net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68 #define XS_TCP_LINGER_TO        (15U * HZ)
  69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71 /*
  72  * We can register our own files under /proc/sys/sunrpc by
  73  * calling register_sysctl_table() again.  The files in that
  74  * directory become the union of all files registered there.
  75  *
  76  * We simply need to make sure that we don't collide with
  77  * someone else's file names!
  78  */
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 181
 182 /*
 183  * TCP idle timeout; client drops the transport socket if it is idle
 184  * for this long.  Note that we also timeout UDP sockets to prevent
 185  * holding port numbers when there is no RPC traffic.
 186  */
 187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 188
 189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 190 # undef  RPC_DEBUG_DATA
 191 # define RPCDBG_FACILITY        RPCDBG_TRANS
 192 #endif
 193
 194 #ifdef RPC_DEBUG_DATA
 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196 {
 197         u8 *buf = (u8 *) packet;
 198         int j;
 199
 200         dprintk("RPC:       %s\n", msg);
 201         for (j = 0; j < count && j < 128; j += 4) {
 202                 if (!(j & 31)) {
 203                         if (j)
 204                                 dprintk("\n");
 205                         dprintk("0x%04x ", j);
 206                 }
 207                 dprintk("%02x%02x%02x%02x ",
 208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209         }
 210         dprintk("\n");
 211 }
 212 #else
 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214 {
 215         /* NOP */
 216 }
 217 #endif
 218
 219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 220 {
 221         return (struct rpc_xprt *) sk->sk_user_data;
 222 }
 223
 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 225 {
 226         return (struct sockaddr *) &xprt->addr;
 227 }
 228
 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 230 {
 231         return (struct sockaddr_un *) &xprt->addr;
 232 }
 233
 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr_in *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_in6 *) &xprt->addr;
 242 }
 243
 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 245 {
 246         struct sockaddr *sap = xs_addr(xprt);
 247         struct sockaddr_in6 *sin6;
 248         struct sockaddr_in *sin;
 249         struct sockaddr_un *sun;
 250         char buf[128];
 251
 252         switch (sap->sa_family) {
 253         case AF_LOCAL:
 254                 sun = xs_addr_un(xprt);
 255                 strlcpy(buf, sun->sun_path, sizeof(buf));
 256                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 257                                                 kstrdup(buf, GFP_KERNEL);
 258                 break;
 259         case AF_INET:
 260                 (void)rpc_ntop(sap, buf, sizeof(buf));
 261                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 262                                                 kstrdup(buf, GFP_KERNEL);
 263                 sin = xs_addr_in(xprt);
 264                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 265                 break;
 266         case AF_INET6:
 267                 (void)rpc_ntop(sap, buf, sizeof(buf));
 268                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 269                                                 kstrdup(buf, GFP_KERNEL);
 270                 sin6 = xs_addr_in6(xprt);
 271                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 272                 break;
 273         default:
 274                 BUG();
 275         }
 276
 277         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 278 }
 279
 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 281 {
 282         struct sockaddr *sap = xs_addr(xprt);
 283         char buf[128];
 284
 285         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 286         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 287
 288         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 289         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 290 }
 291
 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 293                                      const char *protocol,
 294                                      const char *netid)
 295 {
 296         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 297         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 298         xs_format_common_peer_addresses(xprt);
 299         xs_format_common_peer_ports(xprt);
 300 }
 301
 302 static void xs_update_peer_port(struct rpc_xprt *xprt)
 303 {
 304         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 305         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 306
 307         xs_format_common_peer_ports(xprt);
 308 }
 309
 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 311 {
 312         unsigned int i;
 313
 314         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 315                 switch (i) {
 316                 case RPC_DISPLAY_PROTO:
 317                 case RPC_DISPLAY_NETID:
 318                         continue;
 319                 default:
 320                         kfree(xprt->address_strings[i]);
 321                 }
 322 }
 323
 324 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 325
 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 327 {
 328         struct msghdr msg = {
 329                 .msg_name       = addr,
 330                 .msg_namelen    = addrlen,
 331                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 332         };
 333         struct kvec iov = {
 334                 .iov_base       = vec->iov_base + base,
 335                 .iov_len        = vec->iov_len - base,
 336         };
 337
 338         if (iov.iov_len != 0)
 339                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 340         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 341 }
 342
 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 344 {
 345         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 346                         int offset, size_t size, int flags);
 347         struct page **ppage;
 348         unsigned int remainder;
 349         int err;
 350
 351         remainder = xdr->page_len - base;
 352         base += xdr->page_base;
 353         ppage = xdr->pages + (base >> PAGE_SHIFT);
 354         base &= ~PAGE_MASK;
 355         do_sendpage = sock->ops->sendpage;
 356         if (!zerocopy)
 357                 do_sendpage = sock_no_sendpage;
 358         for(;;) {
 359                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 360                 int flags = XS_SENDMSG_FLAGS;
 361
 362                 remainder -= len;
 363                 if (remainder != 0 || more)
 364                         flags |= MSG_MORE;
 365                 err = do_sendpage(sock, *ppage, base, len, flags);
 366                 if (remainder == 0 || err != len)
 367                         break;
 368                 *sent_p += err;
 369                 ppage++;
 370                 base = 0;
 371         }
 372         if (err > 0) {
 373                 *sent_p += err;
 374                 err = 0;
 375         }
 376         return err;
 377 }
 378
 379 /**
 380  * xs_sendpages - write pages directly to a socket
 381  * @sock: socket to send on
 382  * @addr: UDP only -- address of destination
 383  * @addrlen: UDP only -- length of destination address
 384  * @xdr: buffer containing this request
 385  * @base: starting position in the buffer
 386  * @zerocopy: true if it is safe to use sendpage()
 387  * @sent_p: return the total number of bytes successfully queued for sending
 388  *
 389  */
 390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 391 {
 392         unsigned int remainder = xdr->len - base;
 393         int err = 0;
 394         int sent = 0;
 395
 396         if (unlikely(!sock))
 397                 return -ENOTSOCK;
 398
 399         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 400         if (base != 0) {
 401                 addr = NULL;
 402                 addrlen = 0;
 403         }
 404
 405         if (base < xdr->head[0].iov_len || addr != NULL) {
 406                 unsigned int len = xdr->head[0].iov_len - base;
 407                 remainder -= len;
 408                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 409                 if (remainder == 0 || err != len)
 410                         goto out;
 411                 *sent_p += err;
 412                 base = 0;
 413         } else
 414                 base -= xdr->head[0].iov_len;
 415
 416         if (base < xdr->page_len) {
 417                 unsigned int len = xdr->page_len - base;
 418                 remainder -= len;
 419                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 420                 *sent_p += sent;
 421                 if (remainder == 0 || sent != len)
 422                         goto out;
 423                 base = 0;
 424         } else
 425                 base -= xdr->page_len;
 426
 427         if (base >= xdr->tail[0].iov_len)
 428                 return 0;
 429         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 430 out:
 431         if (err > 0) {
 432                 *sent_p += err;
 433                 err = 0;
 434         }
 435         return err;
 436 }
 437
 438 static void xs_nospace_callback(struct rpc_task *task)
 439 {
 440         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 441
 442         transport->inet->sk_write_pending--;
 443         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 444 }
 445
 446 /**
 447  * xs_nospace - place task on wait queue if transmit was incomplete
 448  * @task: task to put to sleep
 449  *
 450  */
 451 static int xs_nospace(struct rpc_task *task)
 452 {
 453         struct rpc_rqst *req = task->tk_rqstp;
 454         struct rpc_xprt *xprt = req->rq_xprt;
 455         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 456         struct sock *sk = transport->inet;
 457         int ret = -EAGAIN;
 458
 459         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 460                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 461                         req->rq_slen);
 462
 463         /* Protect against races with write_space */
 464         spin_lock_bh(&xprt->transport_lock);
 465
 466         /* Don't race with disconnect */
 467         if (xprt_connected(xprt)) {
 468                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 469                         /*
 470                          * Notify TCP that we're limited by the application
 471                          * window size
 472                          */
 473                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 474                         sk->sk_write_pending++;
 475                         /* ...and wait for more buffer space */
 476                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 477                 }
 478         } else {
 479                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 480                 ret = -ENOTCONN;
 481         }
 482
 483         spin_unlock_bh(&xprt->transport_lock);
 484
 485         /* Race breaker in case memory is freed before above code is called */
 486         sk->sk_write_space(sk);
 487         return ret;
 488 }
 489
 490 /*
 491  * Construct a stream transport record marker in @buf.
 492  */
 493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 494 {
 495         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 496         rpc_fraghdr *base = buf->head[0].iov_base;
 497         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 498 }
 499
 500 /**
 501  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 502  * @task: RPC task that manages the state of an RPC request
 503  *
 504  * Return values:
 505  *        0:    The request has been sent
 506  *   EAGAIN:    The socket was blocked, please call again later to
 507  *              complete the request
 508  * ENOTCONN:    Caller needs to invoke connect logic then call again
 509  *    other:    Some other error occured, the request was not sent
 510  */
 511 static int xs_local_send_request(struct rpc_task *task)
 512 {
 513         struct rpc_rqst *req = task->tk_rqstp;
 514         struct rpc_xprt *xprt = req->rq_xprt;
 515         struct sock_xprt *transport =
 516                                 container_of(xprt, struct sock_xprt, xprt);
 517         struct xdr_buf *xdr = &req->rq_snd_buf;
 518         int status;
 519         int sent = 0;
 520
 521         xs_encode_stream_record_marker(&req->rq_snd_buf);
 522
 523         xs_pktdump("packet data:",
 524                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 525
 526         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 527                               true, &sent);
 528         dprintk("RPC:       %s(%u) = %d\n",
 529                         __func__, xdr->len - req->rq_bytes_sent, status);
 530
 531         if (status == -EAGAIN && sock_writeable(transport->inet))
 532                 status = -ENOBUFS;
 533
 534         if (likely(sent > 0) || status == 0) {
 535                 req->rq_bytes_sent += sent;
 536                 req->rq_xmit_bytes_sent += sent;
 537                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 538                         req->rq_bytes_sent = 0;
 539                         return 0;
 540                 }
 541                 status = -EAGAIN;
 542         }
 543
 544         switch (status) {
 545         case -ENOBUFS:
 546                 break;
 547         case -EAGAIN:
 548                 status = xs_nospace(task);
 549                 break;
 550         default:
 551                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 552                         -status);
 553         case -EPIPE:
 554                 xs_close(xprt);
 555                 status = -ENOTCONN;
 556         }
 557
 558         return status;
 559 }
 560
 561 /**
 562  * xs_udp_send_request - write an RPC request to a UDP socket
 563  * @task: address of RPC task that manages the state of an RPC request
 564  *
 565  * Return values:
 566  *        0:    The request has been sent
 567  *   EAGAIN:    The socket was blocked, please call again later to
 568  *              complete the request
 569  * ENOTCONN:    Caller needs to invoke connect logic then call again
 570  *    other:    Some other error occurred, the request was not sent
 571  */
 572 static int xs_udp_send_request(struct rpc_task *task)
 573 {
 574         struct rpc_rqst *req = task->tk_rqstp;
 575         struct rpc_xprt *xprt = req->rq_xprt;
 576         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 577         struct xdr_buf *xdr = &req->rq_snd_buf;
 578         int sent = 0;
 579         int status;
 580
 581         xs_pktdump("packet data:",
 582                                 req->rq_svec->iov_base,
 583                                 req->rq_svec->iov_len);
 584
 585         if (!xprt_bound(xprt))
 586                 return -ENOTCONN;
 587         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 588                               xdr, req->rq_bytes_sent, true, &sent);
 589
 590         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 591                         xdr->len - req->rq_bytes_sent, status);
 592
 593         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 594         if (status == -EPERM)
 595                 goto process_status;
 596
 597         if (status == -EAGAIN && sock_writeable(transport->inet))
 598                 status = -ENOBUFS;
 599
 600         if (sent > 0 || status == 0) {
 601                 req->rq_xmit_bytes_sent += sent;
 602                 if (sent >= req->rq_slen)
 603                         return 0;
 604                 /* Still some bytes left; set up for a retry later. */
 605                 status = -EAGAIN;
 606         }
 607
 608 process_status:
 609         switch (status) {
 610         case -ENOTSOCK:
 611                 status = -ENOTCONN;
 612                 /* Should we call xs_close() here? */
 613                 break;
 614         case -EAGAIN:
 615                 status = xs_nospace(task);
 616                 break;
 617         default:
 618                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 619                         -status);
 620         case -ENETUNREACH:
 621         case -ENOBUFS:
 622         case -EPIPE:
 623         case -ECONNREFUSED:
 624         case -EPERM:
 625                 /* When the server has died, an ICMP port unreachable message
 626                  * prompts ECONNREFUSED. */
 627                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 628         }
 629
 630         return status;
 631 }
 632
 633 /**
 634  * xs_tcp_send_request - write an RPC request to a TCP socket
 635  * @task: address of RPC task that manages the state of an RPC request
 636  *
 637  * Return values:
 638  *        0:    The request has been sent
 639  *   EAGAIN:    The socket was blocked, please call again later to
 640  *              complete the request
 641  * ENOTCONN:    Caller needs to invoke connect logic then call again
 642  *    other:    Some other error occurred, the request was not sent
 643  *
 644  * XXX: In the case of soft timeouts, should we eventually give up
 645  *      if sendmsg is not able to make progress?
 646  */
 647 static int xs_tcp_send_request(struct rpc_task *task)
 648 {
 649         struct rpc_rqst *req = task->tk_rqstp;
 650         struct rpc_xprt *xprt = req->rq_xprt;
 651         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 652         struct xdr_buf *xdr = &req->rq_snd_buf;
 653         bool zerocopy = true;
 654         int status;
 655         int sent;
 656
 657         xs_encode_stream_record_marker(&req->rq_snd_buf);
 658
 659         xs_pktdump("packet data:",
 660                                 req->rq_svec->iov_base,
 661                                 req->rq_svec->iov_len);
 662         /* Don't use zero copy if this is a resend. If the RPC call
 663          * completes while the socket holds a reference to the pages,
 664          * then we may end up resending corrupted data.
 665          */
 666         if (task->tk_flags & RPC_TASK_SENT)
 667                 zerocopy = false;
 668
 669         /* Continue transmitting the packet/record. We must be careful
 670          * to cope with writespace callbacks arriving _after_ we have
 671          * called sendmsg(). */
 672         while (1) {
 673                 sent = 0;
 674                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 675                                       req->rq_bytes_sent, zerocopy, &sent);
 676
 677                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 678                                 xdr->len - req->rq_bytes_sent, status);
 679
 680                 /* If we've sent the entire packet, immediately
 681                  * reset the count of bytes sent. */
 682                 req->rq_bytes_sent += sent;
 683                 req->rq_xmit_bytes_sent += sent;
 684                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 685                         req->rq_bytes_sent = 0;
 686                         return 0;
 687                 }
 688
 689                 if (status < 0)
 690                         break;
 691                 if (sent == 0) {
 692                         status = -EAGAIN;
 693                         break;
 694                 }
 695         }
 696         if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
 697                 status = -ENOBUFS;
 698
 699         switch (status) {
 700         case -ENOTSOCK:
 701                 status = -ENOTCONN;
 702                 /* Should we call xs_close() here? */
 703                 break;
 704         case -EAGAIN:
 705                 status = xs_nospace(task);
 706                 break;
 707         default:
 708                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 709                         -status);
 710         case -ECONNRESET:
 711         case -ECONNREFUSED:
 712         case -ENOTCONN:
 713         case -EADDRINUSE:
 714         case -ENOBUFS:
 715         case -EPIPE:
 716                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 717         }
 718
 719         return status;
 720 }
 721
 722 /**
 723  * xs_tcp_release_xprt - clean up after a tcp transmission
 724  * @xprt: transport
 725  * @task: rpc task
 726  *
 727  * This cleans up if an error causes us to abort the transmission of a request.
 728  * In this case, the socket may need to be reset in order to avoid confusing
 729  * the server.
 730  */
 731 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 732 {
 733         struct rpc_rqst *req;
 734
 735         if (task != xprt->snd_task)
 736                 return;
 737         if (task == NULL)
 738                 goto out_release;
 739         req = task->tk_rqstp;
 740         if (req == NULL)
 741                 goto out_release;
 742         if (req->rq_bytes_sent == 0)
 743                 goto out_release;
 744         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 745                 goto out_release;
 746         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 747 out_release:
 748         xprt_release_xprt(xprt, task);
 749 }
 750
 751 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 752 {
 753         transport->old_data_ready = sk->sk_data_ready;
 754         transport->old_state_change = sk->sk_state_change;
 755         transport->old_write_space = sk->sk_write_space;
 756         transport->old_error_report = sk->sk_error_report;
 757 }
 758
 759 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 760 {
 761         sk->sk_data_ready = transport->old_data_ready;
 762         sk->sk_state_change = transport->old_state_change;
 763         sk->sk_write_space = transport->old_write_space;
 764         sk->sk_error_report = transport->old_error_report;
 765 }
 766
 767 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 768 {
 769         smp_mb__before_atomic();
 770         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 771         clear_bit(XPRT_CLOSING, &xprt->state);
 772         smp_mb__after_atomic();
 773 }
 774
 775 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 776 {
 777         xs_sock_reset_connection_flags(xprt);
 778         /* Mark transport as closed and wake up all pending tasks */
 779         xprt_disconnect_done(xprt);
 780 }
 781
 782 /**
 783  * xs_error_report - callback to handle TCP socket state errors
 784  * @sk: socket
 785  *
 786  * Note: we don't call sock_error() since there may be a rpc_task
 787  * using the socket, and so we don't want to clear sk->sk_err.
 788  */
 789 static void xs_error_report(struct sock *sk)
 790 {
 791         struct rpc_xprt *xprt;
 792         int err;
 793
 794         read_lock_bh(&sk->sk_callback_lock);
 795         if (!(xprt = xprt_from_sock(sk)))
 796                 goto out;
 797
 798         err = -sk->sk_err;
 799         if (err == 0)
 800                 goto out;
 801         /* Is this a reset event? */
 802         if (sk->sk_state == TCP_CLOSE)
 803                 xs_sock_mark_closed(xprt);
 804         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 805                         xprt, -err);
 806         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 807         xprt_wake_pending_tasks(xprt, err);
 808  out:
 809         read_unlock_bh(&sk->sk_callback_lock);
 810 }
 811
 812 static void xs_reset_transport(struct sock_xprt *transport)
 813 {
 814         struct socket *sock = transport->sock;
 815         struct sock *sk = transport->inet;
 816         struct rpc_xprt *xprt = &transport->xprt;
 817
 818         if (sk == NULL)
 819                 return;
 820
 821         if (atomic_read(&transport->xprt.swapper))
 822                 sk_clear_memalloc(sk);
 823
 824         kernel_sock_shutdown(sock, SHUT_RDWR);
 825
 826         write_lock_bh(&sk->sk_callback_lock);
 827         transport->inet = NULL;
 828         transport->sock = NULL;
 829
 830         sk->sk_user_data = NULL;
 831
 832         xs_restore_old_callbacks(transport, sk);
 833         xprt_clear_connected(xprt);
 834         write_unlock_bh(&sk->sk_callback_lock);
 835         xs_sock_reset_connection_flags(xprt);
 836
 837         trace_rpc_socket_close(xprt, sock);
 838         sock_release(sock);
 839 }
 840
 841 /**
 842  * xs_close - close a socket
 843  * @xprt: transport
 844  *
 845  * This is used when all requests are complete; ie, no DRC state remains
 846  * on the server we want to save.
 847  *
 848  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 849  * xs_reset_transport() zeroing the socket from underneath a writer.
 850  */
 851 static void xs_close(struct rpc_xprt *xprt)
 852 {
 853         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 854
 855         dprintk("RPC:       xs_close xprt %p\n", xprt);
 856
 857         xs_reset_transport(transport);
 858         xprt->reestablish_timeout = 0;
 859
 860         xprt_disconnect_done(xprt);
 861 }
 862
 863 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 864 {
 865         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 866                 xprt);
 867         xprt_disconnect_done(xprt);
 868 }
 869
 870 static void xs_xprt_free(struct rpc_xprt *xprt)
 871 {
 872         xs_free_peer_addresses(xprt);
 873         xprt_free(xprt);
 874 }
 875
 876 /**
 877  * xs_destroy - prepare to shutdown a transport
 878  * @xprt: doomed transport
 879  *
 880  */
 881 static void xs_destroy(struct rpc_xprt *xprt)
 882 {
 883         struct sock_xprt *transport = container_of(xprt,
 884                         struct sock_xprt, xprt);
 885         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 886
 887         cancel_delayed_work_sync(&transport->connect_worker);
 888         xs_close(xprt);
 889         xs_xprt_free(xprt);
 890         module_put(THIS_MODULE);
 891 }
 892
 893 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 894 {
 895         struct xdr_skb_reader desc = {
 896                 .skb            = skb,
 897                 .offset         = sizeof(rpc_fraghdr),
 898                 .count          = skb->len - sizeof(rpc_fraghdr),
 899         };
 900
 901         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 902                 return -1;
 903         if (desc.count)
 904                 return -1;
 905         return 0;
 906 }
 907
 908 /**
 909  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 910  * @sk: socket with data to read
 911  *
 912  * Currently this assumes we can read the whole reply in a single gulp.
 913  */
 914 static void xs_local_data_ready(struct sock *sk)
 915 {
 916         struct rpc_task *task;
 917         struct rpc_xprt *xprt;
 918         struct rpc_rqst *rovr;
 919         struct sk_buff *skb;
 920         int err, repsize, copied;
 921         u32 _xid;
 922         __be32 *xp;
 923
 924         read_lock_bh(&sk->sk_callback_lock);
 925         dprintk("RPC:       %s...\n", __func__);
 926         xprt = xprt_from_sock(sk);
 927         if (xprt == NULL)
 928                 goto out;
 929
 930         skb = skb_recv_datagram(sk, 0, 1, &err);
 931         if (skb == NULL)
 932                 goto out;
 933
 934         repsize = skb->len - sizeof(rpc_fraghdr);
 935         if (repsize < 4) {
 936                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 937                 goto dropit;
 938         }
 939
 940         /* Copy the XID from the skb... */
 941         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 942         if (xp == NULL)
 943                 goto dropit;
 944
 945         /* Look up and lock the request corresponding to the given XID */
 946         spin_lock(&xprt->transport_lock);
 947         rovr = xprt_lookup_rqst(xprt, *xp);
 948         if (!rovr)
 949                 goto out_unlock;
 950         task = rovr->rq_task;
 951
 952         copied = rovr->rq_private_buf.buflen;
 953         if (copied > repsize)
 954                 copied = repsize;
 955
 956         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 957                 dprintk("RPC:       sk_buff copy failed\n");
 958                 goto out_unlock;
 959         }
 960
 961         xprt_complete_rqst(task, copied);
 962
 963  out_unlock:
 964         spin_unlock(&xprt->transport_lock);
 965  dropit:
 966         skb_free_datagram(sk, skb);
 967  out:
 968         read_unlock_bh(&sk->sk_callback_lock);
 969 }
 970
 971 /**
 972  * xs_udp_data_ready - "data ready" callback for UDP sockets
 973  * @sk: socket with data to read
 974  *
 975  */
 976 static void xs_udp_data_ready(struct sock *sk)
 977 {
 978         struct rpc_task *task;
 979         struct rpc_xprt *xprt;
 980         struct rpc_rqst *rovr;
 981         struct sk_buff *skb;
 982         int err, repsize, copied;
 983         u32 _xid;
 984         __be32 *xp;
 985
 986         read_lock_bh(&sk->sk_callback_lock);
 987         dprintk("RPC:       xs_udp_data_ready...\n");
 988         if (!(xprt = xprt_from_sock(sk)))
 989                 goto out;
 990
 991         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 992                 goto out;
 993
 994         repsize = skb->len - sizeof(struct udphdr);
 995         if (repsize < 4) {
 996                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 997                 goto dropit;
 998         }
 999
1000         /* Copy the XID from the skb... */
1001         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1002                                 sizeof(_xid), &_xid);
1003         if (xp == NULL)
1004                 goto dropit;
1005
1006         /* Look up and lock the request corresponding to the given XID */
1007         spin_lock(&xprt->transport_lock);
1008         rovr = xprt_lookup_rqst(xprt, *xp);
1009         if (!rovr)
1010                 goto out_unlock;
1011         task = rovr->rq_task;
1012
1013         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1014                 copied = repsize;
1015
1016         /* Suck it into the iovec, verify checksum if not done by hw. */
1017         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1018                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1019                 goto out_unlock;
1020         }
1021
1022         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1023
1024         xprt_adjust_cwnd(xprt, task, copied);
1025         xprt_complete_rqst(task, copied);
1026
1027  out_unlock:
1028         spin_unlock(&xprt->transport_lock);
1029  dropit:
1030         skb_free_datagram(sk, skb);
1031  out:
1032         read_unlock_bh(&sk->sk_callback_lock);
1033 }
1034
1035 /*
1036  * Helper function to force a TCP close if the server is sending
1037  * junk and/or it has put us in CLOSE_WAIT
1038  */
1039 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1040 {
1041         xprt_force_disconnect(xprt);
1042 }
1043
1044 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1045 {
1046         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1047         size_t len, used;
1048         char *p;
1049
1050         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1051         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1052         used = xdr_skb_read_bits(desc, p, len);
1053         transport->tcp_offset += used;
1054         if (used != len)
1055                 return;
1056
1057         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1058         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1059                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1060         else
1061                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1062         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1063
1064         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1065         transport->tcp_offset = 0;
1066
1067         /* Sanity check of the record length */
1068         if (unlikely(transport->tcp_reclen < 8)) {
1069                 dprintk("RPC:       invalid TCP record fragment length\n");
1070                 xs_tcp_force_close(xprt);
1071                 return;
1072         }
1073         dprintk("RPC:       reading TCP record fragment of length %d\n",
1074                         transport->tcp_reclen);
1075 }
1076
1077 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1078 {
1079         if (transport->tcp_offset == transport->tcp_reclen) {
1080                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1081                 transport->tcp_offset = 0;
1082                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1083                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1084                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1085                         transport->tcp_copied = 0;
1086                 }
1087         }
1088 }
1089
1090 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1091 {
1092         size_t len, used;
1093         char *p;
1094
1095         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1096         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1097         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1098         used = xdr_skb_read_bits(desc, p, len);
1099         transport->tcp_offset += used;
1100         if (used != len)
1101                 return;
1102         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1103         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1104         transport->tcp_copied = 4;
1105         dprintk("RPC:       reading %s XID %08x\n",
1106                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1107                                                               : "request with",
1108                         ntohl(transport->tcp_xid));
1109         xs_tcp_check_fraghdr(transport);
1110 }
1111
1112 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1113                                        struct xdr_skb_reader *desc)
1114 {
1115         size_t len, used;
1116         u32 offset;
1117         char *p;
1118
1119         /*
1120          * We want transport->tcp_offset to be 8 at the end of this routine
1121          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1122          * When this function is called for the first time,
1123          * transport->tcp_offset is 4 (after having already read the xid).
1124          */
1125         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1126         len = sizeof(transport->tcp_calldir) - offset;
1127         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1128         p = ((char *) &transport->tcp_calldir) + offset;
1129         used = xdr_skb_read_bits(desc, p, len);
1130         transport->tcp_offset += used;
1131         if (used != len)
1132                 return;
1133         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1134         /*
1135          * We don't yet have the XDR buffer, so we will write the calldir
1136          * out after we get the buffer from the 'struct rpc_rqst'
1137          */
1138         switch (ntohl(transport->tcp_calldir)) {
1139         case RPC_REPLY:
1140                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1141                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1142                 transport->tcp_flags |= TCP_RPC_REPLY;
1143                 break;
1144         case RPC_CALL:
1145                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1146                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1147                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1148                 break;
1149         default:
1150                 dprintk("RPC:       invalid request message type\n");
1151                 xs_tcp_force_close(&transport->xprt);
1152         }
1153         xs_tcp_check_fraghdr(transport);
1154 }
1155
1156 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1157                                      struct xdr_skb_reader *desc,
1158                                      struct rpc_rqst *req)
1159 {
1160         struct sock_xprt *transport =
1161                                 container_of(xprt, struct sock_xprt, xprt);
1162         struct xdr_buf *rcvbuf;
1163         size_t len;
1164         ssize_t r;
1165
1166         rcvbuf = &req->rq_private_buf;
1167
1168         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1169                 /*
1170                  * Save the RPC direction in the XDR buffer
1171                  */
1172                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1173                         &transport->tcp_calldir,
1174                         sizeof(transport->tcp_calldir));
1175                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1176                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1177         }
1178
1179         len = desc->count;
1180         if (len > transport->tcp_reclen - transport->tcp_offset) {
1181                 struct xdr_skb_reader my_desc;
1182
1183                 len = transport->tcp_reclen - transport->tcp_offset;
1184                 memcpy(&my_desc, desc, sizeof(my_desc));
1185                 my_desc.count = len;
1186                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1187                                           &my_desc, xdr_skb_read_bits);
1188                 desc->count -= r;
1189                 desc->offset += r;
1190         } else
1191                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1192                                           desc, xdr_skb_read_bits);
1193
1194         if (r > 0) {
1195                 transport->tcp_copied += r;
1196                 transport->tcp_offset += r;
1197         }
1198         if (r != len) {
1199                 /* Error when copying to the receive buffer,
1200                  * usually because we weren't able to allocate
1201                  * additional buffer pages. All we can do now
1202                  * is turn off TCP_RCV_COPY_DATA, so the request
1203                  * will not receive any additional updates,
1204                  * and time out.
1205                  * Any remaining data from this record will
1206                  * be discarded.
1207                  */
1208                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1209                 dprintk("RPC:       XID %08x truncated request\n",
1210                                 ntohl(transport->tcp_xid));
1211                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1212                                 "tcp_offset = %u, tcp_reclen = %u\n",
1213                                 xprt, transport->tcp_copied,
1214                                 transport->tcp_offset, transport->tcp_reclen);
1215                 return;
1216         }
1217
1218         dprintk("RPC:       XID %08x read %Zd bytes\n",
1219                         ntohl(transport->tcp_xid), r);
1220         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1221                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1222                         transport->tcp_offset, transport->tcp_reclen);
1223
1224         if (transport->tcp_copied == req->rq_private_buf.buflen)
1225                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1226         else if (transport->tcp_offset == transport->tcp_reclen) {
1227                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1228                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1229         }
1230 }
1231
1232 /*
1233  * Finds the request corresponding to the RPC xid and invokes the common
1234  * tcp read code to read the data.
1235  */
1236 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1237                                     struct xdr_skb_reader *desc)
1238 {
1239         struct sock_xprt *transport =
1240                                 container_of(xprt, struct sock_xprt, xprt);
1241         struct rpc_rqst *req;
1242
1243         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1244
1245         /* Find and lock the request corresponding to this xid */
1246         spin_lock(&xprt->transport_lock);
1247         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1248         if (!req) {
1249                 dprintk("RPC:       XID %08x request not found!\n",
1250                                 ntohl(transport->tcp_xid));
1251                 spin_unlock(&xprt->transport_lock);
1252                 return -1;
1253         }
1254
1255         xs_tcp_read_common(xprt, desc, req);
1256
1257         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1258                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1259
1260         spin_unlock(&xprt->transport_lock);
1261         return 0;
1262 }
1263
1264 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1265 /*
1266  * Obtains an rpc_rqst previously allocated and invokes the common
1267  * tcp read code to read the data.  The result is placed in the callback
1268  * queue.
1269  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1270  * connection and return -1.
1271  */
1272 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1273                                        struct xdr_skb_reader *desc)
1274 {
1275         struct sock_xprt *transport =
1276                                 container_of(xprt, struct sock_xprt, xprt);
1277         struct rpc_rqst *req;
1278
1279         /* Look up and lock the request corresponding to the given XID */
1280         spin_lock(&xprt->transport_lock);
1281         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1282         if (req == NULL) {
1283                 spin_unlock(&xprt->transport_lock);
1284                 printk(KERN_WARNING "Callback slot table overflowed\n");
1285                 xprt_force_disconnect(xprt);
1286                 return -1;
1287         }
1288
1289         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1290         xs_tcp_read_common(xprt, desc, req);
1291
1292         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1293                 xprt_complete_bc_request(req, transport->tcp_copied);
1294         spin_unlock(&xprt->transport_lock);
1295
1296         return 0;
1297 }
1298
1299 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1300                                         struct xdr_skb_reader *desc)
1301 {
1302         struct sock_xprt *transport =
1303                                 container_of(xprt, struct sock_xprt, xprt);
1304
1305         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1306                 xs_tcp_read_reply(xprt, desc) :
1307                 xs_tcp_read_callback(xprt, desc);
1308 }
1309 #else
1310 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1311                                         struct xdr_skb_reader *desc)
1312 {
1313         return xs_tcp_read_reply(xprt, desc);
1314 }
1315 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1316
1317 /*
1318  * Read data off the transport.  This can be either an RPC_CALL or an
1319  * RPC_REPLY.  Relay the processing to helper functions.
1320  */
1321 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1322                                     struct xdr_skb_reader *desc)
1323 {
1324         struct sock_xprt *transport =
1325                                 container_of(xprt, struct sock_xprt, xprt);
1326
1327         if (_xs_tcp_read_data(xprt, desc) == 0)
1328                 xs_tcp_check_fraghdr(transport);
1329         else {
1330                 /*
1331                  * The transport_lock protects the request handling.
1332                  * There's no need to hold it to update the tcp_flags.
1333                  */
1334                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1335         }
1336 }
1337
1338 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1339 {
1340         size_t len;
1341
1342         len = transport->tcp_reclen - transport->tcp_offset;
1343         if (len > desc->count)
1344                 len = desc->count;
1345         desc->count -= len;
1346         desc->offset += len;
1347         transport->tcp_offset += len;
1348         dprintk("RPC:       discarded %Zu bytes\n", len);
1349         xs_tcp_check_fraghdr(transport);
1350 }
1351
1352 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1353 {
1354         struct rpc_xprt *xprt = rd_desc->arg.data;
1355         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1356         struct xdr_skb_reader desc = {
1357                 .skb    = skb,
1358                 .offset = offset,
1359                 .count  = len,
1360         };
1361
1362         dprintk("RPC:       xs_tcp_data_recv started\n");
1363         do {
1364                 trace_xs_tcp_data_recv(transport);
1365                 /* Read in a new fragment marker if necessary */
1366                 /* Can we ever really expect to get completely empty fragments? */
1367                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1368                         xs_tcp_read_fraghdr(xprt, &desc);
1369                         continue;
1370                 }
1371                 /* Read in the xid if necessary */
1372                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1373                         xs_tcp_read_xid(transport, &desc);
1374                         continue;
1375                 }
1376                 /* Read in the call/reply flag */
1377                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1378                         xs_tcp_read_calldir(transport, &desc);
1379                         continue;
1380                 }
1381                 /* Read in the request data */
1382                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1383                         xs_tcp_read_data(xprt, &desc);
1384                         continue;
1385                 }
1386                 /* Skip over any trailing bytes on short reads */
1387                 xs_tcp_read_discard(transport, &desc);
1388         } while (desc.count);
1389         trace_xs_tcp_data_recv(transport);
1390         dprintk("RPC:       xs_tcp_data_recv done\n");
1391         return len - desc.count;
1392 }
1393
1394 static void xs_tcp_data_receive(struct sock_xprt *transport)
1395 {
1396         struct rpc_xprt *xprt = &transport->xprt;
1397         struct sock *sk;
1398         read_descriptor_t rd_desc = {
1399                 .count = 2*1024*1024,
1400                 .arg.data = xprt,
1401         };
1402         unsigned long total = 0;
1403         int read = 0;
1404
1405         sk = transport->inet;
1406
1407         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1408         for (;;) {
1409                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1410                 if (read <= 0)
1411                         break;
1412                 total += read;
1413                 rd_desc.count = 65536;
1414         }
1415         trace_xs_tcp_data_ready(xprt, read, total);
1416 }
1417
1418 /**
1419  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1420  * @sk: socket with data to read
1421  *
1422  */
1423 static void xs_tcp_data_ready(struct sock *sk)
1424 {
1425         struct sock_xprt *transport;
1426         struct rpc_xprt *xprt;
1427
1428         dprintk("RPC:       xs_tcp_data_ready...\n");
1429
1430         read_lock_bh(&sk->sk_callback_lock);
1431         if (!(xprt = xprt_from_sock(sk)))
1432                 goto out;
1433         transport = container_of(xprt, struct sock_xprt, xprt);
1434
1435         /* Any data means we had a useful conversation, so
1436          * the we don't need to delay the next reconnect
1437          */
1438         if (xprt->reestablish_timeout)
1439                 xprt->reestablish_timeout = 0;
1440
1441         xs_tcp_data_receive(transport);
1442 out:
1443         read_unlock_bh(&sk->sk_callback_lock);
1444 }
1445
1446 /**
1447  * xs_tcp_state_change - callback to handle TCP socket state changes
1448  * @sk: socket whose state has changed
1449  *
1450  */
1451 static void xs_tcp_state_change(struct sock *sk)
1452 {
1453         struct rpc_xprt *xprt;
1454         struct sock_xprt *transport;
1455
1456         read_lock_bh(&sk->sk_callback_lock);
1457         if (!(xprt = xprt_from_sock(sk)))
1458                 goto out;
1459         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1460         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1461                         sk->sk_state, xprt_connected(xprt),
1462                         sock_flag(sk, SOCK_DEAD),
1463                         sock_flag(sk, SOCK_ZAPPED),
1464                         sk->sk_shutdown);
1465
1466         transport = container_of(xprt, struct sock_xprt, xprt);
1467         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1468         switch (sk->sk_state) {
1469         case TCP_ESTABLISHED:
1470                 spin_lock(&xprt->transport_lock);
1471                 if (!xprt_test_and_set_connected(xprt)) {
1472
1473                         /* Reset TCP record info */
1474                         transport->tcp_offset = 0;
1475                         transport->tcp_reclen = 0;
1476                         transport->tcp_copied = 0;
1477                         transport->tcp_flags =
1478                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1479                         xprt->connect_cookie++;
1480                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1481                         xprt_clear_connecting(xprt);
1482
1483                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1484                 }
1485                 spin_unlock(&xprt->transport_lock);
1486                 break;
1487         case TCP_FIN_WAIT1:
1488                 /* The client initiated a shutdown of the socket */
1489                 xprt->connect_cookie++;
1490                 xprt->reestablish_timeout = 0;
1491                 set_bit(XPRT_CLOSING, &xprt->state);
1492                 smp_mb__before_atomic();
1493                 clear_bit(XPRT_CONNECTED, &xprt->state);
1494                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1495                 smp_mb__after_atomic();
1496                 break;
1497         case TCP_CLOSE_WAIT:
1498                 /* The server initiated a shutdown of the socket */
1499                 xprt->connect_cookie++;
1500                 clear_bit(XPRT_CONNECTED, &xprt->state);
1501                 xs_tcp_force_close(xprt);
1502         case TCP_CLOSING:
1503                 /*
1504                  * If the server closed down the connection, make sure that
1505                  * we back off before reconnecting
1506                  */
1507                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1508                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1509                 break;
1510         case TCP_LAST_ACK:
1511                 set_bit(XPRT_CLOSING, &xprt->state);
1512                 smp_mb__before_atomic();
1513                 clear_bit(XPRT_CONNECTED, &xprt->state);
1514                 smp_mb__after_atomic();
1515                 break;
1516         case TCP_CLOSE:
1517                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1518                                         &transport->sock_state))
1519                         xprt_clear_connecting(xprt);
1520                 xs_sock_mark_closed(xprt);
1521         }
1522  out:
1523         read_unlock_bh(&sk->sk_callback_lock);
1524 }
1525
1526 static void xs_write_space(struct sock *sk)
1527 {
1528         struct socket *sock;
1529         struct rpc_xprt *xprt;
1530
1531         if (unlikely(!(sock = sk->sk_socket)))
1532                 return;
1533         clear_bit(SOCK_NOSPACE, &sock->flags);
1534
1535         if (unlikely(!(xprt = xprt_from_sock(sk))))
1536                 return;
1537         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1538                 return;
1539
1540         xprt_write_space(xprt);
1541 }
1542
1543 /**
1544  * xs_udp_write_space - callback invoked when socket buffer space
1545  *                             becomes available
1546  * @sk: socket whose state has changed
1547  *
1548  * Called when more output buffer space is available for this socket.
1549  * We try not to wake our writers until they can make "significant"
1550  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1551  * with a bunch of small requests.
1552  */
1553 static void xs_udp_write_space(struct sock *sk)
1554 {
1555         read_lock_bh(&sk->sk_callback_lock);
1556
1557         /* from net/core/sock.c:sock_def_write_space */
1558         if (sock_writeable(sk))
1559                 xs_write_space(sk);
1560
1561         read_unlock_bh(&sk->sk_callback_lock);
1562 }
1563
1564 /**
1565  * xs_tcp_write_space - callback invoked when socket buffer space
1566  *                             becomes available
1567  * @sk: socket whose state has changed
1568  *
1569  * Called when more output buffer space is available for this socket.
1570  * We try not to wake our writers until they can make "significant"
1571  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1572  * with a bunch of small requests.
1573  */
1574 static void xs_tcp_write_space(struct sock *sk)
1575 {
1576         read_lock_bh(&sk->sk_callback_lock);
1577
1578         /* from net/core/stream.c:sk_stream_write_space */
1579         if (sk_stream_is_writeable(sk))
1580                 xs_write_space(sk);
1581
1582         read_unlock_bh(&sk->sk_callback_lock);
1583 }
1584
1585 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1586 {
1587         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1588         struct sock *sk = transport->inet;
1589
1590         if (transport->rcvsize) {
1591                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1592                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1593         }
1594         if (transport->sndsize) {
1595                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1596                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1597                 sk->sk_write_space(sk);
1598         }
1599 }
1600
1601 /**
1602  * xs_udp_set_buffer_size - set send and receive limits
1603  * @xprt: generic transport
1604  * @sndsize: requested size of send buffer, in bytes
1605  * @rcvsize: requested size of receive buffer, in bytes
1606  *
1607  * Set socket send and receive buffer size limits.
1608  */
1609 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1610 {
1611         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1612
1613         transport->sndsize = 0;
1614         if (sndsize)
1615                 transport->sndsize = sndsize + 1024;
1616         transport->rcvsize = 0;
1617         if (rcvsize)
1618                 transport->rcvsize = rcvsize + 1024;
1619
1620         xs_udp_do_set_buffer_size(xprt);
1621 }
1622
1623 /**
1624  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1625  * @task: task that timed out
1626  *
1627  * Adjust the congestion window after a retransmit timeout has occurred.
1628  */
1629 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1630 {
1631         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1632 }
1633
1634 static unsigned short xs_get_random_port(void)
1635 {
1636         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1637         unsigned short rand = (unsigned short) prandom_u32() % range;
1638         return rand + xprt_min_resvport;
1639 }
1640
1641 /**
1642  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1643  * @sock: socket
1644  *
1645  * Note that this function has to be called on all sockets that share the
1646  * same port, and it must be called before binding.
1647  */
1648 static void xs_sock_set_reuseport(struct socket *sock)
1649 {
1650         int opt = 1;
1651
1652         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1653                         (char *)&opt, sizeof(opt));
1654 }
1655
1656 static unsigned short xs_sock_getport(struct socket *sock)
1657 {
1658         struct sockaddr_storage buf;
1659         int buflen;
1660         unsigned short port = 0;
1661
1662         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1663                 goto out;
1664         switch (buf.ss_family) {
1665         case AF_INET6:
1666                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1667                 break;
1668         case AF_INET:
1669                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1670         }
1671 out:
1672         return port;
1673 }
1674
1675 /**
1676  * xs_set_port - reset the port number in the remote endpoint address
1677  * @xprt: generic transport
1678  * @port: new port number
1679  *
1680  */
1681 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1682 {
1683         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1684
1685         rpc_set_port(xs_addr(xprt), port);
1686         xs_update_peer_port(xprt);
1687 }
1688
1689 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1690 {
1691         if (transport->srcport == 0)
1692                 transport->srcport = xs_sock_getport(sock);
1693 }
1694
1695 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1696 {
1697         unsigned short port = transport->srcport;
1698
1699         if (port == 0 && transport->xprt.resvport)
1700                 port = xs_get_random_port();
1701         return port;
1702 }
1703
1704 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1705 {
1706         if (transport->srcport != 0)
1707                 transport->srcport = 0;
1708         if (!transport->xprt.resvport)
1709                 return 0;
1710         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1711                 return xprt_max_resvport;
1712         return --port;
1713 }
1714 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1715 {
1716         struct sockaddr_storage myaddr;
1717         int err, nloop = 0;
1718         unsigned short port = xs_get_srcport(transport);
1719         unsigned short last;
1720
1721         /*
1722          * If we are asking for any ephemeral port (i.e. port == 0 &&
1723          * transport->xprt.resvport == 0), don't bind.  Let the local
1724          * port selection happen implicitly when the socket is used
1725          * (for example at connect time).
1726          *
1727          * This ensures that we can continue to establish TCP
1728          * connections even when all local ephemeral ports are already
1729          * a part of some TCP connection.  This makes no difference
1730          * for UDP sockets, but also doens't harm them.
1731          *
1732          * If we're asking for any reserved port (i.e. port == 0 &&
1733          * transport->xprt.resvport == 1) xs_get_srcport above will
1734          * ensure that port is non-zero and we will bind as needed.
1735          */
1736         if (port == 0)
1737                 return 0;
1738
1739         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1740         do {
1741                 rpc_set_port((struct sockaddr *)&myaddr, port);
1742                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1743                                 transport->xprt.addrlen);
1744                 if (err == 0) {
1745                         transport->srcport = port;
1746                         break;
1747                 }
1748                 last = port;
1749                 port = xs_next_srcport(transport, port);
1750                 if (port > last)
1751                         nloop++;
1752         } while (err == -EADDRINUSE && nloop != 2);
1753
1754         if (myaddr.ss_family == AF_INET)
1755                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1756                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1757                                 port, err ? "failed" : "ok", err);
1758         else
1759                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1760                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1761                                 port, err ? "failed" : "ok", err);
1762         return err;
1763 }
1764
1765 /*
1766  * We don't support autobind on AF_LOCAL sockets
1767  */
1768 static void xs_local_rpcbind(struct rpc_task *task)
1769 {
1770         rcu_read_lock();
1771         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1772         rcu_read_unlock();
1773 }
1774
1775 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1776 {
1777 }
1778
1779 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1780 static struct lock_class_key xs_key[2];
1781 static struct lock_class_key xs_slock_key[2];
1782
1783 static inline void xs_reclassify_socketu(struct socket *sock)
1784 {
1785         struct sock *sk = sock->sk;
1786
1787         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1788                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1789 }
1790
1791 static inline void xs_reclassify_socket4(struct socket *sock)
1792 {
1793         struct sock *sk = sock->sk;
1794
1795         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1796                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1797 }
1798
1799 static inline void xs_reclassify_socket6(struct socket *sock)
1800 {
1801         struct sock *sk = sock->sk;
1802
1803         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1804                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1805 }
1806
1807 static inline void xs_reclassify_socket(int family, struct socket *sock)
1808 {
1809         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1810         if (sock_owned_by_user(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_socketu(struct socket *sock)
1827 {
1828 }
1829
1830 static inline void xs_reclassify_socket4(struct socket *sock)
1831 {
1832 }
1833
1834 static inline void xs_reclassify_socket6(struct socket *sock)
1835 {
1836 }
1837
1838 static inline void xs_reclassify_socket(int family, struct socket *sock)
1839 {
1840 }
1841 #endif
1842
1843 static void xs_dummy_setup_socket(struct work_struct *work)
1844 {
1845 }
1846
1847 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1848                 struct sock_xprt *transport, int family, int type,
1849                 int protocol, bool reuseport)
1850 {
1851         struct socket *sock;
1852         int err;
1853
1854         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1855         if (err < 0) {
1856                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1857                                 protocol, -err);
1858                 goto out;
1859         }
1860         xs_reclassify_socket(family, sock);
1861
1862         if (reuseport)
1863                 xs_sock_set_reuseport(sock);
1864
1865         err = xs_bind(transport, sock);
1866         if (err) {
1867                 sock_release(sock);
1868                 goto out;
1869         }
1870
1871         return sock;
1872 out:
1873         return ERR_PTR(err);
1874 }
1875
1876 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1877                                       struct socket *sock)
1878 {
1879         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1880                                                                         xprt);
1881
1882         if (!transport->inet) {
1883                 struct sock *sk = sock->sk;
1884
1885                 write_lock_bh(&sk->sk_callback_lock);
1886
1887                 xs_save_old_callbacks(transport, sk);
1888
1889                 sk->sk_user_data = xprt;
1890                 sk->sk_data_ready = xs_local_data_ready;
1891                 sk->sk_write_space = xs_udp_write_space;
1892                 sk->sk_error_report = xs_error_report;
1893                 sk->sk_allocation = GFP_NOIO;
1894
1895                 xprt_clear_connected(xprt);
1896
1897                 /* Reset to new socket */
1898                 transport->sock = sock;
1899                 transport->inet = sk;
1900
1901                 write_unlock_bh(&sk->sk_callback_lock);
1902         }
1903
1904         /* Tell the socket layer to start connecting... */
1905         xprt->stat.connect_count++;
1906         xprt->stat.connect_start = jiffies;
1907         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1908 }
1909
1910 /**
1911  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1912  * @transport: socket transport to connect
1913  */
1914 static int xs_local_setup_socket(struct sock_xprt *transport)
1915 {
1916         struct rpc_xprt *xprt = &transport->xprt;
1917         struct socket *sock;
1918         int status = -EIO;
1919
1920         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1921                                         SOCK_STREAM, 0, &sock, 1);
1922         if (status < 0) {
1923                 dprintk("RPC:       can't create AF_LOCAL "
1924                         "transport socket (%d).\n", -status);
1925                 goto out;
1926         }
1927         xs_reclassify_socketu(sock);
1928
1929         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1930                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1931
1932         status = xs_local_finish_connecting(xprt, sock);
1933         trace_rpc_socket_connect(xprt, sock, status);
1934         switch (status) {
1935         case 0:
1936                 dprintk("RPC:       xprt %p connected to %s\n",
1937                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1938                 xprt_set_connected(xprt);
1939         case -ENOBUFS:
1940                 break;
1941         case -ENOENT:
1942                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1943                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1944                 break;
1945         case -ECONNREFUSED:
1946                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1947                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1948                 break;
1949         default:
1950                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1951                                 __func__, -status,
1952                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1953         }
1954
1955 out:
1956         xprt_clear_connecting(xprt);
1957         xprt_wake_pending_tasks(xprt, status);
1958         return status;
1959 }
1960
1961 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1962 {
1963         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1964         int ret;
1965
1966          if (RPC_IS_ASYNC(task)) {
1967                 /*
1968                  * We want the AF_LOCAL connect to be resolved in the
1969                  * filesystem namespace of the process making the rpc
1970                  * call.  Thus we connect synchronously.
1971                  *
1972                  * If we want to support asynchronous AF_LOCAL calls,
1973                  * we'll need to figure out how to pass a namespace to
1974                  * connect.
1975                  */
1976                 rpc_exit(task, -ENOTCONN);
1977                 return;
1978         }
1979         ret = xs_local_setup_socket(transport);
1980         if (ret && !RPC_IS_SOFTCONN(task))
1981                 msleep_interruptible(15000);
1982 }
1983
1984 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1985 /*
1986  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1987  * know that we have exclusive access to the socket), to guard against
1988  * races with xs_reset_transport.
1989  */
1990 static void xs_set_memalloc(struct rpc_xprt *xprt)
1991 {
1992         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1993                         xprt);
1994
1995         /*
1996          * If there's no sock, then we have nothing to set. The
1997          * reconnecting process will get it for us.
1998          */
1999         if (!transport->inet)
2000                 return;
2001         if (atomic_read(&xprt->swapper))
2002                 sk_set_memalloc(transport->inet);
2003 }
2004
2005 /**
2006  * xs_enable_swap - Tag this transport as being used for swap.
2007  * @xprt: transport to tag
2008  *
2009  * Take a reference to this transport on behalf of the rpc_clnt, and
2010  * optionally mark it for swapping if it wasn't already.
2011  */
2012 static int
2013 xs_enable_swap(struct rpc_xprt *xprt)
2014 {
2015         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2016
2017         if (atomic_inc_return(&xprt->swapper) != 1)
2018                 return 0;
2019         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2020                 return -ERESTARTSYS;
2021         if (xs->inet)
2022                 sk_set_memalloc(xs->inet);
2023         xprt_release_xprt(xprt, NULL);
2024         return 0;
2025 }
2026
2027 /**
2028  * xs_disable_swap - Untag this transport as being used for swap.
2029  * @xprt: transport to tag
2030  *
2031  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2032  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2033  */
2034 static void
2035 xs_disable_swap(struct rpc_xprt *xprt)
2036 {
2037         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2038
2039         if (!atomic_dec_and_test(&xprt->swapper))
2040                 return;
2041         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2042                 return;
2043         if (xs->inet)
2044                 sk_clear_memalloc(xs->inet);
2045         xprt_release_xprt(xprt, NULL);
2046 }
2047 #else
2048 static void xs_set_memalloc(struct rpc_xprt *xprt)
2049 {
2050 }
2051
2052 static int
2053 xs_enable_swap(struct rpc_xprt *xprt)
2054 {
2055         return -EINVAL;
2056 }
2057
2058 static void
2059 xs_disable_swap(struct rpc_xprt *xprt)
2060 {
2061 }
2062 #endif
2063
2064 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2065 {
2066         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2067
2068         if (!transport->inet) {
2069                 struct sock *sk = sock->sk;
2070
2071                 write_lock_bh(&sk->sk_callback_lock);
2072
2073                 xs_save_old_callbacks(transport, sk);
2074
2075                 sk->sk_user_data = xprt;
2076                 sk->sk_data_ready = xs_udp_data_ready;
2077                 sk->sk_write_space = xs_udp_write_space;
2078                 sk->sk_allocation = GFP_NOIO;
2079
2080                 xprt_set_connected(xprt);
2081
2082                 /* Reset to new socket */
2083                 transport->sock = sock;
2084                 transport->inet = sk;
2085
2086                 xs_set_memalloc(xprt);
2087
2088                 write_unlock_bh(&sk->sk_callback_lock);
2089         }
2090         xs_udp_do_set_buffer_size(xprt);
2091 }
2092
2093 static void xs_udp_setup_socket(struct work_struct *work)
2094 {
2095         struct sock_xprt *transport =
2096                 container_of(work, struct sock_xprt, connect_worker.work);
2097         struct rpc_xprt *xprt = &transport->xprt;
2098         struct socket *sock = transport->sock;
2099         int status = -EIO;
2100
2101         sock = xs_create_sock(xprt, transport,
2102                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2103                         IPPROTO_UDP, false);
2104         if (IS_ERR(sock))
2105                 goto out;
2106
2107         dprintk("RPC:       worker connecting xprt %p via %s to "
2108                                 "%s (port %s)\n", xprt,
2109                         xprt->address_strings[RPC_DISPLAY_PROTO],
2110                         xprt->address_strings[RPC_DISPLAY_ADDR],
2111                         xprt->address_strings[RPC_DISPLAY_PORT]);
2112
2113         xs_udp_finish_connecting(xprt, sock);
2114         trace_rpc_socket_connect(xprt, sock, 0);
2115         status = 0;
2116 out:
2117         xprt_unlock_connect(xprt, transport);
2118         xprt_clear_connecting(xprt);
2119         xprt_wake_pending_tasks(xprt, status);
2120 }
2121
2122 /**
2123  * xs_tcp_shutdown - gracefully shut down a TCP socket
2124  * @xprt: transport
2125  *
2126  * Initiates a graceful shutdown of the TCP socket by calling the
2127  * equivalent of shutdown(SHUT_RDWR);
2128  */
2129 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2130 {
2131         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2132         struct socket *sock = transport->sock;
2133
2134         if (sock == NULL)
2135                 return;
2136         if (xprt_connected(xprt)) {
2137                 kernel_sock_shutdown(sock, SHUT_RDWR);
2138                 trace_rpc_socket_shutdown(xprt, sock);
2139         } else
2140                 xs_reset_transport(transport);
2141 }
2142
2143 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2144 {
2145         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2146         int ret = -ENOTCONN;
2147
2148         if (!transport->inet) {
2149                 struct sock *sk = sock->sk;
2150                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2151                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2152                 unsigned int opt_on = 1;
2153                 unsigned int timeo;
2154
2155                 /* TCP Keepalive options */
2156                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2157                                 (char *)&opt_on, sizeof(opt_on));
2158                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2159                                 (char *)&keepidle, sizeof(keepidle));
2160                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2161                                 (char *)&keepidle, sizeof(keepidle));
2162                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2163                                 (char *)&keepcnt, sizeof(keepcnt));
2164
2165                 /* TCP user timeout (see RFC5482) */
2166                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2167                         (xprt->timeout->to_retries + 1);
2168                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2169                                 (char *)&timeo, sizeof(timeo));
2170
2171                 write_lock_bh(&sk->sk_callback_lock);
2172
2173                 xs_save_old_callbacks(transport, sk);
2174
2175                 sk->sk_user_data = xprt;
2176                 sk->sk_data_ready = xs_tcp_data_ready;
2177                 sk->sk_state_change = xs_tcp_state_change;
2178                 sk->sk_write_space = xs_tcp_write_space;
2179                 sk->sk_error_report = xs_error_report;
2180                 sk->sk_allocation = GFP_NOIO;
2181
2182                 /* socket options */
2183                 sock_reset_flag(sk, SOCK_LINGER);
2184                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2185
2186                 xprt_clear_connected(xprt);
2187
2188                 /* Reset to new socket */
2189                 transport->sock = sock;
2190                 transport->inet = sk;
2191
2192                 write_unlock_bh(&sk->sk_callback_lock);
2193         }
2194
2195         if (!xprt_bound(xprt))
2196                 goto out;
2197
2198         xs_set_memalloc(xprt);
2199
2200         /* Tell the socket layer to start connecting... */
2201         xprt->stat.connect_count++;
2202         xprt->stat.connect_start = jiffies;
2203         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2204         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2205         switch (ret) {
2206         case 0:
2207                 xs_set_srcport(transport, sock);
2208         case -EINPROGRESS:
2209                 /* SYN_SENT! */
2210                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2211                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2212         }
2213 out:
2214         return ret;
2215 }
2216
2217 /**
2218  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2219  *
2220  * Invoked by a work queue tasklet.
2221  */
2222 static void xs_tcp_setup_socket(struct work_struct *work)
2223 {
2224         struct sock_xprt *transport =
2225                 container_of(work, struct sock_xprt, connect_worker.work);
2226         struct socket *sock = transport->sock;
2227         struct rpc_xprt *xprt = &transport->xprt;
2228         int status = -EIO;
2229
2230         if (!sock) {
2231                 sock = xs_create_sock(xprt, transport,
2232                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2233                                 IPPROTO_TCP, true);
2234                 if (IS_ERR(sock)) {
2235                         status = PTR_ERR(sock);
2236                         goto out;
2237                 }
2238         }
2239
2240         dprintk("RPC:       worker connecting xprt %p via %s to "
2241                                 "%s (port %s)\n", xprt,
2242                         xprt->address_strings[RPC_DISPLAY_PROTO],
2243                         xprt->address_strings[RPC_DISPLAY_ADDR],
2244                         xprt->address_strings[RPC_DISPLAY_PORT]);
2245
2246         status = xs_tcp_finish_connecting(xprt, sock);
2247         trace_rpc_socket_connect(xprt, sock, status);
2248         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2249                         xprt, -status, xprt_connected(xprt),
2250                         sock->sk->sk_state);
2251         switch (status) {
2252         default:
2253                 printk("%s: connect returned unhandled error %d\n",
2254                         __func__, status);
2255         case -EADDRNOTAVAIL:
2256                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2257                  * and retry
2258                  */
2259                 xs_tcp_force_close(xprt);
2260                 break;
2261         case 0:
2262         case -EINPROGRESS:
2263         case -EALREADY:
2264                 xprt_unlock_connect(xprt, transport);
2265                 return;
2266         case -EINVAL:
2267                 /* Happens, for instance, if the user specified a link
2268                  * local IPv6 address without a scope-id.
2269                  */
2270         case -ECONNREFUSED:
2271         case -ECONNRESET:
2272         case -ENETUNREACH:
2273         case -EADDRINUSE:
2274         case -ENOBUFS:
2275                 /* retry with existing socket, after a delay */
2276                 xs_tcp_force_close(xprt);
2277                 goto out;
2278         }
2279         status = -EAGAIN;
2280 out:
2281         xprt_unlock_connect(xprt, transport);
2282         xprt_clear_connecting(xprt);
2283         xprt_wake_pending_tasks(xprt, status);
2284 }
2285
2286 /**
2287  * xs_connect - connect a socket to a remote endpoint
2288  * @xprt: pointer to transport structure
2289  * @task: address of RPC task that manages state of connect request
2290  *
2291  * TCP: If the remote end dropped the connection, delay reconnecting.
2292  *
2293  * UDP socket connects are synchronous, but we use a work queue anyway
2294  * to guarantee that even unprivileged user processes can set up a
2295  * socket on a privileged port.
2296  *
2297  * If a UDP socket connect fails, the delay behavior here prevents
2298  * retry floods (hard mounts).
2299  */
2300 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2301 {
2302         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2303
2304         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2305
2306         if (transport->sock != NULL) {
2307                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2308                                 "seconds\n",
2309                                 xprt, xprt->reestablish_timeout / HZ);
2310
2311                 /* Start by resetting any existing state */
2312                 xs_reset_transport(transport);
2313
2314                 queue_delayed_work(rpciod_workqueue,
2315                                    &transport->connect_worker,
2316                                    xprt->reestablish_timeout);
2317                 xprt->reestablish_timeout <<= 1;
2318                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2319                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2320                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2321                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2322         } else {
2323                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2324                 queue_delayed_work(rpciod_workqueue,
2325                                    &transport->connect_worker, 0);
2326         }
2327 }
2328
2329 /**
2330  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2331  * @xprt: rpc_xprt struct containing statistics
2332  * @seq: output file
2333  *
2334  */
2335 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2336 {
2337         long idle_time = 0;
2338
2339         if (xprt_connected(xprt))
2340                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2341
2342         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2343                         "%llu %llu %lu %llu %llu\n",
2344                         xprt->stat.bind_count,
2345                         xprt->stat.connect_count,
2346                         xprt->stat.connect_time,
2347                         idle_time,
2348                         xprt->stat.sends,
2349                         xprt->stat.recvs,
2350                         xprt->stat.bad_xids,
2351                         xprt->stat.req_u,
2352                         xprt->stat.bklog_u,
2353                         xprt->stat.max_slots,
2354                         xprt->stat.sending_u,
2355                         xprt->stat.pending_u);
2356 }
2357
2358 /**
2359  * xs_udp_print_stats - display UDP socket-specifc stats
2360  * @xprt: rpc_xprt struct containing statistics
2361  * @seq: output file
2362  *
2363  */
2364 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2365 {
2366         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2367
2368         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2369                         "%lu %llu %llu\n",
2370                         transport->srcport,
2371                         xprt->stat.bind_count,
2372                         xprt->stat.sends,
2373                         xprt->stat.recvs,
2374                         xprt->stat.bad_xids,
2375                         xprt->stat.req_u,
2376                         xprt->stat.bklog_u,
2377                         xprt->stat.max_slots,
2378                         xprt->stat.sending_u,
2379                         xprt->stat.pending_u);
2380 }
2381
2382 /**
2383  * xs_tcp_print_stats - display TCP socket-specifc stats
2384  * @xprt: rpc_xprt struct containing statistics
2385  * @seq: output file
2386  *
2387  */
2388 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2389 {
2390         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2391         long idle_time = 0;
2392
2393         if (xprt_connected(xprt))
2394                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2395
2396         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2397                         "%llu %llu %lu %llu %llu\n",
2398                         transport->srcport,
2399                         xprt->stat.bind_count,
2400                         xprt->stat.connect_count,
2401                         xprt->stat.connect_time,
2402                         idle_time,
2403                         xprt->stat.sends,
2404                         xprt->stat.recvs,
2405                         xprt->stat.bad_xids,
2406                         xprt->stat.req_u,
2407                         xprt->stat.bklog_u,
2408                         xprt->stat.max_slots,
2409                         xprt->stat.sending_u,
2410                         xprt->stat.pending_u);
2411 }
2412
2413 /*
2414  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2415  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2416  * to use the server side send routines.
2417  */
2418 static void *bc_malloc(struct rpc_task *task, size_t size)
2419 {
2420         struct page *page;
2421         struct rpc_buffer *buf;
2422
2423         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2424         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2425                 return NULL;
2426
2427         page = alloc_page(GFP_KERNEL);
2428         if (!page)
2429                 return NULL;
2430
2431         buf = page_address(page);
2432         buf->len = PAGE_SIZE;
2433
2434         return buf->data;
2435 }
2436
2437 /*
2438  * Free the space allocated in the bc_alloc routine
2439  */
2440 static void bc_free(void *buffer)
2441 {
2442         struct rpc_buffer *buf;
2443
2444         if (!buffer)
2445                 return;
2446
2447         buf = container_of(buffer, struct rpc_buffer, data);
2448         free_page((unsigned long)buf);
2449 }
2450
2451 /*
2452  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2453  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2454  */
2455 static int bc_sendto(struct rpc_rqst *req)
2456 {
2457         int len;
2458         struct xdr_buf *xbufp = &req->rq_snd_buf;
2459         struct rpc_xprt *xprt = req->rq_xprt;
2460         struct sock_xprt *transport =
2461                                 container_of(xprt, struct sock_xprt, xprt);
2462         struct socket *sock = transport->sock;
2463         unsigned long headoff;
2464         unsigned long tailoff;
2465
2466         xs_encode_stream_record_marker(xbufp);
2467
2468         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2469         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2470         len = svc_send_common(sock, xbufp,
2471                               virt_to_page(xbufp->head[0].iov_base), headoff,
2472                               xbufp->tail[0].iov_base, tailoff);
2473
2474         if (len != xbufp->len) {
2475                 printk(KERN_NOTICE "Error sending entire callback!\n");
2476                 len = -EAGAIN;
2477         }
2478
2479         return len;
2480 }
2481
2482 /*
2483  * The send routine. Borrows from svc_send
2484  */
2485 static int bc_send_request(struct rpc_task *task)
2486 {
2487         struct rpc_rqst *req = task->tk_rqstp;
2488         struct svc_xprt *xprt;
2489         u32                     len;
2490
2491         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2492         /*
2493          * Get the server socket associated with this callback xprt
2494          */
2495         xprt = req->rq_xprt->bc_xprt;
2496
2497         /*
2498          * Grab the mutex to serialize data as the connection is shared
2499          * with the fore channel
2500          */
2501         if (!mutex_trylock(&xprt->xpt_mutex)) {
2502                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2503                 if (!mutex_trylock(&xprt->xpt_mutex))
2504                         return -EAGAIN;
2505                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2506         }
2507         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2508                 len = -ENOTCONN;
2509         else
2510                 len = bc_sendto(req);
2511         mutex_unlock(&xprt->xpt_mutex);
2512
2513         if (len > 0)
2514                 len = 0;
2515
2516         return len;
2517 }
2518
2519 /*
2520  * The close routine. Since this is client initiated, we do nothing
2521  */
2522
2523 static void bc_close(struct rpc_xprt *xprt)
2524 {
2525 }
2526
2527 /*
2528  * The xprt destroy routine. Again, because this connection is client
2529  * initiated, we do nothing
2530  */
2531
2532 static void bc_destroy(struct rpc_xprt *xprt)
2533 {
2534         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2535
2536         xs_xprt_free(xprt);
2537         module_put(THIS_MODULE);
2538 }
2539
2540 static struct rpc_xprt_ops xs_local_ops = {
2541         .reserve_xprt           = xprt_reserve_xprt,
2542         .release_xprt           = xs_tcp_release_xprt,
2543         .alloc_slot             = xprt_alloc_slot,
2544         .rpcbind                = xs_local_rpcbind,
2545         .set_port               = xs_local_set_port,
2546         .connect                = xs_local_connect,
2547         .buf_alloc              = rpc_malloc,
2548         .buf_free               = rpc_free,
2549         .send_request           = xs_local_send_request,
2550         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2551         .close                  = xs_close,
2552         .destroy                = xs_destroy,
2553         .print_stats            = xs_local_print_stats,
2554         .enable_swap            = xs_enable_swap,
2555         .disable_swap           = xs_disable_swap,
2556 };
2557
2558 static struct rpc_xprt_ops xs_udp_ops = {
2559         .set_buffer_size        = xs_udp_set_buffer_size,
2560         .reserve_xprt           = xprt_reserve_xprt_cong,
2561         .release_xprt           = xprt_release_xprt_cong,
2562         .alloc_slot             = xprt_alloc_slot,
2563         .rpcbind                = rpcb_getport_async,
2564         .set_port               = xs_set_port,
2565         .connect                = xs_connect,
2566         .buf_alloc              = rpc_malloc,
2567         .buf_free               = rpc_free,
2568         .send_request           = xs_udp_send_request,
2569         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2570         .timer                  = xs_udp_timer,
2571         .release_request        = xprt_release_rqst_cong,
2572         .close                  = xs_close,
2573         .destroy                = xs_destroy,
2574         .print_stats            = xs_udp_print_stats,
2575         .enable_swap            = xs_enable_swap,
2576         .disable_swap           = xs_disable_swap,
2577         .inject_disconnect      = xs_inject_disconnect,
2578 };
2579
2580 static struct rpc_xprt_ops xs_tcp_ops = {
2581         .reserve_xprt           = xprt_reserve_xprt,
2582         .release_xprt           = xs_tcp_release_xprt,
2583         .alloc_slot             = xprt_lock_and_alloc_slot,
2584         .rpcbind                = rpcb_getport_async,
2585         .set_port               = xs_set_port,
2586         .connect                = xs_connect,
2587         .buf_alloc              = rpc_malloc,
2588         .buf_free               = rpc_free,
2589         .send_request           = xs_tcp_send_request,
2590         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2591         .close                  = xs_tcp_shutdown,
2592         .destroy                = xs_destroy,
2593         .print_stats            = xs_tcp_print_stats,
2594         .enable_swap            = xs_enable_swap,
2595         .disable_swap           = xs_disable_swap,
2596         .inject_disconnect      = xs_inject_disconnect,
2597 };
2598
2599 /*
2600  * The rpc_xprt_ops for the server backchannel
2601  */
2602
2603 static struct rpc_xprt_ops bc_tcp_ops = {
2604         .reserve_xprt           = xprt_reserve_xprt,
2605         .release_xprt           = xprt_release_xprt,
2606         .alloc_slot             = xprt_alloc_slot,
2607         .buf_alloc              = bc_malloc,
2608         .buf_free               = bc_free,
2609         .send_request           = bc_send_request,
2610         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2611         .close                  = bc_close,
2612         .destroy                = bc_destroy,
2613         .print_stats            = xs_tcp_print_stats,
2614         .enable_swap            = xs_enable_swap,
2615         .disable_swap           = xs_disable_swap,
2616         .inject_disconnect      = xs_inject_disconnect,
2617 };
2618
2619 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2620 {
2621         static const struct sockaddr_in sin = {
2622                 .sin_family             = AF_INET,
2623                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2624         };
2625         static const struct sockaddr_in6 sin6 = {
2626                 .sin6_family            = AF_INET6,
2627                 .sin6_addr              = IN6ADDR_ANY_INIT,
2628         };
2629
2630         switch (family) {
2631         case AF_LOCAL:
2632                 break;
2633         case AF_INET:
2634                 memcpy(sap, &sin, sizeof(sin));
2635                 break;
2636         case AF_INET6:
2637                 memcpy(sap, &sin6, sizeof(sin6));
2638                 break;
2639         default:
2640                 dprintk("RPC:       %s: Bad address family\n", __func__);
2641                 return -EAFNOSUPPORT;
2642         }
2643         return 0;
2644 }
2645
2646 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2647                                       unsigned int slot_table_size,
2648                                       unsigned int max_slot_table_size)
2649 {
2650         struct rpc_xprt *xprt;
2651         struct sock_xprt *new;
2652
2653         if (args->addrlen > sizeof(xprt->addr)) {
2654                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2655                 return ERR_PTR(-EBADF);
2656         }
2657
2658         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2659                         max_slot_table_size);
2660         if (xprt == NULL) {
2661                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2662                                 "rpc_xprt\n");
2663                 return ERR_PTR(-ENOMEM);
2664         }
2665
2666         new = container_of(xprt, struct sock_xprt, xprt);
2667         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2668         xprt->addrlen = args->addrlen;
2669         if (args->srcaddr)
2670                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2671         else {
2672                 int err;
2673                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2674                                         (struct sockaddr *)&new->srcaddr);
2675                 if (err != 0) {
2676                         xprt_free(xprt);
2677                         return ERR_PTR(err);
2678                 }
2679         }
2680
2681         return xprt;
2682 }
2683
2684 static const struct rpc_timeout xs_local_default_timeout = {
2685         .to_initval = 10 * HZ,
2686         .to_maxval = 10 * HZ,
2687         .to_retries = 2,
2688 };
2689
2690 /**
2691  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2692  * @args: rpc transport creation arguments
2693  *
2694  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2695  */
2696 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2697 {
2698         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2699         struct sock_xprt *transport;
2700         struct rpc_xprt *xprt;
2701         struct rpc_xprt *ret;
2702
2703         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2704                         xprt_max_tcp_slot_table_entries);
2705         if (IS_ERR(xprt))
2706                 return xprt;
2707         transport = container_of(xprt, struct sock_xprt, xprt);
2708
2709         xprt->prot = 0;
2710         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2711         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2712
2713         xprt->bind_timeout = XS_BIND_TO;
2714         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2715         xprt->idle_timeout = XS_IDLE_DISC_TO;
2716
2717         xprt->ops = &xs_local_ops;
2718         xprt->timeout = &xs_local_default_timeout;
2719
2720         INIT_DELAYED_WORK(&transport->connect_worker,
2721                         xs_dummy_setup_socket);
2722
2723         switch (sun->sun_family) {
2724         case AF_LOCAL:
2725                 if (sun->sun_path[0] != '/') {
2726                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2727                                         sun->sun_path);
2728                         ret = ERR_PTR(-EINVAL);
2729                         goto out_err;
2730                 }
2731                 xprt_set_bound(xprt);
2732                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2733                 ret = ERR_PTR(xs_local_setup_socket(transport));
2734                 if (ret)
2735                         goto out_err;
2736                 break;
2737         default:
2738                 ret = ERR_PTR(-EAFNOSUPPORT);
2739                 goto out_err;
2740         }
2741
2742         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2743                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2744
2745         if (try_module_get(THIS_MODULE))
2746                 return xprt;
2747         ret = ERR_PTR(-EINVAL);
2748 out_err:
2749         xs_xprt_free(xprt);
2750         return ret;
2751 }
2752
2753 static const struct rpc_timeout xs_udp_default_timeout = {
2754         .to_initval = 5 * HZ,
2755         .to_maxval = 30 * HZ,
2756         .to_increment = 5 * HZ,
2757         .to_retries = 5,
2758 };
2759
2760 /**
2761  * xs_setup_udp - Set up transport to use a UDP socket
2762  * @args: rpc transport creation arguments
2763  *
2764  */
2765 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2766 {
2767         struct sockaddr *addr = args->dstaddr;
2768         struct rpc_xprt *xprt;
2769         struct sock_xprt *transport;
2770         struct rpc_xprt *ret;
2771
2772         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2773                         xprt_udp_slot_table_entries);
2774         if (IS_ERR(xprt))
2775                 return xprt;
2776         transport = container_of(xprt, struct sock_xprt, xprt);
2777
2778         xprt->prot = IPPROTO_UDP;
2779         xprt->tsh_size = 0;
2780         /* XXX: header size can vary due to auth type, IPv6, etc. */
2781         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2782
2783         xprt->bind_timeout = XS_BIND_TO;
2784         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2785         xprt->idle_timeout = XS_IDLE_DISC_TO;
2786
2787         xprt->ops = &xs_udp_ops;
2788
2789         xprt->timeout = &xs_udp_default_timeout;
2790
2791         switch (addr->sa_family) {
2792         case AF_INET:
2793                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2794                         xprt_set_bound(xprt);
2795
2796                 INIT_DELAYED_WORK(&transport->connect_worker,
2797                                         xs_udp_setup_socket);
2798                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2799                 break;
2800         case AF_INET6:
2801                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2802                         xprt_set_bound(xprt);
2803
2804                 INIT_DELAYED_WORK(&transport->connect_worker,
2805                                         xs_udp_setup_socket);
2806                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2807                 break;
2808         default:
2809                 ret = ERR_PTR(-EAFNOSUPPORT);
2810                 goto out_err;
2811         }
2812
2813         if (xprt_bound(xprt))
2814                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2815                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2816                                 xprt->address_strings[RPC_DISPLAY_PORT],
2817                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2818         else
2819                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2820                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2821                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2822
2823         if (try_module_get(THIS_MODULE))
2824                 return xprt;
2825         ret = ERR_PTR(-EINVAL);
2826 out_err:
2827         xs_xprt_free(xprt);
2828         return ret;
2829 }
2830
2831 static const struct rpc_timeout xs_tcp_default_timeout = {
2832         .to_initval = 60 * HZ,
2833         .to_maxval = 60 * HZ,
2834         .to_retries = 2,
2835 };
2836
2837 /**
2838  * xs_setup_tcp - Set up transport to use a TCP socket
2839  * @args: rpc transport creation arguments
2840  *
2841  */
2842 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2843 {
2844         struct sockaddr *addr = args->dstaddr;
2845         struct rpc_xprt *xprt;
2846         struct sock_xprt *transport;
2847         struct rpc_xprt *ret;
2848         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2849
2850         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2851                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2852
2853         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2854                         max_slot_table_size);
2855         if (IS_ERR(xprt))
2856                 return xprt;
2857         transport = container_of(xprt, struct sock_xprt, xprt);
2858
2859         xprt->prot = IPPROTO_TCP;
2860         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2861         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2862
2863         xprt->bind_timeout = XS_BIND_TO;
2864         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2865         xprt->idle_timeout = XS_IDLE_DISC_TO;
2866
2867         xprt->ops = &xs_tcp_ops;
2868         xprt->timeout = &xs_tcp_default_timeout;
2869
2870         switch (addr->sa_family) {
2871         case AF_INET:
2872                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2873                         xprt_set_bound(xprt);
2874
2875                 INIT_DELAYED_WORK(&transport->connect_worker,
2876                                         xs_tcp_setup_socket);
2877                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2878                 break;
2879         case AF_INET6:
2880                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2881                         xprt_set_bound(xprt);
2882
2883                 INIT_DELAYED_WORK(&transport->connect_worker,
2884                                         xs_tcp_setup_socket);
2885                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2886                 break;
2887         default:
2888                 ret = ERR_PTR(-EAFNOSUPPORT);
2889                 goto out_err;
2890         }
2891
2892         if (xprt_bound(xprt))
2893                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2894                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2895                                 xprt->address_strings[RPC_DISPLAY_PORT],
2896                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2897         else
2898                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2899                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2900                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2901
2902         if (try_module_get(THIS_MODULE))
2903                 return xprt;
2904         ret = ERR_PTR(-EINVAL);
2905 out_err:
2906         xs_xprt_free(xprt);
2907         return ret;
2908 }
2909
2910 /**
2911  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2912  * @args: rpc transport creation arguments
2913  *
2914  */
2915 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2916 {
2917         struct sockaddr *addr = args->dstaddr;
2918         struct rpc_xprt *xprt;
2919         struct sock_xprt *transport;
2920         struct svc_sock *bc_sock;
2921         struct rpc_xprt *ret;
2922
2923         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2924                         xprt_tcp_slot_table_entries);
2925         if (IS_ERR(xprt))
2926                 return xprt;
2927         transport = container_of(xprt, struct sock_xprt, xprt);
2928
2929         xprt->prot = IPPROTO_TCP;
2930         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2931         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2932         xprt->timeout = &xs_tcp_default_timeout;
2933
2934         /* backchannel */
2935         xprt_set_bound(xprt);
2936         xprt->bind_timeout = 0;
2937         xprt->reestablish_timeout = 0;
2938         xprt->idle_timeout = 0;
2939
2940         xprt->ops = &bc_tcp_ops;
2941
2942         switch (addr->sa_family) {
2943         case AF_INET:
2944                 xs_format_peer_addresses(xprt, "tcp",
2945                                          RPCBIND_NETID_TCP);
2946                 break;
2947         case AF_INET6:
2948                 xs_format_peer_addresses(xprt, "tcp",
2949                                    RPCBIND_NETID_TCP6);
2950                 break;
2951         default:
2952                 ret = ERR_PTR(-EAFNOSUPPORT);
2953                 goto out_err;
2954         }
2955
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
2961         /*
2962          * Once we've associated a backchannel xprt with a connection,
2963          * we want to keep it around as long as the connection lasts,
2964          * in case we need to start using it for a backchannel again;
2965          * this reference won't be dropped until bc_xprt is destroyed.
2966          */
2967         xprt_get(xprt);
2968         args->bc_xprt->xpt_bc_xprt = xprt;
2969         xprt->bc_xprt = args->bc_xprt;
2970         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2971         transport->sock = bc_sock->sk_sock;
2972         transport->inet = bc_sock->sk_sk;
2973
2974         /*
2975          * Since we don't want connections for the backchannel, we set
2976          * the xprt status to connected
2977          */
2978         xprt_set_connected(xprt);
2979
2980         if (try_module_get(THIS_MODULE))
2981                 return xprt;
2982
2983         args->bc_xprt->xpt_bc_xprt = NULL;
2984         xprt_put(xprt);
2985         ret = ERR_PTR(-EINVAL);
2986 out_err:
2987         xs_xprt_free(xprt);
2988         return ret;
2989 }
2990
2991 static struct xprt_class        xs_local_transport = {
2992         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2993         .name           = "named UNIX socket",
2994         .owner          = THIS_MODULE,
2995         .ident          = XPRT_TRANSPORT_LOCAL,
2996         .setup          = xs_setup_local,
2997 };
2998
2999 static struct xprt_class        xs_udp_transport = {
3000         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3001         .name           = "udp",
3002         .owner          = THIS_MODULE,
3003         .ident          = XPRT_TRANSPORT_UDP,
3004         .setup          = xs_setup_udp,
3005 };
3006
3007 static struct xprt_class        xs_tcp_transport = {
3008         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3009         .name           = "tcp",
3010         .owner          = THIS_MODULE,
3011         .ident          = XPRT_TRANSPORT_TCP,
3012         .setup          = xs_setup_tcp,
3013 };
3014
3015 static struct xprt_class        xs_bc_tcp_transport = {
3016         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3017         .name           = "tcp NFSv4.1 backchannel",
3018         .owner          = THIS_MODULE,
3019         .ident          = XPRT_TRANSPORT_BC_TCP,
3020         .setup          = xs_setup_bc_tcp,
3021 };
3022
3023 /**
3024  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3025  *
3026  */
3027 int init_socket_xprt(void)
3028 {
3029 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3030         if (!sunrpc_table_header)
3031                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3032 #endif
3033
3034         xprt_register_transport(&xs_local_transport);
3035         xprt_register_transport(&xs_udp_transport);
3036         xprt_register_transport(&xs_tcp_transport);
3037         xprt_register_transport(&xs_bc_tcp_transport);
3038
3039         return 0;
3040 }
3041
3042 /**
3043  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3044  *
3045  */
3046 void cleanup_socket_xprt(void)
3047 {
3048 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3049         if (sunrpc_table_header) {
3050                 unregister_sysctl_table(sunrpc_table_header);
3051                 sunrpc_table_header = NULL;
3052         }
3053 #endif
3054
3055         xprt_unregister_transport(&xs_local_transport);
3056         xprt_unregister_transport(&xs_udp_transport);
3057         xprt_unregister_transport(&xs_tcp_transport);
3058         xprt_unregister_transport(&xs_bc_tcp_transport);
3059 }
3060
3061 static int param_set_uint_minmax(const char *val,
3062                 const struct kernel_param *kp,
3063                 unsigned int min, unsigned int max)
3064 {
3065         unsigned int num;
3066         int ret;
3067
3068         if (!val)
3069                 return -EINVAL;
3070         ret = kstrtouint(val, 0, &num);
3071         if (ret == -EINVAL || num < min || num > max)
3072                 return -EINVAL;
3073         *((unsigned int *)kp->arg) = num;
3074         return 0;
3075 }
3076
3077 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3078 {
3079         return param_set_uint_minmax(val, kp,
3080                         RPC_MIN_RESVPORT,
3081                         RPC_MAX_RESVPORT);
3082 }
3083
3084 static const struct kernel_param_ops param_ops_portnr = {
3085         .set = param_set_portnr,
3086         .get = param_get_uint,
3087 };
3088
3089 #define param_check_portnr(name, p) \
3090         __param_check(name, p, unsigned int);
3091
3092 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3093 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3094
3095 static int param_set_slot_table_size(const char *val,
3096                                      const struct kernel_param *kp)
3097 {
3098         return param_set_uint_minmax(val, kp,
3099                         RPC_MIN_SLOT_TABLE,
3100                         RPC_MAX_SLOT_TABLE);
3101 }
3102
3103 static const struct kernel_param_ops param_ops_slot_table_size = {
3104         .set = param_set_slot_table_size,
3105         .get = param_get_uint,
3106 };
3107
3108 #define param_check_slot_table_size(name, p) \
3109         __param_check(name, p, unsigned int);
3110
3111 static int param_set_max_slot_table_size(const char *val,
3112                                      const struct kernel_param *kp)
3113 {
3114         return param_set_uint_minmax(val, kp,
3115                         RPC_MIN_SLOT_TABLE,
3116                         RPC_MAX_SLOT_TABLE_LIMIT);
3117 }
3118
3119 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3120         .set = param_set_max_slot_table_size,
3121         .get = param_get_uint,
3122 };
3123
3124 #define param_check_max_slot_table_size(name, p) \
3125         __param_check(name, p, unsigned int);
3126
3127 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3128                    slot_table_size, 0644);
3129 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3130                    max_slot_table_size, 0644);
3131 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3132                    slot_table_size, 0644);
3133