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