net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68 #define XS_TCP_LINGER_TO        (15U * HZ)
  69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71 /*
  72  * We can register our own files under /proc/sys/sunrpc by
  73  * calling register_sysctl_table() again.  The files in that
  74  * directory become the union of all files registered there.
  75  *
  76  * We simply need to make sure that we don't collide with
  77  * someone else's file names!
  78  */
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 181
 182 /*
 183  * TCP idle timeout; client drops the transport socket if it is idle
 184  * for this long.  Note that we also timeout UDP sockets to prevent
 185  * holding port numbers when there is no RPC traffic.
 186  */
 187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 188
 189 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 190 # undef  RPC_DEBUG_DATA
 191 # define RPCDBG_FACILITY        RPCDBG_TRANS
 192 #endif
 193
 194 #ifdef RPC_DEBUG_DATA
 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196 {
 197         u8 *buf = (u8 *) packet;
 198         int j;
 199
 200         dprintk("RPC:       %s\n", msg);
 201         for (j = 0; j < count && j < 128; j += 4) {
 202                 if (!(j & 31)) {
 203                         if (j)
 204                                 dprintk("\n");
 205                         dprintk("0x%04x ", j);
 206                 }
 207                 dprintk("%02x%02x%02x%02x ",
 208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209         }
 210         dprintk("\n");
 211 }
 212 #else
 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214 {
 215         /* NOP */
 216 }
 217 #endif
 218
 219 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 220 {
 221         return (struct rpc_xprt *) sk->sk_user_data;
 222 }
 223
 224 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 225 {
 226         return (struct sockaddr *) &xprt->addr;
 227 }
 228
 229 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 230 {
 231         return (struct sockaddr_un *) &xprt->addr;
 232 }
 233
 234 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 235 {
 236         return (struct sockaddr_in *) &xprt->addr;
 237 }
 238
 239 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 240 {
 241         return (struct sockaddr_in6 *) &xprt->addr;
 242 }
 243
 244 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 245 {
 246         struct sockaddr *sap = xs_addr(xprt);
 247         struct sockaddr_in6 *sin6;
 248         struct sockaddr_in *sin;
 249         struct sockaddr_un *sun;
 250         char buf[128];
 251
 252         switch (sap->sa_family) {
 253         case AF_LOCAL:
 254                 sun = xs_addr_un(xprt);
 255                 strlcpy(buf, sun->sun_path, sizeof(buf));
 256                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 257                                                 kstrdup(buf, GFP_KERNEL);
 258                 break;
 259         case AF_INET:
 260                 (void)rpc_ntop(sap, buf, sizeof(buf));
 261                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 262                                                 kstrdup(buf, GFP_KERNEL);
 263                 sin = xs_addr_in(xprt);
 264                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 265                 break;
 266         case AF_INET6:
 267                 (void)rpc_ntop(sap, buf, sizeof(buf));
 268                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 269                                                 kstrdup(buf, GFP_KERNEL);
 270                 sin6 = xs_addr_in6(xprt);
 271                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 272                 break;
 273         default:
 274                 BUG();
 275         }
 276
 277         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 278 }
 279
 280 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 281 {
 282         struct sockaddr *sap = xs_addr(xprt);
 283         char buf[128];
 284
 285         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 286         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 287
 288         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 289         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 290 }
 291
 292 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 293                                      const char *protocol,
 294                                      const char *netid)
 295 {
 296         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 297         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 298         xs_format_common_peer_addresses(xprt);
 299         xs_format_common_peer_ports(xprt);
 300 }
 301
 302 static void xs_update_peer_port(struct rpc_xprt *xprt)
 303 {
 304         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 305         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 306
 307         xs_format_common_peer_ports(xprt);
 308 }
 309
 310 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 311 {
 312         unsigned int i;
 313
 314         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 315                 switch (i) {
 316                 case RPC_DISPLAY_PROTO:
 317                 case RPC_DISPLAY_NETID:
 318                         continue;
 319                 default:
 320                         kfree(xprt->address_strings[i]);
 321                 }
 322 }
 323
 324 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 325
 326 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 327 {
 328         struct msghdr msg = {
 329                 .msg_name       = addr,
 330                 .msg_namelen    = addrlen,
 331                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 332         };
 333         struct kvec iov = {
 334                 .iov_base       = vec->iov_base + base,
 335                 .iov_len        = vec->iov_len - base,
 336         };
 337
 338         if (iov.iov_len != 0)
 339                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 340         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 341 }
 342
 343 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 344 {
 345         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 346                         int offset, size_t size, int flags);
 347         struct page **ppage;
 348         unsigned int remainder;
 349         int err;
 350
 351         remainder = xdr->page_len - base;
 352         base += xdr->page_base;
 353         ppage = xdr->pages + (base >> PAGE_SHIFT);
 354         base &= ~PAGE_MASK;
 355         do_sendpage = sock->ops->sendpage;
 356         if (!zerocopy)
 357                 do_sendpage = sock_no_sendpage;
 358         for(;;) {
 359                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 360                 int flags = XS_SENDMSG_FLAGS;
 361
 362                 remainder -= len;
 363                 if (remainder != 0 || more)
 364                         flags |= MSG_MORE;
 365                 err = do_sendpage(sock, *ppage, base, len, flags);
 366                 if (remainder == 0 || err != len)
 367                         break;
 368                 *sent_p += err;
 369                 ppage++;
 370                 base = 0;
 371         }
 372         if (err > 0) {
 373                 *sent_p += err;
 374                 err = 0;
 375         }
 376         return err;
 377 }
 378
 379 /**
 380  * xs_sendpages - write pages directly to a socket
 381  * @sock: socket to send on
 382  * @addr: UDP only -- address of destination
 383  * @addrlen: UDP only -- length of destination address
 384  * @xdr: buffer containing this request
 385  * @base: starting position in the buffer
 386  * @zerocopy: true if it is safe to use sendpage()
 387  * @sent_p: return the total number of bytes successfully queued for sending
 388  *
 389  */
 390 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 391 {
 392         unsigned int remainder = xdr->len - base;
 393         int err = 0;
 394         int sent = 0;
 395
 396         if (unlikely(!sock))
 397                 return -ENOTSOCK;
 398
 399         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 400         if (base != 0) {
 401                 addr = NULL;
 402                 addrlen = 0;
 403         }
 404
 405         if (base < xdr->head[0].iov_len || addr != NULL) {
 406                 unsigned int len = xdr->head[0].iov_len - base;
 407                 remainder -= len;
 408                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 409                 if (remainder == 0 || err != len)
 410                         goto out;
 411                 *sent_p += err;
 412                 base = 0;
 413         } else
 414                 base -= xdr->head[0].iov_len;
 415
 416         if (base < xdr->page_len) {
 417                 unsigned int len = xdr->page_len - base;
 418                 remainder -= len;
 419                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 420                 *sent_p += sent;
 421                 if (remainder == 0 || sent != len)
 422                         goto out;
 423                 base = 0;
 424         } else
 425                 base -= xdr->page_len;
 426
 427         if (base >= xdr->tail[0].iov_len)
 428                 return 0;
 429         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 430 out:
 431         if (err > 0) {
 432                 *sent_p += err;
 433                 err = 0;
 434         }
 435         return err;
 436 }
 437
 438 static void xs_nospace_callback(struct rpc_task *task)
 439 {
 440         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 441
 442         transport->inet->sk_write_pending--;
 443         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 444 }
 445
 446 /**
 447  * xs_nospace - place task on wait queue if transmit was incomplete
 448  * @task: task to put to sleep
 449  *
 450  */
 451 static int xs_nospace(struct rpc_task *task)
 452 {
 453         struct rpc_rqst *req = task->tk_rqstp;
 454         struct rpc_xprt *xprt = req->rq_xprt;
 455         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 456         struct sock *sk = transport->inet;
 457         int ret = -EAGAIN;
 458
 459         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 460                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 461                         req->rq_slen);
 462
 463         /* Protect against races with write_space */
 464         spin_lock_bh(&xprt->transport_lock);
 465
 466         /* Don't race with disconnect */
 467         if (xprt_connected(xprt)) {
 468                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 469                         /*
 470                          * Notify TCP that we're limited by the application
 471                          * window size
 472                          */
 473                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 474                         sk->sk_write_pending++;
 475                         /* ...and wait for more buffer space */
 476                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 477                 }
 478         } else {
 479                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 480                 ret = -ENOTCONN;
 481         }
 482
 483         spin_unlock_bh(&xprt->transport_lock);
 484
 485         /* Race breaker in case memory is freed before above code is called */
 486         sk->sk_write_space(sk);
 487         return ret;
 488 }
 489
 490 /*
 491  * Construct a stream transport record marker in @buf.
 492  */
 493 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 494 {
 495         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 496         rpc_fraghdr *base = buf->head[0].iov_base;
 497         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 498 }
 499
 500 /**
 501  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 502  * @task: RPC task that manages the state of an RPC request
 503  *
 504  * Return values:
 505  *        0:    The request has been sent
 506  *   EAGAIN:    The socket was blocked, please call again later to
 507  *              complete the request
 508  * ENOTCONN:    Caller needs to invoke connect logic then call again
 509  *    other:    Some other error occured, the request was not sent
 510  */
 511 static int xs_local_send_request(struct rpc_task *task)
 512 {
 513         struct rpc_rqst *req = task->tk_rqstp;
 514         struct rpc_xprt *xprt = req->rq_xprt;
 515         struct sock_xprt *transport =
 516                                 container_of(xprt, struct sock_xprt, xprt);
 517         struct xdr_buf *xdr = &req->rq_snd_buf;
 518         int status;
 519         int sent = 0;
 520
 521         xs_encode_stream_record_marker(&req->rq_snd_buf);
 522
 523         xs_pktdump("packet data:",
 524                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 525
 526         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 527                               true, &sent);
 528         dprintk("RPC:       %s(%u) = %d\n",
 529                         __func__, xdr->len - req->rq_bytes_sent, status);
 530
 531         if (status == -EAGAIN && sock_writeable(transport->inet))
 532                 status = -ENOBUFS;
 533
 534         if (likely(sent > 0) || status == 0) {
 535                 req->rq_bytes_sent += sent;
 536                 req->rq_xmit_bytes_sent += sent;
 537                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 538                         req->rq_bytes_sent = 0;
 539                         return 0;
 540                 }
 541                 status = -EAGAIN;
 542         }
 543
 544         switch (status) {
 545         case -ENOBUFS:
 546                 break;
 547         case -EAGAIN:
 548                 status = xs_nospace(task);
 549                 break;
 550         default:
 551                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 552                         -status);
 553         case -EPIPE:
 554                 xs_close(xprt);
 555                 status = -ENOTCONN;
 556         }
 557
 558         return status;
 559 }
 560
 561 /**
 562  * xs_udp_send_request - write an RPC request to a UDP socket
 563  * @task: address of RPC task that manages the state of an RPC request
 564  *
 565  * Return values:
 566  *        0:    The request has been sent
 567  *   EAGAIN:    The socket was blocked, please call again later to
 568  *              complete the request
 569  * ENOTCONN:    Caller needs to invoke connect logic then call again
 570  *    other:    Some other error occurred, the request was not sent
 571  */
 572 static int xs_udp_send_request(struct rpc_task *task)
 573 {
 574         struct rpc_rqst *req = task->tk_rqstp;
 575         struct rpc_xprt *xprt = req->rq_xprt;
 576         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 577         struct xdr_buf *xdr = &req->rq_snd_buf;
 578         int sent = 0;
 579         int status;
 580
 581         xs_pktdump("packet data:",
 582                                 req->rq_svec->iov_base,
 583                                 req->rq_svec->iov_len);
 584
 585         if (!xprt_bound(xprt))
 586                 return -ENOTCONN;
 587         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 588                               xdr, req->rq_bytes_sent, true, &sent);
 589
 590         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 591                         xdr->len - req->rq_bytes_sent, status);
 592
 593         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
 594         if (status == -EPERM)
 595                 goto process_status;
 596
 597         if (status == -EAGAIN && sock_writeable(transport->inet))
 598                 status = -ENOBUFS;
 599
 600         if (sent > 0 || status == 0) {
 601                 req->rq_xmit_bytes_sent += sent;
 602                 if (sent >= req->rq_slen)
 603                         return 0;
 604                 /* Still some bytes left; set up for a retry later. */
 605                 status = -EAGAIN;
 606         }
 607
 608 process_status:
 609         switch (status) {
 610         case -ENOTSOCK:
 611                 status = -ENOTCONN;
 612                 /* Should we call xs_close() here? */
 613                 break;
 614         case -EAGAIN:
 615                 status = xs_nospace(task);
 616                 break;
 617         default:
 618                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 619                         -status);
 620         case -ENETUNREACH:
 621         case -ENOBUFS:
 622         case -EPIPE:
 623         case -ECONNREFUSED:
 624         case -EPERM:
 625                 /* When the server has died, an ICMP port unreachable message
 626                  * prompts ECONNREFUSED. */
 627                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 628         }
 629
 630         return status;
 631 }
 632
 633 /**
 634  * xs_tcp_send_request - write an RPC request to a TCP socket
 635  * @task: address of RPC task that manages the state of an RPC request
 636  *
 637  * Return values:
 638  *        0:    The request has been sent
 639  *   EAGAIN:    The socket was blocked, please call again later to
 640  *              complete the request
 641  * ENOTCONN:    Caller needs to invoke connect logic then call again
 642  *    other:    Some other error occurred, the request was not sent
 643  *
 644  * XXX: In the case of soft timeouts, should we eventually give up
 645  *      if sendmsg is not able to make progress?
 646  */
 647 static int xs_tcp_send_request(struct rpc_task *task)
 648 {
 649         struct rpc_rqst *req = task->tk_rqstp;
 650         struct rpc_xprt *xprt = req->rq_xprt;
 651         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 652         struct xdr_buf *xdr = &req->rq_snd_buf;
 653         bool zerocopy = true;
 654         int status;
 655         int sent;
 656
 657         xs_encode_stream_record_marker(&req->rq_snd_buf);
 658
 659         xs_pktdump("packet data:",
 660                                 req->rq_svec->iov_base,
 661                                 req->rq_svec->iov_len);
 662         /* Don't use zero copy if this is a resend. If the RPC call
 663          * completes while the socket holds a reference to the pages,
 664          * then we may end up resending corrupted data.
 665          */
 666         if (task->tk_flags & RPC_TASK_SENT)
 667                 zerocopy = false;
 668
 669         /* Continue transmitting the packet/record. We must be careful
 670          * to cope with writespace callbacks arriving _after_ we have
 671          * called sendmsg(). */
 672         while (1) {
 673                 sent = 0;
 674                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
 675                                       req->rq_bytes_sent, zerocopy, &sent);
 676
 677                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 678                                 xdr->len - req->rq_bytes_sent, status);
 679
 680                 /* If we've sent the entire packet, immediately
 681                  * reset the count of bytes sent. */
 682                 req->rq_bytes_sent += sent;
 683                 req->rq_xmit_bytes_sent += sent;
 684                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 685                         req->rq_bytes_sent = 0;
 686                         return 0;
 687                 }
 688
 689                 if (status < 0)
 690                         break;
 691                 if (sent == 0) {
 692                         status = -EAGAIN;
 693                         break;
 694                 }
 695         }
 696         if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
 697                 status = -ENOBUFS;
 698
 699         switch (status) {
 700         case -ENOTSOCK:
 701                 status = -ENOTCONN;
 702                 /* Should we call xs_close() here? */
 703                 break;
 704         case -EAGAIN:
 705                 status = xs_nospace(task);
 706                 break;
 707         default:
 708                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 709                         -status);
 710         case -ECONNRESET:
 711         case -ECONNREFUSED:
 712         case -ENOTCONN:
 713         case -EADDRINUSE:
 714         case -ENOBUFS:
 715         case -EPIPE:
 716                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 717         }
 718
 719         return status;
 720 }
 721
 722 /**
 723  * xs_tcp_release_xprt - clean up after a tcp transmission
 724  * @xprt: transport
 725  * @task: rpc task
 726  *
 727  * This cleans up if an error causes us to abort the transmission of a request.
 728  * In this case, the socket may need to be reset in order to avoid confusing
 729  * the server.
 730  */
 731 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 732 {
 733         struct rpc_rqst *req;
 734
 735         if (task != xprt->snd_task)
 736                 return;
 737         if (task == NULL)
 738                 goto out_release;
 739         req = task->tk_rqstp;
 740         if (req == NULL)
 741                 goto out_release;
 742         if (req->rq_bytes_sent == 0)
 743                 goto out_release;
 744         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 745                 goto out_release;
 746         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 747 out_release:
 748         xprt_release_xprt(xprt, task);
 749 }
 750
 751 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 752 {
 753         transport->old_data_ready = sk->sk_data_ready;
 754         transport->old_state_change = sk->sk_state_change;
 755         transport->old_write_space = sk->sk_write_space;
 756         transport->old_error_report = sk->sk_error_report;
 757 }
 758
 759 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 760 {
 761         sk->sk_data_ready = transport->old_data_ready;
 762         sk->sk_state_change = transport->old_state_change;
 763         sk->sk_write_space = transport->old_write_space;
 764         sk->sk_error_report = transport->old_error_report;
 765 }
 766
 767 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 768 {
 769         smp_mb__before_atomic();
 770         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 771         clear_bit(XPRT_CLOSING, &xprt->state);
 772         smp_mb__after_atomic();
 773 }
 774
 775 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 776 {
 777         xs_sock_reset_connection_flags(xprt);
 778         /* Mark transport as closed and wake up all pending tasks */
 779         xprt_disconnect_done(xprt);
 780 }
 781
 782 /**
 783  * xs_error_report - callback to handle TCP socket state errors
 784  * @sk: socket
 785  *
 786  * Note: we don't call sock_error() since there may be a rpc_task
 787  * using the socket, and so we don't want to clear sk->sk_err.
 788  */
 789 static void xs_error_report(struct sock *sk)
 790 {
 791         struct rpc_xprt *xprt;
 792         int err;
 793
 794         read_lock_bh(&sk->sk_callback_lock);
 795         if (!(xprt = xprt_from_sock(sk)))
 796                 goto out;
 797
 798         err = -sk->sk_err;
 799         if (err == 0)
 800                 goto out;
 801         /* Is this a reset event? */
 802         if (sk->sk_state == TCP_CLOSE)
 803                 xs_sock_mark_closed(xprt);
 804         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 805                         xprt, -err);
 806         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 807         xprt_wake_pending_tasks(xprt, err);
 808  out:
 809         read_unlock_bh(&sk->sk_callback_lock);
 810 }
 811
 812 static void xs_reset_transport(struct sock_xprt *transport)
 813 {
 814         struct socket *sock = transport->sock;
 815         struct sock *sk = transport->inet;
 816         struct rpc_xprt *xprt = &transport->xprt;
 817
 818         if (sk == NULL)
 819                 return;
 820
 821         if (atomic_read(&transport->xprt.swapper))
 822                 sk_clear_memalloc(sk);
 823
 824         kernel_sock_shutdown(sock, SHUT_RDWR);
 825
 826         write_lock_bh(&sk->sk_callback_lock);
 827         transport->inet = NULL;
 828         transport->sock = NULL;
 829
 830         sk->sk_user_data = NULL;
 831
 832         xs_restore_old_callbacks(transport, sk);
 833         xprt_clear_connected(xprt);
 834         write_unlock_bh(&sk->sk_callback_lock);
 835         xs_sock_reset_connection_flags(xprt);
 836
 837         trace_rpc_socket_close(xprt, sock);
 838         sock_release(sock);
 839 }
 840
 841 /**
 842  * xs_close - close a socket
 843  * @xprt: transport
 844  *
 845  * This is used when all requests are complete; ie, no DRC state remains
 846  * on the server we want to save.
 847  *
 848  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 849  * xs_reset_transport() zeroing the socket from underneath a writer.
 850  */
 851 static void xs_close(struct rpc_xprt *xprt)
 852 {
 853         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 854
 855         dprintk("RPC:       xs_close xprt %p\n", xprt);
 856
 857         xs_reset_transport(transport);
 858         xprt->reestablish_timeout = 0;
 859
 860         xprt_disconnect_done(xprt);
 861 }
 862
 863 static void xs_inject_disconnect(struct rpc_xprt *xprt)
 864 {
 865         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 866                 xprt);
 867         xprt_disconnect_done(xprt);
 868 }
 869
 870 static void xs_xprt_free(struct rpc_xprt *xprt)
 871 {
 872         xs_free_peer_addresses(xprt);
 873         xprt_free(xprt);
 874 }
 875
 876 /**
 877  * xs_destroy - prepare to shutdown a transport
 878  * @xprt: doomed transport
 879  *
 880  */
 881 static void xs_destroy(struct rpc_xprt *xprt)
 882 {
 883         struct sock_xprt *transport = container_of(xprt,
 884                         struct sock_xprt, xprt);
 885         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 886
 887         cancel_delayed_work_sync(&transport->connect_worker);
 888         xs_close(xprt);
 889         xs_xprt_free(xprt);
 890         module_put(THIS_MODULE);
 891 }
 892
 893 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 894 {
 895         struct xdr_skb_reader desc = {
 896                 .skb            = skb,
 897                 .offset         = sizeof(rpc_fraghdr),
 898                 .count          = skb->len - sizeof(rpc_fraghdr),
 899         };
 900
 901         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 902                 return -1;
 903         if (desc.count)
 904                 return -1;
 905         return 0;
 906 }
 907
 908 /**
 909  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 910  * @sk: socket with data to read
 911  *
 912  * Currently this assumes we can read the whole reply in a single gulp.
 913  */
 914 static void xs_local_data_ready(struct sock *sk)
 915 {
 916         struct rpc_task *task;
 917         struct rpc_xprt *xprt;
 918         struct rpc_rqst *rovr;
 919         struct sk_buff *skb;
 920         int err, repsize, copied;
 921         u32 _xid;
 922         __be32 *xp;
 923
 924         read_lock_bh(&sk->sk_callback_lock);
 925         dprintk("RPC:       %s...\n", __func__);
 926         xprt = xprt_from_sock(sk);
 927         if (xprt == NULL)
 928                 goto out;
 929
 930         skb = skb_recv_datagram(sk, 0, 1, &err);
 931         if (skb == NULL)
 932                 goto out;
 933
 934         repsize = skb->len - sizeof(rpc_fraghdr);
 935         if (repsize < 4) {
 936                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 937                 goto dropit;
 938         }
 939
 940         /* Copy the XID from the skb... */
 941         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 942         if (xp == NULL)
 943                 goto dropit;
 944
 945         /* Look up and lock the request corresponding to the given XID */
 946         spin_lock(&xprt->transport_lock);
 947         rovr = xprt_lookup_rqst(xprt, *xp);
 948         if (!rovr)
 949                 goto out_unlock;
 950         task = rovr->rq_task;
 951
 952         copied = rovr->rq_private_buf.buflen;
 953         if (copied > repsize)
 954                 copied = repsize;
 955
 956         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 957                 dprintk("RPC:       sk_buff copy failed\n");
 958                 goto out_unlock;
 959         }
 960
 961         xprt_complete_rqst(task, copied);
 962
 963  out_unlock:
 964         spin_unlock(&xprt->transport_lock);
 965  dropit:
 966         skb_free_datagram(sk, skb);
 967  out:
 968         read_unlock_bh(&sk->sk_callback_lock);
 969 }
 970
 971 /**
 972  * xs_udp_data_ready - "data ready" callback for UDP sockets
 973  * @sk: socket with data to read
 974  *
 975  */
 976 static void xs_udp_data_ready(struct sock *sk)
 977 {
 978         struct rpc_task *task;
 979         struct rpc_xprt *xprt;
 980         struct rpc_rqst *rovr;
 981         struct sk_buff *skb;
 982         int err, repsize, copied;
 983         u32 _xid;
 984         __be32 *xp;
 985
 986         read_lock_bh(&sk->sk_callback_lock);
 987         dprintk("RPC:       xs_udp_data_ready...\n");
 988         if (!(xprt = xprt_from_sock(sk)))
 989                 goto out;
 990
 991         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 992                 goto out;
 993
 994         repsize = skb->len - sizeof(struct udphdr);
 995         if (repsize < 4) {
 996                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 997                 goto dropit;
 998         }
 999
1000         /* Copy the XID from the skb... */
1001         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1002                                 sizeof(_xid), &_xid);
1003         if (xp == NULL)
1004                 goto dropit;
1005
1006         /* Look up and lock the request corresponding to the given XID */
1007         spin_lock(&xprt->transport_lock);
1008         rovr = xprt_lookup_rqst(xprt, *xp);
1009         if (!rovr)
1010                 goto out_unlock;
1011         task = rovr->rq_task;
1012
1013         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1014                 copied = repsize;
1015
1016         /* Suck it into the iovec, verify checksum if not done by hw. */
1017         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1018                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1019                 goto out_unlock;
1020         }
1021
1022         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1023
1024         xprt_adjust_cwnd(xprt, task, copied);
1025         xprt_complete_rqst(task, copied);
1026
1027  out_unlock:
1028         spin_unlock(&xprt->transport_lock);
1029  dropit:
1030         skb_free_datagram(sk, skb);
1031  out:
1032         read_unlock_bh(&sk->sk_callback_lock);
1033 }
1034
1035 /*
1036  * Helper function to force a TCP close if the server is sending
1037  * junk and/or it has put us in CLOSE_WAIT
1038  */
1039 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1040 {
1041         xprt_force_disconnect(xprt);
1042 }
1043
1044 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1045 {
1046         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1047         size_t len, used;
1048         char *p;
1049
1050         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1051         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1052         used = xdr_skb_read_bits(desc, p, len);
1053         transport->tcp_offset += used;
1054         if (used != len)
1055                 return;
1056
1057         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1058         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1059                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1060         else
1061                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1062         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1063
1064         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1065         transport->tcp_offset = 0;
1066
1067         /* Sanity check of the record length */
1068         if (unlikely(transport->tcp_reclen < 8)) {
1069                 dprintk("RPC:       invalid TCP record fragment length\n");
1070                 xs_tcp_force_close(xprt);
1071                 return;
1072         }
1073         dprintk("RPC:       reading TCP record fragment of length %d\n",
1074                         transport->tcp_reclen);
1075 }
1076
1077 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1078 {
1079         if (transport->tcp_offset == transport->tcp_reclen) {
1080                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1081                 transport->tcp_offset = 0;
1082                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1083                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1084                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1085                         transport->tcp_copied = 0;
1086                 }
1087         }
1088 }
1089
1090 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1091 {
1092         size_t len, used;
1093         char *p;
1094
1095         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1096         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1097         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1098         used = xdr_skb_read_bits(desc, p, len);
1099         transport->tcp_offset += used;
1100         if (used != len)
1101                 return;
1102         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1103         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1104         transport->tcp_copied = 4;
1105         dprintk("RPC:       reading %s XID %08x\n",
1106                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1107                                                               : "request with",
1108                         ntohl(transport->tcp_xid));
1109         xs_tcp_check_fraghdr(transport);
1110 }
1111
1112 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1113                                        struct xdr_skb_reader *desc)
1114 {
1115         size_t len, used;
1116         u32 offset;
1117         char *p;
1118
1119         /*
1120          * We want transport->tcp_offset to be 8 at the end of this routine
1121          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1122          * When this function is called for the first time,
1123          * transport->tcp_offset is 4 (after having already read the xid).
1124          */
1125         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1126         len = sizeof(transport->tcp_calldir) - offset;
1127         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1128         p = ((char *) &transport->tcp_calldir) + offset;
1129         used = xdr_skb_read_bits(desc, p, len);
1130         transport->tcp_offset += used;
1131         if (used != len)
1132                 return;
1133         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1134         /*
1135          * We don't yet have the XDR buffer, so we will write the calldir
1136          * out after we get the buffer from the 'struct rpc_rqst'
1137          */
1138         switch (ntohl(transport->tcp_calldir)) {
1139         case RPC_REPLY:
1140                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1141                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1142                 transport->tcp_flags |= TCP_RPC_REPLY;
1143                 break;
1144         case RPC_CALL:
1145                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1146                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1147                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1148                 break;
1149         default:
1150                 dprintk("RPC:       invalid request message type\n");
1151                 xs_tcp_force_close(&transport->xprt);
1152         }
1153         xs_tcp_check_fraghdr(transport);
1154 }
1155
1156 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1157                                      struct xdr_skb_reader *desc,
1158                                      struct rpc_rqst *req)
1159 {
1160         struct sock_xprt *transport =
1161                                 container_of(xprt, struct sock_xprt, xprt);
1162         struct xdr_buf *rcvbuf;
1163         size_t len;
1164         ssize_t r;
1165
1166         rcvbuf = &req->rq_private_buf;
1167
1168         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1169                 /*
1170                  * Save the RPC direction in the XDR buffer
1171                  */
1172                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1173                         &transport->tcp_calldir,
1174                         sizeof(transport->tcp_calldir));
1175                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1176                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1177         }
1178
1179         len = desc->count;
1180         if (len > transport->tcp_reclen - transport->tcp_offset) {
1181                 struct xdr_skb_reader my_desc;
1182
1183                 len = transport->tcp_reclen - transport->tcp_offset;
1184                 memcpy(&my_desc, desc, sizeof(my_desc));
1185                 my_desc.count = len;
1186                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1187                                           &my_desc, xdr_skb_read_bits);
1188                 desc->count -= r;
1189                 desc->offset += r;
1190         } else
1191                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1192                                           desc, xdr_skb_read_bits);
1193
1194         if (r > 0) {
1195                 transport->tcp_copied += r;
1196                 transport->tcp_offset += r;
1197         }
1198         if (r != len) {
1199                 /* Error when copying to the receive buffer,
1200                  * usually because we weren't able to allocate
1201                  * additional buffer pages. All we can do now
1202                  * is turn off TCP_RCV_COPY_DATA, so the request
1203                  * will not receive any additional updates,
1204                  * and time out.
1205                  * Any remaining data from this record will
1206                  * be discarded.
1207                  */
1208                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1209                 dprintk("RPC:       XID %08x truncated request\n",
1210                                 ntohl(transport->tcp_xid));
1211                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1212                                 "tcp_offset = %u, tcp_reclen = %u\n",
1213                                 xprt, transport->tcp_copied,
1214                                 transport->tcp_offset, transport->tcp_reclen);
1215                 return;
1216         }
1217
1218         dprintk("RPC:       XID %08x read %Zd bytes\n",
1219                         ntohl(transport->tcp_xid), r);
1220         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1221                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1222                         transport->tcp_offset, transport->tcp_reclen);
1223
1224         if (transport->tcp_copied == req->rq_private_buf.buflen)
1225                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1226         else if (transport->tcp_offset == transport->tcp_reclen) {
1227                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1228                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1229         }
1230 }
1231
1232 /*
1233  * Finds the request corresponding to the RPC xid and invokes the common
1234  * tcp read code to read the data.
1235  */
1236 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1237                                     struct xdr_skb_reader *desc)
1238 {
1239         struct sock_xprt *transport =
1240                                 container_of(xprt, struct sock_xprt, xprt);
1241         struct rpc_rqst *req;
1242
1243         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1244
1245         /* Find and lock the request corresponding to this xid */
1246         spin_lock(&xprt->transport_lock);
1247         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1248         if (!req) {
1249                 dprintk("RPC:       XID %08x request not found!\n",
1250                                 ntohl(transport->tcp_xid));
1251                 spin_unlock(&xprt->transport_lock);
1252                 return -1;
1253         }
1254
1255         xs_tcp_read_common(xprt, desc, req);
1256
1257         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1258                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1259
1260         spin_unlock(&xprt->transport_lock);
1261         return 0;
1262 }
1263
1264 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1265 /*
1266  * Obtains an rpc_rqst previously allocated and invokes the common
1267  * tcp read code to read the data.  The result is placed in the callback
1268  * queue.
1269  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1270  * connection and return -1.
1271  */
1272 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1273                                        struct xdr_skb_reader *desc)
1274 {
1275         struct sock_xprt *transport =
1276                                 container_of(xprt, struct sock_xprt, xprt);
1277         struct rpc_rqst *req;
1278
1279         /* Look up and lock the request corresponding to the given XID */
1280         spin_lock(&xprt->transport_lock);
1281         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1282         if (req == NULL) {
1283                 spin_unlock(&xprt->transport_lock);
1284                 printk(KERN_WARNING "Callback slot table overflowed\n");
1285                 xprt_force_disconnect(xprt);
1286                 return -1;
1287         }
1288
1289         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1290         xs_tcp_read_common(xprt, desc, req);
1291
1292         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1293                 xprt_complete_bc_request(req, transport->tcp_copied);
1294         spin_unlock(&xprt->transport_lock);
1295
1296         return 0;
1297 }
1298
1299 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1300                                         struct xdr_skb_reader *desc)
1301 {
1302         struct sock_xprt *transport =
1303                                 container_of(xprt, struct sock_xprt, xprt);
1304
1305         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1306                 xs_tcp_read_reply(xprt, desc) :
1307                 xs_tcp_read_callback(xprt, desc);
1308 }
1309
1310 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1311 {
1312         int ret;
1313
1314         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1315                               SVC_SOCK_ANONYMOUS);
1316         if (ret < 0)
1317                 return ret;
1318         return 0;
1319 }
1320 #else
1321 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1322                                         struct xdr_skb_reader *desc)
1323 {
1324         return xs_tcp_read_reply(xprt, desc);
1325 }
1326 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1327
1328 /*
1329  * Read data off the transport.  This can be either an RPC_CALL or an
1330  * RPC_REPLY.  Relay the processing to helper functions.
1331  */
1332 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1333                                     struct xdr_skb_reader *desc)
1334 {
1335         struct sock_xprt *transport =
1336                                 container_of(xprt, struct sock_xprt, xprt);
1337
1338         if (_xs_tcp_read_data(xprt, desc) == 0)
1339                 xs_tcp_check_fraghdr(transport);
1340         else {
1341                 /*
1342                  * The transport_lock protects the request handling.
1343                  * There's no need to hold it to update the tcp_flags.
1344                  */
1345                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1346         }
1347 }
1348
1349 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1350 {
1351         size_t len;
1352
1353         len = transport->tcp_reclen - transport->tcp_offset;
1354         if (len > desc->count)
1355                 len = desc->count;
1356         desc->count -= len;
1357         desc->offset += len;
1358         transport->tcp_offset += len;
1359         dprintk("RPC:       discarded %Zu bytes\n", len);
1360         xs_tcp_check_fraghdr(transport);
1361 }
1362
1363 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1364 {
1365         struct rpc_xprt *xprt = rd_desc->arg.data;
1366         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1367         struct xdr_skb_reader desc = {
1368                 .skb    = skb,
1369                 .offset = offset,
1370                 .count  = len,
1371         };
1372
1373         dprintk("RPC:       xs_tcp_data_recv started\n");
1374         do {
1375                 trace_xs_tcp_data_recv(transport);
1376                 /* Read in a new fragment marker if necessary */
1377                 /* Can we ever really expect to get completely empty fragments? */
1378                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1379                         xs_tcp_read_fraghdr(xprt, &desc);
1380                         continue;
1381                 }
1382                 /* Read in the xid if necessary */
1383                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1384                         xs_tcp_read_xid(transport, &desc);
1385                         continue;
1386                 }
1387                 /* Read in the call/reply flag */
1388                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1389                         xs_tcp_read_calldir(transport, &desc);
1390                         continue;
1391                 }
1392                 /* Read in the request data */
1393                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1394                         xs_tcp_read_data(xprt, &desc);
1395                         continue;
1396                 }
1397                 /* Skip over any trailing bytes on short reads */
1398                 xs_tcp_read_discard(transport, &desc);
1399         } while (desc.count);
1400         trace_xs_tcp_data_recv(transport);
1401         dprintk("RPC:       xs_tcp_data_recv done\n");
1402         return len - desc.count;
1403 }
1404
1405 /**
1406  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1407  * @sk: socket with data to read
1408  *
1409  */
1410 static void xs_tcp_data_ready(struct sock *sk)
1411 {
1412         struct rpc_xprt *xprt;
1413         read_descriptor_t rd_desc;
1414         int read;
1415         unsigned long total = 0;
1416
1417         dprintk("RPC:       xs_tcp_data_ready...\n");
1418
1419         read_lock_bh(&sk->sk_callback_lock);
1420         if (!(xprt = xprt_from_sock(sk))) {
1421                 read = 0;
1422                 goto out;
1423         }
1424         /* Any data means we had a useful conversation, so
1425          * the we don't need to delay the next reconnect
1426          */
1427         if (xprt->reestablish_timeout)
1428                 xprt->reestablish_timeout = 0;
1429
1430         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1431         rd_desc.arg.data = xprt;
1432         do {
1433                 rd_desc.count = 65536;
1434                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1435                 if (read > 0)
1436                         total += read;
1437         } while (read > 0);
1438 out:
1439         trace_xs_tcp_data_ready(xprt, read, total);
1440         read_unlock_bh(&sk->sk_callback_lock);
1441 }
1442
1443 /**
1444  * xs_tcp_state_change - callback to handle TCP socket state changes
1445  * @sk: socket whose state has changed
1446  *
1447  */
1448 static void xs_tcp_state_change(struct sock *sk)
1449 {
1450         struct rpc_xprt *xprt;
1451         struct sock_xprt *transport;
1452
1453         read_lock_bh(&sk->sk_callback_lock);
1454         if (!(xprt = xprt_from_sock(sk)))
1455                 goto out;
1456         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1457         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1458                         sk->sk_state, xprt_connected(xprt),
1459                         sock_flag(sk, SOCK_DEAD),
1460                         sock_flag(sk, SOCK_ZAPPED),
1461                         sk->sk_shutdown);
1462
1463         transport = container_of(xprt, struct sock_xprt, xprt);
1464         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1465         switch (sk->sk_state) {
1466         case TCP_ESTABLISHED:
1467                 spin_lock(&xprt->transport_lock);
1468                 if (!xprt_test_and_set_connected(xprt)) {
1469
1470                         /* Reset TCP record info */
1471                         transport->tcp_offset = 0;
1472                         transport->tcp_reclen = 0;
1473                         transport->tcp_copied = 0;
1474                         transport->tcp_flags =
1475                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1476                         xprt->connect_cookie++;
1477                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1478                         xprt_clear_connecting(xprt);
1479
1480                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1481                 }
1482                 spin_unlock(&xprt->transport_lock);
1483                 break;
1484         case TCP_FIN_WAIT1:
1485                 /* The client initiated a shutdown of the socket */
1486                 xprt->connect_cookie++;
1487                 xprt->reestablish_timeout = 0;
1488                 set_bit(XPRT_CLOSING, &xprt->state);
1489                 smp_mb__before_atomic();
1490                 clear_bit(XPRT_CONNECTED, &xprt->state);
1491                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1492                 smp_mb__after_atomic();
1493                 break;
1494         case TCP_CLOSE_WAIT:
1495                 /* The server initiated a shutdown of the socket */
1496                 xprt->connect_cookie++;
1497                 clear_bit(XPRT_CONNECTED, &xprt->state);
1498                 xs_tcp_force_close(xprt);
1499         case TCP_CLOSING:
1500                 /*
1501                  * If the server closed down the connection, make sure that
1502                  * we back off before reconnecting
1503                  */
1504                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1505                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1506                 break;
1507         case TCP_LAST_ACK:
1508                 set_bit(XPRT_CLOSING, &xprt->state);
1509                 smp_mb__before_atomic();
1510                 clear_bit(XPRT_CONNECTED, &xprt->state);
1511                 smp_mb__after_atomic();
1512                 break;
1513         case TCP_CLOSE:
1514                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1515                                         &transport->sock_state))
1516                         xprt_clear_connecting(xprt);
1517                 xs_sock_mark_closed(xprt);
1518         }
1519  out:
1520         read_unlock_bh(&sk->sk_callback_lock);
1521 }
1522
1523 static void xs_write_space(struct sock *sk)
1524 {
1525         struct socket *sock;
1526         struct rpc_xprt *xprt;
1527
1528         if (unlikely(!(sock = sk->sk_socket)))
1529                 return;
1530         clear_bit(SOCK_NOSPACE, &sock->flags);
1531
1532         if (unlikely(!(xprt = xprt_from_sock(sk))))
1533                 return;
1534         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1535                 return;
1536
1537         xprt_write_space(xprt);
1538 }
1539
1540 /**
1541  * xs_udp_write_space - callback invoked when socket buffer space
1542  *                             becomes available
1543  * @sk: socket whose state has changed
1544  *
1545  * Called when more output buffer space is available for this socket.
1546  * We try not to wake our writers until they can make "significant"
1547  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1548  * with a bunch of small requests.
1549  */
1550 static void xs_udp_write_space(struct sock *sk)
1551 {
1552         read_lock_bh(&sk->sk_callback_lock);
1553
1554         /* from net/core/sock.c:sock_def_write_space */
1555         if (sock_writeable(sk))
1556                 xs_write_space(sk);
1557
1558         read_unlock_bh(&sk->sk_callback_lock);
1559 }
1560
1561 /**
1562  * xs_tcp_write_space - callback invoked when socket buffer space
1563  *                             becomes available
1564  * @sk: socket whose state has changed
1565  *
1566  * Called when more output buffer space is available for this socket.
1567  * We try not to wake our writers until they can make "significant"
1568  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1569  * with a bunch of small requests.
1570  */
1571 static void xs_tcp_write_space(struct sock *sk)
1572 {
1573         read_lock_bh(&sk->sk_callback_lock);
1574
1575         /* from net/core/stream.c:sk_stream_write_space */
1576         if (sk_stream_is_writeable(sk))
1577                 xs_write_space(sk);
1578
1579         read_unlock_bh(&sk->sk_callback_lock);
1580 }
1581
1582 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1583 {
1584         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1585         struct sock *sk = transport->inet;
1586
1587         if (transport->rcvsize) {
1588                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1589                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1590         }
1591         if (transport->sndsize) {
1592                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1593                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1594                 sk->sk_write_space(sk);
1595         }
1596 }
1597
1598 /**
1599  * xs_udp_set_buffer_size - set send and receive limits
1600  * @xprt: generic transport
1601  * @sndsize: requested size of send buffer, in bytes
1602  * @rcvsize: requested size of receive buffer, in bytes
1603  *
1604  * Set socket send and receive buffer size limits.
1605  */
1606 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1607 {
1608         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1609
1610         transport->sndsize = 0;
1611         if (sndsize)
1612                 transport->sndsize = sndsize + 1024;
1613         transport->rcvsize = 0;
1614         if (rcvsize)
1615                 transport->rcvsize = rcvsize + 1024;
1616
1617         xs_udp_do_set_buffer_size(xprt);
1618 }
1619
1620 /**
1621  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1622  * @task: task that timed out
1623  *
1624  * Adjust the congestion window after a retransmit timeout has occurred.
1625  */
1626 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1627 {
1628         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1629 }
1630
1631 static unsigned short xs_get_random_port(void)
1632 {
1633         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1634         unsigned short rand = (unsigned short) prandom_u32() % range;
1635         return rand + xprt_min_resvport;
1636 }
1637
1638 /**
1639  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1640  * @sock: socket
1641  *
1642  * Note that this function has to be called on all sockets that share the
1643  * same port, and it must be called before binding.
1644  */
1645 static void xs_sock_set_reuseport(struct socket *sock)
1646 {
1647         int opt = 1;
1648
1649         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1650                         (char *)&opt, sizeof(opt));
1651 }
1652
1653 static unsigned short xs_sock_getport(struct socket *sock)
1654 {
1655         struct sockaddr_storage buf;
1656         int buflen;
1657         unsigned short port = 0;
1658
1659         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1660                 goto out;
1661         switch (buf.ss_family) {
1662         case AF_INET6:
1663                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1664                 break;
1665         case AF_INET:
1666                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1667         }
1668 out:
1669         return port;
1670 }
1671
1672 /**
1673  * xs_set_port - reset the port number in the remote endpoint address
1674  * @xprt: generic transport
1675  * @port: new port number
1676  *
1677  */
1678 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1679 {
1680         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1681
1682         rpc_set_port(xs_addr(xprt), port);
1683         xs_update_peer_port(xprt);
1684 }
1685
1686 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1687 {
1688         if (transport->srcport == 0)
1689                 transport->srcport = xs_sock_getport(sock);
1690 }
1691
1692 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1693 {
1694         unsigned short port = transport->srcport;
1695
1696         if (port == 0 && transport->xprt.resvport)
1697                 port = xs_get_random_port();
1698         return port;
1699 }
1700
1701 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1702 {
1703         if (transport->srcport != 0)
1704                 transport->srcport = 0;
1705         if (!transport->xprt.resvport)
1706                 return 0;
1707         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1708                 return xprt_max_resvport;
1709         return --port;
1710 }
1711 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1712 {
1713         struct sockaddr_storage myaddr;
1714         int err, nloop = 0;
1715         unsigned short port = xs_get_srcport(transport);
1716         unsigned short last;
1717
1718         /*
1719          * If we are asking for any ephemeral port (i.e. port == 0 &&
1720          * transport->xprt.resvport == 0), don't bind.  Let the local
1721          * port selection happen implicitly when the socket is used
1722          * (for example at connect time).
1723          *
1724          * This ensures that we can continue to establish TCP
1725          * connections even when all local ephemeral ports are already
1726          * a part of some TCP connection.  This makes no difference
1727          * for UDP sockets, but also doens't harm them.
1728          *
1729          * If we're asking for any reserved port (i.e. port == 0 &&
1730          * transport->xprt.resvport == 1) xs_get_srcport above will
1731          * ensure that port is non-zero and we will bind as needed.
1732          */
1733         if (port == 0)
1734                 return 0;
1735
1736         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1737         do {
1738                 rpc_set_port((struct sockaddr *)&myaddr, port);
1739                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1740                                 transport->xprt.addrlen);
1741                 if (err == 0) {
1742                         transport->srcport = port;
1743                         break;
1744                 }
1745                 last = port;
1746                 port = xs_next_srcport(transport, port);
1747                 if (port > last)
1748                         nloop++;
1749         } while (err == -EADDRINUSE && nloop != 2);
1750
1751         if (myaddr.ss_family == AF_INET)
1752                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1753                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1754                                 port, err ? "failed" : "ok", err);
1755         else
1756                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1757                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1758                                 port, err ? "failed" : "ok", err);
1759         return err;
1760 }
1761
1762 /*
1763  * We don't support autobind on AF_LOCAL sockets
1764  */
1765 static void xs_local_rpcbind(struct rpc_task *task)
1766 {
1767         rcu_read_lock();
1768         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1769         rcu_read_unlock();
1770 }
1771
1772 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1773 {
1774 }
1775
1776 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1777 static struct lock_class_key xs_key[2];
1778 static struct lock_class_key xs_slock_key[2];
1779
1780 static inline void xs_reclassify_socketu(struct socket *sock)
1781 {
1782         struct sock *sk = sock->sk;
1783
1784         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1785                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1786 }
1787
1788 static inline void xs_reclassify_socket4(struct socket *sock)
1789 {
1790         struct sock *sk = sock->sk;
1791
1792         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1793                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1794 }
1795
1796 static inline void xs_reclassify_socket6(struct socket *sock)
1797 {
1798         struct sock *sk = sock->sk;
1799
1800         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1801                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1802 }
1803
1804 static inline void xs_reclassify_socket(int family, struct socket *sock)
1805 {
1806         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1807         if (sock_owned_by_user(sock->sk))
1808                 return;
1809
1810         switch (family) {
1811         case AF_LOCAL:
1812                 xs_reclassify_socketu(sock);
1813                 break;
1814         case AF_INET:
1815                 xs_reclassify_socket4(sock);
1816                 break;
1817         case AF_INET6:
1818                 xs_reclassify_socket6(sock);
1819                 break;
1820         }
1821 }
1822 #else
1823 static inline void xs_reclassify_socketu(struct socket *sock)
1824 {
1825 }
1826
1827 static inline void xs_reclassify_socket4(struct socket *sock)
1828 {
1829 }
1830
1831 static inline void xs_reclassify_socket6(struct socket *sock)
1832 {
1833 }
1834
1835 static inline void xs_reclassify_socket(int family, struct socket *sock)
1836 {
1837 }
1838 #endif
1839
1840 static void xs_dummy_setup_socket(struct work_struct *work)
1841 {
1842 }
1843
1844 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1845                 struct sock_xprt *transport, int family, int type,
1846                 int protocol, bool reuseport)
1847 {
1848         struct socket *sock;
1849         int err;
1850
1851         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1852         if (err < 0) {
1853                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1854                                 protocol, -err);
1855                 goto out;
1856         }
1857         xs_reclassify_socket(family, sock);
1858
1859         if (reuseport)
1860                 xs_sock_set_reuseport(sock);
1861
1862         err = xs_bind(transport, sock);
1863         if (err) {
1864                 sock_release(sock);
1865                 goto out;
1866         }
1867
1868         return sock;
1869 out:
1870         return ERR_PTR(err);
1871 }
1872
1873 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1874                                       struct socket *sock)
1875 {
1876         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1877                                                                         xprt);
1878
1879         if (!transport->inet) {
1880                 struct sock *sk = sock->sk;
1881
1882                 write_lock_bh(&sk->sk_callback_lock);
1883
1884                 xs_save_old_callbacks(transport, sk);
1885
1886                 sk->sk_user_data = xprt;
1887                 sk->sk_data_ready = xs_local_data_ready;
1888                 sk->sk_write_space = xs_udp_write_space;
1889                 sk->sk_error_report = xs_error_report;
1890                 sk->sk_allocation = GFP_NOIO;
1891
1892                 xprt_clear_connected(xprt);
1893
1894                 /* Reset to new socket */
1895                 transport->sock = sock;
1896                 transport->inet = sk;
1897
1898                 write_unlock_bh(&sk->sk_callback_lock);
1899         }
1900
1901         /* Tell the socket layer to start connecting... */
1902         xprt->stat.connect_count++;
1903         xprt->stat.connect_start = jiffies;
1904         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1905 }
1906
1907 /**
1908  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1909  * @transport: socket transport to connect
1910  */
1911 static int xs_local_setup_socket(struct sock_xprt *transport)
1912 {
1913         struct rpc_xprt *xprt = &transport->xprt;
1914         struct socket *sock;
1915         int status = -EIO;
1916
1917         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1918                                         SOCK_STREAM, 0, &sock, 1);
1919         if (status < 0) {
1920                 dprintk("RPC:       can't create AF_LOCAL "
1921                         "transport socket (%d).\n", -status);
1922                 goto out;
1923         }
1924         xs_reclassify_socketu(sock);
1925
1926         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1927                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1928
1929         status = xs_local_finish_connecting(xprt, sock);
1930         trace_rpc_socket_connect(xprt, sock, status);
1931         switch (status) {
1932         case 0:
1933                 dprintk("RPC:       xprt %p connected to %s\n",
1934                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1935                 xprt_set_connected(xprt);
1936         case -ENOBUFS:
1937                 break;
1938         case -ENOENT:
1939                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1940                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1941                 break;
1942         case -ECONNREFUSED:
1943                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1944                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1945                 break;
1946         default:
1947                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1948                                 __func__, -status,
1949                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1950         }
1951
1952 out:
1953         xprt_clear_connecting(xprt);
1954         xprt_wake_pending_tasks(xprt, status);
1955         return status;
1956 }
1957
1958 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1959 {
1960         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1961         int ret;
1962
1963          if (RPC_IS_ASYNC(task)) {
1964                 /*
1965                  * We want the AF_LOCAL connect to be resolved in the
1966                  * filesystem namespace of the process making the rpc
1967                  * call.  Thus we connect synchronously.
1968                  *
1969                  * If we want to support asynchronous AF_LOCAL calls,
1970                  * we'll need to figure out how to pass a namespace to
1971                  * connect.
1972                  */
1973                 rpc_exit(task, -ENOTCONN);
1974                 return;
1975         }
1976         ret = xs_local_setup_socket(transport);
1977         if (ret && !RPC_IS_SOFTCONN(task))
1978                 msleep_interruptible(15000);
1979 }
1980
1981 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1982 /*
1983  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1984  * know that we have exclusive access to the socket), to guard against
1985  * races with xs_reset_transport.
1986  */
1987 static void xs_set_memalloc(struct rpc_xprt *xprt)
1988 {
1989         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1990                         xprt);
1991
1992         /*
1993          * If there's no sock, then we have nothing to set. The
1994          * reconnecting process will get it for us.
1995          */
1996         if (!transport->inet)
1997                 return;
1998         if (atomic_read(&xprt->swapper))
1999                 sk_set_memalloc(transport->inet);
2000 }
2001
2002 /**
2003  * xs_enable_swap - Tag this transport as being used for swap.
2004  * @xprt: transport to tag
2005  *
2006  * Take a reference to this transport on behalf of the rpc_clnt, and
2007  * optionally mark it for swapping if it wasn't already.
2008  */
2009 static int
2010 xs_enable_swap(struct rpc_xprt *xprt)
2011 {
2012         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2013
2014         if (atomic_inc_return(&xprt->swapper) != 1)
2015                 return 0;
2016         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2017                 return -ERESTARTSYS;
2018         if (xs->inet)
2019                 sk_set_memalloc(xs->inet);
2020         xprt_release_xprt(xprt, NULL);
2021         return 0;
2022 }
2023
2024 /**
2025  * xs_disable_swap - Untag this transport as being used for swap.
2026  * @xprt: transport to tag
2027  *
2028  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2029  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2030  */
2031 static void
2032 xs_disable_swap(struct rpc_xprt *xprt)
2033 {
2034         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2035
2036         if (!atomic_dec_and_test(&xprt->swapper))
2037                 return;
2038         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2039                 return;
2040         if (xs->inet)
2041                 sk_clear_memalloc(xs->inet);
2042         xprt_release_xprt(xprt, NULL);
2043 }
2044 #else
2045 static void xs_set_memalloc(struct rpc_xprt *xprt)
2046 {
2047 }
2048
2049 static int
2050 xs_enable_swap(struct rpc_xprt *xprt)
2051 {
2052         return -EINVAL;
2053 }
2054
2055 static void
2056 xs_disable_swap(struct rpc_xprt *xprt)
2057 {
2058 }
2059 #endif
2060
2061 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2062 {
2063         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2064
2065         if (!transport->inet) {
2066                 struct sock *sk = sock->sk;
2067
2068                 write_lock_bh(&sk->sk_callback_lock);
2069
2070                 xs_save_old_callbacks(transport, sk);
2071
2072                 sk->sk_user_data = xprt;
2073                 sk->sk_data_ready = xs_udp_data_ready;
2074                 sk->sk_write_space = xs_udp_write_space;
2075                 sk->sk_allocation = GFP_NOIO;
2076
2077                 xprt_set_connected(xprt);
2078
2079                 /* Reset to new socket */
2080                 transport->sock = sock;
2081                 transport->inet = sk;
2082
2083                 xs_set_memalloc(xprt);
2084
2085                 write_unlock_bh(&sk->sk_callback_lock);
2086         }
2087         xs_udp_do_set_buffer_size(xprt);
2088 }
2089
2090 static void xs_udp_setup_socket(struct work_struct *work)
2091 {
2092         struct sock_xprt *transport =
2093                 container_of(work, struct sock_xprt, connect_worker.work);
2094         struct rpc_xprt *xprt = &transport->xprt;
2095         struct socket *sock = transport->sock;
2096         int status = -EIO;
2097
2098         sock = xs_create_sock(xprt, transport,
2099                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2100                         IPPROTO_UDP, false);
2101         if (IS_ERR(sock))
2102                 goto out;
2103
2104         dprintk("RPC:       worker connecting xprt %p via %s to "
2105                                 "%s (port %s)\n", xprt,
2106                         xprt->address_strings[RPC_DISPLAY_PROTO],
2107                         xprt->address_strings[RPC_DISPLAY_ADDR],
2108                         xprt->address_strings[RPC_DISPLAY_PORT]);
2109
2110         xs_udp_finish_connecting(xprt, sock);
2111         trace_rpc_socket_connect(xprt, sock, 0);
2112         status = 0;
2113 out:
2114         xprt_unlock_connect(xprt, transport);
2115         xprt_clear_connecting(xprt);
2116         xprt_wake_pending_tasks(xprt, status);
2117 }
2118
2119 /**
2120  * xs_tcp_shutdown - gracefully shut down a TCP socket
2121  * @xprt: transport
2122  *
2123  * Initiates a graceful shutdown of the TCP socket by calling the
2124  * equivalent of shutdown(SHUT_RDWR);
2125  */
2126 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2127 {
2128         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2129         struct socket *sock = transport->sock;
2130
2131         if (sock == NULL)
2132                 return;
2133         if (xprt_connected(xprt)) {
2134                 kernel_sock_shutdown(sock, SHUT_RDWR);
2135                 trace_rpc_socket_shutdown(xprt, sock);
2136         } else
2137                 xs_reset_transport(transport);
2138 }
2139
2140 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2141 {
2142         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2143         int ret = -ENOTCONN;
2144
2145         if (!transport->inet) {
2146                 struct sock *sk = sock->sk;
2147                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2148                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2149                 unsigned int opt_on = 1;
2150                 unsigned int timeo;
2151
2152                 /* TCP Keepalive options */
2153                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2154                                 (char *)&opt_on, sizeof(opt_on));
2155                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2156                                 (char *)&keepidle, sizeof(keepidle));
2157                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2158                                 (char *)&keepidle, sizeof(keepidle));
2159                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2160                                 (char *)&keepcnt, sizeof(keepcnt));
2161
2162                 /* TCP user timeout (see RFC5482) */
2163                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2164                         (xprt->timeout->to_retries + 1);
2165                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2166                                 (char *)&timeo, sizeof(timeo));
2167
2168                 write_lock_bh(&sk->sk_callback_lock);
2169
2170                 xs_save_old_callbacks(transport, sk);
2171
2172                 sk->sk_user_data = xprt;
2173                 sk->sk_data_ready = xs_tcp_data_ready;
2174                 sk->sk_state_change = xs_tcp_state_change;
2175                 sk->sk_write_space = xs_tcp_write_space;
2176                 sk->sk_error_report = xs_error_report;
2177                 sk->sk_allocation = GFP_NOIO;
2178
2179                 /* socket options */
2180                 sock_reset_flag(sk, SOCK_LINGER);
2181                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2182
2183                 xprt_clear_connected(xprt);
2184
2185                 /* Reset to new socket */
2186                 transport->sock = sock;
2187                 transport->inet = sk;
2188
2189                 write_unlock_bh(&sk->sk_callback_lock);
2190         }
2191
2192         if (!xprt_bound(xprt))
2193                 goto out;
2194
2195         xs_set_memalloc(xprt);
2196
2197         /* Tell the socket layer to start connecting... */
2198         xprt->stat.connect_count++;
2199         xprt->stat.connect_start = jiffies;
2200         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2201         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2202         switch (ret) {
2203         case 0:
2204                 xs_set_srcport(transport, sock);
2205         case -EINPROGRESS:
2206                 /* SYN_SENT! */
2207                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2208                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2209         }
2210 out:
2211         return ret;
2212 }
2213
2214 /**
2215  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2216  *
2217  * Invoked by a work queue tasklet.
2218  */
2219 static void xs_tcp_setup_socket(struct work_struct *work)
2220 {
2221         struct sock_xprt *transport =
2222                 container_of(work, struct sock_xprt, connect_worker.work);
2223         struct socket *sock = transport->sock;
2224         struct rpc_xprt *xprt = &transport->xprt;
2225         int status = -EIO;
2226
2227         if (!sock) {
2228                 sock = xs_create_sock(xprt, transport,
2229                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2230                                 IPPROTO_TCP, true);
2231                 if (IS_ERR(sock)) {
2232                         status = PTR_ERR(sock);
2233                         goto out;
2234                 }
2235         }
2236
2237         dprintk("RPC:       worker connecting xprt %p via %s to "
2238                                 "%s (port %s)\n", xprt,
2239                         xprt->address_strings[RPC_DISPLAY_PROTO],
2240                         xprt->address_strings[RPC_DISPLAY_ADDR],
2241                         xprt->address_strings[RPC_DISPLAY_PORT]);
2242
2243         status = xs_tcp_finish_connecting(xprt, sock);
2244         trace_rpc_socket_connect(xprt, sock, status);
2245         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2246                         xprt, -status, xprt_connected(xprt),
2247                         sock->sk->sk_state);
2248         switch (status) {
2249         default:
2250                 printk("%s: connect returned unhandled error %d\n",
2251                         __func__, status);
2252         case -EADDRNOTAVAIL:
2253                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2254                  * and retry
2255                  */
2256                 xs_tcp_force_close(xprt);
2257                 break;
2258         case 0:
2259         case -EINPROGRESS:
2260         case -EALREADY:
2261                 xprt_unlock_connect(xprt, transport);
2262                 return;
2263         case -EINVAL:
2264                 /* Happens, for instance, if the user specified a link
2265                  * local IPv6 address without a scope-id.
2266                  */
2267         case -ECONNREFUSED:
2268         case -ECONNRESET:
2269         case -ENETUNREACH:
2270         case -EADDRINUSE:
2271         case -ENOBUFS:
2272                 /* retry with existing socket, after a delay */
2273                 xs_tcp_force_close(xprt);
2274                 goto out;
2275         }
2276         status = -EAGAIN;
2277 out:
2278         xprt_unlock_connect(xprt, transport);
2279         xprt_clear_connecting(xprt);
2280         xprt_wake_pending_tasks(xprt, status);
2281 }
2282
2283 /**
2284  * xs_connect - connect a socket to a remote endpoint
2285  * @xprt: pointer to transport structure
2286  * @task: address of RPC task that manages state of connect request
2287  *
2288  * TCP: If the remote end dropped the connection, delay reconnecting.
2289  *
2290  * UDP socket connects are synchronous, but we use a work queue anyway
2291  * to guarantee that even unprivileged user processes can set up a
2292  * socket on a privileged port.
2293  *
2294  * If a UDP socket connect fails, the delay behavior here prevents
2295  * retry floods (hard mounts).
2296  */
2297 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2298 {
2299         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2300
2301         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2302
2303         if (transport->sock != NULL) {
2304                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2305                                 "seconds\n",
2306                                 xprt, xprt->reestablish_timeout / HZ);
2307
2308                 /* Start by resetting any existing state */
2309                 xs_reset_transport(transport);
2310
2311                 queue_delayed_work(rpciod_workqueue,
2312                                    &transport->connect_worker,
2313                                    xprt->reestablish_timeout);
2314                 xprt->reestablish_timeout <<= 1;
2315                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2316                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2317                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2318                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2319         } else {
2320                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2321                 queue_delayed_work(rpciod_workqueue,
2322                                    &transport->connect_worker, 0);
2323         }
2324 }
2325
2326 /**
2327  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2328  * @xprt: rpc_xprt struct containing statistics
2329  * @seq: output file
2330  *
2331  */
2332 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2333 {
2334         long idle_time = 0;
2335
2336         if (xprt_connected(xprt))
2337                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2338
2339         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2340                         "%llu %llu %lu %llu %llu\n",
2341                         xprt->stat.bind_count,
2342                         xprt->stat.connect_count,
2343                         xprt->stat.connect_time,
2344                         idle_time,
2345                         xprt->stat.sends,
2346                         xprt->stat.recvs,
2347                         xprt->stat.bad_xids,
2348                         xprt->stat.req_u,
2349                         xprt->stat.bklog_u,
2350                         xprt->stat.max_slots,
2351                         xprt->stat.sending_u,
2352                         xprt->stat.pending_u);
2353 }
2354
2355 /**
2356  * xs_udp_print_stats - display UDP socket-specifc stats
2357  * @xprt: rpc_xprt struct containing statistics
2358  * @seq: output file
2359  *
2360  */
2361 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2362 {
2363         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2364
2365         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2366                         "%lu %llu %llu\n",
2367                         transport->srcport,
2368                         xprt->stat.bind_count,
2369                         xprt->stat.sends,
2370                         xprt->stat.recvs,
2371                         xprt->stat.bad_xids,
2372                         xprt->stat.req_u,
2373                         xprt->stat.bklog_u,
2374                         xprt->stat.max_slots,
2375                         xprt->stat.sending_u,
2376                         xprt->stat.pending_u);
2377 }
2378
2379 /**
2380  * xs_tcp_print_stats - display TCP socket-specifc stats
2381  * @xprt: rpc_xprt struct containing statistics
2382  * @seq: output file
2383  *
2384  */
2385 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2386 {
2387         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2388         long idle_time = 0;
2389
2390         if (xprt_connected(xprt))
2391                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2392
2393         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2394                         "%llu %llu %lu %llu %llu\n",
2395                         transport->srcport,
2396                         xprt->stat.bind_count,
2397                         xprt->stat.connect_count,
2398                         xprt->stat.connect_time,
2399                         idle_time,
2400                         xprt->stat.sends,
2401                         xprt->stat.recvs,
2402                         xprt->stat.bad_xids,
2403                         xprt->stat.req_u,
2404                         xprt->stat.bklog_u,
2405                         xprt->stat.max_slots,
2406                         xprt->stat.sending_u,
2407                         xprt->stat.pending_u);
2408 }
2409
2410 /*
2411  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2412  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2413  * to use the server side send routines.
2414  */
2415 static void *bc_malloc(struct rpc_task *task, size_t size)
2416 {
2417         struct page *page;
2418         struct rpc_buffer *buf;
2419
2420         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2421         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2422                 return NULL;
2423
2424         page = alloc_page(GFP_KERNEL);
2425         if (!page)
2426                 return NULL;
2427
2428         buf = page_address(page);
2429         buf->len = PAGE_SIZE;
2430
2431         return buf->data;
2432 }
2433
2434 /*
2435  * Free the space allocated in the bc_alloc routine
2436  */
2437 static void bc_free(void *buffer)
2438 {
2439         struct rpc_buffer *buf;
2440
2441         if (!buffer)
2442                 return;
2443
2444         buf = container_of(buffer, struct rpc_buffer, data);
2445         free_page((unsigned long)buf);
2446 }
2447
2448 /*
2449  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2450  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2451  */
2452 static int bc_sendto(struct rpc_rqst *req)
2453 {
2454         int len;
2455         struct xdr_buf *xbufp = &req->rq_snd_buf;
2456         struct rpc_xprt *xprt = req->rq_xprt;
2457         struct sock_xprt *transport =
2458                                 container_of(xprt, struct sock_xprt, xprt);
2459         struct socket *sock = transport->sock;
2460         unsigned long headoff;
2461         unsigned long tailoff;
2462
2463         xs_encode_stream_record_marker(xbufp);
2464
2465         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2466         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2467         len = svc_send_common(sock, xbufp,
2468                               virt_to_page(xbufp->head[0].iov_base), headoff,
2469                               xbufp->tail[0].iov_base, tailoff);
2470
2471         if (len != xbufp->len) {
2472                 printk(KERN_NOTICE "Error sending entire callback!\n");
2473                 len = -EAGAIN;
2474         }
2475
2476         return len;
2477 }
2478
2479 /*
2480  * The send routine. Borrows from svc_send
2481  */
2482 static int bc_send_request(struct rpc_task *task)
2483 {
2484         struct rpc_rqst *req = task->tk_rqstp;
2485         struct svc_xprt *xprt;
2486         u32                     len;
2487
2488         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2489         /*
2490          * Get the server socket associated with this callback xprt
2491          */
2492         xprt = req->rq_xprt->bc_xprt;
2493
2494         /*
2495          * Grab the mutex to serialize data as the connection is shared
2496          * with the fore channel
2497          */
2498         if (!mutex_trylock(&xprt->xpt_mutex)) {
2499                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2500                 if (!mutex_trylock(&xprt->xpt_mutex))
2501                         return -EAGAIN;
2502                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2503         }
2504         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2505                 len = -ENOTCONN;
2506         else
2507                 len = bc_sendto(req);
2508         mutex_unlock(&xprt->xpt_mutex);
2509
2510         if (len > 0)
2511                 len = 0;
2512
2513         return len;
2514 }
2515
2516 /*
2517  * The close routine. Since this is client initiated, we do nothing
2518  */
2519
2520 static void bc_close(struct rpc_xprt *xprt)
2521 {
2522 }
2523
2524 /*
2525  * The xprt destroy routine. Again, because this connection is client
2526  * initiated, we do nothing
2527  */
2528
2529 static void bc_destroy(struct rpc_xprt *xprt)
2530 {
2531         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2532
2533         xs_xprt_free(xprt);
2534         module_put(THIS_MODULE);
2535 }
2536
2537 static struct rpc_xprt_ops xs_local_ops = {
2538         .reserve_xprt           = xprt_reserve_xprt,
2539         .release_xprt           = xs_tcp_release_xprt,
2540         .alloc_slot             = xprt_alloc_slot,
2541         .rpcbind                = xs_local_rpcbind,
2542         .set_port               = xs_local_set_port,
2543         .connect                = xs_local_connect,
2544         .buf_alloc              = rpc_malloc,
2545         .buf_free               = rpc_free,
2546         .send_request           = xs_local_send_request,
2547         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2548         .close                  = xs_close,
2549         .destroy                = xs_destroy,
2550         .print_stats            = xs_local_print_stats,
2551         .enable_swap            = xs_enable_swap,
2552         .disable_swap           = xs_disable_swap,
2553 };
2554
2555 static struct rpc_xprt_ops xs_udp_ops = {
2556         .set_buffer_size        = xs_udp_set_buffer_size,
2557         .reserve_xprt           = xprt_reserve_xprt_cong,
2558         .release_xprt           = xprt_release_xprt_cong,
2559         .alloc_slot             = xprt_alloc_slot,
2560         .rpcbind                = rpcb_getport_async,
2561         .set_port               = xs_set_port,
2562         .connect                = xs_connect,
2563         .buf_alloc              = rpc_malloc,
2564         .buf_free               = rpc_free,
2565         .send_request           = xs_udp_send_request,
2566         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2567         .timer                  = xs_udp_timer,
2568         .release_request        = xprt_release_rqst_cong,
2569         .close                  = xs_close,
2570         .destroy                = xs_destroy,
2571         .print_stats            = xs_udp_print_stats,
2572         .enable_swap            = xs_enable_swap,
2573         .disable_swap           = xs_disable_swap,
2574         .inject_disconnect      = xs_inject_disconnect,
2575 };
2576
2577 static struct rpc_xprt_ops xs_tcp_ops = {
2578         .reserve_xprt           = xprt_reserve_xprt,
2579         .release_xprt           = xs_tcp_release_xprt,
2580         .alloc_slot             = xprt_lock_and_alloc_slot,
2581         .rpcbind                = rpcb_getport_async,
2582         .set_port               = xs_set_port,
2583         .connect                = xs_connect,
2584         .buf_alloc              = rpc_malloc,
2585         .buf_free               = rpc_free,
2586         .send_request           = xs_tcp_send_request,
2587         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2588         .close                  = xs_tcp_shutdown,
2589         .destroy                = xs_destroy,
2590         .print_stats            = xs_tcp_print_stats,
2591         .enable_swap            = xs_enable_swap,
2592         .disable_swap           = xs_disable_swap,
2593         .inject_disconnect      = xs_inject_disconnect,
2594 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2595         .bc_setup               = xprt_setup_bc,
2596         .bc_up                  = xs_tcp_bc_up,
2597         .bc_free_rqst           = xprt_free_bc_rqst,
2598         .bc_destroy             = xprt_destroy_bc,
2599 #endif
2600 };
2601
2602 /*
2603  * The rpc_xprt_ops for the server backchannel
2604  */
2605
2606 static struct rpc_xprt_ops bc_tcp_ops = {
2607         .reserve_xprt           = xprt_reserve_xprt,
2608         .release_xprt           = xprt_release_xprt,
2609         .alloc_slot             = xprt_alloc_slot,
2610         .buf_alloc              = bc_malloc,
2611         .buf_free               = bc_free,
2612         .send_request           = bc_send_request,
2613         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2614         .close                  = bc_close,
2615         .destroy                = bc_destroy,
2616         .print_stats            = xs_tcp_print_stats,
2617         .enable_swap            = xs_enable_swap,
2618         .disable_swap           = xs_disable_swap,
2619         .inject_disconnect      = xs_inject_disconnect,
2620 };
2621
2622 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2623 {
2624         static const struct sockaddr_in sin = {
2625                 .sin_family             = AF_INET,
2626                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2627         };
2628         static const struct sockaddr_in6 sin6 = {
2629                 .sin6_family            = AF_INET6,
2630                 .sin6_addr              = IN6ADDR_ANY_INIT,
2631         };
2632
2633         switch (family) {
2634         case AF_LOCAL:
2635                 break;
2636         case AF_INET:
2637                 memcpy(sap, &sin, sizeof(sin));
2638                 break;
2639         case AF_INET6:
2640                 memcpy(sap, &sin6, sizeof(sin6));
2641                 break;
2642         default:
2643                 dprintk("RPC:       %s: Bad address family\n", __func__);
2644                 return -EAFNOSUPPORT;
2645         }
2646         return 0;
2647 }
2648
2649 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2650                                       unsigned int slot_table_size,
2651                                       unsigned int max_slot_table_size)
2652 {
2653         struct rpc_xprt *xprt;
2654         struct sock_xprt *new;
2655
2656         if (args->addrlen > sizeof(xprt->addr)) {
2657                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2658                 return ERR_PTR(-EBADF);
2659         }
2660
2661         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2662                         max_slot_table_size);
2663         if (xprt == NULL) {
2664                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2665                                 "rpc_xprt\n");
2666                 return ERR_PTR(-ENOMEM);
2667         }
2668
2669         new = container_of(xprt, struct sock_xprt, xprt);
2670         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2671         xprt->addrlen = args->addrlen;
2672         if (args->srcaddr)
2673                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2674         else {
2675                 int err;
2676                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2677                                         (struct sockaddr *)&new->srcaddr);
2678                 if (err != 0) {
2679                         xprt_free(xprt);
2680                         return ERR_PTR(err);
2681                 }
2682         }
2683
2684         return xprt;
2685 }
2686
2687 static const struct rpc_timeout xs_local_default_timeout = {
2688         .to_initval = 10 * HZ,
2689         .to_maxval = 10 * HZ,
2690         .to_retries = 2,
2691 };
2692
2693 /**
2694  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2695  * @args: rpc transport creation arguments
2696  *
2697  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2698  */
2699 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2700 {
2701         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2702         struct sock_xprt *transport;
2703         struct rpc_xprt *xprt;
2704         struct rpc_xprt *ret;
2705
2706         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2707                         xprt_max_tcp_slot_table_entries);
2708         if (IS_ERR(xprt))
2709                 return xprt;
2710         transport = container_of(xprt, struct sock_xprt, xprt);
2711
2712         xprt->prot = 0;
2713         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2714         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2715
2716         xprt->bind_timeout = XS_BIND_TO;
2717         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2718         xprt->idle_timeout = XS_IDLE_DISC_TO;
2719
2720         xprt->ops = &xs_local_ops;
2721         xprt->timeout = &xs_local_default_timeout;
2722
2723         INIT_DELAYED_WORK(&transport->connect_worker,
2724                         xs_dummy_setup_socket);
2725
2726         switch (sun->sun_family) {
2727         case AF_LOCAL:
2728                 if (sun->sun_path[0] != '/') {
2729                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2730                                         sun->sun_path);
2731                         ret = ERR_PTR(-EINVAL);
2732                         goto out_err;
2733                 }
2734                 xprt_set_bound(xprt);
2735                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2736                 ret = ERR_PTR(xs_local_setup_socket(transport));
2737                 if (ret)
2738                         goto out_err;
2739                 break;
2740         default:
2741                 ret = ERR_PTR(-EAFNOSUPPORT);
2742                 goto out_err;
2743         }
2744
2745         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2746                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2747
2748         if (try_module_get(THIS_MODULE))
2749                 return xprt;
2750         ret = ERR_PTR(-EINVAL);
2751 out_err:
2752         xs_xprt_free(xprt);
2753         return ret;
2754 }
2755
2756 static const struct rpc_timeout xs_udp_default_timeout = {
2757         .to_initval = 5 * HZ,
2758         .to_maxval = 30 * HZ,
2759         .to_increment = 5 * HZ,
2760         .to_retries = 5,
2761 };
2762
2763 /**
2764  * xs_setup_udp - Set up transport to use a UDP socket
2765  * @args: rpc transport creation arguments
2766  *
2767  */
2768 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2769 {
2770         struct sockaddr *addr = args->dstaddr;
2771         struct rpc_xprt *xprt;
2772         struct sock_xprt *transport;
2773         struct rpc_xprt *ret;
2774
2775         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2776                         xprt_udp_slot_table_entries);
2777         if (IS_ERR(xprt))
2778                 return xprt;
2779         transport = container_of(xprt, struct sock_xprt, xprt);
2780
2781         xprt->prot = IPPROTO_UDP;
2782         xprt->tsh_size = 0;
2783         /* XXX: header size can vary due to auth type, IPv6, etc. */
2784         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2785
2786         xprt->bind_timeout = XS_BIND_TO;
2787         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2788         xprt->idle_timeout = XS_IDLE_DISC_TO;
2789
2790         xprt->ops = &xs_udp_ops;
2791
2792         xprt->timeout = &xs_udp_default_timeout;
2793
2794         switch (addr->sa_family) {
2795         case AF_INET:
2796                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2797                         xprt_set_bound(xprt);
2798
2799                 INIT_DELAYED_WORK(&transport->connect_worker,
2800                                         xs_udp_setup_socket);
2801                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2802                 break;
2803         case AF_INET6:
2804                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2805                         xprt_set_bound(xprt);
2806
2807                 INIT_DELAYED_WORK(&transport->connect_worker,
2808                                         xs_udp_setup_socket);
2809                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2810                 break;
2811         default:
2812                 ret = ERR_PTR(-EAFNOSUPPORT);
2813                 goto out_err;
2814         }
2815
2816         if (xprt_bound(xprt))
2817                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2818                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2819                                 xprt->address_strings[RPC_DISPLAY_PORT],
2820                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2821         else
2822                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2823                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2824                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2825
2826         if (try_module_get(THIS_MODULE))
2827                 return xprt;
2828         ret = ERR_PTR(-EINVAL);
2829 out_err:
2830         xs_xprt_free(xprt);
2831         return ret;
2832 }
2833
2834 static const struct rpc_timeout xs_tcp_default_timeout = {
2835         .to_initval = 60 * HZ,
2836         .to_maxval = 60 * HZ,
2837         .to_retries = 2,
2838 };
2839
2840 /**
2841  * xs_setup_tcp - Set up transport to use a TCP socket
2842  * @args: rpc transport creation arguments
2843  *
2844  */
2845 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2846 {
2847         struct sockaddr *addr = args->dstaddr;
2848         struct rpc_xprt *xprt;
2849         struct sock_xprt *transport;
2850         struct rpc_xprt *ret;
2851         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2852
2853         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2854                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2855
2856         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2857                         max_slot_table_size);
2858         if (IS_ERR(xprt))
2859                 return xprt;
2860         transport = container_of(xprt, struct sock_xprt, xprt);
2861
2862         xprt->prot = IPPROTO_TCP;
2863         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2864         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2865
2866         xprt->bind_timeout = XS_BIND_TO;
2867         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2868         xprt->idle_timeout = XS_IDLE_DISC_TO;
2869
2870         xprt->ops = &xs_tcp_ops;
2871         xprt->timeout = &xs_tcp_default_timeout;
2872
2873         switch (addr->sa_family) {
2874         case AF_INET:
2875                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2876                         xprt_set_bound(xprt);
2877
2878                 INIT_DELAYED_WORK(&transport->connect_worker,
2879                                         xs_tcp_setup_socket);
2880                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2881                 break;
2882         case AF_INET6:
2883                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2884                         xprt_set_bound(xprt);
2885
2886                 INIT_DELAYED_WORK(&transport->connect_worker,
2887                                         xs_tcp_setup_socket);
2888                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2889                 break;
2890         default:
2891                 ret = ERR_PTR(-EAFNOSUPPORT);
2892                 goto out_err;
2893         }
2894
2895         if (xprt_bound(xprt))
2896                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2897                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2898                                 xprt->address_strings[RPC_DISPLAY_PORT],
2899                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2900         else
2901                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2902                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2903                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2904
2905         if (try_module_get(THIS_MODULE))
2906                 return xprt;
2907         ret = ERR_PTR(-EINVAL);
2908 out_err:
2909         xs_xprt_free(xprt);
2910         return ret;
2911 }
2912
2913 /**
2914  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2915  * @args: rpc transport creation arguments
2916  *
2917  */
2918 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2919 {
2920         struct sockaddr *addr = args->dstaddr;
2921         struct rpc_xprt *xprt;
2922         struct sock_xprt *transport;
2923         struct svc_sock *bc_sock;
2924         struct rpc_xprt *ret;
2925
2926         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2927                         xprt_tcp_slot_table_entries);
2928         if (IS_ERR(xprt))
2929                 return xprt;
2930         transport = container_of(xprt, struct sock_xprt, xprt);
2931
2932         xprt->prot = IPPROTO_TCP;
2933         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2934         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2935         xprt->timeout = &xs_tcp_default_timeout;
2936
2937         /* backchannel */
2938         xprt_set_bound(xprt);
2939         xprt->bind_timeout = 0;
2940         xprt->reestablish_timeout = 0;
2941         xprt->idle_timeout = 0;
2942
2943         xprt->ops = &bc_tcp_ops;
2944
2945         switch (addr->sa_family) {
2946         case AF_INET:
2947                 xs_format_peer_addresses(xprt, "tcp",
2948                                          RPCBIND_NETID_TCP);
2949                 break;
2950         case AF_INET6:
2951                 xs_format_peer_addresses(xprt, "tcp",
2952                                    RPCBIND_NETID_TCP6);
2953                 break;
2954         default:
2955                 ret = ERR_PTR(-EAFNOSUPPORT);
2956                 goto out_err;
2957         }
2958
2959         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2960                         xprt->address_strings[RPC_DISPLAY_ADDR],
2961                         xprt->address_strings[RPC_DISPLAY_PORT],
2962                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2963
2964         /*
2965          * Once we've associated a backchannel xprt with a connection,
2966          * we want to keep it around as long as the connection lasts,
2967          * in case we need to start using it for a backchannel again;
2968          * this reference won't be dropped until bc_xprt is destroyed.
2969          */
2970         xprt_get(xprt);
2971         args->bc_xprt->xpt_bc_xprt = xprt;
2972         xprt->bc_xprt = args->bc_xprt;
2973         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2974         transport->sock = bc_sock->sk_sock;
2975         transport->inet = bc_sock->sk_sk;
2976
2977         /*
2978          * Since we don't want connections for the backchannel, we set
2979          * the xprt status to connected
2980          */
2981         xprt_set_connected(xprt);
2982
2983         if (try_module_get(THIS_MODULE))
2984                 return xprt;
2985
2986         args->bc_xprt->xpt_bc_xprt = NULL;
2987         xprt_put(xprt);
2988         ret = ERR_PTR(-EINVAL);
2989 out_err:
2990         xs_xprt_free(xprt);
2991         return ret;
2992 }
2993
2994 static struct xprt_class        xs_local_transport = {
2995         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2996         .name           = "named UNIX socket",
2997         .owner          = THIS_MODULE,
2998         .ident          = XPRT_TRANSPORT_LOCAL,
2999         .setup          = xs_setup_local,
3000 };
3001
3002 static struct xprt_class        xs_udp_transport = {
3003         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3004         .name           = "udp",
3005         .owner          = THIS_MODULE,
3006         .ident          = XPRT_TRANSPORT_UDP,
3007         .setup          = xs_setup_udp,
3008 };
3009
3010 static struct xprt_class        xs_tcp_transport = {
3011         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3012         .name           = "tcp",
3013         .owner          = THIS_MODULE,
3014         .ident          = XPRT_TRANSPORT_TCP,
3015         .setup          = xs_setup_tcp,
3016 };
3017
3018 static struct xprt_class        xs_bc_tcp_transport = {
3019         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3020         .name           = "tcp NFSv4.1 backchannel",
3021         .owner          = THIS_MODULE,
3022         .ident          = XPRT_TRANSPORT_BC_TCP,
3023         .setup          = xs_setup_bc_tcp,
3024 };
3025
3026 /**
3027  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3028  *
3029  */
3030 int init_socket_xprt(void)
3031 {
3032 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3033         if (!sunrpc_table_header)
3034                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3035 #endif
3036
3037         xprt_register_transport(&xs_local_transport);
3038         xprt_register_transport(&xs_udp_transport);
3039         xprt_register_transport(&xs_tcp_transport);
3040         xprt_register_transport(&xs_bc_tcp_transport);
3041
3042         return 0;
3043 }
3044
3045 /**
3046  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3047  *
3048  */
3049 void cleanup_socket_xprt(void)
3050 {
3051 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3052         if (sunrpc_table_header) {
3053                 unregister_sysctl_table(sunrpc_table_header);
3054                 sunrpc_table_header = NULL;
3055         }
3056 #endif
3057
3058         xprt_unregister_transport(&xs_local_transport);
3059         xprt_unregister_transport(&xs_udp_transport);
3060         xprt_unregister_transport(&xs_tcp_transport);
3061         xprt_unregister_transport(&xs_bc_tcp_transport);
3062 }
3063
3064 static int param_set_uint_minmax(const char *val,
3065                 const struct kernel_param *kp,
3066                 unsigned int min, unsigned int max)
3067 {
3068         unsigned int num;
3069         int ret;
3070
3071         if (!val)
3072                 return -EINVAL;
3073         ret = kstrtouint(val, 0, &num);
3074         if (ret == -EINVAL || num < min || num > max)
3075                 return -EINVAL;
3076         *((unsigned int *)kp->arg) = num;
3077         return 0;
3078 }
3079
3080 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3081 {
3082         return param_set_uint_minmax(val, kp,
3083                         RPC_MIN_RESVPORT,
3084                         RPC_MAX_RESVPORT);
3085 }
3086
3087 static const struct kernel_param_ops param_ops_portnr = {
3088         .set = param_set_portnr,
3089         .get = param_get_uint,
3090 };
3091
3092 #define param_check_portnr(name, p) \
3093         __param_check(name, p, unsigned int);
3094
3095 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3096 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3097
3098 static int param_set_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);
3104 }
3105
3106 static const struct kernel_param_ops param_ops_slot_table_size = {
3107         .set = param_set_slot_table_size,
3108         .get = param_get_uint,
3109 };
3110
3111 #define param_check_slot_table_size(name, p) \
3112         __param_check(name, p, unsigned int);
3113
3114 static int param_set_max_slot_table_size(const char *val,
3115                                      const struct kernel_param *kp)
3116 {
3117         return param_set_uint_minmax(val, kp,
3118                         RPC_MIN_SLOT_TABLE,
3119                         RPC_MAX_SLOT_TABLE_LIMIT);
3120 }
3121
3122 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3123         .set = param_set_max_slot_table_size,
3124         .get = param_get_uint,
3125 };
3126
3127 #define param_check_max_slot_table_size(name, p) \
3128         __param_check(name, p, unsigned int);
3129
3130 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3131                    slot_table_size, 0644);
3132 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3133                    max_slot_table_size, 0644);
3134 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3135                    slot_table_size, 0644);
3136