport->tcp_timeout, &to, sizeof(transport->tcp_timeout));
2315         xprt->timeout = &transport->tcp_timeout;
2316         xprt->connect_timeout = connect_timeout;
2317 }
2318
2319 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2320                 unsigned long connect_timeout,
2321                 unsigned long reconnect_timeout)
2322 {
2323         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2324
2325         spin_lock(&xprt->transport_lock);
2326         if (reconnect_timeout < xprt->max_reconnect_timeout)
2327                 xprt->max_reconnect_timeout = reconnect_timeout;
2328         if (connect_timeout < xprt->connect_timeout)
2329                 xs_tcp_do_set_connect_timeout(xprt, connect_timeout);
2330         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2331         spin_unlock(&xprt->transport_lock);
2332 }
2333
2334 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2335 {
2336         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2337
2338         if (!transport->inet) {
2339                 struct sock *sk = sock->sk;
2340
2341                 /* Avoid temporary address, they are bad for long-lived
2342                  * connections such as NFS mounts.
2343                  * RFC4941, section 3.6 suggests that:
2344                  *    Individual applications, which have specific
2345                  *    knowledge about the normal duration of connections,
2346                  *    MAY override this as appropriate.
2347                  */
2348                 if (xs_addr(xprt)->sa_family == PF_INET6) {
2349                         ip6_sock_set_addr_preferences(sk,
2350                                 IPV6_PREFER_SRC_PUBLIC);
2351                 }
2352
2353                 xs_tcp_set_socket_timeouts(xprt, sock);
2354                 tcp_sock_set_nodelay(sk);
2355
2356                 lock_sock(sk);
2357
2358                 xs_save_old_callbacks(transport, sk);
2359
2360                 sk->sk_user_data = xprt;
2361                 sk->sk_data_ready = xs_data_ready;
2362                 sk->sk_state_change = xs_tcp_state_change;
2363                 sk->sk_write_space = xs_tcp_write_space;
2364                 sk->sk_error_report = xs_error_report;
2365                 sk->sk_use_task_frag = false;
2366
2367                 /* socket options */
2368                 sock_reset_flag(sk, SOCK_LINGER);
2369
2370                 xprt_clear_connected(xprt);
2371
2372                 /* Reset to new socket */
2373                 transport->sock = sock;
2374                 transport->inet = sk;
2375
2376                 release_sock(sk);
2377         }
2378
2379         if (!xprt_bound(xprt))
2380                 return -ENOTCONN;
2381
2382         xs_set_memalloc(xprt);
2383
2384         xs_stream_start_connect(transport);
2385
2386         /* Tell the socket layer to start connecting... */
2387         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2388         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2389 }
2390
2391 /**
2392  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2393  * @work: queued work item
2394  *
2395  * Invoked by a work queue tasklet.
2396  */
2397 static void xs_tcp_setup_socket(struct work_struct *work)
2398 {
2399         struct sock_xprt *transport =
2400                 container_of(work, struct sock_xprt, connect_worker.work);
2401         struct socket *sock = transport->sock;
2402         struct rpc_xprt *xprt = &transport->xprt;
2403         int status;
2404         unsigned int pflags = current->flags;
2405
2406         if (atomic_read(&xprt->swapper))
2407                 current->flags |= PF_MEMALLOC;
2408
2409         if (xprt_connected(xprt))
2410                 goto out;
2411         if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2412                                &transport->sock_state) ||
2413             !sock) {
2414                 xs_reset_transport(transport);
2415                 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2416                                       SOCK_STREAM, IPPROTO_TCP, true);
2417                 if (IS_ERR(sock)) {
2418                         xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2419                         goto out;
2420                 }
2421         }
2422
2423         dprintk("RPC:       worker connecting xprt %p via %s to "
2424                                 "%s (port %s)\n", xprt,
2425                         xprt->address_strings[RPC_DISPLAY_PROTO],
2426                         xprt->address_strings[RPC_DISPLAY_ADDR],
2427                         xprt->address_strings[RPC_DISPLAY_PORT]);
2428
2429         status = xs_tcp_finish_connecting(xprt, sock);
2430         trace_rpc_socket_connect(xprt, sock, status);
2431         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2432                         xprt, -status, xprt_connected(xprt),
2433                         sock->sk->sk_state);
2434         switch (status) {
2435         case 0:
2436         case -EINPROGRESS:
2437                 /* SYN_SENT! */
2438                 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2439                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2440                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2441                 fallthrough;
2442         case -EALREADY:
2443                 goto out_unlock;
2444         case -EADDRNOTAVAIL:
2445                 /* Source port number is unavailable. Try a new one! */
2446                 transport->srcport = 0;
2447                 status = -EAGAIN;
2448                 break;
2449         case -EPERM:
2450                 /* Happens, for instance, if a BPF program is preventing
2451                  * the connect. Remap the error so upper layers can better
2452                  * deal with it.
2453                  */
2454                 status = -ECONNREFUSED;
2455                 fallthrough;
2456         case -EINVAL:
2457                 /* Happens, for instance, if the user specified a link
2458                  * local IPv6 address without a scope-id.
2459                  */
2460         case -ECONNREFUSED:
2461         case -ECONNRESET:
2462         case -ENETDOWN:
2463         case -ENETUNREACH:
2464         case -EHOSTUNREACH:
2465         case -EADDRINUSE:
2466         case -ENOBUFS:
2467         case -ENOTCONN:
2468                 break;
2469         default:
2470                 printk("%s: connect returned unhandled error %d\n",
2471                         __func__, status);
2472                 status = -EAGAIN;
2473         }
2474
2475         /* xs_tcp_force_close() wakes tasks with a fixed error code.
2476          * We need to wake them first to ensure the correct error code.
2477          */
2478         xprt_wake_pending_tasks(xprt, status);
2479         xs_tcp_force_close(xprt);
2480 out:
2481         xprt_clear_connecting(xprt);
2482 out_unlock:
2483         xprt_unlock_connect(xprt, transport);
2484         current_restore_flags(pflags, PF_MEMALLOC);
2485 }
2486
2487 /*
2488  * Transfer the connected socket to @upper_transport, then mark that
2489  * xprt CONNECTED.
2490  */
2491 static int xs_tcp_tls_finish_connecting(struct rpc_xprt *lower_xprt,
2492                                         struct sock_xprt *upper_transport)
2493 {
2494         struct sock_xprt *lower_transport =
2495                         container_of(lower_xprt, struct sock_xprt, xprt);
2496         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2497
2498         if (!upper_transport->inet) {
2499                 struct socket *sock = lower_transport->sock;
2500                 struct sock *sk = sock->sk;
2501
2502                 /* Avoid temporary address, they are bad for long-lived
2503                  * connections such as NFS mounts.
2504                  * RFC4941, section 3.6 suggests that:
2505                  *    Individual applications, which have specific
2506                  *    knowledge about the normal duration of connections,
2507                  *    MAY override this as appropriate.
2508                  */
2509                 if (xs_addr(upper_xprt)->sa_family == PF_INET6)
2510                         ip6_sock_set_addr_preferences(sk, IPV6_PREFER_SRC_PUBLIC);
2511
2512                 xs_tcp_set_socket_timeouts(upper_xprt, sock);
2513                 tcp_sock_set_nodelay(sk);
2514
2515                 lock_sock(sk);
2516
2517                 /* @sk is already connected, so it now has the RPC callbacks.
2518                  * Reach into @lower_transport to save the original ones.
2519                  */
2520                 upper_transport->old_data_ready = lower_transport->old_data_ready;
2521                 upper_transport->old_state_change = lower_transport->old_state_change;
2522                 upper_transport->old_write_space = lower_transport->old_write_space;
2523                 upper_transport->old_error_report = lower_transport->old_error_report;
2524                 sk->sk_user_data = upper_xprt;
2525
2526                 /* socket options */
2527                 sock_reset_flag(sk, SOCK_LINGER);
2528
2529                 xprt_clear_connected(upper_xprt);
2530
2531                 upper_transport->sock = sock;
2532                 upper_transport->inet = sk;
2533                 upper_transport->file = lower_transport->file;
2534
2535                 release_sock(sk);
2536
2537                 /* Reset lower_transport before shutting down its clnt */
2538                 mutex_lock(&lower_transport->recv_mutex);
2539                 lower_transport->inet = NULL;
2540                 lower_transport->sock = NULL;
2541                 lower_transport->file = NULL;
2542
2543                 xprt_clear_connected(lower_xprt);
2544                 xs_sock_reset_connection_flags(lower_xprt);
2545                 xs_stream_reset_connect(lower_transport);
2546                 mutex_unlock(&lower_transport->recv_mutex);
2547         }
2548
2549         if (!xprt_bound(upper_xprt))
2550                 return -ENOTCONN;
2551
2552         xs_set_memalloc(upper_xprt);
2553
2554         if (!xprt_test_and_set_connected(upper_xprt)) {
2555                 upper_xprt->connect_cookie++;
2556                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2557                 xprt_clear_connecting(upper_xprt);
2558
2559                 upper_xprt->stat.connect_count++;
2560                 upper_xprt->stat.connect_time += (long)jiffies -
2561                                            upper_xprt->stat.connect_start;
2562                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2563         }
2564         return 0;
2565 }
2566
2567 /**
2568  * xs_tls_handshake_done - TLS handshake completion handler
2569  * @data: address of xprt to wake
2570  * @status: status of handshake
2571  * @peerid: serial number of key containing the remote's identity
2572  *
2573  */
2574 static void xs_tls_handshake_done(void *data, int status, key_serial_t peerid)
2575 {
2576         struct rpc_xprt *lower_xprt = data;
2577         struct sock_xprt *lower_transport =
2578                                 container_of(lower_xprt, struct sock_xprt, xprt);
2579
2580         switch (status) {
2581         case 0:
2582         case -EACCES:
2583         case -ETIMEDOUT:
2584                 lower_transport->xprt_err = status;
2585                 break;
2586         default:
2587                 lower_transport->xprt_err = -EACCES;
2588         }
2589         complete(&lower_transport->handshake_done);
2590         xprt_put(lower_xprt);
2591 }
2592
2593 static int xs_tls_handshake_sync(struct rpc_xprt *lower_xprt, struct xprtsec_parms *xprtsec)
2594 {
2595         struct sock_xprt *lower_transport =
2596                                 container_of(lower_xprt, struct sock_xprt, xprt);
2597         struct tls_handshake_args args = {
2598                 .ta_sock        = lower_transport->sock,
2599                 .ta_done        = xs_tls_handshake_done,
2600                 .ta_data        = xprt_get(lower_xprt),
2601                 .ta_peername    = lower_xprt->servername,
2602         };
2603         struct sock *sk = lower_transport->inet;
2604         int rc;
2605
2606         init_completion(&lower_transport->handshake_done);
2607         set_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2608         lower_transport->xprt_err = -ETIMEDOUT;
2609         switch (xprtsec->policy) {
2610         case RPC_XPRTSEC_TLS_ANON:
2611                 rc = tls_client_hello_anon(&args, GFP_KERNEL);
2612                 if (rc)
2613                         goto out_put_xprt;
2614                 break;
2615         case RPC_XPRTSEC_TLS_X509:
2616                 args.ta_my_cert = xprtsec->cert_serial;
2617                 args.ta_my_privkey = xprtsec->privkey_serial;
2618                 rc = tls_client_hello_x509(&args, GFP_KERNEL);
2619                 if (rc)
2620                         goto out_put_xprt;
2621                 break;
2622         default:
2623                 rc = -EACCES;
2624                 goto out_put_xprt;
2625         }
2626
2627         rc = wait_for_completion_interruptible_timeout(&lower_transport->handshake_done,
2628                                                        XS_TLS_HANDSHAKE_TO);
2629         if (rc <= 0) {
2630                 tls_handshake_cancel(sk);
2631                 if (rc == 0)
2632                         rc = -ETIMEDOUT;
2633                 goto out_put_xprt;
2634         }
2635
2636         rc = lower_transport->xprt_err;
2637
2638 out:
2639         xs_stream_reset_connect(lower_transport);
2640         clear_bit(XPRT_SOCK_IGNORE_RECV, &lower_transport->sock_state);
2641         return rc;
2642
2643 out_put_xprt:
2644         xprt_put(lower_xprt);
2645         goto out;
2646 }
2647
2648 /**
2649  * xs_tcp_tls_setup_socket - establish a TLS session on a TCP socket
2650  * @work: queued work item
2651  *
2652  * Invoked by a work queue tasklet.
2653  *
2654  * For RPC-with-TLS, there is a two-stage connection process.
2655  *
2656  * The "upper-layer xprt" is visible to the RPC consumer. Once it has
2657  * been marked connected, the consumer knows that a TCP connection and
2658  * a TLS session have been established.
2659  *
2660  * A "lower-layer xprt", created in this function, handles the mechanics
2661  * of connecting the TCP socket, performing the RPC_AUTH_TLS probe, and
2662  * then driving the TLS handshake. Once all that is complete, the upper
2663  * layer xprt is marked connected.
2664  */
2665 static void xs_tcp_tls_setup_socket(struct work_struct *work)
2666 {
2667         struct sock_xprt *upper_transport =
2668                 container_of(work, struct sock_xprt, connect_worker.work);
2669         struct rpc_clnt *upper_clnt = upper_transport->clnt;
2670         struct rpc_xprt *upper_xprt = &upper_transport->xprt;
2671         struct rpc_create_args args = {
2672                 .net            = upper_xprt->xprt_net,
2673                 .protocol       = upper_xprt->prot,
2674                 .address        = (struct sockaddr *)&upper_xprt->addr,
2675                 .addrsize       = upper_xprt->addrlen,
2676                 .timeout        = upper_clnt->cl_timeout,
2677                 .servername     = upper_xprt->servername,
2678                 .program        = upper_clnt->cl_program,
2679                 .prognumber     = upper_clnt->cl_prog,
2680                 .version        = upper_clnt->cl_vers,
2681                 .authflavor     = RPC_AUTH_TLS,
2682                 .cred           = upper_clnt->cl_cred,
2683                 .xprtsec        = {
2684                         .policy         = RPC_XPRTSEC_NONE,
2685                 },
2686                 .stats          = upper_clnt->cl_stats,
2687         };
2688         unsigned int pflags = current->flags;
2689         struct rpc_clnt *lower_clnt;
2690         struct rpc_xprt *lower_xprt;
2691         int status;
2692
2693         if (atomic_read(&upper_xprt->swapper))
2694                 current->flags |= PF_MEMALLOC;
2695
2696         xs_stream_start_connect(upper_transport);
2697
2698         /* This implicitly sends an RPC_AUTH_TLS probe */
2699         lower_clnt = rpc_create(&args);
2700         if (IS_ERR(lower_clnt)) {
2701                 trace_rpc_tls_unavailable(upper_clnt, upper_xprt);
2702                 clear_bit(XPRT_SOCK_CONNECTING, &upper_transport->sock_state);
2703                 xprt_clear_connecting(upper_xprt);
2704                 xprt_wake_pending_tasks(upper_xprt, PTR_ERR(lower_clnt));
2705                 xs_run_error_worker(upper_transport, XPRT_SOCK_WAKE_PENDING);
2706                 goto out_unlock;
2707         }
2708
2709         /* RPC_AUTH_TLS probe was successful. Try a TLS handshake on
2710          * the lower xprt.
2711          */
2712         rcu_read_lock();
2713         lower_xprt = rcu_dereference(lower_clnt->cl_xprt);
2714         rcu_read_unlock();
2715
2716         if (wait_on_bit_lock(&lower_xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2717                 goto out_unlock;
2718
2719         status = xs_tls_handshake_sync(lower_xprt, &upper_xprt->xprtsec);
2720         if (status) {
2721                 trace_rpc_tls_not_started(upper_clnt, upper_xprt);
2722                 goto out_close;
2723         }
2724
2725         status = xs_tcp_tls_finish_connecting(lower_xprt, upper_transport);
2726         if (status)
2727                 goto out_close;
2728         xprt_release_write(lower_xprt, NULL);
2729         trace_rpc_socket_connect(upper_xprt, upper_transport->sock, 0);
2730         rpc_shutdown_client(lower_clnt);
2731
2732         /* Check for ingress data that arrived before the socket's
2733          * ->data_ready callback was set up.
2734          */
2735         xs_poll_check_readable(upper_transport);
2736
2737 out_unlock:
2738         current_restore_flags(pflags, PF_MEMALLOC);
2739         upper_transport->clnt = NULL;
2740         xprt_unlock_connect(upper_xprt, upper_transport);
2741         return;
2742
2743 out_close:
2744         xprt_release_write(lower_xprt, NULL);
2745         rpc_shutdown_client(lower_clnt);
2746
2747         /* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
2748          * Wake them first here to ensure they get our tk_status code.
2749          */
2750         xprt_wake_pending_tasks(upper_xprt, status);
2751         xs_tcp_force_close(upper_xprt);
2752         xprt_clear_connecting(upper_xprt);
2753         goto out_unlock;
2754 }
2755
2756 /**
2757  * xs_connect - connect a socket to a remote endpoint
2758  * @xprt: pointer to transport structure
2759  * @task: address of RPC task that manages state of connect request
2760  *
2761  * TCP: If the remote end dropped the connection, delay reconnecting.
2762  *
2763  * UDP socket connects are synchronous, but we use a work queue anyway
2764  * to guarantee that even unprivileged user processes can set up a
2765  * socket on a privileged port.
2766  *
2767  * If a UDP socket connect fails, the delay behavior here prevents
2768  * retry floods (hard mounts).
2769  */
2770 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2771 {
2772         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2773         unsigned long delay = 0;
2774
2775         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2776
2777         if (transport->sock != NULL) {
2778                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2779                         "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2780
2781                 delay = xprt_reconnect_delay(xprt);
2782                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2783
2784         } else
2785                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2786
2787         transport->clnt = task->tk_client;
2788         queue_delayed_work(xprtiod_workqueue,
2789                         &transport->connect_worker,
2790                         delay);
2791 }
2792
2793 static void xs_wake_disconnect(struct sock_xprt *transport)
2794 {
2795         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2796                 xs_tcp_force_close(&transport->xprt);
2797 }
2798
2799 static void xs_wake_write(struct sock_xprt *transport)
2800 {
2801         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2802                 xprt_write_space(&transport->xprt);
2803 }
2804
2805 static void xs_wake_error(struct sock_xprt *transport)
2806 {
2807         int sockerr;
2808
2809         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2810                 return;
2811         sockerr = xchg(&transport->xprt_err, 0);
2812         if (sockerr < 0) {
2813                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2814                 xs_tcp_force_close(&transport->xprt);
2815         }
2816 }
2817
2818 static void xs_wake_pending(struct sock_xprt *transport)
2819 {
2820         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2821                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2822 }
2823
2824 static void xs_error_handle(struct work_struct *work)
2825 {
2826         struct sock_xprt *transport = container_of(work,
2827                         struct sock_xprt, error_worker);
2828
2829         xs_wake_disconnect(transport);
2830         xs_wake_write(transport);
2831         xs_wake_error(transport);
2832         xs_wake_pending(transport);
2833 }
2834
2835 /**
2836  * xs_local_print_stats - display AF_LOCAL socket-specific stats
2837  * @xprt: rpc_xprt struct containing statistics
2838  * @seq: output file
2839  *
2840  */
2841 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2842 {
2843         long idle_time = 0;
2844
2845         if (xprt_connected(xprt))
2846                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2847
2848         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2849                         "%llu %llu %lu %llu %llu\n",
2850                         xprt->stat.bind_count,
2851                         xprt->stat.connect_count,
2852                         xprt->stat.connect_time / HZ,
2853                         idle_time,
2854                         xprt->stat.sends,
2855                         xprt->stat.recvs,
2856                         xprt->stat.bad_xids,
2857                         xprt->stat.req_u,
2858                         xprt->stat.bklog_u,
2859                         xprt->stat.max_slots,
2860                         xprt->stat.sending_u,
2861                         xprt->stat.pending_u);
2862 }
2863
2864 /**
2865  * xs_udp_print_stats - display UDP socket-specific stats
2866  * @xprt: rpc_xprt struct containing statistics
2867  * @seq: output file
2868  *
2869  */
2870 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2871 {
2872         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2873
2874         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2875                         "%lu %llu %llu\n",
2876                         transport->srcport,
2877                         xprt->stat.bind_count,
2878                         xprt->stat.sends,
2879                         xprt->stat.recvs,
2880                         xprt->stat.bad_xids,
2881                         xprt->stat.req_u,
2882                         xprt->stat.bklog_u,
2883                         xprt->stat.max_slots,
2884                         xprt->stat.sending_u,
2885                         xprt->stat.pending_u);
2886 }
2887
2888 /**
2889  * xs_tcp_print_stats - display TCP socket-specific stats
2890  * @xprt: rpc_xprt struct containing statistics
2891  * @seq: output file
2892  *
2893  */
2894 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2895 {
2896         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2897         long idle_time = 0;
2898
2899         if (xprt_connected(xprt))
2900                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2901
2902         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2903                         "%llu %llu %lu %llu %llu\n",
2904                         transport->srcport,
2905                         xprt->stat.bind_count,
2906                         xprt->stat.connect_count,
2907                         xprt->stat.connect_time / HZ,
2908                         idle_time,
2909                         xprt->stat.sends,
2910                         xprt->stat.recvs,
2911                         xprt->stat.bad_xids,
2912                         xprt->stat.req_u,
2913                         xprt->stat.bklog_u,
2914                         xprt->stat.max_slots,
2915                         xprt->stat.sending_u,
2916                         xprt->stat.pending_u);
2917 }
2918
2919 /*
2920  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2921  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2922  * to use the server side send routines.
2923  */
2924 static int bc_malloc(struct rpc_task *task)
2925 {
2926         struct rpc_rqst *rqst = task->tk_rqstp;
2927         size_t size = rqst->rq_callsize;
2928         struct page *page;
2929         struct rpc_buffer *buf;
2930
2931         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2932                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2933                           size);
2934                 return -EINVAL;
2935         }
2936
2937         page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2938         if (!page)
2939                 return -ENOMEM;
2940
2941         buf = page_address(page);
2942         buf->len = PAGE_SIZE;
2943
2944         rqst->rq_buffer = buf->data;
2945         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2946         return 0;
2947 }
2948
2949 /*
2950  * Free the space allocated in the bc_alloc routine
2951  */
2952 static void bc_free(struct rpc_task *task)
2953 {
2954         void *buffer = task->tk_rqstp->rq_buffer;
2955         struct rpc_buffer *buf;
2956
2957         buf = container_of(buffer, struct rpc_buffer, data);
2958         free_page((unsigned long)buf);
2959 }
2960
2961 static int bc_sendto(struct rpc_rqst *req)
2962 {
2963         struct xdr_buf *xdr = &req->rq_snd_buf;
2964         struct sock_xprt *transport =
2965                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2966         struct msghdr msg = {
2967                 .msg_flags      = 0,
2968         };
2969         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2970                                          (u32)xdr->len);
2971         unsigned int sent = 0;
2972         int err;
2973
2974         req->rq_xtime = ktime_get();
2975         err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2976         if (err < 0)
2977                 return err;
2978         err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2979         xdr_free_bvec(xdr);
2980         if (err < 0 || sent != (xdr->len + sizeof(marker)))
2981                 return -EAGAIN;
2982         return sent;
2983 }
2984
2985 /**
2986  * bc_send_request - Send a backchannel Call on a TCP socket
2987  * @req: rpc_rqst containing Call message to be sent
2988  *
2989  * xpt_mutex ensures @rqstp's whole message is written to the socket
2990  * without interruption.
2991  *
2992  * Return values:
2993  *   %0 if the message was sent successfully
2994  *   %ENOTCONN if the message was not sent
2995  */
2996 static int bc_send_request(struct rpc_rqst *req)
2997 {
2998         struct svc_xprt *xprt;
2999         int len;
3000
3001         /*
3002          * Get the server socket associated with this callback xprt
3003          */
3004         xprt = req->rq_xprt->bc_xprt;
3005
3006         /*
3007          * Grab the mutex to serialize data as the connection is shared
3008          * with the fore channel
3009          */
3010         mutex_lock(&xprt->xpt_mutex);
3011         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
3012                 len = -ENOTCONN;
3013         else
3014                 len = bc_sendto(req);
3015         mutex_unlock(&xprt->xpt_mutex);
3016
3017         if (len > 0)
3018                 len = 0;
3019
3020         return len;
3021 }
3022
3023 static void bc_close(struct rpc_xprt *xprt)
3024 {
3025         xprt_disconnect_done(xprt);
3026 }
3027
3028 static void bc_destroy(struct rpc_xprt *xprt)
3029 {
3030         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
3031
3032         xs_xprt_free(xprt);
3033         module_put(THIS_MODULE);
3034 }
3035
3036 static const struct rpc_xprt_ops xs_local_ops = {
3037         .reserve_xprt           = xprt_reserve_xprt,
3038         .release_xprt           = xprt_release_xprt,
3039         .alloc_slot             = xprt_alloc_slot,
3040         .free_slot              = xprt_free_slot,
3041         .rpcbind                = xs_local_rpcbind,
3042         .set_port               = xs_local_set_port,
3043         .connect                = xs_local_connect,
3044         .buf_alloc              = rpc_malloc,
3045         .buf_free               = rpc_free,
3046         .prepare_request        = xs_stream_prepare_request,
3047         .send_request           = xs_local_send_request,
3048         .abort_send_request     = xs_stream_abort_send_request,
3049         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3050         .close                  = xs_close,
3051         .destroy                = xs_destroy,
3052         .print_stats            = xs_local_print_stats,
3053         .enable_swap            = xs_enable_swap,
3054         .disable_swap           = xs_disable_swap,
3055 };
3056
3057 static const struct rpc_xprt_ops xs_udp_ops = {
3058         .set_buffer_size        = xs_udp_set_buffer_size,
3059         .reserve_xprt           = xprt_reserve_xprt_cong,
3060         .release_xprt           = xprt_release_xprt_cong,
3061         .alloc_slot             = xprt_alloc_slot,
3062         .free_slot              = xprt_free_slot,
3063         .rpcbind                = rpcb_getport_async,
3064         .set_port               = xs_set_port,
3065         .connect                = xs_connect,
3066         .get_srcaddr            = xs_sock_srcaddr,
3067         .get_srcport            = xs_sock_srcport,
3068         .buf_alloc              = rpc_malloc,
3069         .buf_free               = rpc_free,
3070         .send_request           = xs_udp_send_request,
3071         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
3072         .timer                  = xs_udp_timer,
3073         .release_request        = xprt_release_rqst_cong,
3074         .close                  = xs_close,
3075         .destroy                = xs_destroy,
3076         .print_stats            = xs_udp_print_stats,
3077         .enable_swap            = xs_enable_swap,
3078         .disable_swap           = xs_disable_swap,
3079         .inject_disconnect      = xs_inject_disconnect,
3080 };
3081
3082 static const struct rpc_xprt_ops xs_tcp_ops = {
3083         .reserve_xprt           = xprt_reserve_xprt,
3084         .release_xprt           = xprt_release_xprt,
3085         .alloc_slot             = xprt_alloc_slot,
3086         .free_slot              = xprt_free_slot,
3087         .rpcbind                = rpcb_getport_async,
3088         .set_port               = xs_set_port,
3089         .connect                = xs_connect,
3090         .get_srcaddr            = xs_sock_srcaddr,
3091         .get_srcport            = xs_sock_srcport,
3092         .buf_alloc              = rpc_malloc,
3093         .buf_free               = rpc_free,
3094         .prepare_request        = xs_stream_prepare_request,
3095         .send_request           = xs_tcp_send_request,
3096         .abort_send_request     = xs_stream_abort_send_request,
3097         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3098         .close                  = xs_tcp_shutdown,
3099         .destroy                = xs_destroy,
3100         .set_connect_timeout    = xs_tcp_set_connect_timeout,
3101         .print_stats            = xs_tcp_print_stats,
3102         .enable_swap            = xs_enable_swap,
3103         .disable_swap           = xs_disable_swap,
3104         .inject_disconnect      = xs_inject_disconnect,
3105 #ifdef CONFIG_SUNRPC_BACKCHANNEL
3106         .bc_setup               = xprt_setup_bc,
3107         .bc_maxpayload          = xs_tcp_bc_maxpayload,
3108         .bc_num_slots           = xprt_bc_max_slots,
3109         .bc_free_rqst           = xprt_free_bc_rqst,
3110         .bc_destroy             = xprt_destroy_bc,
3111 #endif
3112 };
3113
3114 /*
3115  * The rpc_xprt_ops for the server backchannel
3116  */
3117
3118 static const struct rpc_xprt_ops bc_tcp_ops = {
3119         .reserve_xprt           = xprt_reserve_xprt,
3120         .release_xprt           = xprt_release_xprt,
3121         .alloc_slot             = xprt_alloc_slot,
3122         .free_slot              = xprt_free_slot,
3123         .buf_alloc              = bc_malloc,
3124         .buf_free               = bc_free,
3125         .send_request           = bc_send_request,
3126         .wait_for_reply_request = xprt_wait_for_reply_request_def,
3127         .close                  = bc_close,
3128         .destroy                = bc_destroy,
3129         .print_stats            = xs_tcp_print_stats,
3130         .enable_swap            = xs_enable_swap,
3131         .disable_swap           = xs_disable_swap,
3132         .inject_disconnect      = xs_inject_disconnect,
3133 };
3134
3135 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
3136 {
3137         static const struct sockaddr_in sin = {
3138                 .sin_family             = AF_INET,
3139                 .sin_addr.s_addr        = htonl(INADDR_ANY),
3140         };
3141         static const struct sockaddr_in6 sin6 = {
3142                 .sin6_family            = AF_INET6,
3143                 .sin6_addr              = IN6ADDR_ANY_INIT,
3144         };
3145
3146         switch (family) {
3147         case AF_LOCAL:
3148                 break;
3149         case AF_INET:
3150                 memcpy(sap, &sin, sizeof(sin));
3151                 break;
3152         case AF_INET6:
3153                 memcpy(sap, &sin6, sizeof(sin6));
3154                 break;
3155         default:
3156                 dprintk("RPC:       %s: Bad address family\n", __func__);
3157                 return -EAFNOSUPPORT;
3158         }
3159         return 0;
3160 }
3161
3162 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
3163                                       unsigned int slot_table_size,
3164                                       unsigned int max_slot_table_size)
3165 {
3166         struct rpc_xprt *xprt;
3167         struct sock_xprt *new;
3168
3169         if (args->addrlen > sizeof(xprt->addr)) {
3170                 dprintk("RPC:       xs_setup_xprt: address too large\n");
3171                 return ERR_PTR(-EBADF);
3172         }
3173
3174         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
3175                         max_slot_table_size);
3176         if (xprt == NULL) {
3177                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
3178                                 "rpc_xprt\n");
3179                 return ERR_PTR(-ENOMEM);
3180         }
3181
3182         new = container_of(xprt, struct sock_xprt, xprt);
3183         mutex_init(&new->recv_mutex);
3184         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
3185         xprt->addrlen = args->addrlen;
3186         if (args->srcaddr)
3187                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
3188         else {
3189                 int err;
3190                 err = xs_init_anyaddr(args->dstaddr->sa_family,
3191                                         (struct sockaddr *)&new->srcaddr);
3192                 if (err != 0) {
3193                         xprt_free(xprt);
3194                         return ERR_PTR(err);
3195                 }
3196         }
3197
3198         return xprt;
3199 }
3200
3201 static const struct rpc_timeout xs_local_default_timeout = {
3202         .to_initval = 10 * HZ,
3203         .to_maxval = 10 * HZ,
3204         .to_retries = 2,
3205 };
3206
3207 /**
3208  * xs_setup_local - Set up transport to use an AF_LOCAL socket
3209  * @args: rpc transport creation arguments
3210  *
3211  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
3212  */
3213 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
3214 {
3215         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
3216         struct sock_xprt *transport;
3217         struct rpc_xprt *xprt;
3218         struct rpc_xprt *ret;
3219
3220         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3221                         xprt_max_tcp_slot_table_entries);
3222         if (IS_ERR(xprt))
3223                 return xprt;
3224         transport = container_of(xprt, struct sock_xprt, xprt);
3225
3226         xprt->prot = 0;
3227         xprt->xprt_class = &xs_local_transport;
3228         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3229
3230         xprt->bind_timeout = XS_BIND_TO;
3231         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3232         xprt->idle_timeout = XS_IDLE_DISC_TO;
3233
3234         xprt->ops = &xs_local_ops;
3235         xprt->timeout = &xs_local_default_timeout;
3236
3237         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3238         INIT_WORK(&transport->error_worker, xs_error_handle);
3239         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
3240
3241         switch (sun->sun_family) {
3242         case AF_LOCAL:
3243                 if (sun->sun_path[0] != '/' && sun->sun_path[0] != '\0') {
3244                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
3245                                         sun->sun_path);
3246                         ret = ERR_PTR(-EINVAL);
3247                         goto out_err;
3248                 }
3249                 xprt_set_bound(xprt);
3250                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
3251                 break;
3252         default:
3253                 ret = ERR_PTR(-EAFNOSUPPORT);
3254                 goto out_err;
3255         }
3256
3257         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
3258                         xprt->address_strings[RPC_DISPLAY_ADDR]);
3259
3260         if (try_module_get(THIS_MODULE))
3261                 return xprt;
3262         ret = ERR_PTR(-EINVAL);
3263 out_err:
3264         xs_xprt_free(xprt);
3265         return ret;
3266 }
3267
3268 static const struct rpc_timeout xs_udp_default_timeout = {
3269         .to_initval = 5 * HZ,
3270         .to_maxval = 30 * HZ,
3271         .to_increment = 5 * HZ,
3272         .to_retries = 5,
3273 };
3274
3275 /**
3276  * xs_setup_udp - Set up transport to use a UDP socket
3277  * @args: rpc transport creation arguments
3278  *
3279  */
3280 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
3281 {
3282         struct sockaddr *addr = args->dstaddr;
3283         struct rpc_xprt *xprt;
3284         struct sock_xprt *transport;
3285         struct rpc_xprt *ret;
3286
3287         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3288                         xprt_udp_slot_table_entries);
3289         if (IS_ERR(xprt))
3290                 return xprt;
3291         transport = container_of(xprt, struct sock_xprt, xprt);
3292
3293         xprt->prot = IPPROTO_UDP;
3294         xprt->xprt_class = &xs_udp_transport;
3295         /* XXX: header size can vary due to auth type, IPv6, etc. */
3296         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3297
3298         xprt->bind_timeout = XS_BIND_TO;
3299         xprt->reestablish_timeout = XS_UDP_REEST_TO;
3300         xprt->idle_timeout = XS_IDLE_DISC_TO;
3301
3302         xprt->ops = &xs_udp_ops;
3303
3304         xprt->timeout = &xs_udp_default_timeout;
3305
3306         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3307         INIT_WORK(&transport->error_worker, xs_error_handle);
3308         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3309
3310         switch (addr->sa_family) {
3311         case AF_INET:
3312                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3313                         xprt_set_bound(xprt);
3314
3315                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3316                 break;
3317         case AF_INET6:
3318                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3319                         xprt_set_bound(xprt);
3320
3321                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3322                 break;
3323         default:
3324                 ret = ERR_PTR(-EAFNOSUPPORT);
3325                 goto out_err;
3326         }
3327
3328         if (xprt_bound(xprt))
3329                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3330                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3331                                 xprt->address_strings[RPC_DISPLAY_PORT],
3332                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3333         else
3334                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3335                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3336                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3337
3338         if (try_module_get(THIS_MODULE))
3339                 return xprt;
3340         ret = ERR_PTR(-EINVAL);
3341 out_err:
3342         xs_xprt_free(xprt);
3343         return ret;
3344 }
3345
3346 static const struct rpc_timeout xs_tcp_default_timeout = {
3347         .to_initval = 60 * HZ,
3348         .to_maxval = 60 * HZ,
3349         .to_retries = 2,
3350 };
3351
3352 /**
3353  * xs_setup_tcp - Set up transport to use a TCP socket
3354  * @args: rpc transport creation arguments
3355  *
3356  */
3357 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3358 {
3359         struct sockaddr *addr = args->dstaddr;
3360         struct rpc_xprt *xprt;
3361         struct sock_xprt *transport;
3362         struct rpc_xprt *ret;
3363         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3364
3365         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3366                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3367
3368         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3369                         max_slot_table_size);
3370         if (IS_ERR(xprt))
3371                 return xprt;
3372         transport = container_of(xprt, struct sock_xprt, xprt);
3373
3374         xprt->prot = IPPROTO_TCP;
3375         xprt->xprt_class = &xs_tcp_transport;
3376         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3377
3378         xprt->bind_timeout = XS_BIND_TO;
3379         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3380         xprt->idle_timeout = XS_IDLE_DISC_TO;
3381
3382         xprt->ops = &xs_tcp_ops;
3383         xprt->timeout = &xs_tcp_default_timeout;
3384
3385         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3386         if (args->reconnect_timeout)
3387                 xprt->max_reconnect_timeout = args->reconnect_timeout;
3388
3389         xprt->connect_timeout = xprt->timeout->to_initval *
3390                 (xprt->timeout->to_retries + 1);
3391         if (args->connect_timeout)
3392                 xs_tcp_do_set_connect_timeout(xprt, args->connect_timeout);
3393
3394         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3395         INIT_WORK(&transport->error_worker, xs_error_handle);
3396         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3397
3398         switch (addr->sa_family) {
3399         case AF_INET:
3400                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3401                         xprt_set_bound(xprt);
3402
3403                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3404                 break;
3405         case AF_INET6:
3406                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3407                         xprt_set_bound(xprt);
3408
3409                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3410                 break;
3411         default:
3412                 ret = ERR_PTR(-EAFNOSUPPORT);
3413                 goto out_err;
3414         }
3415
3416         if (xprt_bound(xprt))
3417                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3418                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3419                                 xprt->address_strings[RPC_DISPLAY_PORT],
3420                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3421         else
3422                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3423                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3424                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3425
3426         if (try_module_get(THIS_MODULE))
3427                 return xprt;
3428         ret = ERR_PTR(-EINVAL);
3429 out_err:
3430         xs_xprt_free(xprt);
3431         return ret;
3432 }
3433
3434 /**
3435  * xs_setup_tcp_tls - Set up transport to use a TCP with TLS
3436  * @args: rpc transport creation arguments
3437  *
3438  */
3439 static struct rpc_xprt *xs_setup_tcp_tls(struct xprt_create *args)
3440 {
3441         struct sockaddr *addr = args->dstaddr;
3442         struct rpc_xprt *xprt;
3443         struct sock_xprt *transport;
3444         struct rpc_xprt *ret;
3445         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3446
3447         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3448                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3449
3450         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3451                              max_slot_table_size);
3452         if (IS_ERR(xprt))
3453                 return xprt;
3454         transport = container_of(xprt, struct sock_xprt, xprt);
3455
3456         xprt->prot = IPPROTO_TCP;
3457         xprt->xprt_class = &xs_tcp_transport;
3458         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3459
3460         xprt->bind_timeout = XS_BIND_TO;
3461         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3462         xprt->idle_timeout = XS_IDLE_DISC_TO;
3463
3464         xprt->ops = &xs_tcp_ops;
3465         xprt->timeout = &xs_tcp_default_timeout;
3466
3467         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3468         xprt->connect_timeout = xprt->timeout->to_initval *
3469                 (xprt->timeout->to_retries + 1);
3470
3471         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3472         INIT_WORK(&transport->error_worker, xs_error_handle);
3473
3474         switch (args->xprtsec.policy) {
3475         case RPC_XPRTSEC_TLS_ANON:
3476         case RPC_XPRTSEC_TLS_X509:
3477                 xprt->xprtsec = args->xprtsec;
3478                 INIT_DELAYED_WORK(&transport->connect_worker,
3479                                   xs_tcp_tls_setup_socket);
3480                 break;
3481         default:
3482                 ret = ERR_PTR(-EACCES);
3483                 goto out_err;
3484         }
3485
3486         switch (addr->sa_family) {
3487         case AF_INET:
3488                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3489                         xprt_set_bound(xprt);
3490
3491                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3492                 break;
3493         case AF_INET6:
3494                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3495                         xprt_set_bound(xprt);
3496
3497                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3498                 break;
3499         default:
3500                 ret = ERR_PTR(-EAFNOSUPPORT);
3501                 goto out_err;
3502         }
3503
3504         if (xprt_bound(xprt))
3505                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3506                         xprt->address_strings[RPC_DISPLAY_ADDR],
3507                         xprt->address_strings[RPC_DISPLAY_PORT],
3508                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3509         else
3510                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3511                         xprt->address_strings[RPC_DISPLAY_ADDR],
3512                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3513
3514         if (try_module_get(THIS_MODULE))
3515                 return xprt;
3516         ret = ERR_PTR(-EINVAL);
3517 out_err:
3518         xs_xprt_free(xprt);
3519         return ret;
3520 }
3521
3522 /**
3523  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3524  * @args: rpc transport creation arguments
3525  *
3526  */
3527 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3528 {
3529         struct sockaddr *addr = args->dstaddr;
3530         struct rpc_xprt *xprt;
3531         struct sock_xprt *transport;
3532         struct svc_sock *bc_sock;
3533         struct rpc_xprt *ret;
3534
3535         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3536                         xprt_tcp_slot_table_entries);
3537         if (IS_ERR(xprt))
3538                 return xprt;
3539         transport = container_of(xprt, struct sock_xprt, xprt);
3540
3541         xprt->prot = IPPROTO_TCP;
3542         xprt->xprt_class = &xs_bc_tcp_transport;
3543         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3544         xprt->timeout = &xs_tcp_default_timeout;
3545
3546         /* backchannel */
3547         xprt_set_bound(xprt);
3548         xprt->bind_timeout = 0;
3549         xprt->reestablish_timeout = 0;
3550         xprt->idle_timeout = 0;
3551
3552         xprt->ops = &bc_tcp_ops;
3553
3554         switch (addr->sa_family) {
3555         case AF_INET:
3556                 xs_format_peer_addresses(xprt, "tcp",
3557                                          RPCBIND_NETID_TCP);
3558                 break;
3559         case AF_INET6:
3560                 xs_format_peer_addresses(xprt, "tcp",
3561                                    RPCBIND_NETID_TCP6);
3562                 break;
3563         default:
3564                 ret = ERR_PTR(-EAFNOSUPPORT);
3565                 goto out_err;
3566         }
3567
3568         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3569                         xprt->address_strings[RPC_DISPLAY_ADDR],
3570                         xprt->address_strings[RPC_DISPLAY_PORT],
3571                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3572
3573         /*
3574          * Once we've associated a backchannel xprt with a connection,
3575          * we want to keep it around as long as the connection lasts,
3576          * in case we need to start using it for a backchannel again;
3577          * this reference won't be dropped until bc_xprt is destroyed.
3578          */
3579         xprt_get(xprt);
3580         args->bc_xprt->xpt_bc_xprt = xprt;
3581         xprt->bc_xprt = args->bc_xprt;
3582         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3583         transport->sock = bc_sock->sk_sock;
3584         transport->inet = bc_sock->sk_sk;
3585
3586         /*
3587          * Since we don't want connections for the backchannel, we set
3588          * the xprt status to connected
3589          */
3590         xprt_set_connected(xprt);
3591
3592         if (try_module_get(THIS_MODULE))
3593                 return xprt;
3594
3595         args->bc_xprt->xpt_bc_xprt = NULL;
3596         args->bc_xprt->xpt_bc_xps = NULL;
3597         xprt_put(xprt);
3598         ret = ERR_PTR(-EINVAL);
3599 out_err:
3600         xs_xprt_free(xprt);
3601         return ret;
3602 }
3603
3604 static struct xprt_class        xs_local_transport = {
3605         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3606         .name           = "named UNIX socket",
3607         .owner          = THIS_MODULE,
3608         .ident          = XPRT_TRANSPORT_LOCAL,
3609         .setup          = xs_setup_local,
3610         .netid          = { "" },
3611 };
3612
3613 static struct xprt_class        xs_udp_transport = {
3614         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3615         .name           = "udp",
3616         .owner          = THIS_MODULE,
3617         .ident          = XPRT_TRANSPORT_UDP,
3618         .setup          = xs_setup_udp,
3619         .netid          = { "udp", "udp6", "" },
3620 };
3621
3622 static struct xprt_class        xs_tcp_transport = {
3623         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3624         .name           = "tcp",
3625         .owner          = THIS_MODULE,
3626         .ident          = XPRT_TRANSPORT_TCP,
3627         .setup          = xs_setup_tcp,
3628         .netid          = { "tcp", "tcp6", "" },
3629 };
3630
3631 static struct xprt_class        xs_tcp_tls_transport = {
3632         .list           = LIST_HEAD_INIT(xs_tcp_tls_transport.list),
3633         .name           = "tcp-with-tls",
3634         .owner          = THIS_MODULE,
3635         .ident          = XPRT_TRANSPORT_TCP_TLS,
3636         .setup          = xs_setup_tcp_tls,
3637         .netid          = { "tcp", "tcp6", "" },
3638 };
3639
3640 static struct xprt_class        xs_bc_tcp_transport = {
3641         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3642         .name           = "tcp NFSv4.1 backchannel",
3643         .owner          = THIS_MODULE,
3644         .ident          = XPRT_TRANSPORT_BC_TCP,
3645         .setup          = xs_setup_bc_tcp,
3646         .netid          = { "" },
3647 };
3648
3649 /**
3650  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3651  *
3652  */
3653 int init_socket_xprt(void)
3654 {
3655         if (!sunrpc_table_header)
3656                 sunrpc_table_header = register_sysctl("sunrpc", xs_tunables_table);
3657
3658         xprt_register_transport(&xs_local_transport);
3659         xprt_register_transport(&xs_udp_transport);
3660         xprt_register_transport(&xs_tcp_transport);
3661         xprt_register_transport(&xs_tcp_tls_transport);
3662         xprt_register_transport(&xs_bc_tcp_transport);
3663
3664         return 0;
3665 }
3666
3667 /**
3668  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3669  *
3670  */
3671 void cleanup_socket_xprt(void)
3672 {
3673         if (sunrpc_table_header) {
3674                 unregister_sysctl_table(sunrpc_table_header);
3675                 sunrpc_table_header = NULL;
3676         }
3677
3678         xprt_unregister_transport(&xs_local_transport);
3679         xprt_unregister_transport(&xs_udp_transport);
3680         xprt_unregister_transport(&xs_tcp_transport);
3681         xprt_unregister_transport(&xs_tcp_tls_transport);
3682         xprt_unregister_transport(&xs_bc_tcp_transport);
3683 }
3684
3685 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3686 {
3687         return param_set_uint_minmax(val, kp,
3688                         RPC_MIN_RESVPORT,
3689                         RPC_MAX_RESVPORT);
3690 }
3691
3692 static const struct kernel_param_ops param_ops_portnr = {
3693         .set = param_set_portnr,
3694         .get = param_get_uint,
3695 };
3696
3697 #define param_check_portnr(name, p) \
3698         __param_check(name, p, unsigned int);
3699
3700 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3701 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3702
3703 static int param_set_slot_table_size(const char *val,
3704                                      const struct kernel_param *kp)
3705 {
3706         return param_set_uint_minmax(val, kp,
3707                         RPC_MIN_SLOT_TABLE,
3708                         RPC_MAX_SLOT_TABLE);
3709 }
3710
3711 static const struct kernel_param_ops param_ops_slot_table_size = {
3712         .set = param_set_slot_table_size,
3713         .get = param_get_uint,
3714 };
3715
3716 #define param_check_slot_table_size(name, p) \
3717         __param_check(name, p, unsigned int);
3718
3719 static int param_set_max_slot_table_size(const char *val,
3720                                      const struct kernel_param *kp)
3721 {
3722         return param_set_uint_minmax(val, kp,
3723                         RPC_MIN_SLOT_TABLE,
3724                         RPC_MAX_SLOT_TABLE_LIMIT);
3725 }
3726
3727 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3728         .set = param_set_max_slot_table_size,
3729         .get = param_get_uint,
3730 };
3731
3732 #define param_check_max_slot_table_size(name, p) \
3733         __param_check(name, p, unsigned int);
3734
3735 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3736                    slot_table_size, 0644);
3737 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3738                    max_slot_table_size, 0644);
3739 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3740                    slot_table_size, 0644);