net/sunrpc/xprtsock.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/net/sunrpc/xprtsock.c
   4  *
   5  * Client-side transport implementation for sockets.
   6  *
   7  * TCP callback races fixes (C) 1998 Red Hat
   8  * TCP send fixes (C) 1998 Red Hat
   9  * TCP NFS related read + write fixes
  10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  11  *
  12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  13  * Fix behaviour when socket buffer is full.
  14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  15  *
  16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  17  *
  18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  19  *   <gilles.quillard@bull.net>
  20  */
  21
  22 #include <linux/types.h>
  23 #include <linux/string.h>
  24 #include <linux/slab.h>
  25 #include <linux/module.h>
  26 #include <linux/capability.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/errno.h>
  29 #include <linux/socket.h>
  30 #include <linux/in.h>
  31 #include <linux/net.h>
  32 #include <linux/mm.h>
  33 #include <linux/un.h>
  34 #include <linux/udp.h>
  35 #include <linux/tcp.h>
  36 #include <linux/sunrpc/clnt.h>
  37 #include <linux/sunrpc/addr.h>
  38 #include <linux/sunrpc/sched.h>
  39 #include <linux/sunrpc/svcsock.h>
  40 #include <linux/sunrpc/xprtsock.h>
  41 #include <linux/file.h>
  42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  43 #include <linux/sunrpc/bc_xprt.h>
  44 #endif
  45
  46 #include <net/sock.h>
  47 #include <net/checksum.h>
  48 #include <net/udp.h>
  49 #include <net/tcp.h>
  50 #include <linux/bvec.h>
  51 #include <linux/highmem.h>
  52 #include <linux/uio.h>
  53 #include <linux/sched/mm.h>
  54
  55 #include <trace/events/sock.h>
  56 #include <trace/events/sunrpc.h>
  57
  58 #include "socklib.h"
  59 #include "sunrpc.h"
  60
  61 static void xs_close(struct rpc_xprt *xprt);
  62 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock);
  63 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  64                 struct socket *sock);
  65
  66 /*
  67  * xprtsock tunables
  68  */
  69 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  70 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  71 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  72
  73 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  74 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  75
  76 #define XS_TCP_LINGER_TO        (15U * HZ)
  77 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  78
  79 /*
  80  * We can register our own files under /proc/sys/sunrpc by
  81  * calling register_sysctl_table() again.  The files in that
  82  * directory become the union of all files registered there.
  83  *
  84  * We simply need to make sure that we don't collide with
  85  * someone else's file names!
  86  */
  87
  88 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  89 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  90 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  91 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  92 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  93
  94 static struct ctl_table_header *sunrpc_table_header;
  95
  96 static struct xprt_class xs_local_transport;
  97 static struct xprt_class xs_udp_transport;
  98 static struct xprt_class xs_tcp_transport;
  99 static struct xprt_class xs_bc_tcp_transport;
 100
 101 /*
 102  * FIXME: changing the UDP slot table size should also resize the UDP
 103  *        socket buffers for existing UDP transports
 104  */
 105 static struct ctl_table xs_tunables_table[] = {
 106         {
 107                 .procname       = "udp_slot_table_entries",
 108                 .data           = &xprt_udp_slot_table_entries,
 109                 .maxlen         = sizeof(unsigned int),
 110                 .mode           = 0644,
 111                 .proc_handler   = proc_dointvec_minmax,
 112                 .extra1         = &min_slot_table_size,
 113                 .extra2         = &max_slot_table_size
 114         },
 115         {
 116                 .procname       = "tcp_slot_table_entries",
 117                 .data           = &xprt_tcp_slot_table_entries,
 118                 .maxlen         = sizeof(unsigned int),
 119                 .mode           = 0644,
 120                 .proc_handler   = proc_dointvec_minmax,
 121                 .extra1         = &min_slot_table_size,
 122                 .extra2         = &max_slot_table_size
 123         },
 124         {
 125                 .procname       = "tcp_max_slot_table_entries",
 126                 .data           = &xprt_max_tcp_slot_table_entries,
 127                 .maxlen         = sizeof(unsigned int),
 128                 .mode           = 0644,
 129                 .proc_handler   = proc_dointvec_minmax,
 130                 .extra1         = &min_slot_table_size,
 131                 .extra2         = &max_tcp_slot_table_limit
 132         },
 133         {
 134                 .procname       = "min_resvport",
 135                 .data           = &xprt_min_resvport,
 136                 .maxlen         = sizeof(unsigned int),
 137                 .mode           = 0644,
 138                 .proc_handler   = proc_dointvec_minmax,
 139                 .extra1         = &xprt_min_resvport_limit,
 140                 .extra2         = &xprt_max_resvport_limit
 141         },
 142         {
 143                 .procname       = "max_resvport",
 144                 .data           = &xprt_max_resvport,
 145                 .maxlen         = sizeof(unsigned int),
 146                 .mode           = 0644,
 147                 .proc_handler   = proc_dointvec_minmax,
 148                 .extra1         = &xprt_min_resvport_limit,
 149                 .extra2         = &xprt_max_resvport_limit
 150         },
 151         {
 152                 .procname       = "tcp_fin_timeout",
 153                 .data           = &xs_tcp_fin_timeout,
 154                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 155                 .mode           = 0644,
 156                 .proc_handler   = proc_dointvec_jiffies,
 157         },
 158         { },
 159 };
 160
 161 static struct ctl_table sunrpc_table[] = {
 162         {
 163                 .procname       = "sunrpc",
 164                 .mode           = 0555,
 165                 .child          = xs_tunables_table
 166         },
 167         { },
 168 };
 169
 170 /*
 171  * Wait duration for a reply from the RPC portmapper.
 172  */
 173 #define XS_BIND_TO              (60U * HZ)
 174
 175 /*
 176  * Delay if a UDP socket connect error occurs.  This is most likely some
 177  * kind of resource problem on the local host.
 178  */
 179 #define XS_UDP_REEST_TO         (2U * HZ)
 180
 181 /*
 182  * The reestablish timeout allows clients to delay for a bit before attempting
 183  * to reconnect to a server that just dropped our connection.
 184  *
 185  * We implement an exponential backoff when trying to reestablish a TCP
 186  * transport connection with the server.  Some servers like to drop a TCP
 187  * connection when they are overworked, so we start with a short timeout and
 188  * increase over time if the server is down or not responding.
 189  */
 190 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 191
 192 /*
 193  * TCP idle timeout; client drops the transport socket if it is idle
 194  * for this long.  Note that we also timeout UDP sockets to prevent
 195  * holding port numbers when there is no RPC traffic.
 196  */
 197 #define XS_IDLE_DISC_TO         (5U * 60 * 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                 strscpy(buf, sun->sun_path, sizeof(buf));
 266                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 267                                                 kstrdup(buf, GFP_KERNEL);
 268                 break;
 269         case AF_INET:
 270                 (void)rpc_ntop(sap, buf, sizeof(buf));
 271                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 272                                                 kstrdup(buf, GFP_KERNEL);
 273                 sin = xs_addr_in(xprt);
 274                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 275                 break;
 276         case AF_INET6:
 277                 (void)rpc_ntop(sap, buf, sizeof(buf));
 278                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 279                                                 kstrdup(buf, GFP_KERNEL);
 280                 sin6 = xs_addr_in6(xprt);
 281                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 282                 break;
 283         default:
 284                 BUG();
 285         }
 286
 287         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 288 }
 289
 290 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 291 {
 292         struct sockaddr *sap = xs_addr(xprt);
 293         char buf[128];
 294
 295         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 296         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 297
 298         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 299         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 300 }
 301
 302 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 303                                      const char *protocol,
 304                                      const char *netid)
 305 {
 306         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 307         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 308         xs_format_common_peer_addresses(xprt);
 309         xs_format_common_peer_ports(xprt);
 310 }
 311
 312 static void xs_update_peer_port(struct rpc_xprt *xprt)
 313 {
 314         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 315         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 316
 317         xs_format_common_peer_ports(xprt);
 318 }
 319
 320 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 321 {
 322         unsigned int i;
 323
 324         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 325                 switch (i) {
 326                 case RPC_DISPLAY_PROTO:
 327                 case RPC_DISPLAY_NETID:
 328                         continue;
 329                 default:
 330                         kfree(xprt->address_strings[i]);
 331                 }
 332 }
 333
 334 static size_t
 335 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 336 {
 337         size_t i,n;
 338
 339         if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 340                 return want;
 341         n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 342         for (i = 0; i < n; i++) {
 343                 if (buf->pages[i])
 344                         continue;
 345                 buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 346                 if (!buf->pages[i]) {
 347                         i *= PAGE_SIZE;
 348                         return i > buf->page_base ? i - buf->page_base : 0;
 349                 }
 350         }
 351         return want;
 352 }
 353
 354 static ssize_t
 355 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 356 {
 357         ssize_t ret;
 358         if (seek != 0)
 359                 iov_iter_advance(&msg->msg_iter, seek);
 360         ret = sock_recvmsg(sock, msg, flags);
 361         return ret > 0 ? ret + seek : ret;
 362 }
 363
 364 static ssize_t
 365 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 366                 struct kvec *kvec, size_t count, size_t seek)
 367 {
 368         iov_iter_kvec(&msg->msg_iter, ITER_DEST, kvec, 1, count);
 369         return xs_sock_recvmsg(sock, msg, flags, seek);
 370 }
 371
 372 static ssize_t
 373 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 374                 struct bio_vec *bvec, unsigned long nr, size_t count,
 375                 size_t seek)
 376 {
 377         iov_iter_bvec(&msg->msg_iter, ITER_DEST, bvec, nr, count);
 378         return xs_sock_recvmsg(sock, msg, flags, seek);
 379 }
 380
 381 static ssize_t
 382 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 383                 size_t count)
 384 {
 385         iov_iter_discard(&msg->msg_iter, ITER_DEST, count);
 386         return sock_recvmsg(sock, msg, flags);
 387 }
 388
 389 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 390 static void
 391 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 392 {
 393         struct bvec_iter bi = {
 394                 .bi_size = count,
 395         };
 396         struct bio_vec bv;
 397
 398         bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 399         for_each_bvec(bv, bvec, bi, bi)
 400                 flush_dcache_page(bv.bv_page);
 401 }
 402 #else
 403 static inline void
 404 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 405 {
 406 }
 407 #endif
 408
 409 static ssize_t
 410 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 411                 struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 412 {
 413         size_t want, seek_init = seek, offset = 0;
 414         ssize_t ret;
 415
 416         want = min_t(size_t, count, buf->head[0].iov_len);
 417         if (seek < want) {
 418                 ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 419                 if (ret <= 0)
 420                         goto sock_err;
 421                 offset += ret;
 422                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 423                         goto out;
 424                 if (ret != want)
 425                         goto out;
 426                 seek = 0;
 427         } else {
 428                 seek -= want;
 429                 offset += want;
 430         }
 431
 432         want = xs_alloc_sparse_pages(
 433                 buf, min_t(size_t, count - offset, buf->page_len),
 434                 GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
 435         if (seek < want) {
 436                 ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 437                                 xdr_buf_pagecount(buf),
 438                                 want + buf->page_base,
 439                                 seek + buf->page_base);
 440                 if (ret <= 0)
 441                         goto sock_err;
 442                 xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 443                 ret -= buf->page_base;
 444                 offset += ret;
 445                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 446                         goto out;
 447                 if (ret != want)
 448                         goto out;
 449                 seek = 0;
 450         } else {
 451                 seek -= want;
 452                 offset += want;
 453         }
 454
 455         want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 456         if (seek < want) {
 457                 ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 458                 if (ret <= 0)
 459                         goto sock_err;
 460                 offset += ret;
 461                 if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 462                         goto out;
 463                 if (ret != want)
 464                         goto out;
 465         } else if (offset < seek_init)
 466                 offset = seek_init;
 467         ret = -EMSGSIZE;
 468 out:
 469         *read = offset - seek_init;
 470         return ret;
 471 sock_err:
 472         offset += seek;
 473         goto out;
 474 }
 475
 476 static void
 477 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 478 {
 479         if (!transport->recv.copied) {
 480                 if (buf->head[0].iov_len >= transport->recv.offset)
 481                         memcpy(buf->head[0].iov_base,
 482                                         &transport->recv.xid,
 483                                         transport->recv.offset);
 484                 transport->recv.copied = transport->recv.offset;
 485         }
 486 }
 487
 488 static bool
 489 xs_read_stream_request_done(struct sock_xprt *transport)
 490 {
 491         return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 492 }
 493
 494 static void
 495 xs_read_stream_check_eor(struct sock_xprt *transport,
 496                 struct msghdr *msg)
 497 {
 498         if (xs_read_stream_request_done(transport))
 499                 msg->msg_flags |= MSG_EOR;
 500 }
 501
 502 static ssize_t
 503 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 504                 int flags, struct rpc_rqst *req)
 505 {
 506         struct xdr_buf *buf = &req->rq_private_buf;
 507         size_t want, read;
 508         ssize_t ret;
 509
 510         xs_read_header(transport, buf);
 511
 512         want = transport->recv.len - transport->recv.offset;
 513         if (want != 0) {
 514                 ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 515                                 transport->recv.copied + want,
 516                                 transport->recv.copied,
 517                                 &read);
 518                 transport->recv.offset += read;
 519                 transport->recv.copied += read;
 520         }
 521
 522         if (transport->recv.offset == transport->recv.len)
 523                 xs_read_stream_check_eor(transport, msg);
 524
 525         if (want == 0)
 526                 return 0;
 527
 528         switch (ret) {
 529         default:
 530                 break;
 531         case -EFAULT:
 532         case -EMSGSIZE:
 533                 msg->msg_flags |= MSG_TRUNC;
 534                 return read;
 535         case 0:
 536                 return -ESHUTDOWN;
 537         }
 538         return ret < 0 ? ret : read;
 539 }
 540
 541 static size_t
 542 xs_read_stream_headersize(bool isfrag)
 543 {
 544         if (isfrag)
 545                 return sizeof(__be32);
 546         return 3 * sizeof(__be32);
 547 }
 548
 549 static ssize_t
 550 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 551                 int flags, size_t want, size_t seek)
 552 {
 553         struct kvec kvec = {
 554                 .iov_base = &transport->recv.fraghdr,
 555                 .iov_len = want,
 556         };
 557         return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 558 }
 559
 560 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 561 static ssize_t
 562 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 563 {
 564         struct rpc_xprt *xprt = &transport->xprt;
 565         struct rpc_rqst *req;
 566         ssize_t ret;
 567
 568         /* Is this transport associated with the backchannel? */
 569         if (!xprt->bc_serv)
 570                 return -ESHUTDOWN;
 571
 572         /* Look up and lock the request corresponding to the given XID */
 573         req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 574         if (!req) {
 575                 printk(KERN_WARNING "Callback slot table overflowed\n");
 576                 return -ESHUTDOWN;
 577         }
 578         if (transport->recv.copied && !req->rq_private_buf.len)
 579                 return -ESHUTDOWN;
 580
 581         ret = xs_read_stream_request(transport, msg, flags, req);
 582         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 583                 xprt_complete_bc_request(req, transport->recv.copied);
 584         else
 585                 req->rq_private_buf.len = transport->recv.copied;
 586
 587         return ret;
 588 }
 589 #else /* CONFIG_SUNRPC_BACKCHANNEL */
 590 static ssize_t
 591 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 592 {
 593         return -ESHUTDOWN;
 594 }
 595 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 596
 597 static ssize_t
 598 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 599 {
 600         struct rpc_xprt *xprt = &transport->xprt;
 601         struct rpc_rqst *req;
 602         ssize_t ret = 0;
 603
 604         /* Look up and lock the request corresponding to the given XID */
 605         spin_lock(&xprt->queue_lock);
 606         req = xprt_lookup_rqst(xprt, transport->recv.xid);
 607         if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 608                 msg->msg_flags |= MSG_TRUNC;
 609                 goto out;
 610         }
 611         xprt_pin_rqst(req);
 612         spin_unlock(&xprt->queue_lock);
 613
 614         ret = xs_read_stream_request(transport, msg, flags, req);
 615
 616         spin_lock(&xprt->queue_lock);
 617         if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 618                 xprt_complete_rqst(req->rq_task, transport->recv.copied);
 619         else
 620                 req->rq_private_buf.len = transport->recv.copied;
 621         xprt_unpin_rqst(req);
 622 out:
 623         spin_unlock(&xprt->queue_lock);
 624         return ret;
 625 }
 626
 627 static ssize_t
 628 xs_read_stream(struct sock_xprt *transport, int flags)
 629 {
 630         struct msghdr msg = { 0 };
 631         size_t want, read = 0;
 632         ssize_t ret = 0;
 633
 634         if (transport->recv.len == 0) {
 635                 want = xs_read_stream_headersize(transport->recv.copied != 0);
 636                 ret = xs_read_stream_header(transport, &msg, flags, want,
 637                                 transport->recv.offset);
 638                 if (ret <= 0)
 639                         goto out_err;
 640                 transport->recv.offset = ret;
 641                 if (transport->recv.offset != want)
 642                         return transport->recv.offset;
 643                 transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 644                         RPC_FRAGMENT_SIZE_MASK;
 645                 transport->recv.offset -= sizeof(transport->recv.fraghdr);
 646                 read = ret;
 647         }
 648
 649         switch (be32_to_cpu(transport->recv.calldir)) {
 650         default:
 651                 msg.msg_flags |= MSG_TRUNC;
 652                 break;
 653         case RPC_CALL:
 654                 ret = xs_read_stream_call(transport, &msg, flags);
 655                 break;
 656         case RPC_REPLY:
 657                 ret = xs_read_stream_reply(transport, &msg, flags);
 658         }
 659         if (msg.msg_flags & MSG_TRUNC) {
 660                 transport->recv.calldir = cpu_to_be32(-1);
 661                 transport->recv.copied = -1;
 662         }
 663         if (ret < 0)
 664                 goto out_err;
 665         read += ret;
 666         if (transport->recv.offset < transport->recv.len) {
 667                 if (!(msg.msg_flags & MSG_TRUNC))
 668                         return read;
 669                 msg.msg_flags = 0;
 670                 ret = xs_read_discard(transport->sock, &msg, flags,
 671                                 transport->recv.len - transport->recv.offset);
 672                 if (ret <= 0)
 673                         goto out_err;
 674                 transport->recv.offset += ret;
 675                 read += ret;
 676                 if (transport->recv.offset != transport->recv.len)
 677                         return read;
 678         }
 679         if (xs_read_stream_request_done(transport)) {
 680                 trace_xs_stream_read_request(transport);
 681                 transport->recv.copied = 0;
 682         }
 683         transport->recv.offset = 0;
 684         transport->recv.len = 0;
 685         return read;
 686 out_err:
 687         return ret != 0 ? ret : -ESHUTDOWN;
 688 }
 689
 690 static __poll_t xs_poll_socket(struct sock_xprt *transport)
 691 {
 692         return transport->sock->ops->poll(transport->file, transport->sock,
 693                         NULL);
 694 }
 695
 696 static bool xs_poll_socket_readable(struct sock_xprt *transport)
 697 {
 698         __poll_t events = xs_poll_socket(transport);
 699
 700         return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 701 }
 702
 703 static void xs_poll_check_readable(struct sock_xprt *transport)
 704 {
 705
 706         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 707         if (!xs_poll_socket_readable(transport))
 708                 return;
 709         if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 710                 queue_work(xprtiod_workqueue, &transport->recv_worker);
 711 }
 712
 713 static void xs_stream_data_receive(struct sock_xprt *transport)
 714 {
 715         size_t read = 0;
 716         ssize_t ret = 0;
 717
 718         mutex_lock(&transport->recv_mutex);
 719         if (transport->sock == NULL)
 720                 goto out;
 721         for (;;) {
 722                 ret = xs_read_stream(transport, MSG_DONTWAIT);
 723                 if (ret < 0)
 724                         break;
 725                 read += ret;
 726                 cond_resched();
 727         }
 728         if (ret == -ESHUTDOWN)
 729                 kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 730         else
 731                 xs_poll_check_readable(transport);
 732 out:
 733         mutex_unlock(&transport->recv_mutex);
 734         trace_xs_stream_read_data(&transport->xprt, ret, read);
 735 }
 736
 737 static void xs_stream_data_receive_workfn(struct work_struct *work)
 738 {
 739         struct sock_xprt *transport =
 740                 container_of(work, struct sock_xprt, recv_worker);
 741         unsigned int pflags = memalloc_nofs_save();
 742
 743         xs_stream_data_receive(transport);
 744         memalloc_nofs_restore(pflags);
 745 }
 746
 747 static void
 748 xs_stream_reset_connect(struct sock_xprt *transport)
 749 {
 750         transport->recv.offset = 0;
 751         transport->recv.len = 0;
 752         transport->recv.copied = 0;
 753         transport->xmit.offset = 0;
 754 }
 755
 756 static void
 757 xs_stream_start_connect(struct sock_xprt *transport)
 758 {
 759         transport->xprt.stat.connect_count++;
 760         transport->xprt.stat.connect_start = jiffies;
 761 }
 762
 763 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 764
 765 /**
 766  * xs_nospace - handle transmit was incomplete
 767  * @req: pointer to RPC request
 768  * @transport: pointer to struct sock_xprt
 769  *
 770  */
 771 static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 772 {
 773         struct rpc_xprt *xprt = &transport->xprt;
 774         struct sock *sk = transport->inet;
 775         int ret = -EAGAIN;
 776
 777         trace_rpc_socket_nospace(req, transport);
 778
 779         /* Protect against races with write_space */
 780         spin_lock(&xprt->transport_lock);
 781
 782         /* Don't race with disconnect */
 783         if (xprt_connected(xprt)) {
 784                 /* wait for more buffer space */
 785                 set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
 786                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 787                 sk->sk_write_pending++;
 788                 xprt_wait_for_buffer_space(xprt);
 789         } else
 790                 ret = -ENOTCONN;
 791
 792         spin_unlock(&xprt->transport_lock);
 793         return ret;
 794 }
 795
 796 static int xs_sock_nospace(struct rpc_rqst *req)
 797 {
 798         struct sock_xprt *transport =
 799                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 800         struct sock *sk = transport->inet;
 801         int ret = -EAGAIN;
 802
 803         lock_sock(sk);
 804         if (!sock_writeable(sk))
 805                 ret = xs_nospace(req, transport);
 806         release_sock(sk);
 807         return ret;
 808 }
 809
 810 static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
 811 {
 812         struct sock_xprt *transport =
 813                 container_of(req->rq_xprt, struct sock_xprt, xprt);
 814         struct sock *sk = transport->inet;
 815         int ret = -EAGAIN;
 816
 817         if (vm_wait)
 818                 return -ENOBUFS;
 819         lock_sock(sk);
 820         if (!sk_stream_memory_free(sk))
 821                 ret = xs_nospace(req, transport);
 822         release_sock(sk);
 823         return ret;
 824 }
 825
 826 static int xs_stream_prepare_request(struct rpc_rqst *req, struct xdr_buf *buf)
 827 {
 828         return xdr_alloc_bvec(buf, rpc_task_gfp_mask());
 829 }
 830
 831 /*
 832  * Determine if the previous message in the stream was aborted before it
 833  * could complete transmission.
 834  */
 835 static bool
 836 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 837 {
 838         return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 839 }
 840
 841 /*
 842  * Return the stream record marker field for a record of length < 2^31-1
 843  */
 844 static rpc_fraghdr
 845 xs_stream_record_marker(struct xdr_buf *xdr)
 846 {
 847         if (!xdr->len)
 848                 return 0;
 849         return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 850 }
 851
 852 /**
 853  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 854  * @req: pointer to RPC request
 855  *
 856  * Return values:
 857  *        0:    The request has been sent
 858  *   EAGAIN:    The socket was blocked, please call again later to
 859  *              complete the request
 860  * ENOTCONN:    Caller needs to invoke connect logic then call again
 861  *    other:    Some other error occurred, the request was not sent
 862  */
 863 static int xs_local_send_request(struct rpc_rqst *req)
 864 {
 865         struct rpc_xprt *xprt = req->rq_xprt;
 866         struct sock_xprt *transport =
 867                                 container_of(xprt, struct sock_xprt, xprt);
 868         struct xdr_buf *xdr = &req->rq_snd_buf;
 869         rpc_fraghdr rm = xs_stream_record_marker(xdr);
 870         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 871         struct msghdr msg = {
 872                 .msg_flags      = XS_SENDMSG_FLAGS,
 873         };
 874         bool vm_wait;
 875         unsigned int sent;
 876         int status;
 877
 878         /* Close the stream if the previous transmission was incomplete */
 879         if (xs_send_request_was_aborted(transport, req)) {
 880                 xprt_force_disconnect(xprt);
 881                 return -ENOTCONN;
 882         }
 883
 884         xs_pktdump("packet data:",
 885                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 886
 887         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
 888
 889         req->rq_xtime = ktime_get();
 890         status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 891                                    transport->xmit.offset, rm, &sent);
 892         dprintk("RPC:       %s(%u) = %d\n",
 893                         __func__, xdr->len - transport->xmit.offset, status);
 894
 895         if (likely(sent > 0) || status == 0) {
 896                 transport->xmit.offset += sent;
 897                 req->rq_bytes_sent = transport->xmit.offset;
 898                 if (likely(req->rq_bytes_sent >= msglen)) {
 899                         req->rq_xmit_bytes_sent += transport->xmit.offset;
 900                         transport->xmit.offset = 0;
 901                         return 0;
 902                 }
 903                 status = -EAGAIN;
 904                 vm_wait = false;
 905         }
 906
 907         switch (status) {
 908         case -EAGAIN:
 909                 status = xs_stream_nospace(req, vm_wait);
 910                 break;
 911         default:
 912                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 913                         -status);
 914                 fallthrough;
 915         case -EPIPE:
 916                 xprt_force_disconnect(xprt);
 917                 status = -ENOTCONN;
 918         }
 919
 920         return status;
 921 }
 922
 923 /**
 924  * xs_udp_send_request - write an RPC request to a UDP socket
 925  * @req: pointer to RPC request
 926  *
 927  * Return values:
 928  *        0:    The request has been sent
 929  *   EAGAIN:    The socket was blocked, please call again later to
 930  *              complete the request
 931  * ENOTCONN:    Caller needs to invoke connect logic then call again
 932  *    other:    Some other error occurred, the request was not sent
 933  */
 934 static int xs_udp_send_request(struct rpc_rqst *req)
 935 {
 936         struct rpc_xprt *xprt = req->rq_xprt;
 937         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 938         struct xdr_buf *xdr = &req->rq_snd_buf;
 939         struct msghdr msg = {
 940                 .msg_name       = xs_addr(xprt),
 941                 .msg_namelen    = xprt->addrlen,
 942                 .msg_flags      = XS_SENDMSG_FLAGS,
 943         };
 944         unsigned int sent;
 945         int status;
 946
 947         xs_pktdump("packet data:",
 948                                 req->rq_svec->iov_base,
 949                                 req->rq_svec->iov_len);
 950
 951         if (!xprt_bound(xprt))
 952                 return -ENOTCONN;
 953
 954         if (!xprt_request_get_cong(xprt, req))
 955                 return -EBADSLT;
 956
 957         status = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
 958         if (status < 0)
 959                 return status;
 960         req->rq_xtime = ktime_get();
 961         status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
 962
 963         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 964                         xdr->len, status);
 965
 966         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 967         if (status == -EPERM)
 968                 goto process_status;
 969
 970         if (status == -EAGAIN && sock_writeable(transport->inet))
 971                 status = -ENOBUFS;
 972
 973         if (sent > 0 || status == 0) {
 974                 req->rq_xmit_bytes_sent += sent;
 975                 if (sent >= req->rq_slen)
 976                         return 0;
 977                 /* Still some bytes left; set up for a retry later. */
 978                 status = -EAGAIN;
 979         }
 980
 981 process_status:
 982         switch (status) {
 983         case -ENOTSOCK:
 984                 status = -ENOTCONN;
 985                 /* Should we call xs_close() here? */
 986                 break;
 987         case -EAGAIN:
 988                 status = xs_sock_nospace(req);
 989                 break;
 990         case -ENETUNREACH:
 991         case -ENOBUFS:
 992         case -EPIPE:
 993         case -ECONNREFUSED:
 994         case -EPERM:
 995                 /* When the server has died, an ICMP port unreachable message
 996                  * prompts ECONNREFUSED. */
 997                 break;
 998         default:
 999                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1000                         -status);
1001         }
1002
1003         return status;
1004 }
1005
1006 /**
1007  * xs_tcp_send_request - write an RPC request to a TCP socket
1008  * @req: pointer to RPC request
1009  *
1010  * Return values:
1011  *        0:    The request has been sent
1012  *   EAGAIN:    The socket was blocked, please call again later to
1013  *              complete the request
1014  * ENOTCONN:    Caller needs to invoke connect logic then call again
1015  *    other:    Some other error occurred, the request was not sent
1016  *
1017  * XXX: In the case of soft timeouts, should we eventually give up
1018  *      if sendmsg is not able to make progress?
1019  */
1020 static int xs_tcp_send_request(struct rpc_rqst *req)
1021 {
1022         struct rpc_xprt *xprt = req->rq_xprt;
1023         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1024         struct xdr_buf *xdr = &req->rq_snd_buf;
1025         rpc_fraghdr rm = xs_stream_record_marker(xdr);
1026         unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1027         struct msghdr msg = {
1028                 .msg_flags      = XS_SENDMSG_FLAGS,
1029         };
1030         bool vm_wait;
1031         unsigned int sent;
1032         int status;
1033
1034         /* Close the stream if the previous transmission was incomplete */
1035         if (xs_send_request_was_aborted(transport, req)) {
1036                 if (transport->sock != NULL)
1037                         kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1038                 return -ENOTCONN;
1039         }
1040         if (!transport->inet)
1041                 return -ENOTCONN;
1042
1043         xs_pktdump("packet data:",
1044                                 req->rq_svec->iov_base,
1045                                 req->rq_svec->iov_len);
1046
1047         if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1048                 xs_tcp_set_socket_timeouts(xprt, transport->sock);
1049
1050         xs_set_srcport(transport, transport->sock);
1051
1052         /* Continue transmitting the packet/record. We must be careful
1053          * to cope with writespace callbacks arriving _after_ we have
1054          * called sendmsg(). */
1055         req->rq_xtime = ktime_get();
1056         tcp_sock_set_cork(transport->inet, true);
1057
1058         vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
1059
1060         do {
1061                 status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1062                                            transport->xmit.offset, rm, &sent);
1063
1064                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1065                                 xdr->len - transport->xmit.offset, status);
1066
1067                 /* If we've sent the entire packet, immediately
1068                  * reset the count of bytes sent. */
1069                 transport->xmit.offset += sent;
1070                 req->rq_bytes_sent = transport->xmit.offset;
1071                 if (likely(req->rq_bytes_sent >= msglen)) {
1072                         req->rq_xmit_bytes_sent += transport->xmit.offset;
1073                         transport->xmit.offset = 0;
1074                         if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1075                                 tcp_sock_set_cork(transport->inet, false);
1076                         return 0;
1077                 }
1078
1079                 WARN_ON_ONCE(sent == 0 && status == 0);
1080
1081                 if (sent > 0)
1082                         vm_wait = false;
1083
1084         } while (status == 0);
1085
1086         switch (status) {
1087         case -ENOTSOCK:
1088                 status = -ENOTCONN;
1089                 /* Should we call xs_close() here? */
1090                 break;
1091         case -EAGAIN:
1092                 status = xs_stream_nospace(req, vm_wait);
1093                 break;
1094         case -ECONNRESET:
1095         case -ECONNREFUSED:
1096         case -ENOTCONN:
1097         case -EADDRINUSE:
1098         case -ENOBUFS:
1099         case -EPIPE:
1100                 break;
1101         default:
1102                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1103                         -status);
1104         }
1105
1106         return status;
1107 }
1108
1109 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1110 {
1111         transport->old_data_ready = sk->sk_data_ready;
1112         transport->old_state_change = sk->sk_state_change;
1113         transport->old_write_space = sk->sk_write_space;
1114         transport->old_error_report = sk->sk_error_report;
1115 }
1116
1117 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1118 {
1119         sk->sk_data_ready = transport->old_data_ready;
1120         sk->sk_state_change = transport->old_state_change;
1121         sk->sk_write_space = transport->old_write_space;
1122         sk->sk_error_report = transport->old_error_report;
1123 }
1124
1125 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1126 {
1127         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1128
1129         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1130         clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1131         clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1132         clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1133         clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
1134 }
1135
1136 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1137 {
1138         set_bit(nr, &transport->sock_state);
1139         queue_work(xprtiod_workqueue, &transport->error_worker);
1140 }
1141
1142 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1143 {
1144         xprt->connect_cookie++;
1145         smp_mb__before_atomic();
1146         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1147         clear_bit(XPRT_CLOSING, &xprt->state);
1148         xs_sock_reset_state_flags(xprt);
1149         smp_mb__after_atomic();
1150 }
1151
1152 /**
1153  * xs_error_report - callback to handle TCP socket state errors
1154  * @sk: socket
1155  *
1156  * Note: we don't call sock_error() since there may be a rpc_task
1157  * using the socket, and so we don't want to clear sk->sk_err.
1158  */
1159 static void xs_error_report(struct sock *sk)
1160 {
1161         struct sock_xprt *transport;
1162         struct rpc_xprt *xprt;
1163
1164         if (!(xprt = xprt_from_sock(sk)))
1165                 return;
1166
1167         transport = container_of(xprt, struct sock_xprt, xprt);
1168         transport->xprt_err = -sk->sk_err;
1169         if (transport->xprt_err == 0)
1170                 return;
1171         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1172                         xprt, -transport->xprt_err);
1173         trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1174
1175         /* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1176         smp_mb__before_atomic();
1177         xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1178 }
1179
1180 static void xs_reset_transport(struct sock_xprt *transport)
1181 {
1182         struct socket *sock = transport->sock;
1183         struct sock *sk = transport->inet;
1184         struct rpc_xprt *xprt = &transport->xprt;
1185         struct file *filp = transport->file;
1186
1187         if (sk == NULL)
1188                 return;
1189         /*
1190          * Make sure we're calling this in a context from which it is safe
1191          * to call __fput_sync(). In practice that means rpciod and the
1192          * system workqueue.
1193          */
1194         if (!(current->flags & PF_WQ_WORKER)) {
1195                 WARN_ON_ONCE(1);
1196                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
1197                 return;
1198         }
1199
1200         if (atomic_read(&transport->xprt.swapper))
1201                 sk_clear_memalloc(sk);
1202
1203         kernel_sock_shutdown(sock, SHUT_RDWR);
1204
1205         mutex_lock(&transport->recv_mutex);
1206         lock_sock(sk);
1207         transport->inet = NULL;
1208         transport->sock = NULL;
1209         transport->file = NULL;
1210
1211         sk->sk_user_data = NULL;
1212
1213         xs_restore_old_callbacks(transport, sk);
1214         xprt_clear_connected(xprt);
1215         xs_sock_reset_connection_flags(xprt);
1216         /* Reset stream record info */
1217         xs_stream_reset_connect(transport);
1218         release_sock(sk);
1219         mutex_unlock(&transport->recv_mutex);
1220
1221         trace_rpc_socket_close(xprt, sock);
1222         __fput_sync(filp);
1223
1224         xprt_disconnect_done(xprt);
1225 }
1226
1227 /**
1228  * xs_close - close a socket
1229  * @xprt: transport
1230  *
1231  * This is used when all requests are complete; ie, no DRC state remains
1232  * on the server we want to save.
1233  *
1234  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1235  * xs_reset_transport() zeroing the socket from underneath a writer.
1236  */
1237 static void xs_close(struct rpc_xprt *xprt)
1238 {
1239         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1240
1241         dprintk("RPC:       xs_close xprt %p\n", xprt);
1242
1243         xs_reset_transport(transport);
1244         xprt->reestablish_timeout = 0;
1245 }
1246
1247 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1248 {
1249         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1250                 xprt);
1251         xprt_disconnect_done(xprt);
1252 }
1253
1254 static void xs_xprt_free(struct rpc_xprt *xprt)
1255 {
1256         xs_free_peer_addresses(xprt);
1257         xprt_free(xprt);
1258 }
1259
1260 /**
1261  * xs_destroy - prepare to shutdown a transport
1262  * @xprt: doomed transport
1263  *
1264  */
1265 static void xs_destroy(struct rpc_xprt *xprt)
1266 {
1267         struct sock_xprt *transport = container_of(xprt,
1268                         struct sock_xprt, xprt);
1269         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1270
1271         cancel_delayed_work_sync(&transport->connect_worker);
1272         xs_close(xprt);
1273         cancel_work_sync(&transport->recv_worker);
1274         cancel_work_sync(&transport->error_worker);
1275         xs_xprt_free(xprt);
1276         module_put(THIS_MODULE);
1277 }
1278
1279 /**
1280  * xs_udp_data_read_skb - receive callback for UDP sockets
1281  * @xprt: transport
1282  * @sk: socket
1283  * @skb: skbuff
1284  *
1285  */
1286 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1287                 struct sock *sk,
1288                 struct sk_buff *skb)
1289 {
1290         struct rpc_task *task;
1291         struct rpc_rqst *rovr;
1292         int repsize, copied;
1293         u32 _xid;
1294         __be32 *xp;
1295
1296         repsize = skb->len;
1297         if (repsize < 4) {
1298                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1299                 return;
1300         }
1301
1302         /* Copy the XID from the skb... */
1303         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1304         if (xp == NULL)
1305                 return;
1306
1307         /* Look up and lock the request corresponding to the given XID */
1308         spin_lock(&xprt->queue_lock);
1309         rovr = xprt_lookup_rqst(xprt, *xp);
1310         if (!rovr)
1311                 goto out_unlock;
1312         xprt_pin_rqst(rovr);
1313         xprt_update_rtt(rovr->rq_task);
1314         spin_unlock(&xprt->queue_lock);
1315         task = rovr->rq_task;
1316
1317         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1318                 copied = repsize;
1319
1320         /* Suck it into the iovec, verify checksum if not done by hw. */
1321         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1322                 spin_lock(&xprt->queue_lock);
1323                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1324                 goto out_unpin;
1325         }
1326
1327
1328         spin_lock(&xprt->transport_lock);
1329         xprt_adjust_cwnd(xprt, task, copied);
1330         spin_unlock(&xprt->transport_lock);
1331         spin_lock(&xprt->queue_lock);
1332         xprt_complete_rqst(task, copied);
1333         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1334 out_unpin:
1335         xprt_unpin_rqst(rovr);
1336  out_unlock:
1337         spin_unlock(&xprt->queue_lock);
1338 }
1339
1340 static void xs_udp_data_receive(struct sock_xprt *transport)
1341 {
1342         struct sk_buff *skb;
1343         struct sock *sk;
1344         int err;
1345
1346         mutex_lock(&transport->recv_mutex);
1347         sk = transport->inet;
1348         if (sk == NULL)
1349                 goto out;
1350         for (;;) {
1351                 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1352                 if (skb == NULL)
1353                         break;
1354                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1355                 consume_skb(skb);
1356                 cond_resched();
1357         }
1358         xs_poll_check_readable(transport);
1359 out:
1360         mutex_unlock(&transport->recv_mutex);
1361 }
1362
1363 static void xs_udp_data_receive_workfn(struct work_struct *work)
1364 {
1365         struct sock_xprt *transport =
1366                 container_of(work, struct sock_xprt, recv_worker);
1367         unsigned int pflags = memalloc_nofs_save();
1368
1369         xs_udp_data_receive(transport);
1370         memalloc_nofs_restore(pflags);
1371 }
1372
1373 /**
1374  * xs_data_ready - "data ready" callback for sockets
1375  * @sk: socket with data to read
1376  *
1377  */
1378 static void xs_data_ready(struct sock *sk)
1379 {
1380         struct rpc_xprt *xprt;
1381
1382         trace_sk_data_ready(sk);
1383
1384         xprt = xprt_from_sock(sk);
1385         if (xprt != NULL) {
1386                 struct sock_xprt *transport = container_of(xprt,
1387                                 struct sock_xprt, xprt);
1388
1389                 trace_xs_data_ready(xprt);
1390
1391                 transport->old_data_ready(sk);
1392                 /* Any data means we had a useful conversation, so
1393                  * then we don't need to delay the next reconnect
1394                  */
1395                 if (xprt->reestablish_timeout)
1396                         xprt->reestablish_timeout = 0;
1397                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1398                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1399         }
1400 }
1401
1402 /*
1403  * Helper function to force a TCP close if the server is sending
1404  * junk and/or it has put us in CLOSE_WAIT
1405  */
1406 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1407 {
1408         xprt_force_disconnect(xprt);
1409 }
1410
1411 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1412 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1413 {
1414         return PAGE_SIZE;
1415 }
1416 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1417
1418 /**
1419  * xs_local_state_change - callback to handle AF_LOCAL socket state changes
1420  * @sk: socket whose state has changed
1421  *
1422  */
1423 static void xs_local_state_change(struct sock *sk)
1424 {
1425         struct rpc_xprt *xprt;
1426         struct sock_xprt *transport;
1427
1428         if (!(xprt = xprt_from_sock(sk)))
1429                 return;
1430         transport = container_of(xprt, struct sock_xprt, xprt);
1431         if (sk->sk_shutdown & SHUTDOWN_MASK) {
1432                 clear_bit(XPRT_CONNECTED, &xprt->state);
1433                 /* Trigger the socket release */
1434                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1435         }
1436 }
1437
1438 /**
1439  * xs_tcp_state_change - callback to handle TCP socket state changes
1440  * @sk: socket whose state has changed
1441  *
1442  */
1443 static void xs_tcp_state_change(struct sock *sk)
1444 {
1445         struct rpc_xprt *xprt;
1446         struct sock_xprt *transport;
1447
1448         if (!(xprt = xprt_from_sock(sk)))
1449                 return;
1450         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1451         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1452                         sk->sk_state, xprt_connected(xprt),
1453                         sock_flag(sk, SOCK_DEAD),
1454                         sock_flag(sk, SOCK_ZAPPED),
1455                         sk->sk_shutdown);
1456
1457         transport = container_of(xprt, struct sock_xprt, xprt);
1458         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1459         switch (sk->sk_state) {
1460         case TCP_ESTABLISHED:
1461                 if (!xprt_test_and_set_connected(xprt)) {
1462                         xprt->connect_cookie++;
1463                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1464                         xprt_clear_connecting(xprt);
1465
1466                         xprt->stat.connect_count++;
1467                         xprt->stat.connect_time += (long)jiffies -
1468                                                    xprt->stat.connect_start;
1469                         xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1470                 }
1471                 break;
1472         case TCP_FIN_WAIT1:
1473                 /* The client initiated a shutdown of the socket */
1474                 xprt->connect_cookie++;
1475                 xprt->reestablish_timeout = 0;
1476                 set_bit(XPRT_CLOSING, &xprt->state);
1477                 smp_mb__before_atomic();
1478                 clear_bit(XPRT_CONNECTED, &xprt->state);
1479                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1480                 smp_mb__after_atomic();
1481                 break;
1482         case TCP_CLOSE_WAIT:
1483                 /* The server initiated a shutdown of the socket */
1484                 xprt->connect_cookie++;
1485                 clear_bit(XPRT_CONNECTED, &xprt->state);
1486                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1487                 fallthrough;
1488         case TCP_CLOSING:
1489                 /*
1490                  * If the server closed down the connection, make sure that
1491                  * we back off before reconnecting
1492                  */
1493                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1494                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1495                 break;
1496         case TCP_LAST_ACK:
1497                 set_bit(XPRT_CLOSING, &xprt->state);
1498                 smp_mb__before_atomic();
1499                 clear_bit(XPRT_CONNECTED, &xprt->state);
1500                 smp_mb__after_atomic();
1501                 break;
1502         case TCP_CLOSE:
1503                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1504                                         &transport->sock_state))
1505                         xprt_clear_connecting(xprt);
1506                 clear_bit(XPRT_CLOSING, &xprt->state);
1507                 /* Trigger the socket release */
1508                 xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1509         }
1510 }
1511
1512 static void xs_write_space(struct sock *sk)
1513 {
1514         struct sock_xprt *transport;
1515         struct rpc_xprt *xprt;
1516
1517         if (!sk->sk_socket)
1518                 return;
1519         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1520
1521         if (unlikely(!(xprt = xprt_from_sock(sk))))
1522                 return;
1523         transport = container_of(xprt, struct sock_xprt, xprt);
1524         if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
1525                 return;
1526         xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1527         sk->sk_write_pending--;
1528 }
1529
1530 /**
1531  * xs_udp_write_space - callback invoked when socket buffer space
1532  *                             becomes available
1533  * @sk: socket whose state has changed
1534  *
1535  * Called when more output buffer space is available for this socket.
1536  * We try not to wake our writers until they can make "significant"
1537  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1538  * with a bunch of small requests.
1539  */
1540 static void xs_udp_write_space(struct sock *sk)
1541 {
1542         /* from net/core/sock.c:sock_def_write_space */
1543         if (sock_writeable(sk))
1544                 xs_write_space(sk);
1545 }
1546
1547 /**
1548  * xs_tcp_write_space - callback invoked when socket buffer space
1549  *                             becomes available
1550  * @sk: socket whose state has changed
1551  *
1552  * Called when more output buffer space is available for this socket.
1553  * We try not to wake our writers until they can make "significant"
1554  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1555  * with a bunch of small requests.
1556  */
1557 static void xs_tcp_write_space(struct sock *sk)
1558 {
1559         /* from net/core/stream.c:sk_stream_write_space */
1560         if (sk_stream_is_writeable(sk))
1561                 xs_write_space(sk);
1562 }
1563
1564 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1565 {
1566         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1567         struct sock *sk = transport->inet;
1568
1569         if (transport->rcvsize) {
1570                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1571                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1572         }
1573         if (transport->sndsize) {
1574                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1575                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1576                 sk->sk_write_space(sk);
1577         }
1578 }
1579
1580 /**
1581  * xs_udp_set_buffer_size - set send and receive limits
1582  * @xprt: generic transport
1583  * @sndsize: requested size of send buffer, in bytes
1584  * @rcvsize: requested size of receive buffer, in bytes
1585  *
1586  * Set socket send and receive buffer size limits.
1587  */
1588 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1589 {
1590         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1591
1592         transport->sndsize = 0;
1593         if (sndsize)
1594                 transport->sndsize = sndsize + 1024;
1595         transport->rcvsize = 0;
1596         if (rcvsize)
1597                 transport->rcvsize = rcvsize + 1024;
1598
1599         xs_udp_do_set_buffer_size(xprt);
1600 }
1601
1602 /**
1603  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1604  * @xprt: controlling transport
1605  * @task: task that timed out
1606  *
1607  * Adjust the congestion window after a retransmit timeout has occurred.
1608  */
1609 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1610 {
1611         spin_lock(&xprt->transport_lock);
1612         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1613         spin_unlock(&xprt->transport_lock);
1614 }
1615
1616 static int xs_get_random_port(void)
1617 {
1618         unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1619         unsigned short range;
1620         unsigned short rand;
1621
1622         if (max < min)
1623                 return -EADDRINUSE;
1624         range = max - min + 1;
1625         rand = get_random_u32_below(range);
1626         return rand + min;
1627 }
1628
1629 static unsigned short xs_sock_getport(struct socket *sock)
1630 {
1631         struct sockaddr_storage buf;
1632         unsigned short port = 0;
1633
1634         if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1635                 goto out;
1636         switch (buf.ss_family) {
1637         case AF_INET6:
1638                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1639                 break;
1640         case AF_INET:
1641                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1642         }
1643 out:
1644         return port;
1645 }
1646
1647 /**
1648  * xs_set_port - reset the port number in the remote endpoint address
1649  * @xprt: generic transport
1650  * @port: new port number
1651  *
1652  */
1653 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1654 {
1655         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1656
1657         rpc_set_port(xs_addr(xprt), port);
1658         xs_update_peer_port(xprt);
1659 }
1660
1661 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1662 {
1663         if (transport->srcport == 0 && transport->xprt.reuseport)
1664                 transport->srcport = xs_sock_getport(sock);
1665 }
1666
1667 static int xs_get_srcport(struct sock_xprt *transport)
1668 {
1669         int port = transport->srcport;
1670
1671         if (port == 0 && transport->xprt.resvport)
1672                 port = xs_get_random_port();
1673         return port;
1674 }
1675
1676 static unsigned short xs_sock_srcport(struct rpc_xprt *xprt)
1677 {
1678         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1679         unsigned short ret = 0;
1680         mutex_lock(&sock->recv_mutex);
1681         if (sock->sock)
1682                 ret = xs_sock_getport(sock->sock);
1683         mutex_unlock(&sock->recv_mutex);
1684         return ret;
1685 }
1686
1687 static int xs_sock_srcaddr(struct rpc_xprt *xprt, char *buf, size_t buflen)
1688 {
1689         struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1690         union {
1691                 struct sockaddr sa;
1692                 struct sockaddr_storage st;
1693         } saddr;
1694         int ret = -ENOTCONN;
1695
1696         mutex_lock(&sock->recv_mutex);
1697         if (sock->sock) {
1698                 ret = kernel_getsockname(sock->sock, &saddr.sa);
1699                 if (ret >= 0)
1700                         ret = snprintf(buf, buflen, "%pISc", &saddr.sa);
1701         }
1702         mutex_unlock(&sock->recv_mutex);
1703         return ret;
1704 }
1705
1706 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1707 {
1708         if (transport->srcport != 0)
1709                 transport->srcport = 0;
1710         if (!transport->xprt.resvport)
1711                 return 0;
1712         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1713                 return xprt_max_resvport;
1714         return --port;
1715 }
1716 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1717 {
1718         struct sockaddr_storage myaddr;
1719         int err, nloop = 0;
1720         int port = xs_get_srcport(transport);
1721         unsigned short last;
1722
1723         /*
1724          * If we are asking for any ephemeral port (i.e. port == 0 &&
1725          * transport->xprt.resvport == 0), don't bind.  Let the local
1726          * port selection happen implicitly when the socket is used
1727          * (for example at connect time).
1728          *
1729          * This ensures that we can continue to establish TCP
1730          * connections even when all local ephemeral ports are already
1731          * a part of some TCP connection.  This makes no difference
1732          * for UDP sockets, but also doesn't harm them.
1733          *
1734          * If we're asking for any reserved port (i.e. port == 0 &&
1735          * transport->xprt.resvport == 1) xs_get_srcport above will
1736          * ensure that port is non-zero and we will bind as needed.
1737          */
1738         if (port <= 0)
1739                 return port;
1740
1741         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1742         do {
1743                 rpc_set_port((struct sockaddr *)&myaddr, port);
1744                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1745                                 transport->xprt.addrlen);
1746                 if (err == 0) {
1747                         if (transport->xprt.reuseport)
1748                                 transport->srcport = port;
1749                         break;
1750                 }
1751                 last = port;
1752                 port = xs_next_srcport(transport, port);
1753                 if (port > last)
1754                         nloop++;
1755         } while (err == -EADDRINUSE && nloop != 2);
1756
1757         if (myaddr.ss_family == AF_INET)
1758                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1759                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1760                                 port, err ? "failed" : "ok", err);
1761         else
1762                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1763                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1764                                 port, err ? "failed" : "ok", err);
1765         return err;
1766 }
1767
1768 /*
1769  * We don't support autobind on AF_LOCAL sockets
1770  */
1771 static void xs_local_rpcbind(struct rpc_task *task)
1772 {
1773         xprt_set_bound(task->tk_xprt);
1774 }
1775
1776 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1777 {
1778 }
1779
1780 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1781 static struct lock_class_key xs_key[3];
1782 static struct lock_class_key xs_slock_key[3];
1783
1784 static inline void xs_reclassify_socketu(struct socket *sock)
1785 {
1786         struct sock *sk = sock->sk;
1787
1788         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1789                 &xs_slock_key[0], "sk_lock-AF_LOCAL-RPC", &xs_key[0]);
1790 }
1791
1792 static inline void xs_reclassify_socket4(struct socket *sock)
1793 {
1794         struct sock *sk = sock->sk;
1795
1796         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1797                 &xs_slock_key[1], "sk_lock-AF_INET-RPC", &xs_key[1]);
1798 }
1799
1800 static inline void xs_reclassify_socket6(struct socket *sock)
1801 {
1802         struct sock *sk = sock->sk;
1803
1804         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1805                 &xs_slock_key[2], "sk_lock-AF_INET6-RPC", &xs_key[2]);
1806 }
1807
1808 static inline void xs_reclassify_socket(int family, struct socket *sock)
1809 {
1810         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1811                 return;
1812
1813         switch (family) {
1814         case AF_LOCAL:
1815                 xs_reclassify_socketu(sock);
1816                 break;
1817         case AF_INET:
1818                 xs_reclassify_socket4(sock);
1819                 break;
1820         case AF_INET6:
1821                 xs_reclassify_socket6(sock);
1822                 break;
1823         }
1824 }
1825 #else
1826 static inline void xs_reclassify_socket(int family, struct socket *sock)
1827 {
1828 }
1829 #endif
1830
1831 static void xs_dummy_setup_socket(struct work_struct *work)
1832 {
1833 }
1834
1835 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1836                 struct sock_xprt *transport, int family, int type,
1837                 int protocol, bool reuseport)
1838 {
1839         struct file *filp;
1840         struct socket *sock;
1841         int err;
1842
1843         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1844         if (err < 0) {
1845                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1846                                 protocol, -err);
1847                 goto out;
1848         }
1849         xs_reclassify_socket(family, sock);
1850
1851         if (reuseport)
1852                 sock_set_reuseport(sock->sk);
1853
1854         err = xs_bind(transport, sock);
1855         if (err) {
1856                 sock_release(sock);
1857                 goto out;
1858         }
1859
1860         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1861         if (IS_ERR(filp))
1862                 return ERR_CAST(filp);
1863         transport->file = filp;
1864
1865         return sock;
1866 out:
1867         return ERR_PTR(err);
1868 }
1869
1870 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1871                                       struct socket *sock)
1872 {
1873         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1874                                                                         xprt);
1875
1876         if (!transport->inet) {
1877                 struct sock *sk = sock->sk;
1878
1879                 lock_sock(sk);
1880
1881                 xs_save_old_callbacks(transport, sk);
1882
1883                 sk->sk_user_data = xprt;
1884                 sk->sk_data_ready = xs_data_ready;
1885                 sk->sk_write_space = xs_udp_write_space;
1886                 sk->sk_state_change = xs_local_state_change;
1887                 sk->sk_error_report = xs_error_report;
1888                 sk->sk_use_task_frag = false;
1889
1890                 xprt_clear_connected(xprt);
1891
1892                 /* Reset to new socket */
1893                 transport->sock = sock;
1894                 transport->inet = sk;
1895
1896                 release_sock(sk);
1897         }
1898
1899         xs_stream_start_connect(transport);
1900
1901         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1902 }
1903
1904 /**
1905  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1906  * @transport: socket transport to connect
1907  */
1908 static int xs_local_setup_socket(struct sock_xprt *transport)
1909 {
1910         struct rpc_xprt *xprt = &transport->xprt;
1911         struct file *filp;
1912         struct socket *sock;
1913         int status;
1914
1915         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1916                                         SOCK_STREAM, 0, &sock, 1);
1917         if (status < 0) {
1918                 dprintk("RPC:       can't create AF_LOCAL "
1919                         "transport socket (%d).\n", -status);
1920                 goto out;
1921         }
1922         xs_reclassify_socket(AF_LOCAL, sock);
1923
1924         filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1925         if (IS_ERR(filp)) {
1926                 status = PTR_ERR(filp);
1927                 goto out;
1928         }
1929         transport->file = filp;
1930
1931         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1932                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1933
1934         status = xs_local_finish_connecting(xprt, sock);
1935         trace_rpc_socket_connect(xprt, sock, status);
1936         switch (status) {
1937         case 0:
1938                 dprintk("RPC:       xprt %p connected to %s\n",
1939                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1940                 xprt->stat.connect_count++;
1941                 xprt->stat.connect_time += (long)jiffies -
1942                                            xprt->stat.connect_start;
1943                 xprt_set_connected(xprt);
1944                 break;
1945         case -ENOBUFS:
1946                 break;
1947         case -ENOENT:
1948                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1949                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1950                 break;
1951         case -ECONNREFUSED:
1952                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1953                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1954                 break;
1955         default:
1956                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1957                                 __func__, -status,
1958                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1959         }
1960
1961 out:
1962         xprt_clear_connecting(xprt);
1963         xprt_wake_pending_tasks(xprt, status);
1964         return status;
1965 }
1966
1967 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1968 {
1969         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1970         int ret;
1971
1972         if (transport->file)
1973                 goto force_disconnect;
1974
1975         if (RPC_IS_ASYNC(task)) {
1976                 /*
1977                  * We want the AF_LOCAL connect to be resolved in the
1978                  * filesystem namespace of the process making the rpc
1979                  * call.  Thus we connect synchronously.
1980                  *
1981                  * If we want to support asynchronous AF_LOCAL calls,
1982                  * we'll need to figure out how to pass a namespace to
1983                  * connect.
1984                  */
1985                 rpc_task_set_rpc_status(task, -ENOTCONN);
1986                 goto out_wake;
1987         }
1988         ret = xs_local_setup_socket(transport);
1989         if (ret && !RPC_IS_SOFTCONN(task))
1990                 msleep_interruptible(15000);
1991         return;
1992 force_disconnect:
1993         xprt_force_disconnect(xprt);
1994 out_wake:
1995         xprt_clear_connecting(xprt);
1996         xprt_wake_pending_tasks(xprt, -ENOTCONN);
1997 }
1998
1999 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2000 /*
2001  * Note that this should be called with XPRT_LOCKED held, or recv_mutex
2002  * held, or when we otherwise know that we have exclusive access to the
2003  * socket, to guard against races with xs_reset_transport.
2004  */
2005 static void xs_set_memalloc(struct rpc_xprt *xprt)
2006 {
2007         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2008                         xprt);
2009
2010         /*
2011          * If there's no sock, then we have nothing to set. The
2012          * reconnecting process will get it for us.
2013          */
2014         if (!transport->inet)
2015                 return;
2016         if (atomic_read(&xprt->swapper))
2017                 sk_set_memalloc(transport->inet);
2018 }
2019
2020 /**
2021  * xs_enable_swap - Tag this transport as being used for swap.
2022  * @xprt: transport to tag
2023  *
2024  * Take a reference to this transport on behalf of the rpc_clnt, and
2025  * optionally mark it for swapping if it wasn't already.
2026  */
2027 static int
2028 xs_enable_swap(struct rpc_xprt *xprt)
2029 {
2030         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2031
2032         mutex_lock(&xs->recv_mutex);
2033         if (atomic_inc_return(&xprt->swapper) == 1 &&
2034             xs->inet)
2035                 sk_set_memalloc(xs->inet);
2036         mutex_unlock(&xs->recv_mutex);
2037         return 0;
2038 }
2039
2040 /**
2041  * xs_disable_swap - Untag this transport as being used for swap.
2042  * @xprt: transport to tag
2043  *
2044  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2045  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2046  */
2047 static void
2048 xs_disable_swap(struct rpc_xprt *xprt)
2049 {
2050         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2051
2052         mutex_lock(&xs->recv_mutex);
2053         if (atomic_dec_and_test(&xprt->swapper) &&
2054             xs->inet)
2055                 sk_clear_memalloc(xs->inet);
2056         mutex_unlock(&xs->recv_mutex);
2057 }
2058 #else
2059 static void xs_set_memalloc(struct rpc_xprt *xprt)
2060 {
2061 }
2062
2063 static int
2064 xs_enable_swap(struct rpc_xprt *xprt)
2065 {
2066         return -EINVAL;
2067 }
2068
2069 static void
2070 xs_disable_swap(struct rpc_xprt *xprt)
2071 {
2072 }
2073 #endif
2074
2075 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2076 {
2077         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2078
2079         if (!transport->inet) {
2080                 struct sock *sk = sock->sk;
2081
2082                 lock_sock(sk);
2083
2084                 xs_save_old_callbacks(transport, sk);
2085
2086                 sk->sk_user_data = xprt;
2087                 sk->sk_data_ready = xs_data_ready;
2088                 sk->sk_write_space = xs_udp_write_space;
2089                 sk->sk_use_task_frag = false;
2090
2091                 xprt_set_connected(xprt);
2092
2093                 /* Reset to new socket */
2094                 transport->sock = sock;
2095                 transport->inet = sk;
2096
2097                 xs_set_memalloc(xprt);
2098
2099                 release_sock(sk);
2100         }
2101         xs_udp_do_set_buffer_size(xprt);
2102
2103         xprt->stat.connect_start = jiffies;
2104 }
2105
2106 static void xs_udp_setup_socket(struct work_struct *work)
2107 {
2108         struct sock_xprt *transport =
2109                 container_of(work, struct sock_xprt, connect_worker.work);
2110         struct rpc_xprt *xprt = &transport->xprt;
2111         struct socket *sock;
2112         int status = -EIO;
2113         unsigned int pflags = current->flags;
2114
2115         if (atomic_read(&xprt->swapper))
2116                 current->flags |= PF_MEMALLOC;
2117         sock = xs_create_sock(xprt, transport,
2118                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2119                         IPPROTO_UDP, false);
2120         if (IS_ERR(sock))
2121                 goto out;
2122
2123         dprintk("RPC:       worker connecting xprt %p via %s to "
2124                                 "%s (port %s)\n", xprt,
2125                         xprt->address_strings[RPC_DISPLAY_PROTO],
2126                         xprt->address_strings[RPC_DISPLAY_ADDR],
2127                         xprt->address_strings[RPC_DISPLAY_PORT]);
2128
2129         xs_udp_finish_connecting(xprt, sock);
2130         trace_rpc_socket_connect(xprt, sock, 0);
2131         status = 0;
2132 out:
2133         xprt_clear_connecting(xprt);
2134         xprt_unlock_connect(xprt, transport);
2135         xprt_wake_pending_tasks(xprt, status);
2136         current_restore_flags(pflags, PF_MEMALLOC);
2137 }
2138
2139 /**
2140  * xs_tcp_shutdown - gracefully shut down a TCP socket
2141  * @xprt: transport
2142  *
2143  * Initiates a graceful shutdown of the TCP socket by calling the
2144  * equivalent of shutdown(SHUT_RDWR);
2145  */
2146 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2147 {
2148         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2149         struct socket *sock = transport->sock;
2150         int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2151
2152         if (sock == NULL)
2153                 return;
2154         if (!xprt->reuseport) {
2155                 xs_close(xprt);
2156                 return;
2157         }
2158         switch (skst) {
2159         case TCP_FIN_WAIT1:
2160         case TCP_FIN_WAIT2:
2161         case TCP_LAST_ACK:
2162                 break;
2163         case TCP_ESTABLISHED:
2164         case TCP_CLOSE_WAIT:
2165                 kernel_sock_shutdown(sock, SHUT_RDWR);
2166                 trace_rpc_socket_shutdown(xprt, sock);
2167                 break;
2168         default:
2169                 xs_reset_transport(transport);
2170         }
2171 }
2172
2173 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2174                 struct socket *sock)
2175 {
2176         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2177         unsigned int keepidle;
2178         unsigned int keepcnt;
2179         unsigned int timeo;
2180
2181         spin_lock(&xprt->transport_lock);
2182         keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2183         keepcnt = xprt->timeout->to_retries + 1;
2184         timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2185                 (xprt->timeout->to_retries + 1);
2186         clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2187         spin_unlock(&xprt->transport_lock);
2188
2189         /* TCP Keepalive options */
2190         sock_set_keepalive(sock->sk);
2191         tcp_sock_set_keepidle(sock->sk, keepidle);
2192         tcp_sock_set_keepintvl(sock->sk, keepidle);
2193         tcp_sock_set_keepcnt(sock->sk, keepcnt);
2194
2195         /* TCP user timeout (see RFC5482) */
2196         tcp_sock_set_user_timeout(sock->sk, timeo);
2197 }
2198
2199 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2200                 unsigned long connect_timeout,
2201                 unsigned long reconnect_timeout)
2202 {
2203         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2204         struct rpc_timeout to;
2205         unsigned long initval;
2206
2207         spin_lock(&xprt->transport_lock);
2208         if (reconnect_timeout < xprt->max_reconnect_timeout)
2209                 xprt->max_reconnect_timeout = reconnect_timeout;
2210         if (connect_timeout < xprt->connect_timeout) {
2211                 memcpy(&to, xprt->timeout, sizeof(to));
2212                 initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2213                 /* Arbitrary lower limit */
2214                 if (initval <  XS_TCP_INIT_REEST_TO << 1)
2215                         initval = XS_TCP_INIT_REEST_TO << 1;
2216                 to.to_initval = initval;
2217                 to.to_maxval = initval;
2218                 memcpy(&transport->tcp_timeout, &to,
2219                                 sizeof(transport->tcp_timeout));
2220                 xprt->timeout = &transport->tcp_timeout;
2221                 xprt->connect_timeout = connect_timeout;
2222         }
2223         set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2224         spin_unlock(&xprt->transport_lock);
2225 }
2226
2227 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2228 {
2229         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2230
2231         if (!transport->inet) {
2232                 struct sock *sk = sock->sk;
2233
2234                 /* Avoid temporary address, they are bad for long-lived
2235                  * connections such as NFS mounts.
2236                  * RFC4941, section 3.6 suggests that:
2237                  *    Individual applications, which have specific
2238                  *    knowledge about the normal duration of connections,
2239                  *    MAY override this as appropriate.
2240                  */
2241                 if (xs_addr(xprt)->sa_family == PF_INET6) {
2242                         ip6_sock_set_addr_preferences(sk,
2243                                 IPV6_PREFER_SRC_PUBLIC);
2244                 }
2245
2246                 xs_tcp_set_socket_timeouts(xprt, sock);
2247                 tcp_sock_set_nodelay(sk);
2248
2249                 lock_sock(sk);
2250
2251                 xs_save_old_callbacks(transport, sk);
2252
2253                 sk->sk_user_data = xprt;
2254                 sk->sk_data_ready = xs_data_ready;
2255                 sk->sk_state_change = xs_tcp_state_change;
2256                 sk->sk_write_space = xs_tcp_write_space;
2257                 sk->sk_error_report = xs_error_report;
2258                 sk->sk_use_task_frag = false;
2259
2260                 /* socket options */
2261                 sock_reset_flag(sk, SOCK_LINGER);
2262
2263                 xprt_clear_connected(xprt);
2264
2265                 /* Reset to new socket */
2266                 transport->sock = sock;
2267                 transport->inet = sk;
2268
2269                 release_sock(sk);
2270         }
2271
2272         if (!xprt_bound(xprt))
2273                 return -ENOTCONN;
2274
2275         xs_set_memalloc(xprt);
2276
2277         xs_stream_start_connect(transport);
2278
2279         /* Tell the socket layer to start connecting... */
2280         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2281         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2282 }
2283
2284 /**
2285  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2286  * @work: queued work item
2287  *
2288  * Invoked by a work queue tasklet.
2289  */
2290 static void xs_tcp_setup_socket(struct work_struct *work)
2291 {
2292         struct sock_xprt *transport =
2293                 container_of(work, struct sock_xprt, connect_worker.work);
2294         struct socket *sock = transport->sock;
2295         struct rpc_xprt *xprt = &transport->xprt;
2296         int status;
2297         unsigned int pflags = current->flags;
2298
2299         if (atomic_read(&xprt->swapper))
2300                 current->flags |= PF_MEMALLOC;
2301
2302         if (xprt_connected(xprt))
2303                 goto out;
2304         if (test_and_clear_bit(XPRT_SOCK_CONNECT_SENT,
2305                                &transport->sock_state) ||
2306             !sock) {
2307                 xs_reset_transport(transport);
2308                 sock = xs_create_sock(xprt, transport, xs_addr(xprt)->sa_family,
2309                                       SOCK_STREAM, IPPROTO_TCP, true);
2310                 if (IS_ERR(sock)) {
2311                         xprt_wake_pending_tasks(xprt, PTR_ERR(sock));
2312                         goto out;
2313                 }
2314         }
2315
2316         dprintk("RPC:       worker connecting xprt %p via %s to "
2317                                 "%s (port %s)\n", xprt,
2318                         xprt->address_strings[RPC_DISPLAY_PROTO],
2319                         xprt->address_strings[RPC_DISPLAY_ADDR],
2320                         xprt->address_strings[RPC_DISPLAY_PORT]);
2321
2322         status = xs_tcp_finish_connecting(xprt, sock);
2323         trace_rpc_socket_connect(xprt, sock, status);
2324         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2325                         xprt, -status, xprt_connected(xprt),
2326                         sock->sk->sk_state);
2327         switch (status) {
2328         case 0:
2329         case -EINPROGRESS:
2330                 /* SYN_SENT! */
2331                 set_bit(XPRT_SOCK_CONNECT_SENT, &transport->sock_state);
2332                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2333                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2334                 fallthrough;
2335         case -EALREADY:
2336                 goto out_unlock;
2337         case -EADDRNOTAVAIL:
2338                 /* Source port number is unavailable. Try a new one! */
2339                 transport->srcport = 0;
2340                 status = -EAGAIN;
2341                 break;
2342         case -EINVAL:
2343                 /* Happens, for instance, if the user specified a link
2344                  * local IPv6 address without a scope-id.
2345                  */
2346         case -ECONNREFUSED:
2347         case -ECONNRESET:
2348         case -ENETDOWN:
2349         case -ENETUNREACH:
2350         case -EHOSTUNREACH:
2351         case -EADDRINUSE:
2352         case -ENOBUFS:
2353                 break;
2354         default:
2355                 printk("%s: connect returned unhandled error %d\n",
2356                         __func__, status);
2357                 status = -EAGAIN;
2358         }
2359
2360         /* xs_tcp_force_close() wakes tasks with a fixed error code.
2361          * We need to wake them first to ensure the correct error code.
2362          */
2363         xprt_wake_pending_tasks(xprt, status);
2364         xs_tcp_force_close(xprt);
2365 out:
2366         xprt_clear_connecting(xprt);
2367 out_unlock:
2368         xprt_unlock_connect(xprt, transport);
2369         current_restore_flags(pflags, PF_MEMALLOC);
2370 }
2371
2372 /**
2373  * xs_connect - connect a socket to a remote endpoint
2374  * @xprt: pointer to transport structure
2375  * @task: address of RPC task that manages state of connect request
2376  *
2377  * TCP: If the remote end dropped the connection, delay reconnecting.
2378  *
2379  * UDP socket connects are synchronous, but we use a work queue anyway
2380  * to guarantee that even unprivileged user processes can set up a
2381  * socket on a privileged port.
2382  *
2383  * If a UDP socket connect fails, the delay behavior here prevents
2384  * retry floods (hard mounts).
2385  */
2386 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2387 {
2388         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2389         unsigned long delay = 0;
2390
2391         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2392
2393         if (transport->sock != NULL) {
2394                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2395                         "seconds\n", xprt, xprt->reestablish_timeout / HZ);
2396
2397                 delay = xprt_reconnect_delay(xprt);
2398                 xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2399
2400         } else
2401                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2402
2403         queue_delayed_work(xprtiod_workqueue,
2404                         &transport->connect_worker,
2405                         delay);
2406 }
2407
2408 static void xs_wake_disconnect(struct sock_xprt *transport)
2409 {
2410         if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2411                 xs_tcp_force_close(&transport->xprt);
2412 }
2413
2414 static void xs_wake_write(struct sock_xprt *transport)
2415 {
2416         if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2417                 xprt_write_space(&transport->xprt);
2418 }
2419
2420 static void xs_wake_error(struct sock_xprt *transport)
2421 {
2422         int sockerr;
2423
2424         if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2425                 return;
2426         mutex_lock(&transport->recv_mutex);
2427         if (transport->sock == NULL)
2428                 goto out;
2429         if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2430                 goto out;
2431         sockerr = xchg(&transport->xprt_err, 0);
2432         if (sockerr < 0)
2433                 xprt_wake_pending_tasks(&transport->xprt, sockerr);
2434 out:
2435         mutex_unlock(&transport->recv_mutex);
2436 }
2437
2438 static void xs_wake_pending(struct sock_xprt *transport)
2439 {
2440         if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2441                 xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2442 }
2443
2444 static void xs_error_handle(struct work_struct *work)
2445 {
2446         struct sock_xprt *transport = container_of(work,
2447                         struct sock_xprt, error_worker);
2448
2449         xs_wake_disconnect(transport);
2450         xs_wake_write(transport);
2451         xs_wake_error(transport);
2452         xs_wake_pending(transport);
2453 }
2454
2455 /**
2456  * xs_local_print_stats - display AF_LOCAL socket-specific stats
2457  * @xprt: rpc_xprt struct containing statistics
2458  * @seq: output file
2459  *
2460  */
2461 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2462 {
2463         long idle_time = 0;
2464
2465         if (xprt_connected(xprt))
2466                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2467
2468         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2469                         "%llu %llu %lu %llu %llu\n",
2470                         xprt->stat.bind_count,
2471                         xprt->stat.connect_count,
2472                         xprt->stat.connect_time / HZ,
2473                         idle_time,
2474                         xprt->stat.sends,
2475                         xprt->stat.recvs,
2476                         xprt->stat.bad_xids,
2477                         xprt->stat.req_u,
2478                         xprt->stat.bklog_u,
2479                         xprt->stat.max_slots,
2480                         xprt->stat.sending_u,
2481                         xprt->stat.pending_u);
2482 }
2483
2484 /**
2485  * xs_udp_print_stats - display UDP socket-specific stats
2486  * @xprt: rpc_xprt struct containing statistics
2487  * @seq: output file
2488  *
2489  */
2490 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2491 {
2492         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2493
2494         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2495                         "%lu %llu %llu\n",
2496                         transport->srcport,
2497                         xprt->stat.bind_count,
2498                         xprt->stat.sends,
2499                         xprt->stat.recvs,
2500                         xprt->stat.bad_xids,
2501                         xprt->stat.req_u,
2502                         xprt->stat.bklog_u,
2503                         xprt->stat.max_slots,
2504                         xprt->stat.sending_u,
2505                         xprt->stat.pending_u);
2506 }
2507
2508 /**
2509  * xs_tcp_print_stats - display TCP socket-specific stats
2510  * @xprt: rpc_xprt struct containing statistics
2511  * @seq: output file
2512  *
2513  */
2514 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2515 {
2516         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2517         long idle_time = 0;
2518
2519         if (xprt_connected(xprt))
2520                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2521
2522         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2523                         "%llu %llu %lu %llu %llu\n",
2524                         transport->srcport,
2525                         xprt->stat.bind_count,
2526                         xprt->stat.connect_count,
2527                         xprt->stat.connect_time / HZ,
2528                         idle_time,
2529                         xprt->stat.sends,
2530                         xprt->stat.recvs,
2531                         xprt->stat.bad_xids,
2532                         xprt->stat.req_u,
2533                         xprt->stat.bklog_u,
2534                         xprt->stat.max_slots,
2535                         xprt->stat.sending_u,
2536                         xprt->stat.pending_u);
2537 }
2538
2539 /*
2540  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2541  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2542  * to use the server side send routines.
2543  */
2544 static int bc_malloc(struct rpc_task *task)
2545 {
2546         struct rpc_rqst *rqst = task->tk_rqstp;
2547         size_t size = rqst->rq_callsize;
2548         struct page *page;
2549         struct rpc_buffer *buf;
2550
2551         if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2552                 WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2553                           size);
2554                 return -EINVAL;
2555         }
2556
2557         page = alloc_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
2558         if (!page)
2559                 return -ENOMEM;
2560
2561         buf = page_address(page);
2562         buf->len = PAGE_SIZE;
2563
2564         rqst->rq_buffer = buf->data;
2565         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2566         return 0;
2567 }
2568
2569 /*
2570  * Free the space allocated in the bc_alloc routine
2571  */
2572 static void bc_free(struct rpc_task *task)
2573 {
2574         void *buffer = task->tk_rqstp->rq_buffer;
2575         struct rpc_buffer *buf;
2576
2577         buf = container_of(buffer, struct rpc_buffer, data);
2578         free_page((unsigned long)buf);
2579 }
2580
2581 static int bc_sendto(struct rpc_rqst *req)
2582 {
2583         struct xdr_buf *xdr = &req->rq_snd_buf;
2584         struct sock_xprt *transport =
2585                         container_of(req->rq_xprt, struct sock_xprt, xprt);
2586         struct msghdr msg = {
2587                 .msg_flags      = 0,
2588         };
2589         rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2590                                          (u32)xdr->len);
2591         unsigned int sent = 0;
2592         int err;
2593
2594         req->rq_xtime = ktime_get();
2595         err = xdr_alloc_bvec(xdr, rpc_task_gfp_mask());
2596         if (err < 0)
2597                 return err;
2598         err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2599         xdr_free_bvec(xdr);
2600         if (err < 0 || sent != (xdr->len + sizeof(marker)))
2601                 return -EAGAIN;
2602         return sent;
2603 }
2604
2605 /**
2606  * bc_send_request - Send a backchannel Call on a TCP socket
2607  * @req: rpc_rqst containing Call message to be sent
2608  *
2609  * xpt_mutex ensures @rqstp's whole message is written to the socket
2610  * without interruption.
2611  *
2612  * Return values:
2613  *   %0 if the message was sent successfully
2614  *   %ENOTCONN if the message was not sent
2615  */
2616 static int bc_send_request(struct rpc_rqst *req)
2617 {
2618         struct svc_xprt *xprt;
2619         int len;
2620
2621         /*
2622          * Get the server socket associated with this callback xprt
2623          */
2624         xprt = req->rq_xprt->bc_xprt;
2625
2626         /*
2627          * Grab the mutex to serialize data as the connection is shared
2628          * with the fore channel
2629          */
2630         mutex_lock(&xprt->xpt_mutex);
2631         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2632                 len = -ENOTCONN;
2633         else
2634                 len = bc_sendto(req);
2635         mutex_unlock(&xprt->xpt_mutex);
2636
2637         if (len > 0)
2638                 len = 0;
2639
2640         return len;
2641 }
2642
2643 /*
2644  * The close routine. Since this is client initiated, we do nothing
2645  */
2646
2647 static void bc_close(struct rpc_xprt *xprt)
2648 {
2649         xprt_disconnect_done(xprt);
2650 }
2651
2652 /*
2653  * The xprt destroy routine. Again, because this connection is client
2654  * initiated, we do nothing
2655  */
2656
2657 static void bc_destroy(struct rpc_xprt *xprt)
2658 {
2659         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2660
2661         xs_xprt_free(xprt);
2662         module_put(THIS_MODULE);
2663 }
2664
2665 static const struct rpc_xprt_ops xs_local_ops = {
2666         .reserve_xprt           = xprt_reserve_xprt,
2667         .release_xprt           = xprt_release_xprt,
2668         .alloc_slot             = xprt_alloc_slot,
2669         .free_slot              = xprt_free_slot,
2670         .rpcbind                = xs_local_rpcbind,
2671         .set_port               = xs_local_set_port,
2672         .connect                = xs_local_connect,
2673         .buf_alloc              = rpc_malloc,
2674         .buf_free               = rpc_free,
2675         .prepare_request        = xs_stream_prepare_request,
2676         .send_request           = xs_local_send_request,
2677         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2678         .close                  = xs_close,
2679         .destroy                = xs_destroy,
2680         .print_stats            = xs_local_print_stats,
2681         .enable_swap            = xs_enable_swap,
2682         .disable_swap           = xs_disable_swap,
2683 };
2684
2685 static const struct rpc_xprt_ops xs_udp_ops = {
2686         .set_buffer_size        = xs_udp_set_buffer_size,
2687         .reserve_xprt           = xprt_reserve_xprt_cong,
2688         .release_xprt           = xprt_release_xprt_cong,
2689         .alloc_slot             = xprt_alloc_slot,
2690         .free_slot              = xprt_free_slot,
2691         .rpcbind                = rpcb_getport_async,
2692         .set_port               = xs_set_port,
2693         .connect                = xs_connect,
2694         .get_srcaddr            = xs_sock_srcaddr,
2695         .get_srcport            = xs_sock_srcport,
2696         .buf_alloc              = rpc_malloc,
2697         .buf_free               = rpc_free,
2698         .send_request           = xs_udp_send_request,
2699         .wait_for_reply_request = xprt_wait_for_reply_request_rtt,
2700         .timer                  = xs_udp_timer,
2701         .release_request        = xprt_release_rqst_cong,
2702         .close                  = xs_close,
2703         .destroy                = xs_destroy,
2704         .print_stats            = xs_udp_print_stats,
2705         .enable_swap            = xs_enable_swap,
2706         .disable_swap           = xs_disable_swap,
2707         .inject_disconnect      = xs_inject_disconnect,
2708 };
2709
2710 static const struct rpc_xprt_ops xs_tcp_ops = {
2711         .reserve_xprt           = xprt_reserve_xprt,
2712         .release_xprt           = xprt_release_xprt,
2713         .alloc_slot             = xprt_alloc_slot,
2714         .free_slot              = xprt_free_slot,
2715         .rpcbind                = rpcb_getport_async,
2716         .set_port               = xs_set_port,
2717         .connect                = xs_connect,
2718         .get_srcaddr            = xs_sock_srcaddr,
2719         .get_srcport            = xs_sock_srcport,
2720         .buf_alloc              = rpc_malloc,
2721         .buf_free               = rpc_free,
2722         .prepare_request        = xs_stream_prepare_request,
2723         .send_request           = xs_tcp_send_request,
2724         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2725         .close                  = xs_tcp_shutdown,
2726         .destroy                = xs_destroy,
2727         .set_connect_timeout    = xs_tcp_set_connect_timeout,
2728         .print_stats            = xs_tcp_print_stats,
2729         .enable_swap            = xs_enable_swap,
2730         .disable_swap           = xs_disable_swap,
2731         .inject_disconnect      = xs_inject_disconnect,
2732 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2733         .bc_setup               = xprt_setup_bc,
2734         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2735         .bc_num_slots           = xprt_bc_max_slots,
2736         .bc_free_rqst           = xprt_free_bc_rqst,
2737         .bc_destroy             = xprt_destroy_bc,
2738 #endif
2739 };
2740
2741 /*
2742  * The rpc_xprt_ops for the server backchannel
2743  */
2744
2745 static const struct rpc_xprt_ops bc_tcp_ops = {
2746         .reserve_xprt           = xprt_reserve_xprt,
2747         .release_xprt           = xprt_release_xprt,
2748         .alloc_slot             = xprt_alloc_slot,
2749         .free_slot              = xprt_free_slot,
2750         .buf_alloc              = bc_malloc,
2751         .buf_free               = bc_free,
2752         .send_request           = bc_send_request,
2753         .wait_for_reply_request = xprt_wait_for_reply_request_def,
2754         .close                  = bc_close,
2755         .destroy                = bc_destroy,
2756         .print_stats            = xs_tcp_print_stats,
2757         .enable_swap            = xs_enable_swap,
2758         .disable_swap           = xs_disable_swap,
2759         .inject_disconnect      = xs_inject_disconnect,
2760 };
2761
2762 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2763 {
2764         static const struct sockaddr_in sin = {
2765                 .sin_family             = AF_INET,
2766                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2767         };
2768         static const struct sockaddr_in6 sin6 = {
2769                 .sin6_family            = AF_INET6,
2770                 .sin6_addr              = IN6ADDR_ANY_INIT,
2771         };
2772
2773         switch (family) {
2774         case AF_LOCAL:
2775                 break;
2776         case AF_INET:
2777                 memcpy(sap, &sin, sizeof(sin));
2778                 break;
2779         case AF_INET6:
2780                 memcpy(sap, &sin6, sizeof(sin6));
2781                 break;
2782         default:
2783                 dprintk("RPC:       %s: Bad address family\n", __func__);
2784                 return -EAFNOSUPPORT;
2785         }
2786         return 0;
2787 }
2788
2789 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2790                                       unsigned int slot_table_size,
2791                                       unsigned int max_slot_table_size)
2792 {
2793         struct rpc_xprt *xprt;
2794         struct sock_xprt *new;
2795
2796         if (args->addrlen > sizeof(xprt->addr)) {
2797                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2798                 return ERR_PTR(-EBADF);
2799         }
2800
2801         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2802                         max_slot_table_size);
2803         if (xprt == NULL) {
2804                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2805                                 "rpc_xprt\n");
2806                 return ERR_PTR(-ENOMEM);
2807         }
2808
2809         new = container_of(xprt, struct sock_xprt, xprt);
2810         mutex_init(&new->recv_mutex);
2811         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2812         xprt->addrlen = args->addrlen;
2813         if (args->srcaddr)
2814                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2815         else {
2816                 int err;
2817                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2818                                         (struct sockaddr *)&new->srcaddr);
2819                 if (err != 0) {
2820                         xprt_free(xprt);
2821                         return ERR_PTR(err);
2822                 }
2823         }
2824
2825         return xprt;
2826 }
2827
2828 static const struct rpc_timeout xs_local_default_timeout = {
2829         .to_initval = 10 * HZ,
2830         .to_maxval = 10 * HZ,
2831         .to_retries = 2,
2832 };
2833
2834 /**
2835  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2836  * @args: rpc transport creation arguments
2837  *
2838  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2839  */
2840 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2841 {
2842         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2843         struct sock_xprt *transport;
2844         struct rpc_xprt *xprt;
2845         struct rpc_xprt *ret;
2846
2847         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2848                         xprt_max_tcp_slot_table_entries);
2849         if (IS_ERR(xprt))
2850                 return xprt;
2851         transport = container_of(xprt, struct sock_xprt, xprt);
2852
2853         xprt->prot = 0;
2854         xprt->xprt_class = &xs_local_transport;
2855         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2856
2857         xprt->bind_timeout = XS_BIND_TO;
2858         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2859         xprt->idle_timeout = XS_IDLE_DISC_TO;
2860
2861         xprt->ops = &xs_local_ops;
2862         xprt->timeout = &xs_local_default_timeout;
2863
2864         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2865         INIT_WORK(&transport->error_worker, xs_error_handle);
2866         INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2867
2868         switch (sun->sun_family) {
2869         case AF_LOCAL:
2870                 if (sun->sun_path[0] != '/') {
2871                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2872                                         sun->sun_path);
2873                         ret = ERR_PTR(-EINVAL);
2874                         goto out_err;
2875                 }
2876                 xprt_set_bound(xprt);
2877                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2878                 break;
2879         default:
2880                 ret = ERR_PTR(-EAFNOSUPPORT);
2881                 goto out_err;
2882         }
2883
2884         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2885                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2886
2887         if (try_module_get(THIS_MODULE))
2888                 return xprt;
2889         ret = ERR_PTR(-EINVAL);
2890 out_err:
2891         xs_xprt_free(xprt);
2892         return ret;
2893 }
2894
2895 static const struct rpc_timeout xs_udp_default_timeout = {
2896         .to_initval = 5 * HZ,
2897         .to_maxval = 30 * HZ,
2898         .to_increment = 5 * HZ,
2899         .to_retries = 5,
2900 };
2901
2902 /**
2903  * xs_setup_udp - Set up transport to use a UDP socket
2904  * @args: rpc transport creation arguments
2905  *
2906  */
2907 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2908 {
2909         struct sockaddr *addr = args->dstaddr;
2910         struct rpc_xprt *xprt;
2911         struct sock_xprt *transport;
2912         struct rpc_xprt *ret;
2913
2914         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2915                         xprt_udp_slot_table_entries);
2916         if (IS_ERR(xprt))
2917                 return xprt;
2918         transport = container_of(xprt, struct sock_xprt, xprt);
2919
2920         xprt->prot = IPPROTO_UDP;
2921         xprt->xprt_class = &xs_udp_transport;
2922         /* XXX: header size can vary due to auth type, IPv6, etc. */
2923         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2924
2925         xprt->bind_timeout = XS_BIND_TO;
2926         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2927         xprt->idle_timeout = XS_IDLE_DISC_TO;
2928
2929         xprt->ops = &xs_udp_ops;
2930
2931         xprt->timeout = &xs_udp_default_timeout;
2932
2933         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2934         INIT_WORK(&transport->error_worker, xs_error_handle);
2935         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2936
2937         switch (addr->sa_family) {
2938         case AF_INET:
2939                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2940                         xprt_set_bound(xprt);
2941
2942                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2943                 break;
2944         case AF_INET6:
2945                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2946                         xprt_set_bound(xprt);
2947
2948                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2949                 break;
2950         default:
2951                 ret = ERR_PTR(-EAFNOSUPPORT);
2952                 goto out_err;
2953         }
2954
2955         if (xprt_bound(xprt))
2956                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2957                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2958                                 xprt->address_strings[RPC_DISPLAY_PORT],
2959                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2960         else
2961                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2962                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2963                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2964
2965         if (try_module_get(THIS_MODULE))
2966                 return xprt;
2967         ret = ERR_PTR(-EINVAL);
2968 out_err:
2969         xs_xprt_free(xprt);
2970         return ret;
2971 }
2972
2973 static const struct rpc_timeout xs_tcp_default_timeout = {
2974         .to_initval = 60 * HZ,
2975         .to_maxval = 60 * HZ,
2976         .to_retries = 2,
2977 };
2978
2979 /**
2980  * xs_setup_tcp - Set up transport to use a TCP socket
2981  * @args: rpc transport creation arguments
2982  *
2983  */
2984 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2985 {
2986         struct sockaddr *addr = args->dstaddr;
2987         struct rpc_xprt *xprt;
2988         struct sock_xprt *transport;
2989         struct rpc_xprt *ret;
2990         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2991
2992         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2993                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2994
2995         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2996                         max_slot_table_size);
2997         if (IS_ERR(xprt))
2998                 return xprt;
2999         transport = container_of(xprt, struct sock_xprt, xprt);
3000
3001         xprt->prot = IPPROTO_TCP;
3002         xprt->xprt_class = &xs_tcp_transport;
3003         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3004
3005         xprt->bind_timeout = XS_BIND_TO;
3006         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3007         xprt->idle_timeout = XS_IDLE_DISC_TO;
3008
3009         xprt->ops = &xs_tcp_ops;
3010         xprt->timeout = &xs_tcp_default_timeout;
3011
3012         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3013         xprt->connect_timeout = xprt->timeout->to_initval *
3014                 (xprt->timeout->to_retries + 1);
3015
3016         INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3017         INIT_WORK(&transport->error_worker, xs_error_handle);
3018         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3019
3020         switch (addr->sa_family) {
3021         case AF_INET:
3022                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3023                         xprt_set_bound(xprt);
3024
3025                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3026                 break;
3027         case AF_INET6:
3028                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3029                         xprt_set_bound(xprt);
3030
3031                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3032                 break;
3033         default:
3034                 ret = ERR_PTR(-EAFNOSUPPORT);
3035                 goto out_err;
3036         }
3037
3038         if (xprt_bound(xprt))
3039                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3040                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3041                                 xprt->address_strings[RPC_DISPLAY_PORT],
3042                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3043         else
3044                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3045                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3046                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3047
3048         if (try_module_get(THIS_MODULE))
3049                 return xprt;
3050         ret = ERR_PTR(-EINVAL);
3051 out_err:
3052         xs_xprt_free(xprt);
3053         return ret;
3054 }
3055
3056 /**
3057  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3058  * @args: rpc transport creation arguments
3059  *
3060  */
3061 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3062 {
3063         struct sockaddr *addr = args->dstaddr;
3064         struct rpc_xprt *xprt;
3065         struct sock_xprt *transport;
3066         struct svc_sock *bc_sock;
3067         struct rpc_xprt *ret;
3068
3069         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3070                         xprt_tcp_slot_table_entries);
3071         if (IS_ERR(xprt))
3072                 return xprt;
3073         transport = container_of(xprt, struct sock_xprt, xprt);
3074
3075         xprt->prot = IPPROTO_TCP;
3076         xprt->xprt_class = &xs_bc_tcp_transport;
3077         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3078         xprt->timeout = &xs_tcp_default_timeout;
3079
3080         /* backchannel */
3081         xprt_set_bound(xprt);
3082         xprt->bind_timeout = 0;
3083         xprt->reestablish_timeout = 0;
3084         xprt->idle_timeout = 0;
3085
3086         xprt->ops = &bc_tcp_ops;
3087
3088         switch (addr->sa_family) {
3089         case AF_INET:
3090                 xs_format_peer_addresses(xprt, "tcp",
3091                                          RPCBIND_NETID_TCP);
3092                 break;
3093         case AF_INET6:
3094                 xs_format_peer_addresses(xprt, "tcp",
3095                                    RPCBIND_NETID_TCP6);
3096                 break;
3097         default:
3098                 ret = ERR_PTR(-EAFNOSUPPORT);
3099                 goto out_err;
3100         }
3101
3102         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3103                         xprt->address_strings[RPC_DISPLAY_ADDR],
3104                         xprt->address_strings[RPC_DISPLAY_PORT],
3105                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3106
3107         /*
3108          * Once we've associated a backchannel xprt with a connection,
3109          * we want to keep it around as long as the connection lasts,
3110          * in case we need to start using it for a backchannel again;
3111          * this reference won't be dropped until bc_xprt is destroyed.
3112          */
3113         xprt_get(xprt);
3114         args->bc_xprt->xpt_bc_xprt = xprt;
3115         xprt->bc_xprt = args->bc_xprt;
3116         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3117         transport->sock = bc_sock->sk_sock;
3118         transport->inet = bc_sock->sk_sk;
3119
3120         /*
3121          * Since we don't want connections for the backchannel, we set
3122          * the xprt status to connected
3123          */
3124         xprt_set_connected(xprt);
3125
3126         if (try_module_get(THIS_MODULE))
3127                 return xprt;
3128
3129         args->bc_xprt->xpt_bc_xprt = NULL;
3130         args->bc_xprt->xpt_bc_xps = NULL;
3131         xprt_put(xprt);
3132         ret = ERR_PTR(-EINVAL);
3133 out_err:
3134         xs_xprt_free(xprt);
3135         return ret;
3136 }
3137
3138 static struct xprt_class        xs_local_transport = {
3139         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3140         .name           = "named UNIX socket",
3141         .owner          = THIS_MODULE,
3142         .ident          = XPRT_TRANSPORT_LOCAL,
3143         .setup          = xs_setup_local,
3144         .netid          = { "" },
3145 };
3146
3147 static struct xprt_class        xs_udp_transport = {
3148         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3149         .name           = "udp",
3150         .owner          = THIS_MODULE,
3151         .ident          = XPRT_TRANSPORT_UDP,
3152         .setup          = xs_setup_udp,
3153         .netid          = { "udp", "udp6", "" },
3154 };
3155
3156 static struct xprt_class        xs_tcp_transport = {
3157         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3158         .name           = "tcp",
3159         .owner          = THIS_MODULE,
3160         .ident          = XPRT_TRANSPORT_TCP,
3161         .setup          = xs_setup_tcp,
3162         .netid          = { "tcp", "tcp6", "" },
3163 };
3164
3165 static struct xprt_class        xs_bc_tcp_transport = {
3166         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3167         .name           = "tcp NFSv4.1 backchannel",
3168         .owner          = THIS_MODULE,
3169         .ident          = XPRT_TRANSPORT_BC_TCP,
3170         .setup          = xs_setup_bc_tcp,
3171         .netid          = { "" },
3172 };
3173
3174 /**
3175  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3176  *
3177  */
3178 int init_socket_xprt(void)
3179 {
3180         if (!sunrpc_table_header)
3181                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3182
3183         xprt_register_transport(&xs_local_transport);
3184         xprt_register_transport(&xs_udp_transport);
3185         xprt_register_transport(&xs_tcp_transport);
3186         xprt_register_transport(&xs_bc_tcp_transport);
3187
3188         return 0;
3189 }
3190
3191 /**
3192  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3193  *
3194  */
3195 void cleanup_socket_xprt(void)
3196 {
3197         if (sunrpc_table_header) {
3198                 unregister_sysctl_table(sunrpc_table_header);
3199                 sunrpc_table_header = NULL;
3200         }
3201
3202         xprt_unregister_transport(&xs_local_transport);
3203         xprt_unregister_transport(&xs_udp_transport);
3204         xprt_unregister_transport(&xs_tcp_transport);
3205         xprt_unregister_transport(&xs_bc_tcp_transport);
3206 }
3207
3208 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3209 {
3210         return param_set_uint_minmax(val, kp,
3211                         RPC_MIN_RESVPORT,
3212                         RPC_MAX_RESVPORT);
3213 }
3214
3215 static const struct kernel_param_ops param_ops_portnr = {
3216         .set = param_set_portnr,
3217         .get = param_get_uint,
3218 };
3219
3220 #define param_check_portnr(name, p) \
3221         __param_check(name, p, unsigned int);
3222
3223 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3224 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3225
3226 static int param_set_slot_table_size(const char *val,
3227                                      const struct kernel_param *kp)
3228 {
3229         return param_set_uint_minmax(val, kp,
3230                         RPC_MIN_SLOT_TABLE,
3231                         RPC_MAX_SLOT_TABLE);
3232 }
3233
3234 static const struct kernel_param_ops param_ops_slot_table_size = {
3235         .set = param_set_slot_table_size,
3236         .get = param_get_uint,
3237 };
3238
3239 #define param_check_slot_table_size(name, p) \
3240         __param_check(name, p, unsigned int);
3241
3242 static int param_set_max_slot_table_size(const char *val,
3243                                      const struct kernel_param *kp)
3244 {
3245         return param_set_uint_minmax(val, kp,
3246                         RPC_MIN_SLOT_TABLE,
3247                         RPC_MAX_SLOT_TABLE_LIMIT);
3248 }
3249
3250 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3251         .set = param_set_max_slot_table_size,
3252         .get = param_get_uint,
3253 };
3254
3255 #define param_check_max_slot_table_size(name, p) \
3256         __param_check(name, p, unsigned int);
3257
3258 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3259                    slot_table_size, 0644);
3260 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3261                    max_slot_table_size, 0644);
3262 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3263                    slot_table_size, 0644);