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         xprt_force_disconnect(xprt);
 781 }
 782
 783 /**
 784  * xs_error_report - callback to handle TCP socket state errors
 785  * @sk: socket
 786  *
 787  * Note: we don't call sock_error() since there may be a rpc_task
 788  * using the socket, and so we don't want to clear sk->sk_err.
 789  */
 790 static void xs_error_report(struct sock *sk)
 791 {
 792         struct rpc_xprt *xprt;
 793         int err;
 794
 795         read_lock_bh(&sk->sk_callback_lock);
 796         if (!(xprt = xprt_from_sock(sk)))
 797                 goto out;
 798
 799         err = -sk->sk_err;
 800         if (err == 0)
 801                 goto out;
 802         /* Is this a reset event? */
 803         if (sk->sk_state == TCP_CLOSE)
 804                 xs_sock_mark_closed(xprt);
 805         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 806                         xprt, -err);
 807         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 808         xprt_wake_pending_tasks(xprt, err);
 809  out:
 810         read_unlock_bh(&sk->sk_callback_lock);
 811 }
 812
 813 static void xs_reset_transport(struct sock_xprt *transport)
 814 {
 815         struct socket *sock = transport->sock;
 816         struct sock *sk = transport->inet;
 817         struct rpc_xprt *xprt = &transport->xprt;
 818
 819         if (sk == NULL)
 820                 return;
 821
 822         if (atomic_read(&transport->xprt.swapper))
 823                 sk_clear_memalloc(sk);
 824
 825         kernel_sock_shutdown(sock, SHUT_RDWR);
 826
 827         write_lock_bh(&sk->sk_callback_lock);
 828         transport->inet = NULL;
 829         transport->sock = NULL;
 830
 831         sk->sk_user_data = NULL;
 832
 833         xs_restore_old_callbacks(transport, sk);
 834         xprt_clear_connected(xprt);
 835         write_unlock_bh(&sk->sk_callback_lock);
 836         xs_sock_reset_connection_flags(xprt);
 837
 838         trace_rpc_socket_close(xprt, sock);
 839         sock_release(sock);
 840 }
 841
 842 /**
 843  * xs_close - close a socket
 844  * @xprt: transport
 845  *
 846  * This is used when all requests are complete; ie, no DRC state remains
 847  * on the server we want to save.
 848  *
 849  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 850  * xs_reset_transport() zeroing the socket from underneath a writer.
 851  */
 852 static void xs_close(struct rpc_xprt *xprt)
 853 {
 854         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 855
 856         dprintk("RPC:       xs_close xprt %p\n", xprt);
 857
 858         xs_reset_transport(transport);
 859         xprt->reestablish_timeout = 0;
 860
 861         xprt_disconnect_done(xprt);
 862 }
 863
 864 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 865 {
 866         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 867                 xprt);
 868         xprt_disconnect_done(xprt);
 869 }
 870
 871 static void xs_xprt_free(struct rpc_xprt *xprt)
 872 {
 873         xs_free_peer_addresses(xprt);
 874         xprt_free(xprt);
 875 }
 876
 877 /**
 878  * xs_destroy - prepare to shutdown a transport
 879  * @xprt: doomed transport
 880  *
 881  */
 882 static void xs_destroy(struct rpc_xprt *xprt)
 883 {
 884         struct sock_xprt *transport = container_of(xprt,
 885                         struct sock_xprt, xprt);
 886         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 887
 888         cancel_delayed_work_sync(&transport->connect_worker);
 889         xs_close(xprt);
 890         xs_xprt_free(xprt);
 891         module_put(THIS_MODULE);
 892 }
 893
 894 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 895 {
 896         struct xdr_skb_reader desc = {
 897                 .skb            = skb,
 898                 .offset         = sizeof(rpc_fraghdr),
 899                 .count          = skb->len - sizeof(rpc_fraghdr),
 900         };
 901
 902         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 903                 return -1;
 904         if (desc.count)
 905                 return -1;
 906         return 0;
 907 }
 908
 909 /**
 910  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 911  * @sk: socket with data to read
 912  *
 913  * Currently this assumes we can read the whole reply in a single gulp.
 914  */
 915 static void xs_local_data_ready(struct sock *sk)
 916 {
 917         struct rpc_task *task;
 918         struct rpc_xprt *xprt;
 919         struct rpc_rqst *rovr;
 920         struct sk_buff *skb;
 921         int err, repsize, copied;
 922         u32 _xid;
 923         __be32 *xp;
 924
 925         read_lock_bh(&sk->sk_callback_lock);
 926         dprintk("RPC:       %s...\n", __func__);
 927         xprt = xprt_from_sock(sk);
 928         if (xprt == NULL)
 929                 goto out;
 930
 931         skb = skb_recv_datagram(sk, 0, 1, &err);
 932         if (skb == NULL)
 933                 goto out;
 934
 935         repsize = skb->len - sizeof(rpc_fraghdr);
 936         if (repsize < 4) {
 937                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 938                 goto dropit;
 939         }
 940
 941         /* Copy the XID from the skb... */
 942         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 943         if (xp == NULL)
 944                 goto dropit;
 945
 946         /* Look up and lock the request corresponding to the given XID */
 947         spin_lock(&xprt->transport_lock);
 948         rovr = xprt_lookup_rqst(xprt, *xp);
 949         if (!rovr)
 950                 goto out_unlock;
 951         task = rovr->rq_task;
 952
 953         copied = rovr->rq_private_buf.buflen;
 954         if (copied > repsize)
 955                 copied = repsize;
 956
 957         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 958                 dprintk("RPC:       sk_buff copy failed\n");
 959                 goto out_unlock;
 960         }
 961
 962         xprt_complete_rqst(task, copied);
 963
 964  out_unlock:
 965         spin_unlock(&xprt->transport_lock);
 966  dropit:
 967         skb_free_datagram(sk, skb);
 968  out:
 969         read_unlock_bh(&sk->sk_callback_lock);
 970 }
 971
 972 /**
 973  * xs_udp_data_ready - "data ready" callback for UDP sockets
 974  * @sk: socket with data to read
 975  *
 976  */
 977 static void xs_udp_data_ready(struct sock *sk)
 978 {
 979         struct rpc_task *task;
 980         struct rpc_xprt *xprt;
 981         struct rpc_rqst *rovr;
 982         struct sk_buff *skb;
 983         int err, repsize, copied;
 984         u32 _xid;
 985         __be32 *xp;
 986
 987         read_lock_bh(&sk->sk_callback_lock);
 988         dprintk("RPC:       xs_udp_data_ready...\n");
 989         if (!(xprt = xprt_from_sock(sk)))
 990                 goto out;
 991
 992         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 993                 goto out;
 994
 995         repsize = skb->len - sizeof(struct udphdr);
 996         if (repsize < 4) {
 997                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 998                 goto dropit;
 999         }
1000
1001         /* Copy the XID from the skb... */
1002         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1003                                 sizeof(_xid), &_xid);
1004         if (xp == NULL)
1005                 goto dropit;
1006
1007         /* Look up and lock the request corresponding to the given XID */
1008         spin_lock(&xprt->transport_lock);
1009         rovr = xprt_lookup_rqst(xprt, *xp);
1010         if (!rovr)
1011                 goto out_unlock;
1012         task = rovr->rq_task;
1013
1014         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1015                 copied = repsize;
1016
1017         /* Suck it into the iovec, verify checksum if not done by hw. */
1018         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1019                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1020                 goto out_unlock;
1021         }
1022
1023         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1024
1025         xprt_adjust_cwnd(xprt, task, copied);
1026         xprt_complete_rqst(task, copied);
1027
1028  out_unlock:
1029         spin_unlock(&xprt->transport_lock);
1030  dropit:
1031         skb_free_datagram(sk, skb);
1032  out:
1033         read_unlock_bh(&sk->sk_callback_lock);
1034 }
1035
1036 /*
1037  * Helper function to force a TCP close if the server is sending
1038  * junk and/or it has put us in CLOSE_WAIT
1039  */
1040 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1041 {
1042         xprt_force_disconnect(xprt);
1043 }
1044
1045 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1046 {
1047         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1048         size_t len, used;
1049         char *p;
1050
1051         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1052         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1053         used = xdr_skb_read_bits(desc, p, len);
1054         transport->tcp_offset += used;
1055         if (used != len)
1056                 return;
1057
1058         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1059         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1060                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1061         else
1062                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1063         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1064
1065         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1066         transport->tcp_offset = 0;
1067
1068         /* Sanity check of the record length */
1069         if (unlikely(transport->tcp_reclen < 8)) {
1070                 dprintk("RPC:       invalid TCP record fragment length\n");
1071                 xs_tcp_force_close(xprt);
1072                 return;
1073         }
1074         dprintk("RPC:       reading TCP record fragment of length %d\n",
1075                         transport->tcp_reclen);
1076 }
1077
1078 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1079 {
1080         if (transport->tcp_offset == transport->tcp_reclen) {
1081                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1082                 transport->tcp_offset = 0;
1083                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1084                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1085                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1086                         transport->tcp_copied = 0;
1087                 }
1088         }
1089 }
1090
1091 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1092 {
1093         size_t len, used;
1094         char *p;
1095
1096         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1097         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1098         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1099         used = xdr_skb_read_bits(desc, p, len);
1100         transport->tcp_offset += used;
1101         if (used != len)
1102                 return;
1103         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1104         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1105         transport->tcp_copied = 4;
1106         dprintk("RPC:       reading %s XID %08x\n",
1107                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1108                                                               : "request with",
1109                         ntohl(transport->tcp_xid));
1110         xs_tcp_check_fraghdr(transport);
1111 }
1112
1113 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1114                                        struct xdr_skb_reader *desc)
1115 {
1116         size_t len, used;
1117         u32 offset;
1118         char *p;
1119
1120         /*
1121          * We want transport->tcp_offset to be 8 at the end of this routine
1122          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1123          * When this function is called for the first time,
1124          * transport->tcp_offset is 4 (after having already read the xid).
1125          */
1126         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1127         len = sizeof(transport->tcp_calldir) - offset;
1128         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1129         p = ((char *) &transport->tcp_calldir) + offset;
1130         used = xdr_skb_read_bits(desc, p, len);
1131         transport->tcp_offset += used;
1132         if (used != len)
1133                 return;
1134         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1135         /*
1136          * We don't yet have the XDR buffer, so we will write the calldir
1137          * out after we get the buffer from the 'struct rpc_rqst'
1138          */
1139         switch (ntohl(transport->tcp_calldir)) {
1140         case RPC_REPLY:
1141                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1142                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1143                 transport->tcp_flags |= TCP_RPC_REPLY;
1144                 break;
1145         case RPC_CALL:
1146                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1147                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1148                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1149                 break;
1150         default:
1151                 dprintk("RPC:       invalid request message type\n");
1152                 xs_tcp_force_close(&transport->xprt);
1153         }
1154         xs_tcp_check_fraghdr(transport);
1155 }
1156
1157 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1158                                      struct xdr_skb_reader *desc,
1159                                      struct rpc_rqst *req)
1160 {
1161         struct sock_xprt *transport =
1162                                 container_of(xprt, struct sock_xprt, xprt);
1163         struct xdr_buf *rcvbuf;
1164         size_t len;
1165         ssize_t r;
1166
1167         rcvbuf = &req->rq_private_buf;
1168
1169         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1170                 /*
1171                  * Save the RPC direction in the XDR buffer
1172                  */
1173                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1174                         &transport->tcp_calldir,
1175                         sizeof(transport->tcp_calldir));
1176                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1177                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1178         }
1179
1180         len = desc->count;
1181         if (len > transport->tcp_reclen - transport->tcp_offset) {
1182                 struct xdr_skb_reader my_desc;
1183
1184                 len = transport->tcp_reclen - transport->tcp_offset;
1185                 memcpy(&my_desc, desc, sizeof(my_desc));
1186                 my_desc.count = len;
1187                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1188                                           &my_desc, xdr_skb_read_bits);
1189                 desc->count -= r;
1190                 desc->offset += r;
1191         } else
1192                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1193                                           desc, xdr_skb_read_bits);
1194
1195         if (r > 0) {
1196                 transport->tcp_copied += r;
1197                 transport->tcp_offset += r;
1198         }
1199         if (r != len) {
1200                 /* Error when copying to the receive buffer,
1201                  * usually because we weren't able to allocate
1202                  * additional buffer pages. All we can do now
1203                  * is turn off TCP_RCV_COPY_DATA, so the request
1204                  * will not receive any additional updates,
1205                  * and time out.
1206                  * Any remaining data from this record will
1207                  * be discarded.
1208                  */
1209                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1210                 dprintk("RPC:       XID %08x truncated request\n",
1211                                 ntohl(transport->tcp_xid));
1212                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1213                                 "tcp_offset = %u, tcp_reclen = %u\n",
1214                                 xprt, transport->tcp_copied,
1215                                 transport->tcp_offset, transport->tcp_reclen);
1216                 return;
1217         }
1218
1219         dprintk("RPC:       XID %08x read %Zd bytes\n",
1220                         ntohl(transport->tcp_xid), r);
1221         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1222                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1223                         transport->tcp_offset, transport->tcp_reclen);
1224
1225         if (transport->tcp_copied == req->rq_private_buf.buflen)
1226                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1227         else if (transport->tcp_offset == transport->tcp_reclen) {
1228                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1229                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1230         }
1231 }
1232
1233 /*
1234  * Finds the request corresponding to the RPC xid and invokes the common
1235  * tcp read code to read the data.
1236  */
1237 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1238                                     struct xdr_skb_reader *desc)
1239 {
1240         struct sock_xprt *transport =
1241                                 container_of(xprt, struct sock_xprt, xprt);
1242         struct rpc_rqst *req;
1243
1244         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1245
1246         /* Find and lock the request corresponding to this xid */
1247         spin_lock(&xprt->transport_lock);
1248         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1249         if (!req) {
1250                 dprintk("RPC:       XID %08x request not found!\n",
1251                                 ntohl(transport->tcp_xid));
1252                 spin_unlock(&xprt->transport_lock);
1253                 return -1;
1254         }
1255
1256         xs_tcp_read_common(xprt, desc, req);
1257
1258         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1259                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1260
1261         spin_unlock(&xprt->transport_lock);
1262         return 0;
1263 }
1264
1265 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1266 /*
1267  * Obtains an rpc_rqst previously allocated and invokes the common
1268  * tcp read code to read the data.  The result is placed in the callback
1269  * queue.
1270  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1271  * connection and return -1.
1272  */
1273 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1274                                        struct xdr_skb_reader *desc)
1275 {
1276         struct sock_xprt *transport =
1277                                 container_of(xprt, struct sock_xprt, xprt);
1278         struct rpc_rqst *req;
1279
1280         /* Look up and lock the request corresponding to the given XID */
1281         spin_lock(&xprt->transport_lock);
1282         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1283         if (req == NULL) {
1284                 spin_unlock(&xprt->transport_lock);
1285                 printk(KERN_WARNING "Callback slot table overflowed\n");
1286                 xprt_force_disconnect(xprt);
1287                 return -1;
1288         }
1289
1290         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1291         xs_tcp_read_common(xprt, desc, req);
1292
1293         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1294                 xprt_complete_bc_request(req, transport->tcp_copied);
1295         spin_unlock(&xprt->transport_lock);
1296
1297         return 0;
1298 }
1299
1300 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1301                                         struct xdr_skb_reader *desc)
1302 {
1303         struct sock_xprt *transport =
1304                                 container_of(xprt, struct sock_xprt, xprt);
1305
1306         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1307                 xs_tcp_read_reply(xprt, desc) :
1308                 xs_tcp_read_callback(xprt, desc);
1309 }
1310 #else
1311 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1312                                         struct xdr_skb_reader *desc)
1313 {
1314         return xs_tcp_read_reply(xprt, desc);
1315 }
1316 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1317
1318 /*
1319  * Read data off the transport.  This can be either an RPC_CALL or an
1320  * RPC_REPLY.  Relay the processing to helper functions.
1321  */
1322 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1323                                     struct xdr_skb_reader *desc)
1324 {
1325         struct sock_xprt *transport =
1326                                 container_of(xprt, struct sock_xprt, xprt);
1327
1328         if (_xs_tcp_read_data(xprt, desc) == 0)
1329                 xs_tcp_check_fraghdr(transport);
1330         else {
1331                 /*
1332                  * The transport_lock protects the request handling.
1333                  * There's no need to hold it to update the tcp_flags.
1334                  */
1335                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1336         }
1337 }
1338
1339 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1340 {
1341         size_t len;
1342
1343         len = transport->tcp_reclen - transport->tcp_offset;
1344         if (len > desc->count)
1345                 len = desc->count;
1346         desc->count -= len;
1347         desc->offset += len;
1348         transport->tcp_offset += len;
1349         dprintk("RPC:       discarded %Zu bytes\n", len);
1350         xs_tcp_check_fraghdr(transport);
1351 }
1352
1353 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1354 {
1355         struct rpc_xprt *xprt = rd_desc->arg.data;
1356         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1357         struct xdr_skb_reader desc = {
1358                 .skb    = skb,
1359                 .offset = offset,
1360                 .count  = len,
1361         };
1362
1363         dprintk("RPC:       xs_tcp_data_recv started\n");
1364         do {
1365                 trace_xs_tcp_data_recv(transport);
1366                 /* Read in a new fragment marker if necessary */
1367                 /* Can we ever really expect to get completely empty fragments? */
1368                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1369                         xs_tcp_read_fraghdr(xprt, &desc);
1370                         continue;
1371                 }
1372                 /* Read in the xid if necessary */
1373                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1374                         xs_tcp_read_xid(transport, &desc);
1375                         continue;
1376                 }
1377                 /* Read in the call/reply flag */
1378                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1379                         xs_tcp_read_calldir(transport, &desc);
1380                         continue;
1381                 }
1382                 /* Read in the request data */
1383                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1384                         xs_tcp_read_data(xprt, &desc);
1385                         continue;
1386                 }
1387                 /* Skip over any trailing bytes on short reads */
1388                 xs_tcp_read_discard(transport, &desc);
1389         } while (desc.count);
1390         trace_xs_tcp_data_recv(transport);
1391         dprintk("RPC:       xs_tcp_data_recv done\n");
1392         return len - desc.count;
1393 }
1394
1395 /**
1396  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1397  * @sk: socket with data to read
1398  *
1399  */
1400 static void xs_tcp_data_ready(struct sock *sk)
1401 {
1402         struct rpc_xprt *xprt;
1403         read_descriptor_t rd_desc;
1404         int read;
1405         unsigned long total = 0;
1406
1407         dprintk("RPC:       xs_tcp_data_ready...\n");
1408
1409         read_lock_bh(&sk->sk_callback_lock);
1410         if (!(xprt = xprt_from_sock(sk))) {
1411                 read = 0;
1412                 goto out;
1413         }
1414         /* Any data means we had a useful conversation, so
1415          * the we don't need to delay the next reconnect
1416          */
1417         if (xprt->reestablish_timeout)
1418                 xprt->reestablish_timeout = 0;
1419
1420         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1421         rd_desc.arg.data = xprt;
1422         do {
1423                 rd_desc.count = 65536;
1424                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1425                 if (read > 0)
1426                         total += read;
1427         } while (read > 0);
1428 out:
1429         trace_xs_tcp_data_ready(xprt, read, total);
1430         read_unlock_bh(&sk->sk_callback_lock);
1431 }
1432
1433 /**
1434  * xs_tcp_state_change - callback to handle TCP socket state changes
1435  * @sk: socket whose state has changed
1436  *
1437  */
1438 static void xs_tcp_state_change(struct sock *sk)
1439 {
1440         struct rpc_xprt *xprt;
1441         struct sock_xprt *transport;
1442
1443         read_lock_bh(&sk->sk_callback_lock);
1444         if (!(xprt = xprt_from_sock(sk)))
1445                 goto out;
1446         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1447         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1448                         sk->sk_state, xprt_connected(xprt),
1449                         sock_flag(sk, SOCK_DEAD),
1450                         sock_flag(sk, SOCK_ZAPPED),
1451                         sk->sk_shutdown);
1452
1453         transport = container_of(xprt, struct sock_xprt, xprt);
1454         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1455         switch (sk->sk_state) {
1456         case TCP_ESTABLISHED:
1457                 spin_lock(&xprt->transport_lock);
1458                 if (!xprt_test_and_set_connected(xprt)) {
1459
1460                         /* Reset TCP record info */
1461                         transport->tcp_offset = 0;
1462                         transport->tcp_reclen = 0;
1463                         transport->tcp_copied = 0;
1464                         transport->tcp_flags =
1465                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1466                         xprt->connect_cookie++;
1467                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1468                         xprt_clear_connecting(xprt);
1469
1470                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1471                 }
1472                 spin_unlock(&xprt->transport_lock);
1473                 break;
1474         case TCP_FIN_WAIT1:
1475                 /* The client initiated a shutdown of the socket */
1476                 xprt->connect_cookie++;
1477                 xprt->reestablish_timeout = 0;
1478                 set_bit(XPRT_CLOSING, &xprt->state);
1479                 smp_mb__before_atomic();
1480                 clear_bit(XPRT_CONNECTED, &xprt->state);
1481                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1482                 smp_mb__after_atomic();
1483                 break;
1484         case TCP_CLOSE_WAIT:
1485                 /* The server initiated a shutdown of the socket */
1486                 xprt->connect_cookie++;
1487                 clear_bit(XPRT_CONNECTED, &xprt->state);
1488                 xs_tcp_force_close(xprt);
1489         case TCP_CLOSING:
1490                 /*
1491                  * If the server closed down the connection, make sure that
1492                  * we back off before reconnecting
1493                  */
1494                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1495                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1496                 break;
1497         case TCP_LAST_ACK:
1498                 set_bit(XPRT_CLOSING, &xprt->state);
1499                 smp_mb__before_atomic();
1500                 clear_bit(XPRT_CONNECTED, &xprt->state);
1501                 smp_mb__after_atomic();
1502                 break;
1503         case TCP_CLOSE:
1504                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1505                                         &transport->sock_state))
1506                         xprt_clear_connecting(xprt);
1507                 xs_sock_mark_closed(xprt);
1508         }
1509  out:
1510         read_unlock_bh(&sk->sk_callback_lock);
1511 }
1512
1513 static void xs_write_space(struct sock *sk)
1514 {
1515         struct socket *sock;
1516         struct rpc_xprt *xprt;
1517
1518         if (unlikely(!(sock = sk->sk_socket)))
1519                 return;
1520         clear_bit(SOCK_NOSPACE, &sock->flags);
1521
1522         if (unlikely(!(xprt = xprt_from_sock(sk))))
1523                 return;
1524         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1525                 return;
1526
1527         xprt_write_space(xprt);
1528 }
1529
1530 /**
1531  * xs_udp_write_space - callback invoked when socket buffer space
1532  *                             becomes available
1533  * @sk: socket whose state has changed
1534  *
1535  * Called when more output buffer space is available for this socket.
1536  * We try not to wake our writers until they can make "significant"
1537  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1538  * with a bunch of small requests.
1539  */
1540 static void xs_udp_write_space(struct sock *sk)
1541 {
1542         read_lock_bh(&sk->sk_callback_lock);
1543
1544         /* from net/core/sock.c:sock_def_write_space */
1545         if (sock_writeable(sk))
1546                 xs_write_space(sk);
1547
1548         read_unlock_bh(&sk->sk_callback_lock);
1549 }
1550
1551 /**
1552  * xs_tcp_write_space - callback invoked when socket buffer space
1553  *                             becomes available
1554  * @sk: socket whose state has changed
1555  *
1556  * Called when more output buffer space is available for this socket.
1557  * We try not to wake our writers until they can make "significant"
1558  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1559  * with a bunch of small requests.
1560  */
1561 static void xs_tcp_write_space(struct sock *sk)
1562 {
1563         read_lock_bh(&sk->sk_callback_lock);
1564
1565         /* from net/core/stream.c:sk_stream_write_space */
1566         if (sk_stream_is_writeable(sk))
1567                 xs_write_space(sk);
1568
1569         read_unlock_bh(&sk->sk_callback_lock);
1570 }
1571
1572 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1573 {
1574         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1575         struct sock *sk = transport->inet;
1576
1577         if (transport->rcvsize) {
1578                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1579                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1580         }
1581         if (transport->sndsize) {
1582                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1583                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1584                 sk->sk_write_space(sk);
1585         }
1586 }
1587
1588 /**
1589  * xs_udp_set_buffer_size - set send and receive limits
1590  * @xprt: generic transport
1591  * @sndsize: requested size of send buffer, in bytes
1592  * @rcvsize: requested size of receive buffer, in bytes
1593  *
1594  * Set socket send and receive buffer size limits.
1595  */
1596 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1597 {
1598         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1599
1600         transport->sndsize = 0;
1601         if (sndsize)
1602                 transport->sndsize = sndsize + 1024;
1603         transport->rcvsize = 0;
1604         if (rcvsize)
1605                 transport->rcvsize = rcvsize + 1024;
1606
1607         xs_udp_do_set_buffer_size(xprt);
1608 }
1609
1610 /**
1611  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1612  * @task: task that timed out
1613  *
1614  * Adjust the congestion window after a retransmit timeout has occurred.
1615  */
1616 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1617 {
1618         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1619 }
1620
1621 static unsigned short xs_get_random_port(void)
1622 {
1623         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1624         unsigned short rand = (unsigned short) prandom_u32() % range;
1625         return rand + xprt_min_resvport;
1626 }
1627
1628 /**
1629  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1630  * @sock: socket
1631  *
1632  * Note that this function has to be called on all sockets that share the
1633  * same port, and it must be called before binding.
1634  */
1635 static void xs_sock_set_reuseport(struct socket *sock)
1636 {
1637         int opt = 1;
1638
1639         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1640                         (char *)&opt, sizeof(opt));
1641 }
1642
1643 static unsigned short xs_sock_getport(struct socket *sock)
1644 {
1645         struct sockaddr_storage buf;
1646         int buflen;
1647         unsigned short port = 0;
1648
1649         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1650                 goto out;
1651         switch (buf.ss_family) {
1652         case AF_INET6:
1653                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1654                 break;
1655         case AF_INET:
1656                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1657         }
1658 out:
1659         return port;
1660 }
1661
1662 /**
1663  * xs_set_port - reset the port number in the remote endpoint address
1664  * @xprt: generic transport
1665  * @port: new port number
1666  *
1667  */
1668 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1669 {
1670         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1671
1672         rpc_set_port(xs_addr(xprt), port);
1673         xs_update_peer_port(xprt);
1674 }
1675
1676 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1677 {
1678         if (transport->srcport == 0)
1679                 transport->srcport = xs_sock_getport(sock);
1680 }
1681
1682 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1683 {
1684         unsigned short port = transport->srcport;
1685
1686         if (port == 0 && transport->xprt.resvport)
1687                 port = xs_get_random_port();
1688         return port;
1689 }
1690
1691 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1692 {
1693         if (transport->srcport != 0)
1694                 transport->srcport = 0;
1695         if (!transport->xprt.resvport)
1696                 return 0;
1697         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1698                 return xprt_max_resvport;
1699         return --port;
1700 }
1701 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1702 {
1703         struct sockaddr_storage myaddr;
1704         int err, nloop = 0;
1705         unsigned short port = xs_get_srcport(transport);
1706         unsigned short last;
1707
1708         /*
1709          * If we are asking for any ephemeral port (i.e. port == 0 &&
1710          * transport->xprt.resvport == 0), don't bind.  Let the local
1711          * port selection happen implicitly when the socket is used
1712          * (for example at connect time).
1713          *
1714          * This ensures that we can continue to establish TCP
1715          * connections even when all local ephemeral ports are already
1716          * a part of some TCP connection.  This makes no difference
1717          * for UDP sockets, but also doens't harm them.
1718          *
1719          * If we're asking for any reserved port (i.e. port == 0 &&
1720          * transport->xprt.resvport == 1) xs_get_srcport above will
1721          * ensure that port is non-zero and we will bind as needed.
1722          */
1723         if (port == 0)
1724                 return 0;
1725
1726         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1727         do {
1728                 rpc_set_port((struct sockaddr *)&myaddr, port);
1729                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1730                                 transport->xprt.addrlen);
1731                 if (err == 0) {
1732                         transport->srcport = port;
1733                         break;
1734                 }
1735                 last = port;
1736                 port = xs_next_srcport(transport, port);
1737                 if (port > last)
1738                         nloop++;
1739         } while (err == -EADDRINUSE && nloop != 2);
1740
1741         if (myaddr.ss_family == AF_INET)
1742                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1743                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1744                                 port, err ? "failed" : "ok", err);
1745         else
1746                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1747                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1748                                 port, err ? "failed" : "ok", err);
1749         return err;
1750 }
1751
1752 /*
1753  * We don't support autobind on AF_LOCAL sockets
1754  */
1755 static void xs_local_rpcbind(struct rpc_task *task)
1756 {
1757         rcu_read_lock();
1758         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1759         rcu_read_unlock();
1760 }
1761
1762 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1763 {
1764 }
1765
1766 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1767 static struct lock_class_key xs_key[2];
1768 static struct lock_class_key xs_slock_key[2];
1769
1770 static inline void xs_reclassify_socketu(struct socket *sock)
1771 {
1772         struct sock *sk = sock->sk;
1773
1774         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1775                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1776 }
1777
1778 static inline void xs_reclassify_socket4(struct socket *sock)
1779 {
1780         struct sock *sk = sock->sk;
1781
1782         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1783                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1784 }
1785
1786 static inline void xs_reclassify_socket6(struct socket *sock)
1787 {
1788         struct sock *sk = sock->sk;
1789
1790         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1791                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1792 }
1793
1794 static inline void xs_reclassify_socket(int family, struct socket *sock)
1795 {
1796         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1797         if (sock_owned_by_user(sock->sk))
1798                 return;
1799
1800         switch (family) {
1801         case AF_LOCAL:
1802                 xs_reclassify_socketu(sock);
1803                 break;
1804         case AF_INET:
1805                 xs_reclassify_socket4(sock);
1806                 break;
1807         case AF_INET6:
1808                 xs_reclassify_socket6(sock);
1809                 break;
1810         }
1811 }
1812 #else
1813 static inline void xs_reclassify_socketu(struct socket *sock)
1814 {
1815 }
1816
1817 static inline void xs_reclassify_socket4(struct socket *sock)
1818 {
1819 }
1820
1821 static inline void xs_reclassify_socket6(struct socket *sock)
1822 {
1823 }
1824
1825 static inline void xs_reclassify_socket(int family, struct socket *sock)
1826 {
1827 }
1828 #endif
1829
1830 static void xs_dummy_setup_socket(struct work_struct *work)
1831 {
1832 }
1833
1834 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1835                 struct sock_xprt *transport, int family, int type,
1836                 int protocol, bool reuseport)
1837 {
1838         struct socket *sock;
1839         int err;
1840
1841         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1842         if (err < 0) {
1843                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1844                                 protocol, -err);
1845                 goto out;
1846         }
1847         xs_reclassify_socket(family, sock);
1848
1849         if (reuseport)
1850                 xs_sock_set_reuseport(sock);
1851
1852         err = xs_bind(transport, sock);
1853         if (err) {
1854                 sock_release(sock);
1855                 goto out;
1856         }
1857
1858         return sock;
1859 out:
1860         return ERR_PTR(err);
1861 }
1862
1863 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1864                                       struct socket *sock)
1865 {
1866         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1867                                                                         xprt);
1868
1869         if (!transport->inet) {
1870                 struct sock *sk = sock->sk;
1871
1872                 write_lock_bh(&sk->sk_callback_lock);
1873
1874                 xs_save_old_callbacks(transport, sk);
1875
1876                 sk->sk_user_data = xprt;
1877                 sk->sk_data_ready = xs_local_data_ready;
1878                 sk->sk_write_space = xs_udp_write_space;
1879                 sk->sk_error_report = xs_error_report;
1880                 sk->sk_allocation = GFP_NOIO;
1881
1882                 xprt_clear_connected(xprt);
1883
1884                 /* Reset to new socket */
1885                 transport->sock = sock;
1886                 transport->inet = sk;
1887
1888                 write_unlock_bh(&sk->sk_callback_lock);
1889         }
1890
1891         /* Tell the socket layer to start connecting... */
1892         xprt->stat.connect_count++;
1893         xprt->stat.connect_start = jiffies;
1894         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1895 }
1896
1897 /**
1898  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1899  * @transport: socket transport to connect
1900  */
1901 static int xs_local_setup_socket(struct sock_xprt *transport)
1902 {
1903         struct rpc_xprt *xprt = &transport->xprt;
1904         struct socket *sock;
1905         int status = -EIO;
1906
1907         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1908                                         SOCK_STREAM, 0, &sock, 1);
1909         if (status < 0) {
1910                 dprintk("RPC:       can't create AF_LOCAL "
1911                         "transport socket (%d).\n", -status);
1912                 goto out;
1913         }
1914         xs_reclassify_socketu(sock);
1915
1916         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1917                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1918
1919         status = xs_local_finish_connecting(xprt, sock);
1920         trace_rpc_socket_connect(xprt, sock, status);
1921         switch (status) {
1922         case 0:
1923                 dprintk("RPC:       xprt %p connected to %s\n",
1924                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1925                 xprt_set_connected(xprt);
1926         case -ENOBUFS:
1927                 break;
1928         case -ENOENT:
1929                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1930                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1931                 break;
1932         case -ECONNREFUSED:
1933                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1934                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1935                 break;
1936         default:
1937                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1938                                 __func__, -status,
1939                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1940         }
1941
1942 out:
1943         xprt_clear_connecting(xprt);
1944         xprt_wake_pending_tasks(xprt, status);
1945         return status;
1946 }
1947
1948 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1949 {
1950         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1951         int ret;
1952
1953          if (RPC_IS_ASYNC(task)) {
1954                 /*
1955                  * We want the AF_LOCAL connect to be resolved in the
1956                  * filesystem namespace of the process making the rpc
1957                  * call.  Thus we connect synchronously.
1958                  *
1959                  * If we want to support asynchronous AF_LOCAL calls,
1960                  * we'll need to figure out how to pass a namespace to
1961                  * connect.
1962                  */
1963                 rpc_exit(task, -ENOTCONN);
1964                 return;
1965         }
1966         ret = xs_local_setup_socket(transport);
1967         if (ret && !RPC_IS_SOFTCONN(task))
1968                 msleep_interruptible(15000);
1969 }
1970
1971 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1972 /*
1973  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1974  * know that we have exclusive access to the socket), to guard against
1975  * races with xs_reset_transport.
1976  */
1977 static void xs_set_memalloc(struct rpc_xprt *xprt)
1978 {
1979         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1980                         xprt);
1981
1982         /*
1983          * If there's no sock, then we have nothing to set. The
1984          * reconnecting process will get it for us.
1985          */
1986         if (!transport->inet)
1987                 return;
1988         if (atomic_read(&xprt->swapper))
1989                 sk_set_memalloc(transport->inet);
1990 }
1991
1992 /**
1993  * xs_enable_swap - Tag this transport as being used for swap.
1994  * @xprt: transport to tag
1995  *
1996  * Take a reference to this transport on behalf of the rpc_clnt, and
1997  * optionally mark it for swapping if it wasn't already.
1998  */
1999 static int
2000 xs_enable_swap(struct rpc_xprt *xprt)
2001 {
2002         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2003
2004         if (atomic_inc_return(&xprt->swapper) != 1)
2005                 return 0;
2006         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2007                 return -ERESTARTSYS;
2008         if (xs->inet)
2009                 sk_set_memalloc(xs->inet);
2010         xprt_release_xprt(xprt, NULL);
2011         return 0;
2012 }
2013
2014 /**
2015  * xs_disable_swap - Untag this transport as being used for swap.
2016  * @xprt: transport to tag
2017  *
2018  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2019  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2020  */
2021 static void
2022 xs_disable_swap(struct rpc_xprt *xprt)
2023 {
2024         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2025
2026         if (!atomic_dec_and_test(&xprt->swapper))
2027                 return;
2028         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2029                 return;
2030         if (xs->inet)
2031                 sk_clear_memalloc(xs->inet);
2032         xprt_release_xprt(xprt, NULL);
2033 }
2034 #else
2035 static void xs_set_memalloc(struct rpc_xprt *xprt)
2036 {
2037 }
2038
2039 static int
2040 xs_enable_swap(struct rpc_xprt *xprt)
2041 {
2042         return -EINVAL;
2043 }
2044
2045 static void
2046 xs_disable_swap(struct rpc_xprt *xprt)
2047 {
2048 }
2049 #endif
2050
2051 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2052 {
2053         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2054
2055         if (!transport->inet) {
2056                 struct sock *sk = sock->sk;
2057
2058                 write_lock_bh(&sk->sk_callback_lock);
2059
2060                 xs_save_old_callbacks(transport, sk);
2061
2062                 sk->sk_user_data = xprt;
2063                 sk->sk_data_ready = xs_udp_data_ready;
2064                 sk->sk_write_space = xs_udp_write_space;
2065                 sk->sk_allocation = GFP_NOIO;
2066
2067                 xprt_set_connected(xprt);
2068
2069                 /* Reset to new socket */
2070                 transport->sock = sock;
2071                 transport->inet = sk;
2072
2073                 xs_set_memalloc(xprt);
2074
2075                 write_unlock_bh(&sk->sk_callback_lock);
2076         }
2077         xs_udp_do_set_buffer_size(xprt);
2078 }
2079
2080 static void xs_udp_setup_socket(struct work_struct *work)
2081 {
2082         struct sock_xprt *transport =
2083                 container_of(work, struct sock_xprt, connect_worker.work);
2084         struct rpc_xprt *xprt = &transport->xprt;
2085         struct socket *sock = transport->sock;
2086         int status = -EIO;
2087
2088         sock = xs_create_sock(xprt, transport,
2089                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2090                         IPPROTO_UDP, false);
2091         if (IS_ERR(sock))
2092                 goto out;
2093
2094         dprintk("RPC:       worker connecting xprt %p via %s to "
2095                                 "%s (port %s)\n", xprt,
2096                         xprt->address_strings[RPC_DISPLAY_PROTO],
2097                         xprt->address_strings[RPC_DISPLAY_ADDR],
2098                         xprt->address_strings[RPC_DISPLAY_PORT]);
2099
2100         xs_udp_finish_connecting(xprt, sock);
2101         trace_rpc_socket_connect(xprt, sock, 0);
2102         status = 0;
2103 out:
2104         xprt_unlock_connect(xprt, transport);
2105         xprt_clear_connecting(xprt);
2106         xprt_wake_pending_tasks(xprt, status);
2107 }
2108
2109 /**
2110  * xs_tcp_shutdown - gracefully shut down a TCP socket
2111  * @xprt: transport
2112  *
2113  * Initiates a graceful shutdown of the TCP socket by calling the
2114  * equivalent of shutdown(SHUT_RDWR);
2115  */
2116 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2117 {
2118         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2119         struct socket *sock = transport->sock;
2120
2121         if (sock == NULL)
2122                 return;
2123         if (xprt_connected(xprt)) {
2124                 kernel_sock_shutdown(sock, SHUT_RDWR);
2125                 trace_rpc_socket_shutdown(xprt, sock);
2126         } else
2127                 xs_reset_transport(transport);
2128 }
2129
2130 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2131 {
2132         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2133         int ret = -ENOTCONN;
2134
2135         if (!transport->inet) {
2136                 struct sock *sk = sock->sk;
2137                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2138                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2139                 unsigned int opt_on = 1;
2140                 unsigned int timeo;
2141
2142                 /* TCP Keepalive options */
2143                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2144                                 (char *)&opt_on, sizeof(opt_on));
2145                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2146                                 (char *)&keepidle, sizeof(keepidle));
2147                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2148                                 (char *)&keepidle, sizeof(keepidle));
2149                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2150                                 (char *)&keepcnt, sizeof(keepcnt));
2151
2152                 /* TCP user timeout (see RFC5482) */
2153                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2154                         (xprt->timeout->to_retries + 1);
2155                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2156                                 (char *)&timeo, sizeof(timeo));
2157
2158                 write_lock_bh(&sk->sk_callback_lock);
2159
2160                 xs_save_old_callbacks(transport, sk);
2161
2162                 sk->sk_user_data = xprt;
2163                 sk->sk_data_ready = xs_tcp_data_ready;
2164                 sk->sk_state_change = xs_tcp_state_change;
2165                 sk->sk_write_space = xs_tcp_write_space;
2166                 sk->sk_error_report = xs_error_report;
2167                 sk->sk_allocation = GFP_NOIO;
2168
2169                 /* socket options */
2170                 sock_reset_flag(sk, SOCK_LINGER);
2171                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2172
2173                 xprt_clear_connected(xprt);
2174
2175                 /* Reset to new socket */
2176                 transport->sock = sock;
2177                 transport->inet = sk;
2178
2179                 write_unlock_bh(&sk->sk_callback_lock);
2180         }
2181
2182         if (!xprt_bound(xprt))
2183                 goto out;
2184
2185         xs_set_memalloc(xprt);
2186
2187         /* Tell the socket layer to start connecting... */
2188         xprt->stat.connect_count++;
2189         xprt->stat.connect_start = jiffies;
2190         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2191         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2192         switch (ret) {
2193         case 0:
2194                 xs_set_srcport(transport, sock);
2195         case -EINPROGRESS:
2196                 /* SYN_SENT! */
2197                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2198                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2199         }
2200 out:
2201         return ret;
2202 }
2203
2204 /**
2205  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2206  *
2207  * Invoked by a work queue tasklet.
2208  */
2209 static void xs_tcp_setup_socket(struct work_struct *work)
2210 {
2211         struct sock_xprt *transport =
2212                 container_of(work, struct sock_xprt, connect_worker.work);
2213         struct socket *sock = transport->sock;
2214         struct rpc_xprt *xprt = &transport->xprt;
2215         int status = -EIO;
2216
2217         if (!sock) {
2218                 sock = xs_create_sock(xprt, transport,
2219                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2220                                 IPPROTO_TCP, true);
2221                 if (IS_ERR(sock)) {
2222                         status = PTR_ERR(sock);
2223                         goto out;
2224                 }
2225         }
2226
2227         dprintk("RPC:       worker connecting xprt %p via %s to "
2228                                 "%s (port %s)\n", xprt,
2229                         xprt->address_strings[RPC_DISPLAY_PROTO],
2230                         xprt->address_strings[RPC_DISPLAY_ADDR],
2231                         xprt->address_strings[RPC_DISPLAY_PORT]);
2232
2233         status = xs_tcp_finish_connecting(xprt, sock);
2234         trace_rpc_socket_connect(xprt, sock, status);
2235         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2236                         xprt, -status, xprt_connected(xprt),
2237                         sock->sk->sk_state);
2238         switch (status) {
2239         default:
2240                 printk("%s: connect returned unhandled error %d\n",
2241                         __func__, status);
2242         case -EADDRNOTAVAIL:
2243                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2244                  * and retry
2245                  */
2246                 xs_tcp_force_close(xprt);
2247                 break;
2248         case 0:
2249         case -EINPROGRESS:
2250         case -EALREADY:
2251                 xprt_unlock_connect(xprt, transport);
2252                 return;
2253         case -EINVAL:
2254                 /* Happens, for instance, if the user specified a link
2255                  * local IPv6 address without a scope-id.
2256                  */
2257         case -ECONNREFUSED:
2258         case -ECONNRESET:
2259         case -ENETUNREACH:
2260         case -EADDRINUSE:
2261         case -ENOBUFS:
2262                 /* retry with existing socket, after a delay */
2263                 xs_tcp_force_close(xprt);
2264                 goto out;
2265         }
2266         status = -EAGAIN;
2267 out:
2268         xprt_unlock_connect(xprt, transport);
2269         xprt_clear_connecting(xprt);
2270         xprt_wake_pending_tasks(xprt, status);
2271 }
2272
2273 /**
2274  * xs_connect - connect a socket to a remote endpoint
2275  * @xprt: pointer to transport structure
2276  * @task: address of RPC task that manages state of connect request
2277  *
2278  * TCP: If the remote end dropped the connection, delay reconnecting.
2279  *
2280  * UDP socket connects are synchronous, but we use a work queue anyway
2281  * to guarantee that even unprivileged user processes can set up a
2282  * socket on a privileged port.
2283  *
2284  * If a UDP socket connect fails, the delay behavior here prevents
2285  * retry floods (hard mounts).
2286  */
2287 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2288 {
2289         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2290
2291         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2292
2293         if (transport->sock != NULL) {
2294                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2295                                 "seconds\n",
2296                                 xprt, xprt->reestablish_timeout / HZ);
2297
2298                 /* Start by resetting any existing state */
2299                 xs_reset_transport(transport);
2300
2301                 queue_delayed_work(rpciod_workqueue,
2302                                    &transport->connect_worker,
2303                                    xprt->reestablish_timeout);
2304                 xprt->reestablish_timeout <<= 1;
2305                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2306                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2307                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2308                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2309         } else {
2310                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2311                 queue_delayed_work(rpciod_workqueue,
2312                                    &transport->connect_worker, 0);
2313         }
2314 }
2315
2316 /**
2317  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2318  * @xprt: rpc_xprt struct containing statistics
2319  * @seq: output file
2320  *
2321  */
2322 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2323 {
2324         long idle_time = 0;
2325
2326         if (xprt_connected(xprt))
2327                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2328
2329         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2330                         "%llu %llu %lu %llu %llu\n",
2331                         xprt->stat.bind_count,
2332                         xprt->stat.connect_count,
2333                         xprt->stat.connect_time,
2334                         idle_time,
2335                         xprt->stat.sends,
2336                         xprt->stat.recvs,
2337                         xprt->stat.bad_xids,
2338                         xprt->stat.req_u,
2339                         xprt->stat.bklog_u,
2340                         xprt->stat.max_slots,
2341                         xprt->stat.sending_u,
2342                         xprt->stat.pending_u);
2343 }
2344
2345 /**
2346  * xs_udp_print_stats - display UDP socket-specifc stats
2347  * @xprt: rpc_xprt struct containing statistics
2348  * @seq: output file
2349  *
2350  */
2351 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2352 {
2353         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2354
2355         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2356                         "%lu %llu %llu\n",
2357                         transport->srcport,
2358                         xprt->stat.bind_count,
2359                         xprt->stat.sends,
2360                         xprt->stat.recvs,
2361                         xprt->stat.bad_xids,
2362                         xprt->stat.req_u,
2363                         xprt->stat.bklog_u,
2364                         xprt->stat.max_slots,
2365                         xprt->stat.sending_u,
2366                         xprt->stat.pending_u);
2367 }
2368
2369 /**
2370  * xs_tcp_print_stats - display TCP socket-specifc stats
2371  * @xprt: rpc_xprt struct containing statistics
2372  * @seq: output file
2373  *
2374  */
2375 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2376 {
2377         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2378         long idle_time = 0;
2379
2380         if (xprt_connected(xprt))
2381                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2382
2383         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2384                         "%llu %llu %lu %llu %llu\n",
2385                         transport->srcport,
2386                         xprt->stat.bind_count,
2387                         xprt->stat.connect_count,
2388                         xprt->stat.connect_time,
2389                         idle_time,
2390                         xprt->stat.sends,
2391                         xprt->stat.recvs,
2392                         xprt->stat.bad_xids,
2393                         xprt->stat.req_u,
2394                         xprt->stat.bklog_u,
2395                         xprt->stat.max_slots,
2396                         xprt->stat.sending_u,
2397                         xprt->stat.pending_u);
2398 }
2399
2400 /*
2401  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2402  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2403  * to use the server side send routines.
2404  */
2405 static void *bc_malloc(struct rpc_task *task, size_t size)
2406 {
2407         struct page *page;
2408         struct rpc_buffer *buf;
2409
2410         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2411         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2412                 return NULL;
2413
2414         page = alloc_page(GFP_KERNEL);
2415         if (!page)
2416                 return NULL;
2417
2418         buf = page_address(page);
2419         buf->len = PAGE_SIZE;
2420
2421         return buf->data;
2422 }
2423
2424 /*
2425  * Free the space allocated in the bc_alloc routine
2426  */
2427 static void bc_free(void *buffer)
2428 {
2429         struct rpc_buffer *buf;
2430
2431         if (!buffer)
2432                 return;
2433
2434         buf = container_of(buffer, struct rpc_buffer, data);
2435         free_page((unsigned long)buf);
2436 }
2437
2438 /*
2439  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2440  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2441  */
2442 static int bc_sendto(struct rpc_rqst *req)
2443 {
2444         int len;
2445         struct xdr_buf *xbufp = &req->rq_snd_buf;
2446         struct rpc_xprt *xprt = req->rq_xprt;
2447         struct sock_xprt *transport =
2448                                 container_of(xprt, struct sock_xprt, xprt);
2449         struct socket *sock = transport->sock;
2450         unsigned long headoff;
2451         unsigned long tailoff;
2452
2453         xs_encode_stream_record_marker(xbufp);
2454
2455         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2456         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2457         len = svc_send_common(sock, xbufp,
2458                               virt_to_page(xbufp->head[0].iov_base), headoff,
2459                               xbufp->tail[0].iov_base, tailoff);
2460
2461         if (len != xbufp->len) {
2462                 printk(KERN_NOTICE "Error sending entire callback!\n");
2463                 len = -EAGAIN;
2464         }
2465
2466         return len;
2467 }
2468
2469 /*
2470  * The send routine. Borrows from svc_send
2471  */
2472 static int bc_send_request(struct rpc_task *task)
2473 {
2474         struct rpc_rqst *req = task->tk_rqstp;
2475         struct svc_xprt *xprt;
2476         u32                     len;
2477
2478         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2479         /*
2480          * Get the server socket associated with this callback xprt
2481          */
2482         xprt = req->rq_xprt->bc_xprt;
2483
2484         /*
2485          * Grab the mutex to serialize data as the connection is shared
2486          * with the fore channel
2487          */
2488         if (!mutex_trylock(&xprt->xpt_mutex)) {
2489                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2490                 if (!mutex_trylock(&xprt->xpt_mutex))
2491                         return -EAGAIN;
2492                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2493         }
2494         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2495                 len = -ENOTCONN;
2496         else
2497                 len = bc_sendto(req);
2498         mutex_unlock(&xprt->xpt_mutex);
2499
2500         if (len > 0)
2501                 len = 0;
2502
2503         return len;
2504 }
2505
2506 /*
2507  * The close routine. Since this is client initiated, we do nothing
2508  */
2509
2510 static void bc_close(struct rpc_xprt *xprt)
2511 {
2512 }
2513
2514 /*
2515  * The xprt destroy routine. Again, because this connection is client
2516  * initiated, we do nothing
2517  */
2518
2519 static void bc_destroy(struct rpc_xprt *xprt)
2520 {
2521         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2522
2523         xs_xprt_free(xprt);
2524         module_put(THIS_MODULE);
2525 }
2526
2527 static struct rpc_xprt_ops xs_local_ops = {
2528         .reserve_xprt           = xprt_reserve_xprt,
2529         .release_xprt           = xs_tcp_release_xprt,
2530         .alloc_slot             = xprt_alloc_slot,
2531         .rpcbind                = xs_local_rpcbind,
2532         .set_port               = xs_local_set_port,
2533         .connect                = xs_local_connect,
2534         .buf_alloc              = rpc_malloc,
2535         .buf_free               = rpc_free,
2536         .send_request           = xs_local_send_request,
2537         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2538         .close                  = xs_close,
2539         .destroy                = xs_destroy,
2540         .print_stats            = xs_local_print_stats,
2541         .enable_swap            = xs_enable_swap,
2542         .disable_swap           = xs_disable_swap,
2543 };
2544
2545 static struct rpc_xprt_ops xs_udp_ops = {
2546         .set_buffer_size        = xs_udp_set_buffer_size,
2547         .reserve_xprt           = xprt_reserve_xprt_cong,
2548         .release_xprt           = xprt_release_xprt_cong,
2549         .alloc_slot             = xprt_alloc_slot,
2550         .rpcbind                = rpcb_getport_async,
2551         .set_port               = xs_set_port,
2552         .connect                = xs_connect,
2553         .buf_alloc              = rpc_malloc,
2554         .buf_free               = rpc_free,
2555         .send_request           = xs_udp_send_request,
2556         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2557         .timer                  = xs_udp_timer,
2558         .release_request        = xprt_release_rqst_cong,
2559         .close                  = xs_close,
2560         .destroy                = xs_destroy,
2561         .print_stats            = xs_udp_print_stats,
2562         .enable_swap            = xs_enable_swap,
2563         .disable_swap           = xs_disable_swap,
2564         .inject_disconnect      = xs_inject_disconnect,
2565 };
2566
2567 static struct rpc_xprt_ops xs_tcp_ops = {
2568         .reserve_xprt           = xprt_reserve_xprt,
2569         .release_xprt           = xs_tcp_release_xprt,
2570         .alloc_slot             = xprt_lock_and_alloc_slot,
2571         .rpcbind                = rpcb_getport_async,
2572         .set_port               = xs_set_port,
2573         .connect                = xs_connect,
2574         .buf_alloc              = rpc_malloc,
2575         .buf_free               = rpc_free,
2576         .send_request           = xs_tcp_send_request,
2577         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2578         .close                  = xs_tcp_shutdown,
2579         .destroy                = xs_destroy,
2580         .print_stats            = xs_tcp_print_stats,
2581         .enable_swap            = xs_enable_swap,
2582         .disable_swap           = xs_disable_swap,
2583         .inject_disconnect      = xs_inject_disconnect,
2584 };
2585
2586 /*
2587  * The rpc_xprt_ops for the server backchannel
2588  */
2589
2590 static struct rpc_xprt_ops bc_tcp_ops = {
2591         .reserve_xprt           = xprt_reserve_xprt,
2592         .release_xprt           = xprt_release_xprt,
2593         .alloc_slot             = xprt_alloc_slot,
2594         .buf_alloc              = bc_malloc,
2595         .buf_free               = bc_free,
2596         .send_request           = bc_send_request,
2597         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2598         .close                  = bc_close,
2599         .destroy                = bc_destroy,
2600         .print_stats            = xs_tcp_print_stats,
2601         .enable_swap            = xs_enable_swap,
2602         .disable_swap           = xs_disable_swap,
2603         .inject_disconnect      = xs_inject_disconnect,
2604 };
2605
2606 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2607 {
2608         static const struct sockaddr_in sin = {
2609                 .sin_family             = AF_INET,
2610                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2611         };
2612         static const struct sockaddr_in6 sin6 = {
2613                 .sin6_family            = AF_INET6,
2614                 .sin6_addr              = IN6ADDR_ANY_INIT,
2615         };
2616
2617         switch (family) {
2618         case AF_LOCAL:
2619                 break;
2620         case AF_INET:
2621                 memcpy(sap, &sin, sizeof(sin));
2622                 break;
2623         case AF_INET6:
2624                 memcpy(sap, &sin6, sizeof(sin6));
2625                 break;
2626         default:
2627                 dprintk("RPC:       %s: Bad address family\n", __func__);
2628                 return -EAFNOSUPPORT;
2629         }
2630         return 0;
2631 }
2632
2633 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2634                                       unsigned int slot_table_size,
2635                                       unsigned int max_slot_table_size)
2636 {
2637         struct rpc_xprt *xprt;
2638         struct sock_xprt *new;
2639
2640         if (args->addrlen > sizeof(xprt->addr)) {
2641                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2642                 return ERR_PTR(-EBADF);
2643         }
2644
2645         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2646                         max_slot_table_size);
2647         if (xprt == NULL) {
2648                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2649                                 "rpc_xprt\n");
2650                 return ERR_PTR(-ENOMEM);
2651         }
2652
2653         new = container_of(xprt, struct sock_xprt, xprt);
2654         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2655         xprt->addrlen = args->addrlen;
2656         if (args->srcaddr)
2657                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2658         else {
2659                 int err;
2660                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2661                                         (struct sockaddr *)&new->srcaddr);
2662                 if (err != 0) {
2663                         xprt_free(xprt);
2664                         return ERR_PTR(err);
2665                 }
2666         }
2667
2668         return xprt;
2669 }
2670
2671 static const struct rpc_timeout xs_local_default_timeout = {
2672         .to_initval = 10 * HZ,
2673         .to_maxval = 10 * HZ,
2674         .to_retries = 2,
2675 };
2676
2677 /**
2678  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2679  * @args: rpc transport creation arguments
2680  *
2681  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2682  */
2683 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2684 {
2685         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2686         struct sock_xprt *transport;
2687         struct rpc_xprt *xprt;
2688         struct rpc_xprt *ret;
2689
2690         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2691                         xprt_max_tcp_slot_table_entries);
2692         if (IS_ERR(xprt))
2693                 return xprt;
2694         transport = container_of(xprt, struct sock_xprt, xprt);
2695
2696         xprt->prot = 0;
2697         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2698         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2699
2700         xprt->bind_timeout = XS_BIND_TO;
2701         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2702         xprt->idle_timeout = XS_IDLE_DISC_TO;
2703
2704         xprt->ops = &xs_local_ops;
2705         xprt->timeout = &xs_local_default_timeout;
2706
2707         INIT_DELAYED_WORK(&transport->connect_worker,
2708                         xs_dummy_setup_socket);
2709
2710         switch (sun->sun_family) {
2711         case AF_LOCAL:
2712                 if (sun->sun_path[0] != '/') {
2713                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2714                                         sun->sun_path);
2715                         ret = ERR_PTR(-EINVAL);
2716                         goto out_err;
2717                 }
2718                 xprt_set_bound(xprt);
2719                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2720                 ret = ERR_PTR(xs_local_setup_socket(transport));
2721                 if (ret)
2722                         goto out_err;
2723                 break;
2724         default:
2725                 ret = ERR_PTR(-EAFNOSUPPORT);
2726                 goto out_err;
2727         }
2728
2729         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2730                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2731
2732         if (try_module_get(THIS_MODULE))
2733                 return xprt;
2734         ret = ERR_PTR(-EINVAL);
2735 out_err:
2736         xs_xprt_free(xprt);
2737         return ret;
2738 }
2739
2740 static const struct rpc_timeout xs_udp_default_timeout = {
2741         .to_initval = 5 * HZ,
2742         .to_maxval = 30 * HZ,
2743         .to_increment = 5 * HZ,
2744         .to_retries = 5,
2745 };
2746
2747 /**
2748  * xs_setup_udp - Set up transport to use a UDP socket
2749  * @args: rpc transport creation arguments
2750  *
2751  */
2752 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2753 {
2754         struct sockaddr *addr = args->dstaddr;
2755         struct rpc_xprt *xprt;
2756         struct sock_xprt *transport;
2757         struct rpc_xprt *ret;
2758
2759         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2760                         xprt_udp_slot_table_entries);
2761         if (IS_ERR(xprt))
2762                 return xprt;
2763         transport = container_of(xprt, struct sock_xprt, xprt);
2764
2765         xprt->prot = IPPROTO_UDP;
2766         xprt->tsh_size = 0;
2767         /* XXX: header size can vary due to auth type, IPv6, etc. */
2768         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2769
2770         xprt->bind_timeout = XS_BIND_TO;
2771         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2772         xprt->idle_timeout = XS_IDLE_DISC_TO;
2773
2774         xprt->ops = &xs_udp_ops;
2775
2776         xprt->timeout = &xs_udp_default_timeout;
2777
2778         switch (addr->sa_family) {
2779         case AF_INET:
2780                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2781                         xprt_set_bound(xprt);
2782
2783                 INIT_DELAYED_WORK(&transport->connect_worker,
2784                                         xs_udp_setup_socket);
2785                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2786                 break;
2787         case AF_INET6:
2788                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2789                         xprt_set_bound(xprt);
2790
2791                 INIT_DELAYED_WORK(&transport->connect_worker,
2792                                         xs_udp_setup_socket);
2793                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2794                 break;
2795         default:
2796                 ret = ERR_PTR(-EAFNOSUPPORT);
2797                 goto out_err;
2798         }
2799
2800         if (xprt_bound(xprt))
2801                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2802                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2803                                 xprt->address_strings[RPC_DISPLAY_PORT],
2804                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2805         else
2806                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2807                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2808                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2809
2810         if (try_module_get(THIS_MODULE))
2811                 return xprt;
2812         ret = ERR_PTR(-EINVAL);
2813 out_err:
2814         xs_xprt_free(xprt);
2815         return ret;
2816 }
2817
2818 static const struct rpc_timeout xs_tcp_default_timeout = {
2819         .to_initval = 60 * HZ,
2820         .to_maxval = 60 * HZ,
2821         .to_retries = 2,
2822 };
2823
2824 /**
2825  * xs_setup_tcp - Set up transport to use a TCP socket
2826  * @args: rpc transport creation arguments
2827  *
2828  */
2829 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2830 {
2831         struct sockaddr *addr = args->dstaddr;
2832         struct rpc_xprt *xprt;
2833         struct sock_xprt *transport;
2834         struct rpc_xprt *ret;
2835         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2836
2837         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2838                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2839
2840         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2841                         max_slot_table_size);
2842         if (IS_ERR(xprt))
2843                 return xprt;
2844         transport = container_of(xprt, struct sock_xprt, xprt);
2845
2846         xprt->prot = IPPROTO_TCP;
2847         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2848         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2849
2850         xprt->bind_timeout = XS_BIND_TO;
2851         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2852         xprt->idle_timeout = XS_IDLE_DISC_TO;
2853
2854         xprt->ops = &xs_tcp_ops;
2855         xprt->timeout = &xs_tcp_default_timeout;
2856
2857         switch (addr->sa_family) {
2858         case AF_INET:
2859                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2860                         xprt_set_bound(xprt);
2861
2862                 INIT_DELAYED_WORK(&transport->connect_worker,
2863                                         xs_tcp_setup_socket);
2864                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2865                 break;
2866         case AF_INET6:
2867                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2868                         xprt_set_bound(xprt);
2869
2870                 INIT_DELAYED_WORK(&transport->connect_worker,
2871                                         xs_tcp_setup_socket);
2872                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2873                 break;
2874         default:
2875                 ret = ERR_PTR(-EAFNOSUPPORT);
2876                 goto out_err;
2877         }
2878
2879         if (xprt_bound(xprt))
2880                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2881                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2882                                 xprt->address_strings[RPC_DISPLAY_PORT],
2883                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2884         else
2885                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2886                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2887                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2888
2889         if (try_module_get(THIS_MODULE))
2890                 return xprt;
2891         ret = ERR_PTR(-EINVAL);
2892 out_err:
2893         xs_xprt_free(xprt);
2894         return ret;
2895 }
2896
2897 /**
2898  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2899  * @args: rpc transport creation arguments
2900  *
2901  */
2902 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2903 {
2904         struct sockaddr *addr = args->dstaddr;
2905         struct rpc_xprt *xprt;
2906         struct sock_xprt *transport;
2907         struct svc_sock *bc_sock;
2908         struct rpc_xprt *ret;
2909
2910         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2911                         xprt_tcp_slot_table_entries);
2912         if (IS_ERR(xprt))
2913                 return xprt;
2914         transport = container_of(xprt, struct sock_xprt, xprt);
2915
2916         xprt->prot = IPPROTO_TCP;
2917         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2918         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2919         xprt->timeout = &xs_tcp_default_timeout;
2920
2921         /* backchannel */
2922         xprt_set_bound(xprt);
2923         xprt->bind_timeout = 0;
2924         xprt->reestablish_timeout = 0;
2925         xprt->idle_timeout = 0;
2926
2927         xprt->ops = &bc_tcp_ops;
2928
2929         switch (addr->sa_family) {
2930         case AF_INET:
2931                 xs_format_peer_addresses(xprt, "tcp",
2932                                          RPCBIND_NETID_TCP);
2933                 break;
2934         case AF_INET6:
2935                 xs_format_peer_addresses(xprt, "tcp",
2936                                    RPCBIND_NETID_TCP6);
2937                 break;
2938         default:
2939                 ret = ERR_PTR(-EAFNOSUPPORT);
2940                 goto out_err;
2941         }
2942
2943         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2944                         xprt->address_strings[RPC_DISPLAY_ADDR],
2945                         xprt->address_strings[RPC_DISPLAY_PORT],
2946                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2947
2948         /*
2949          * Once we've associated a backchannel xprt with a connection,
2950          * we want to keep it around as long as the connection lasts,
2951          * in case we need to start using it for a backchannel again;
2952          * this reference won't be dropped until bc_xprt is destroyed.
2953          */
2954         xprt_get(xprt);
2955         args->bc_xprt->xpt_bc_xprt = xprt;
2956         xprt->bc_xprt = args->bc_xprt;
2957         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2958         transport->sock = bc_sock->sk_sock;
2959         transport->inet = bc_sock->sk_sk;
2960
2961         /*
2962          * Since we don't want connections for the backchannel, we set
2963          * the xprt status to connected
2964          */
2965         xprt_set_connected(xprt);
2966
2967         if (try_module_get(THIS_MODULE))
2968                 return xprt;
2969
2970         args->bc_xprt->xpt_bc_xprt = NULL;
2971         xprt_put(xprt);
2972         ret = ERR_PTR(-EINVAL);
2973 out_err:
2974         xs_xprt_free(xprt);
2975         return ret;
2976 }
2977
2978 static struct xprt_class        xs_local_transport = {
2979         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2980         .name           = "named UNIX socket",
2981         .owner          = THIS_MODULE,
2982         .ident          = XPRT_TRANSPORT_LOCAL,
2983         .setup          = xs_setup_local,
2984 };
2985
2986 static struct xprt_class        xs_udp_transport = {
2987         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2988         .name           = "udp",
2989         .owner          = THIS_MODULE,
2990         .ident          = XPRT_TRANSPORT_UDP,
2991         .setup          = xs_setup_udp,
2992 };
2993
2994 static struct xprt_class        xs_tcp_transport = {
2995         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2996         .name           = "tcp",
2997         .owner          = THIS_MODULE,
2998         .ident          = XPRT_TRANSPORT_TCP,
2999         .setup          = xs_setup_tcp,
3000 };
3001
3002 static struct xprt_class        xs_bc_tcp_transport = {
3003         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3004         .name           = "tcp NFSv4.1 backchannel",
3005         .owner          = THIS_MODULE,
3006         .ident          = XPRT_TRANSPORT_BC_TCP,
3007         .setup          = xs_setup_bc_tcp,
3008 };
3009
3010 /**
3011  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3012  *
3013  */
3014 int init_socket_xprt(void)
3015 {
3016 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3017         if (!sunrpc_table_header)
3018                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3019 #endif
3020
3021         xprt_register_transport(&xs_local_transport);
3022         xprt_register_transport(&xs_udp_transport);
3023         xprt_register_transport(&xs_tcp_transport);
3024         xprt_register_transport(&xs_bc_tcp_transport);
3025
3026         return 0;
3027 }
3028
3029 /**
3030  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3031  *
3032  */
3033 void cleanup_socket_xprt(void)
3034 {
3035 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3036         if (sunrpc_table_header) {
3037                 unregister_sysctl_table(sunrpc_table_header);
3038                 sunrpc_table_header = NULL;
3039         }
3040 #endif
3041
3042         xprt_unregister_transport(&xs_local_transport);
3043         xprt_unregister_transport(&xs_udp_transport);
3044         xprt_unregister_transport(&xs_tcp_transport);
3045         xprt_unregister_transport(&xs_bc_tcp_transport);
3046 }
3047
3048 static int param_set_uint_minmax(const char *val,
3049                 const struct kernel_param *kp,
3050                 unsigned int min, unsigned int max)
3051 {
3052         unsigned int num;
3053         int ret;
3054
3055         if (!val)
3056                 return -EINVAL;
3057         ret = kstrtouint(val, 0, &num);
3058         if (ret == -EINVAL || num < min || num > max)
3059                 return -EINVAL;
3060         *((unsigned int *)kp->arg) = num;
3061         return 0;
3062 }
3063
3064 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3065 {
3066         return param_set_uint_minmax(val, kp,
3067                         RPC_MIN_RESVPORT,
3068                         RPC_MAX_RESVPORT);
3069 }
3070
3071 static const struct kernel_param_ops param_ops_portnr = {
3072         .set = param_set_portnr,
3073         .get = param_get_uint,
3074 };
3075
3076 #define param_check_portnr(name, p) \
3077         __param_check(name, p, unsigned int);
3078
3079 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3080 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3081
3082 static int param_set_slot_table_size(const char *val,
3083                                      const struct kernel_param *kp)
3084 {
3085         return param_set_uint_minmax(val, kp,
3086                         RPC_MIN_SLOT_TABLE,
3087                         RPC_MAX_SLOT_TABLE);
3088 }
3089
3090 static const struct kernel_param_ops param_ops_slot_table_size = {
3091         .set = param_set_slot_table_size,
3092         .get = param_get_uint,
3093 };
3094
3095 #define param_check_slot_table_size(name, p) \
3096         __param_check(name, p, unsigned int);
3097
3098 static int param_set_max_slot_table_size(const char *val,
3099                                      const struct kernel_param *kp)
3100 {
3101         return param_set_uint_minmax(val, kp,
3102                         RPC_MIN_SLOT_TABLE,
3103                         RPC_MAX_SLOT_TABLE_LIMIT);
3104 }
3105
3106 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3107         .set = param_set_max_slot_table_size,
3108         .get = param_get_uint,
3109 };
3110
3111 #define param_check_max_slot_table_size(name, p) \
3112         __param_check(name, p, unsigned int);
3113
3114 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3115                    slot_table_size, 0644);
3116 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3117                    max_slot_table_size, 0644);
3118 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3119                    slot_table_size, 0644);
3120