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 "sunrpc.h"
  51
  52 static void xs_close(struct rpc_xprt *xprt);
  53
  54 /*
  55  * xprtsock tunables
  56  */
  57 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  58 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  59 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  60
  61 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  62 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  63
  64 #define XS_TCP_LINGER_TO        (15U * HZ)
  65 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  66
  67 /*
  68  * We can register our own files under /proc/sys/sunrpc by
  69  * calling register_sysctl_table() again.  The files in that
  70  * directory become the union of all files registered there.
  71  *
  72  * We simply need to make sure that we don't collide with
  73  * someone else's file names!
  74  */
  75
  76 #ifdef RPC_DEBUG
  77
  78 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  79 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  80 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  81 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  82 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  83
  84 static struct ctl_table_header *sunrpc_table_header;
  85
  86 /*
  87  * FIXME: changing the UDP slot table size should also resize the UDP
  88  *        socket buffers for existing UDP transports
  89  */
  90 static ctl_table xs_tunables_table[] = {
  91         {
  92                 .procname       = "udp_slot_table_entries",
  93                 .data           = &xprt_udp_slot_table_entries,
  94                 .maxlen         = sizeof(unsigned int),
  95                 .mode           = 0644,
  96                 .proc_handler   = proc_dointvec_minmax,
  97                 .extra1         = &min_slot_table_size,
  98                 .extra2         = &max_slot_table_size
  99         },
 100         {
 101                 .procname       = "tcp_slot_table_entries",
 102                 .data           = &xprt_tcp_slot_table_entries,
 103                 .maxlen         = sizeof(unsigned int),
 104                 .mode           = 0644,
 105                 .proc_handler   = proc_dointvec_minmax,
 106                 .extra1         = &min_slot_table_size,
 107                 .extra2         = &max_slot_table_size
 108         },
 109         {
 110                 .procname       = "tcp_max_slot_table_entries",
 111                 .data           = &xprt_max_tcp_slot_table_entries,
 112                 .maxlen         = sizeof(unsigned int),
 113                 .mode           = 0644,
 114                 .proc_handler   = proc_dointvec_minmax,
 115                 .extra1         = &min_slot_table_size,
 116                 .extra2         = &max_tcp_slot_table_limit
 117         },
 118         {
 119                 .procname       = "min_resvport",
 120                 .data           = &xprt_min_resvport,
 121                 .maxlen         = sizeof(unsigned int),
 122                 .mode           = 0644,
 123                 .proc_handler   = proc_dointvec_minmax,
 124                 .extra1         = &xprt_min_resvport_limit,
 125                 .extra2         = &xprt_max_resvport_limit
 126         },
 127         {
 128                 .procname       = "max_resvport",
 129                 .data           = &xprt_max_resvport,
 130                 .maxlen         = sizeof(unsigned int),
 131                 .mode           = 0644,
 132                 .proc_handler   = proc_dointvec_minmax,
 133                 .extra1         = &xprt_min_resvport_limit,
 134                 .extra2         = &xprt_max_resvport_limit
 135         },
 136         {
 137                 .procname       = "tcp_fin_timeout",
 138                 .data           = &xs_tcp_fin_timeout,
 139                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 140                 .mode           = 0644,
 141                 .proc_handler   = proc_dointvec_jiffies,
 142         },
 143         { },
 144 };
 145
 146 static ctl_table sunrpc_table[] = {
 147         {
 148                 .procname       = "sunrpc",
 149                 .mode           = 0555,
 150                 .child          = xs_tunables_table
 151         },
 152         { },
 153 };
 154
 155 #endif
 156
 157 /*
 158  * Wait duration for a reply from the RPC portmapper.
 159  */
 160 #define XS_BIND_TO              (60U * HZ)
 161
 162 /*
 163  * Delay if a UDP socket connect error occurs.  This is most likely some
 164  * kind of resource problem on the local host.
 165  */
 166 #define XS_UDP_REEST_TO         (2U * HZ)
 167
 168 /*
 169  * The reestablish timeout allows clients to delay for a bit before attempting
 170  * to reconnect to a server that just dropped our connection.
 171  *
 172  * We implement an exponential backoff when trying to reestablish a TCP
 173  * transport connection with the server.  Some servers like to drop a TCP
 174  * connection when they are overworked, so we start with a short timeout and
 175  * increase over time if the server is down or not responding.
 176  */
 177 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 178 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 179
 180 /*
 181  * TCP idle timeout; client drops the transport socket if it is idle
 182  * for this long.  Note that we also timeout UDP sockets to prevent
 183  * holding port numbers when there is no RPC traffic.
 184  */
 185 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 186
 187 #ifdef RPC_DEBUG
 188 # undef  RPC_DEBUG_DATA
 189 # define RPCDBG_FACILITY        RPCDBG_TRANS
 190 #endif
 191
 192 #ifdef RPC_DEBUG_DATA
 193 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 194 {
 195         u8 *buf = (u8 *) packet;
 196         int j;
 197
 198         dprintk("RPC:       %s\n", msg);
 199         for (j = 0; j < count && j < 128; j += 4) {
 200                 if (!(j & 31)) {
 201                         if (j)
 202                                 dprintk("\n");
 203                         dprintk("0x%04x ", j);
 204                 }
 205                 dprintk("%02x%02x%02x%02x ",
 206                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 207         }
 208         dprintk("\n");
 209 }
 210 #else
 211 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 212 {
 213         /* NOP */
 214 }
 215 #endif
 216
 217 struct sock_xprt {
 218         struct rpc_xprt         xprt;
 219
 220         /*
 221          * Network layer
 222          */
 223         struct socket *         sock;
 224         struct sock *           inet;
 225
 226         /*
 227          * State of TCP reply receive
 228          */
 229         __be32                  tcp_fraghdr,
 230                                 tcp_xid,
 231                                 tcp_calldir;
 232
 233         u32                     tcp_offset,
 234                                 tcp_reclen;
 235
 236         unsigned long           tcp_copied,
 237                                 tcp_flags;
 238
 239         /*
 240          * Connection of transports
 241          */
 242         struct delayed_work     connect_worker;
 243         struct sockaddr_storage srcaddr;
 244         unsigned short          srcport;
 245
 246         /*
 247          * UDP socket buffer size parameters
 248          */
 249         size_t                  rcvsize,
 250                                 sndsize;
 251
 252         /*
 253          * Saved socket callback addresses
 254          */
 255         void                    (*old_data_ready)(struct sock *, int);
 256         void                    (*old_state_change)(struct sock *);
 257         void                    (*old_write_space)(struct sock *);
 258 };
 259
 260 /*
 261  * TCP receive state flags
 262  */
 263 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 264 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 265 #define TCP_RCV_COPY_XID        (1UL << 2)
 266 #define TCP_RCV_COPY_DATA       (1UL << 3)
 267 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 268 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 269
 270 /*
 271  * TCP RPC flags
 272  */
 273 #define TCP_RPC_REPLY           (1UL << 6)
 274
 275 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 276 {
 277         return (struct sockaddr *) &xprt->addr;
 278 }
 279
 280 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 281 {
 282         return (struct sockaddr_un *) &xprt->addr;
 283 }
 284
 285 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 286 {
 287         return (struct sockaddr_in *) &xprt->addr;
 288 }
 289
 290 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 291 {
 292         return (struct sockaddr_in6 *) &xprt->addr;
 293 }
 294
 295 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 296 {
 297         struct sockaddr *sap = xs_addr(xprt);
 298         struct sockaddr_in6 *sin6;
 299         struct sockaddr_in *sin;
 300         struct sockaddr_un *sun;
 301         char buf[128];
 302
 303         switch (sap->sa_family) {
 304         case AF_LOCAL:
 305                 sun = xs_addr_un(xprt);
 306                 strlcpy(buf, sun->sun_path, sizeof(buf));
 307                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 308                                                 kstrdup(buf, GFP_KERNEL);
 309                 break;
 310         case AF_INET:
 311                 (void)rpc_ntop(sap, buf, sizeof(buf));
 312                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 313                                                 kstrdup(buf, GFP_KERNEL);
 314                 sin = xs_addr_in(xprt);
 315                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 316                 break;
 317         case AF_INET6:
 318                 (void)rpc_ntop(sap, buf, sizeof(buf));
 319                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 320                                                 kstrdup(buf, GFP_KERNEL);
 321                 sin6 = xs_addr_in6(xprt);
 322                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 323                 break;
 324         default:
 325                 BUG();
 326         }
 327
 328         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 329 }
 330
 331 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 332 {
 333         struct sockaddr *sap = xs_addr(xprt);
 334         char buf[128];
 335
 336         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 337         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 338
 339         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 340         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 341 }
 342
 343 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 344                                      const char *protocol,
 345                                      const char *netid)
 346 {
 347         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 348         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 349         xs_format_common_peer_addresses(xprt);
 350         xs_format_common_peer_ports(xprt);
 351 }
 352
 353 static void xs_update_peer_port(struct rpc_xprt *xprt)
 354 {
 355         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 356         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 357
 358         xs_format_common_peer_ports(xprt);
 359 }
 360
 361 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 362 {
 363         unsigned int i;
 364
 365         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 366                 switch (i) {
 367                 case RPC_DISPLAY_PROTO:
 368                 case RPC_DISPLAY_NETID:
 369                         continue;
 370                 default:
 371                         kfree(xprt->address_strings[i]);
 372                 }
 373 }
 374
 375 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 376
 377 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 378 {
 379         struct msghdr msg = {
 380                 .msg_name       = addr,
 381                 .msg_namelen    = addrlen,
 382                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 383         };
 384         struct kvec iov = {
 385                 .iov_base       = vec->iov_base + base,
 386                 .iov_len        = vec->iov_len - base,
 387         };
 388
 389         if (iov.iov_len != 0)
 390                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 391         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 392 }
 393
 394 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
 395 {
 396         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 397                         int offset, size_t size, int flags);
 398         struct page **ppage;
 399         unsigned int remainder;
 400         int err, sent = 0;
 401
 402         remainder = xdr->page_len - base;
 403         base += xdr->page_base;
 404         ppage = xdr->pages + (base >> PAGE_SHIFT);
 405         base &= ~PAGE_MASK;
 406         do_sendpage = sock->ops->sendpage;
 407         if (!zerocopy)
 408                 do_sendpage = sock_no_sendpage;
 409         for(;;) {
 410                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 411                 int flags = XS_SENDMSG_FLAGS;
 412
 413                 remainder -= len;
 414                 if (remainder != 0 || more)
 415                         flags |= MSG_MORE;
 416                 err = do_sendpage(sock, *ppage, base, len, flags);
 417                 if (remainder == 0 || err != len)
 418                         break;
 419                 sent += err;
 420                 ppage++;
 421                 base = 0;
 422         }
 423         if (sent == 0)
 424                 return err;
 425         if (err > 0)
 426                 sent += err;
 427         return sent;
 428 }
 429
 430 /**
 431  * xs_sendpages - write pages directly to a socket
 432  * @sock: socket to send on
 433  * @addr: UDP only -- address of destination
 434  * @addrlen: UDP only -- length of destination address
 435  * @xdr: buffer containing this request
 436  * @base: starting position in the buffer
 437  * @zerocopy: true if it is safe to use sendpage()
 438  *
 439  */
 440 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
 441 {
 442         unsigned int remainder = xdr->len - base;
 443         int err, sent = 0;
 444
 445         if (unlikely(!sock))
 446                 return -ENOTSOCK;
 447
 448         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 449         if (base != 0) {
 450                 addr = NULL;
 451                 addrlen = 0;
 452         }
 453
 454         if (base < xdr->head[0].iov_len || addr != NULL) {
 455                 unsigned int len = xdr->head[0].iov_len - base;
 456                 remainder -= len;
 457                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 458                 if (remainder == 0 || err != len)
 459                         goto out;
 460                 sent += err;
 461                 base = 0;
 462         } else
 463                 base -= xdr->head[0].iov_len;
 464
 465         if (base < xdr->page_len) {
 466                 unsigned int len = xdr->page_len - base;
 467                 remainder -= len;
 468                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
 469                 if (remainder == 0 || err != len)
 470                         goto out;
 471                 sent += err;
 472                 base = 0;
 473         } else
 474                 base -= xdr->page_len;
 475
 476         if (base >= xdr->tail[0].iov_len)
 477                 return sent;
 478         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 479 out:
 480         if (sent == 0)
 481                 return err;
 482         if (err > 0)
 483                 sent += err;
 484         return sent;
 485 }
 486
 487 static void xs_nospace_callback(struct rpc_task *task)
 488 {
 489         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 490
 491         transport->inet->sk_write_pending--;
 492         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 493 }
 494
 495 /**
 496  * xs_nospace - place task on wait queue if transmit was incomplete
 497  * @task: task to put to sleep
 498  *
 499  */
 500 static int xs_nospace(struct rpc_task *task)
 501 {
 502         struct rpc_rqst *req = task->tk_rqstp;
 503         struct rpc_xprt *xprt = req->rq_xprt;
 504         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 505         struct sock *sk = transport->inet;
 506         int ret = -EAGAIN;
 507
 508         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 509                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 510                         req->rq_slen);
 511
 512         /* Protect against races with write_space */
 513         spin_lock_bh(&xprt->transport_lock);
 514
 515         /* Don't race with disconnect */
 516         if (xprt_connected(xprt)) {
 517                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 518                         /*
 519                          * Notify TCP that we're limited by the application
 520                          * window size
 521                          */
 522                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 523                         sk->sk_write_pending++;
 524                         /* ...and wait for more buffer space */
 525                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 526                 }
 527         } else {
 528                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 529                 ret = -ENOTCONN;
 530         }
 531
 532         spin_unlock_bh(&xprt->transport_lock);
 533
 534         /* Race breaker in case memory is freed before above code is called */
 535         sk->sk_write_space(sk);
 536         return ret;
 537 }
 538
 539 /*
 540  * Construct a stream transport record marker in @buf.
 541  */
 542 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 543 {
 544         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 545         rpc_fraghdr *base = buf->head[0].iov_base;
 546         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 547 }
 548
 549 /**
 550  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 551  * @task: RPC task that manages the state of an RPC request
 552  *
 553  * Return values:
 554  *        0:    The request has been sent
 555  *   EAGAIN:    The socket was blocked, please call again later to
 556  *              complete the request
 557  * ENOTCONN:    Caller needs to invoke connect logic then call again
 558  *    other:    Some other error occured, the request was not sent
 559  */
 560 static int xs_local_send_request(struct rpc_task *task)
 561 {
 562         struct rpc_rqst *req = task->tk_rqstp;
 563         struct rpc_xprt *xprt = req->rq_xprt;
 564         struct sock_xprt *transport =
 565                                 container_of(xprt, struct sock_xprt, xprt);
 566         struct xdr_buf *xdr = &req->rq_snd_buf;
 567         int status;
 568
 569         xs_encode_stream_record_marker(&req->rq_snd_buf);
 570
 571         xs_pktdump("packet data:",
 572                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 573
 574         status = xs_sendpages(transport->sock, NULL, 0,
 575                                                 xdr, req->rq_bytes_sent, true);
 576         dprintk("RPC:       %s(%u) = %d\n",
 577                         __func__, xdr->len - req->rq_bytes_sent, status);
 578         if (likely(status >= 0)) {
 579                 req->rq_bytes_sent += status;
 580                 req->rq_xmit_bytes_sent += status;
 581                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 582                         req->rq_bytes_sent = 0;
 583                         return 0;
 584                 }
 585                 status = -EAGAIN;
 586         }
 587
 588         switch (status) {
 589         case -EAGAIN:
 590                 status = xs_nospace(task);
 591                 break;
 592         default:
 593                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 594                         -status);
 595         case -EPIPE:
 596                 xs_close(xprt);
 597                 status = -ENOTCONN;
 598         }
 599
 600         return status;
 601 }
 602
 603 /**
 604  * xs_udp_send_request - write an RPC request to a UDP socket
 605  * @task: address of RPC task that manages the state of an RPC request
 606  *
 607  * Return values:
 608  *        0:    The request has been sent
 609  *   EAGAIN:    The socket was blocked, please call again later to
 610  *              complete the request
 611  * ENOTCONN:    Caller needs to invoke connect logic then call again
 612  *    other:    Some other error occurred, the request was not sent
 613  */
 614 static int xs_udp_send_request(struct rpc_task *task)
 615 {
 616         struct rpc_rqst *req = task->tk_rqstp;
 617         struct rpc_xprt *xprt = req->rq_xprt;
 618         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 619         struct xdr_buf *xdr = &req->rq_snd_buf;
 620         int status;
 621
 622         xs_pktdump("packet data:",
 623                                 req->rq_svec->iov_base,
 624                                 req->rq_svec->iov_len);
 625
 626         if (!xprt_bound(xprt))
 627                 return -ENOTCONN;
 628         status = xs_sendpages(transport->sock,
 629                               xs_addr(xprt),
 630                               xprt->addrlen, xdr,
 631                               req->rq_bytes_sent, true);
 632
 633         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 634                         xdr->len - req->rq_bytes_sent, status);
 635
 636         if (status >= 0) {
 637                 req->rq_xmit_bytes_sent += status;
 638                 if (status >= req->rq_slen)
 639                         return 0;
 640                 /* Still some bytes left; set up for a retry later. */
 641                 status = -EAGAIN;
 642         }
 643
 644         switch (status) {
 645         case -ENOTSOCK:
 646                 status = -ENOTCONN;
 647                 /* Should we call xs_close() here? */
 648                 break;
 649         case -EAGAIN:
 650                 status = xs_nospace(task);
 651                 break;
 652         default:
 653                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 654                         -status);
 655         case -ENETUNREACH:
 656         case -EPIPE:
 657         case -ECONNREFUSED:
 658                 /* When the server has died, an ICMP port unreachable message
 659                  * prompts ECONNREFUSED. */
 660                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 661         }
 662
 663         return status;
 664 }
 665
 666 /**
 667  * xs_tcp_shutdown - gracefully shut down a TCP socket
 668  * @xprt: transport
 669  *
 670  * Initiates a graceful shutdown of the TCP socket by calling the
 671  * equivalent of shutdown(SHUT_WR);
 672  */
 673 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 674 {
 675         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 676         struct socket *sock = transport->sock;
 677
 678         if (sock != NULL)
 679                 kernel_sock_shutdown(sock, SHUT_WR);
 680 }
 681
 682 /**
 683  * xs_tcp_send_request - write an RPC request to a TCP socket
 684  * @task: address of RPC task that manages the state of an RPC request
 685  *
 686  * Return values:
 687  *        0:    The request has been sent
 688  *   EAGAIN:    The socket was blocked, please call again later to
 689  *              complete the request
 690  * ENOTCONN:    Caller needs to invoke connect logic then call again
 691  *    other:    Some other error occurred, the request was not sent
 692  *
 693  * XXX: In the case of soft timeouts, should we eventually give up
 694  *      if sendmsg is not able to make progress?
 695  */
 696 static int xs_tcp_send_request(struct rpc_task *task)
 697 {
 698         struct rpc_rqst *req = task->tk_rqstp;
 699         struct rpc_xprt *xprt = req->rq_xprt;
 700         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 701         struct xdr_buf *xdr = &req->rq_snd_buf;
 702         bool zerocopy = true;
 703         int status;
 704
 705         xs_encode_stream_record_marker(&req->rq_snd_buf);
 706
 707         xs_pktdump("packet data:",
 708                                 req->rq_svec->iov_base,
 709                                 req->rq_svec->iov_len);
 710         /* Don't use zero copy if this is a resend. If the RPC call
 711          * completes while the socket holds a reference to the pages,
 712          * then we may end up resending corrupted data.
 713          */
 714         if (task->tk_flags & RPC_TASK_SENT)
 715                 zerocopy = false;
 716
 717         /* Continue transmitting the packet/record. We must be careful
 718          * to cope with writespace callbacks arriving _after_ we have
 719          * called sendmsg(). */
 720         while (1) {
 721                 status = xs_sendpages(transport->sock,
 722                                         NULL, 0, xdr, req->rq_bytes_sent,
 723                                         zerocopy);
 724
 725                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 726                                 xdr->len - req->rq_bytes_sent, status);
 727
 728                 if (unlikely(status < 0))
 729                         break;
 730
 731                 /* If we've sent the entire packet, immediately
 732                  * reset the count of bytes sent. */
 733                 req->rq_bytes_sent += status;
 734                 req->rq_xmit_bytes_sent += status;
 735                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 736                         req->rq_bytes_sent = 0;
 737                         return 0;
 738                 }
 739
 740                 if (status != 0)
 741                         continue;
 742                 status = -EAGAIN;
 743                 break;
 744         }
 745
 746         switch (status) {
 747         case -ENOTSOCK:
 748                 status = -ENOTCONN;
 749                 /* Should we call xs_close() here? */
 750                 break;
 751         case -EAGAIN:
 752                 status = xs_nospace(task);
 753                 break;
 754         default:
 755                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 756                         -status);
 757         case -ECONNRESET:
 758                 xs_tcp_shutdown(xprt);
 759         case -ECONNREFUSED:
 760         case -ENOTCONN:
 761         case -EPIPE:
 762                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 763         }
 764
 765         return status;
 766 }
 767
 768 /**
 769  * xs_tcp_release_xprt - clean up after a tcp transmission
 770  * @xprt: transport
 771  * @task: rpc task
 772  *
 773  * This cleans up if an error causes us to abort the transmission of a request.
 774  * In this case, the socket may need to be reset in order to avoid confusing
 775  * the server.
 776  */
 777 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 778 {
 779         struct rpc_rqst *req;
 780
 781         if (task != xprt->snd_task)
 782                 return;
 783         if (task == NULL)
 784                 goto out_release;
 785         req = task->tk_rqstp;
 786         if (req == NULL)
 787                 goto out_release;
 788         if (req->rq_bytes_sent == 0)
 789                 goto out_release;
 790         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 791                 goto out_release;
 792         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 793 out_release:
 794         xprt_release_xprt(xprt, task);
 795 }
 796
 797 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 798 {
 799         transport->old_data_ready = sk->sk_data_ready;
 800         transport->old_state_change = sk->sk_state_change;
 801         transport->old_write_space = sk->sk_write_space;
 802 }
 803
 804 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 805 {
 806         sk->sk_data_ready = transport->old_data_ready;
 807         sk->sk_state_change = transport->old_state_change;
 808         sk->sk_write_space = transport->old_write_space;
 809 }
 810
 811 static void xs_reset_transport(struct sock_xprt *transport)
 812 {
 813         struct socket *sock = transport->sock;
 814         struct sock *sk = transport->inet;
 815
 816         if (sk == NULL)
 817                 return;
 818
 819         transport->srcport = 0;
 820
 821         write_lock_bh(&sk->sk_callback_lock);
 822         transport->inet = NULL;
 823         transport->sock = NULL;
 824
 825         sk->sk_user_data = NULL;
 826
 827         xs_restore_old_callbacks(transport, sk);
 828         write_unlock_bh(&sk->sk_callback_lock);
 829
 830         sk->sk_no_check = 0;
 831
 832         sock_release(sock);
 833 }
 834
 835 /**
 836  * xs_close - close a socket
 837  * @xprt: transport
 838  *
 839  * This is used when all requests are complete; ie, no DRC state remains
 840  * on the server we want to save.
 841  *
 842  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 843  * xs_reset_transport() zeroing the socket from underneath a writer.
 844  */
 845 static void xs_close(struct rpc_xprt *xprt)
 846 {
 847         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 848
 849         dprintk("RPC:       xs_close xprt %p\n", xprt);
 850
 851         xs_reset_transport(transport);
 852         xprt->reestablish_timeout = 0;
 853
 854         smp_mb__before_clear_bit();
 855         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 856         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 857         clear_bit(XPRT_CLOSING, &xprt->state);
 858         smp_mb__after_clear_bit();
 859         xprt_disconnect_done(xprt);
 860 }
 861
 862 static void xs_tcp_close(struct rpc_xprt *xprt)
 863 {
 864         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 865                 xs_close(xprt);
 866         else
 867                 xs_tcp_shutdown(xprt);
 868 }
 869
 870 static void xs_local_destroy(struct rpc_xprt *xprt)
 871 {
 872         xs_close(xprt);
 873         xs_free_peer_addresses(xprt);
 874         xprt_free(xprt);
 875         module_put(THIS_MODULE);
 876 }
 877
 878 /**
 879  * xs_destroy - prepare to shutdown a transport
 880  * @xprt: doomed transport
 881  *
 882  */
 883 static void xs_destroy(struct rpc_xprt *xprt)
 884 {
 885         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 886
 887         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 888
 889         cancel_delayed_work_sync(&transport->connect_worker);
 890
 891         xs_local_destroy(xprt);
 892 }
 893
 894 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 895 {
 896         return (struct rpc_xprt *) sk->sk_user_data;
 897 }
 898
 899 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 900 {
 901         struct xdr_skb_reader desc = {
 902                 .skb            = skb,
 903                 .offset         = sizeof(rpc_fraghdr),
 904                 .count          = skb->len - sizeof(rpc_fraghdr),
 905         };
 906
 907         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 908                 return -1;
 909         if (desc.count)
 910                 return -1;
 911         return 0;
 912 }
 913
 914 /**
 915  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 916  * @sk: socket with data to read
 917  * @len: how much data to read
 918  *
 919  * Currently this assumes we can read the whole reply in a single gulp.
 920  */
 921 static void xs_local_data_ready(struct sock *sk, int len)
 922 {
 923         struct rpc_task *task;
 924         struct rpc_xprt *xprt;
 925         struct rpc_rqst *rovr;
 926         struct sk_buff *skb;
 927         int err, repsize, copied;
 928         u32 _xid;
 929         __be32 *xp;
 930
 931         read_lock_bh(&sk->sk_callback_lock);
 932         dprintk("RPC:       %s...\n", __func__);
 933         xprt = xprt_from_sock(sk);
 934         if (xprt == NULL)
 935                 goto out;
 936
 937         skb = skb_recv_datagram(sk, 0, 1, &err);
 938         if (skb == NULL)
 939                 goto out;
 940
 941         repsize = skb->len - sizeof(rpc_fraghdr);
 942         if (repsize < 4) {
 943                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 944                 goto dropit;
 945         }
 946
 947         /* Copy the XID from the skb... */
 948         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 949         if (xp == NULL)
 950                 goto dropit;
 951
 952         /* Look up and lock the request corresponding to the given XID */
 953         spin_lock(&xprt->transport_lock);
 954         rovr = xprt_lookup_rqst(xprt, *xp);
 955         if (!rovr)
 956                 goto out_unlock;
 957         task = rovr->rq_task;
 958
 959         copied = rovr->rq_private_buf.buflen;
 960         if (copied > repsize)
 961                 copied = repsize;
 962
 963         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 964                 dprintk("RPC:       sk_buff copy failed\n");
 965                 goto out_unlock;
 966         }
 967
 968         xprt_complete_rqst(task, copied);
 969
 970  out_unlock:
 971         spin_unlock(&xprt->transport_lock);
 972  dropit:
 973         skb_free_datagram(sk, skb);
 974  out:
 975         read_unlock_bh(&sk->sk_callback_lock);
 976 }
 977
 978 /**
 979  * xs_udp_data_ready - "data ready" callback for UDP sockets
 980  * @sk: socket with data to read
 981  * @len: how much data to read
 982  *
 983  */
 984 static void xs_udp_data_ready(struct sock *sk, int len)
 985 {
 986         struct rpc_task *task;
 987         struct rpc_xprt *xprt;
 988         struct rpc_rqst *rovr;
 989         struct sk_buff *skb;
 990         int err, repsize, copied;
 991         u32 _xid;
 992         __be32 *xp;
 993
 994         read_lock_bh(&sk->sk_callback_lock);
 995         dprintk("RPC:       xs_udp_data_ready...\n");
 996         if (!(xprt = xprt_from_sock(sk)))
 997                 goto out;
 998
 999         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
1000                 goto out;
1001
1002         repsize = skb->len - sizeof(struct udphdr);
1003         if (repsize < 4) {
1004                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1005                 goto dropit;
1006         }
1007
1008         /* Copy the XID from the skb... */
1009         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1010                                 sizeof(_xid), &_xid);
1011         if (xp == NULL)
1012                 goto dropit;
1013
1014         /* Look up and lock the request corresponding to the given XID */
1015         spin_lock(&xprt->transport_lock);
1016         rovr = xprt_lookup_rqst(xprt, *xp);
1017         if (!rovr)
1018                 goto out_unlock;
1019         task = rovr->rq_task;
1020
1021         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1022                 copied = repsize;
1023
1024         /* Suck it into the iovec, verify checksum if not done by hw. */
1025         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1026                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1027                 goto out_unlock;
1028         }
1029
1030         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1031
1032         xprt_adjust_cwnd(xprt, task, copied);
1033         xprt_complete_rqst(task, copied);
1034
1035  out_unlock:
1036         spin_unlock(&xprt->transport_lock);
1037  dropit:
1038         skb_free_datagram(sk, skb);
1039  out:
1040         read_unlock_bh(&sk->sk_callback_lock);
1041 }
1042
1043 /*
1044  * Helper function to force a TCP close if the server is sending
1045  * junk and/or it has put us in CLOSE_WAIT
1046  */
1047 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1048 {
1049         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1050         xprt_force_disconnect(xprt);
1051 }
1052
1053 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1054 {
1055         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1056         size_t len, used;
1057         char *p;
1058
1059         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1060         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1061         used = xdr_skb_read_bits(desc, p, len);
1062         transport->tcp_offset += used;
1063         if (used != len)
1064                 return;
1065
1066         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1067         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1068                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1069         else
1070                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1071         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1072
1073         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1074         transport->tcp_offset = 0;
1075
1076         /* Sanity check of the record length */
1077         if (unlikely(transport->tcp_reclen < 8)) {
1078                 dprintk("RPC:       invalid TCP record fragment length\n");
1079                 xs_tcp_force_close(xprt);
1080                 return;
1081         }
1082         dprintk("RPC:       reading TCP record fragment of length %d\n",
1083                         transport->tcp_reclen);
1084 }
1085
1086 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1087 {
1088         if (transport->tcp_offset == transport->tcp_reclen) {
1089                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1090                 transport->tcp_offset = 0;
1091                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1092                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1093                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1094                         transport->tcp_copied = 0;
1095                 }
1096         }
1097 }
1098
1099 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1100 {
1101         size_t len, used;
1102         char *p;
1103
1104         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1105         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1106         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1107         used = xdr_skb_read_bits(desc, p, len);
1108         transport->tcp_offset += used;
1109         if (used != len)
1110                 return;
1111         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1112         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1113         transport->tcp_copied = 4;
1114         dprintk("RPC:       reading %s XID %08x\n",
1115                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1116                                                               : "request with",
1117                         ntohl(transport->tcp_xid));
1118         xs_tcp_check_fraghdr(transport);
1119 }
1120
1121 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1122                                        struct xdr_skb_reader *desc)
1123 {
1124         size_t len, used;
1125         u32 offset;
1126         char *p;
1127
1128         /*
1129          * We want transport->tcp_offset to be 8 at the end of this routine
1130          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1131          * When this function is called for the first time,
1132          * transport->tcp_offset is 4 (after having already read the xid).
1133          */
1134         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1135         len = sizeof(transport->tcp_calldir) - offset;
1136         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1137         p = ((char *) &transport->tcp_calldir) + offset;
1138         used = xdr_skb_read_bits(desc, p, len);
1139         transport->tcp_offset += used;
1140         if (used != len)
1141                 return;
1142         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1143         /*
1144          * We don't yet have the XDR buffer, so we will write the calldir
1145          * out after we get the buffer from the 'struct rpc_rqst'
1146          */
1147         switch (ntohl(transport->tcp_calldir)) {
1148         case RPC_REPLY:
1149                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1150                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1151                 transport->tcp_flags |= TCP_RPC_REPLY;
1152                 break;
1153         case RPC_CALL:
1154                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1155                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1156                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1157                 break;
1158         default:
1159                 dprintk("RPC:       invalid request message type\n");
1160                 xs_tcp_force_close(&transport->xprt);
1161         }
1162         xs_tcp_check_fraghdr(transport);
1163 }
1164
1165 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1166                                      struct xdr_skb_reader *desc,
1167                                      struct rpc_rqst *req)
1168 {
1169         struct sock_xprt *transport =
1170                                 container_of(xprt, struct sock_xprt, xprt);
1171         struct xdr_buf *rcvbuf;
1172         size_t len;
1173         ssize_t r;
1174
1175         rcvbuf = &req->rq_private_buf;
1176
1177         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1178                 /*
1179                  * Save the RPC direction in the XDR buffer
1180                  */
1181                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1182                         &transport->tcp_calldir,
1183                         sizeof(transport->tcp_calldir));
1184                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1185                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1186         }
1187
1188         len = desc->count;
1189         if (len > transport->tcp_reclen - transport->tcp_offset) {
1190                 struct xdr_skb_reader my_desc;
1191
1192                 len = transport->tcp_reclen - transport->tcp_offset;
1193                 memcpy(&my_desc, desc, sizeof(my_desc));
1194                 my_desc.count = len;
1195                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1196                                           &my_desc, xdr_skb_read_bits);
1197                 desc->count -= r;
1198                 desc->offset += r;
1199         } else
1200                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1201                                           desc, xdr_skb_read_bits);
1202
1203         if (r > 0) {
1204                 transport->tcp_copied += r;
1205                 transport->tcp_offset += r;
1206         }
1207         if (r != len) {
1208                 /* Error when copying to the receive buffer,
1209                  * usually because we weren't able to allocate
1210                  * additional buffer pages. All we can do now
1211                  * is turn off TCP_RCV_COPY_DATA, so the request
1212                  * will not receive any additional updates,
1213                  * and time out.
1214                  * Any remaining data from this record will
1215                  * be discarded.
1216                  */
1217                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1218                 dprintk("RPC:       XID %08x truncated request\n",
1219                                 ntohl(transport->tcp_xid));
1220                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1221                                 "tcp_offset = %u, tcp_reclen = %u\n",
1222                                 xprt, transport->tcp_copied,
1223                                 transport->tcp_offset, transport->tcp_reclen);
1224                 return;
1225         }
1226
1227         dprintk("RPC:       XID %08x read %Zd bytes\n",
1228                         ntohl(transport->tcp_xid), r);
1229         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1230                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1231                         transport->tcp_offset, transport->tcp_reclen);
1232
1233         if (transport->tcp_copied == req->rq_private_buf.buflen)
1234                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1235         else if (transport->tcp_offset == transport->tcp_reclen) {
1236                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1237                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1238         }
1239 }
1240
1241 /*
1242  * Finds the request corresponding to the RPC xid and invokes the common
1243  * tcp read code to read the data.
1244  */
1245 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1246                                     struct xdr_skb_reader *desc)
1247 {
1248         struct sock_xprt *transport =
1249                                 container_of(xprt, struct sock_xprt, xprt);
1250         struct rpc_rqst *req;
1251
1252         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1253
1254         /* Find and lock the request corresponding to this xid */
1255         spin_lock(&xprt->transport_lock);
1256         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1257         if (!req) {
1258                 dprintk("RPC:       XID %08x request not found!\n",
1259                                 ntohl(transport->tcp_xid));
1260                 spin_unlock(&xprt->transport_lock);
1261                 return -1;
1262         }
1263
1264         xs_tcp_read_common(xprt, desc, req);
1265
1266         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1267                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1268
1269         spin_unlock(&xprt->transport_lock);
1270         return 0;
1271 }
1272
1273 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1274 /*
1275  * Obtains an rpc_rqst previously allocated and invokes the common
1276  * tcp read code to read the data.  The result is placed in the callback
1277  * queue.
1278  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1279  * connection and return -1.
1280  */
1281 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1282                                        struct xdr_skb_reader *desc)
1283 {
1284         struct sock_xprt *transport =
1285                                 container_of(xprt, struct sock_xprt, xprt);
1286         struct rpc_rqst *req;
1287
1288         req = xprt_alloc_bc_request(xprt);
1289         if (req == NULL) {
1290                 printk(KERN_WARNING "Callback slot table overflowed\n");
1291                 xprt_force_disconnect(xprt);
1292                 return -1;
1293         }
1294
1295         req->rq_xid = transport->tcp_xid;
1296         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1297         xs_tcp_read_common(xprt, desc, req);
1298
1299         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1300                 struct svc_serv *bc_serv = xprt->bc_serv;
1301
1302                 /*
1303                  * Add callback request to callback list.  The callback
1304                  * service sleeps on the sv_cb_waitq waiting for new
1305                  * requests.  Wake it up after adding enqueing the
1306                  * request.
1307                  */
1308                 dprintk("RPC:       add callback request to list\n");
1309                 spin_lock(&bc_serv->sv_cb_lock);
1310                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1311                 spin_unlock(&bc_serv->sv_cb_lock);
1312                 wake_up(&bc_serv->sv_cb_waitq);
1313         }
1314
1315         req->rq_private_buf.len = transport->tcp_copied;
1316
1317         return 0;
1318 }
1319
1320 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1321                                         struct xdr_skb_reader *desc)
1322 {
1323         struct sock_xprt *transport =
1324                                 container_of(xprt, struct sock_xprt, xprt);
1325
1326         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1327                 xs_tcp_read_reply(xprt, desc) :
1328                 xs_tcp_read_callback(xprt, desc);
1329 }
1330 #else
1331 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1332                                         struct xdr_skb_reader *desc)
1333 {
1334         return xs_tcp_read_reply(xprt, desc);
1335 }
1336 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1337
1338 /*
1339  * Read data off the transport.  This can be either an RPC_CALL or an
1340  * RPC_REPLY.  Relay the processing to helper functions.
1341  */
1342 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1343                                     struct xdr_skb_reader *desc)
1344 {
1345         struct sock_xprt *transport =
1346                                 container_of(xprt, struct sock_xprt, xprt);
1347
1348         if (_xs_tcp_read_data(xprt, desc) == 0)
1349                 xs_tcp_check_fraghdr(transport);
1350         else {
1351                 /*
1352                  * The transport_lock protects the request handling.
1353                  * There's no need to hold it to update the tcp_flags.
1354                  */
1355                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1356         }
1357 }
1358
1359 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1360 {
1361         size_t len;
1362
1363         len = transport->tcp_reclen - transport->tcp_offset;
1364         if (len > desc->count)
1365                 len = desc->count;
1366         desc->count -= len;
1367         desc->offset += len;
1368         transport->tcp_offset += len;
1369         dprintk("RPC:       discarded %Zu bytes\n", len);
1370         xs_tcp_check_fraghdr(transport);
1371 }
1372
1373 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1374 {
1375         struct rpc_xprt *xprt = rd_desc->arg.data;
1376         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1377         struct xdr_skb_reader desc = {
1378                 .skb    = skb,
1379                 .offset = offset,
1380                 .count  = len,
1381         };
1382
1383         dprintk("RPC:       xs_tcp_data_recv started\n");
1384         do {
1385                 /* Read in a new fragment marker if necessary */
1386                 /* Can we ever really expect to get completely empty fragments? */
1387                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1388                         xs_tcp_read_fraghdr(xprt, &desc);
1389                         continue;
1390                 }
1391                 /* Read in the xid if necessary */
1392                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1393                         xs_tcp_read_xid(transport, &desc);
1394                         continue;
1395                 }
1396                 /* Read in the call/reply flag */
1397                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1398                         xs_tcp_read_calldir(transport, &desc);
1399                         continue;
1400                 }
1401                 /* Read in the request data */
1402                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1403                         xs_tcp_read_data(xprt, &desc);
1404                         continue;
1405                 }
1406                 /* Skip over any trailing bytes on short reads */
1407                 xs_tcp_read_discard(transport, &desc);
1408         } while (desc.count);
1409         dprintk("RPC:       xs_tcp_data_recv done\n");
1410         return len - desc.count;
1411 }
1412
1413 /**
1414  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1415  * @sk: socket with data to read
1416  * @bytes: how much data to read
1417  *
1418  */
1419 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1420 {
1421         struct rpc_xprt *xprt;
1422         read_descriptor_t rd_desc;
1423         int read;
1424
1425         dprintk("RPC:       xs_tcp_data_ready...\n");
1426
1427         read_lock_bh(&sk->sk_callback_lock);
1428         if (!(xprt = xprt_from_sock(sk)))
1429                 goto out;
1430         /* Any data means we had a useful conversation, so
1431          * the we don't need to delay the next reconnect
1432          */
1433         if (xprt->reestablish_timeout)
1434                 xprt->reestablish_timeout = 0;
1435
1436         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1437         rd_desc.arg.data = xprt;
1438         do {
1439                 rd_desc.count = 65536;
1440                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1441         } while (read > 0);
1442 out:
1443         read_unlock_bh(&sk->sk_callback_lock);
1444 }
1445
1446 /*
1447  * Do the equivalent of linger/linger2 handling for dealing with
1448  * broken servers that don't close the socket in a timely
1449  * fashion
1450  */
1451 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1452                 unsigned long timeout)
1453 {
1454         struct sock_xprt *transport;
1455
1456         if (xprt_test_and_set_connecting(xprt))
1457                 return;
1458         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1459         transport = container_of(xprt, struct sock_xprt, xprt);
1460         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1461                            timeout);
1462 }
1463
1464 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1465 {
1466         struct sock_xprt *transport;
1467
1468         transport = container_of(xprt, struct sock_xprt, xprt);
1469
1470         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1471             !cancel_delayed_work(&transport->connect_worker))
1472                 return;
1473         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1474         xprt_clear_connecting(xprt);
1475 }
1476
1477 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1478 {
1479         smp_mb__before_clear_bit();
1480         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1481         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1482         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1483         clear_bit(XPRT_CLOSING, &xprt->state);
1484         smp_mb__after_clear_bit();
1485 }
1486
1487 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1488 {
1489         xs_sock_reset_connection_flags(xprt);
1490         /* Mark transport as closed and wake up all pending tasks */
1491         xprt_disconnect_done(xprt);
1492 }
1493
1494 /**
1495  * xs_tcp_state_change - callback to handle TCP socket state changes
1496  * @sk: socket whose state has changed
1497  *
1498  */
1499 static void xs_tcp_state_change(struct sock *sk)
1500 {
1501         struct rpc_xprt *xprt;
1502
1503         read_lock_bh(&sk->sk_callback_lock);
1504         if (!(xprt = xprt_from_sock(sk)))
1505                 goto out;
1506         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1507         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1508                         sk->sk_state, xprt_connected(xprt),
1509                         sock_flag(sk, SOCK_DEAD),
1510                         sock_flag(sk, SOCK_ZAPPED),
1511                         sk->sk_shutdown);
1512
1513         switch (sk->sk_state) {
1514         case TCP_ESTABLISHED:
1515                 spin_lock(&xprt->transport_lock);
1516                 if (!xprt_test_and_set_connected(xprt)) {
1517                         struct sock_xprt *transport = container_of(xprt,
1518                                         struct sock_xprt, xprt);
1519
1520                         /* Reset TCP record info */
1521                         transport->tcp_offset = 0;
1522                         transport->tcp_reclen = 0;
1523                         transport->tcp_copied = 0;
1524                         transport->tcp_flags =
1525                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1526
1527                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1528                 }
1529                 spin_unlock(&xprt->transport_lock);
1530                 break;
1531         case TCP_FIN_WAIT1:
1532                 /* The client initiated a shutdown of the socket */
1533                 xprt->connect_cookie++;
1534                 xprt->reestablish_timeout = 0;
1535                 set_bit(XPRT_CLOSING, &xprt->state);
1536                 smp_mb__before_clear_bit();
1537                 clear_bit(XPRT_CONNECTED, &xprt->state);
1538                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1539                 smp_mb__after_clear_bit();
1540                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1541                 break;
1542         case TCP_CLOSE_WAIT:
1543                 /* The server initiated a shutdown of the socket */
1544                 xprt->connect_cookie++;
1545                 clear_bit(XPRT_CONNECTED, &xprt->state);
1546                 xs_tcp_force_close(xprt);
1547         case TCP_CLOSING:
1548                 /*
1549                  * If the server closed down the connection, make sure that
1550                  * we back off before reconnecting
1551                  */
1552                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1553                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1554                 break;
1555         case TCP_LAST_ACK:
1556                 set_bit(XPRT_CLOSING, &xprt->state);
1557                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1558                 smp_mb__before_clear_bit();
1559                 clear_bit(XPRT_CONNECTED, &xprt->state);
1560                 smp_mb__after_clear_bit();
1561                 break;
1562         case TCP_CLOSE:
1563                 xs_tcp_cancel_linger_timeout(xprt);
1564                 xs_sock_mark_closed(xprt);
1565         }
1566  out:
1567         read_unlock_bh(&sk->sk_callback_lock);
1568 }
1569
1570 static void xs_write_space(struct sock *sk)
1571 {
1572         struct socket *sock;
1573         struct rpc_xprt *xprt;
1574
1575         if (unlikely(!(sock = sk->sk_socket)))
1576                 return;
1577         clear_bit(SOCK_NOSPACE, &sock->flags);
1578
1579         if (unlikely(!(xprt = xprt_from_sock(sk))))
1580                 return;
1581         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1582                 return;
1583
1584         xprt_write_space(xprt);
1585 }
1586
1587 /**
1588  * xs_udp_write_space - callback invoked when socket buffer space
1589  *                             becomes available
1590  * @sk: socket whose state has changed
1591  *
1592  * Called when more output buffer space is available for this socket.
1593  * We try not to wake our writers until they can make "significant"
1594  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1595  * with a bunch of small requests.
1596  */
1597 static void xs_udp_write_space(struct sock *sk)
1598 {
1599         read_lock_bh(&sk->sk_callback_lock);
1600
1601         /* from net/core/sock.c:sock_def_write_space */
1602         if (sock_writeable(sk))
1603                 xs_write_space(sk);
1604
1605         read_unlock_bh(&sk->sk_callback_lock);
1606 }
1607
1608 /**
1609  * xs_tcp_write_space - callback invoked when socket buffer space
1610  *                             becomes available
1611  * @sk: socket whose state has changed
1612  *
1613  * Called when more output buffer space is available for this socket.
1614  * We try not to wake our writers until they can make "significant"
1615  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1616  * with a bunch of small requests.
1617  */
1618 static void xs_tcp_write_space(struct sock *sk)
1619 {
1620         read_lock_bh(&sk->sk_callback_lock);
1621
1622         /* from net/core/stream.c:sk_stream_write_space */
1623         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1624                 xs_write_space(sk);
1625
1626         read_unlock_bh(&sk->sk_callback_lock);
1627 }
1628
1629 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1630 {
1631         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1632         struct sock *sk = transport->inet;
1633
1634         if (transport->rcvsize) {
1635                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1636                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1637         }
1638         if (transport->sndsize) {
1639                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1640                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1641                 sk->sk_write_space(sk);
1642         }
1643 }
1644
1645 /**
1646  * xs_udp_set_buffer_size - set send and receive limits
1647  * @xprt: generic transport
1648  * @sndsize: requested size of send buffer, in bytes
1649  * @rcvsize: requested size of receive buffer, in bytes
1650  *
1651  * Set socket send and receive buffer size limits.
1652  */
1653 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1654 {
1655         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1656
1657         transport->sndsize = 0;
1658         if (sndsize)
1659                 transport->sndsize = sndsize + 1024;
1660         transport->rcvsize = 0;
1661         if (rcvsize)
1662                 transport->rcvsize = rcvsize + 1024;
1663
1664         xs_udp_do_set_buffer_size(xprt);
1665 }
1666
1667 /**
1668  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1669  * @task: task that timed out
1670  *
1671  * Adjust the congestion window after a retransmit timeout has occurred.
1672  */
1673 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1674 {
1675         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1676 }
1677
1678 static unsigned short xs_get_random_port(void)
1679 {
1680         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1681         unsigned short rand = (unsigned short) net_random() % range;
1682         return rand + xprt_min_resvport;
1683 }
1684
1685 /**
1686  * xs_set_port - reset the port number in the remote endpoint address
1687  * @xprt: generic transport
1688  * @port: new port number
1689  *
1690  */
1691 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1692 {
1693         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1694
1695         rpc_set_port(xs_addr(xprt), port);
1696         xs_update_peer_port(xprt);
1697 }
1698
1699 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1700 {
1701         unsigned short port = transport->srcport;
1702
1703         if (port == 0 && transport->xprt.resvport)
1704                 port = xs_get_random_port();
1705         return port;
1706 }
1707
1708 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1709 {
1710         if (transport->srcport != 0)
1711                 transport->srcport = 0;
1712         if (!transport->xprt.resvport)
1713                 return 0;
1714         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1715                 return xprt_max_resvport;
1716         return --port;
1717 }
1718 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1719 {
1720         struct sockaddr_storage myaddr;
1721         int err, nloop = 0;
1722         unsigned short port = xs_get_srcport(transport);
1723         unsigned short last;
1724
1725         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1726         do {
1727                 rpc_set_port((struct sockaddr *)&myaddr, port);
1728                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1729                                 transport->xprt.addrlen);
1730                 if (port == 0)
1731                         break;
1732                 if (err == 0) {
1733                         transport->srcport = port;
1734                         break;
1735                 }
1736                 last = port;
1737                 port = xs_next_srcport(transport, port);
1738                 if (port > last)
1739                         nloop++;
1740         } while (err == -EADDRINUSE && nloop != 2);
1741
1742         if (myaddr.ss_family == AF_INET)
1743                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1744                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1745                                 port, err ? "failed" : "ok", err);
1746         else
1747                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1748                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1749                                 port, err ? "failed" : "ok", err);
1750         return err;
1751 }
1752
1753 /*
1754  * We don't support autobind on AF_LOCAL sockets
1755  */
1756 static void xs_local_rpcbind(struct rpc_task *task)
1757 {
1758         rcu_read_lock();
1759         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1760         rcu_read_unlock();
1761 }
1762
1763 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1764 {
1765 }
1766
1767 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1768 static struct lock_class_key xs_key[2];
1769 static struct lock_class_key xs_slock_key[2];
1770
1771 static inline void xs_reclassify_socketu(struct socket *sock)
1772 {
1773         struct sock *sk = sock->sk;
1774
1775         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1776                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1777 }
1778
1779 static inline void xs_reclassify_socket4(struct socket *sock)
1780 {
1781         struct sock *sk = sock->sk;
1782
1783         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1784                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1785 }
1786
1787 static inline void xs_reclassify_socket6(struct socket *sock)
1788 {
1789         struct sock *sk = sock->sk;
1790
1791         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1792                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1793 }
1794
1795 static inline void xs_reclassify_socket(int family, struct socket *sock)
1796 {
1797         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1798         if (sock_owned_by_user(sock->sk))
1799                 return;
1800
1801         switch (family) {
1802         case AF_LOCAL:
1803                 xs_reclassify_socketu(sock);
1804                 break;
1805         case AF_INET:
1806                 xs_reclassify_socket4(sock);
1807                 break;
1808         case AF_INET6:
1809                 xs_reclassify_socket6(sock);
1810                 break;
1811         }
1812 }
1813 #else
1814 static inline void xs_reclassify_socketu(struct socket *sock)
1815 {
1816 }
1817
1818 static inline void xs_reclassify_socket4(struct socket *sock)
1819 {
1820 }
1821
1822 static inline void xs_reclassify_socket6(struct socket *sock)
1823 {
1824 }
1825
1826 static inline void xs_reclassify_socket(int family, struct socket *sock)
1827 {
1828 }
1829 #endif
1830
1831 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1832                 struct sock_xprt *transport, int family, int type, int protocol)
1833 {
1834         struct socket *sock;
1835         int err;
1836
1837         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1838         if (err < 0) {
1839                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1840                                 protocol, -err);
1841                 goto out;
1842         }
1843         xs_reclassify_socket(family, sock);
1844
1845         err = xs_bind(transport, sock);
1846         if (err) {
1847                 sock_release(sock);
1848                 goto out;
1849         }
1850
1851         return sock;
1852 out:
1853         return ERR_PTR(err);
1854 }
1855
1856 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1857                                       struct socket *sock)
1858 {
1859         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1860                                                                         xprt);
1861
1862         if (!transport->inet) {
1863                 struct sock *sk = sock->sk;
1864
1865                 write_lock_bh(&sk->sk_callback_lock);
1866
1867                 xs_save_old_callbacks(transport, sk);
1868
1869                 sk->sk_user_data = xprt;
1870                 sk->sk_data_ready = xs_local_data_ready;
1871                 sk->sk_write_space = xs_udp_write_space;
1872                 sk->sk_allocation = GFP_ATOMIC;
1873
1874                 xprt_clear_connected(xprt);
1875
1876                 /* Reset to new socket */
1877                 transport->sock = sock;
1878                 transport->inet = sk;
1879
1880                 write_unlock_bh(&sk->sk_callback_lock);
1881         }
1882
1883         /* Tell the socket layer to start connecting... */
1884         xprt->stat.connect_count++;
1885         xprt->stat.connect_start = jiffies;
1886         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1887 }
1888
1889 /**
1890  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1891  * @xprt: RPC transport to connect
1892  * @transport: socket transport to connect
1893  * @create_sock: function to create a socket of the correct type
1894  */
1895 static int xs_local_setup_socket(struct sock_xprt *transport)
1896 {
1897         struct rpc_xprt *xprt = &transport->xprt;
1898         struct socket *sock;
1899         int status = -EIO;
1900
1901         current->flags |= PF_FSTRANS;
1902
1903         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1904         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1905                                         SOCK_STREAM, 0, &sock, 1);
1906         if (status < 0) {
1907                 dprintk("RPC:       can't create AF_LOCAL "
1908                         "transport socket (%d).\n", -status);
1909                 goto out;
1910         }
1911         xs_reclassify_socketu(sock);
1912
1913         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1914                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915
1916         status = xs_local_finish_connecting(xprt, sock);
1917         switch (status) {
1918         case 0:
1919                 dprintk("RPC:       xprt %p connected to %s\n",
1920                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1921                 xprt_set_connected(xprt);
1922                 break;
1923         case -ENOENT:
1924                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1925                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1926                 break;
1927         case -ECONNREFUSED:
1928                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1929                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1930                 break;
1931         default:
1932                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1933                                 __func__, -status,
1934                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1935         }
1936
1937 out:
1938         xprt_clear_connecting(xprt);
1939         xprt_wake_pending_tasks(xprt, status);
1940         current->flags &= ~PF_FSTRANS;
1941         return status;
1942 }
1943
1944 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1945 {
1946         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1947         int ret;
1948
1949          if (RPC_IS_ASYNC(task)) {
1950                 /*
1951                  * We want the AF_LOCAL connect to be resolved in the
1952                  * filesystem namespace of the process making the rpc
1953                  * call.  Thus we connect synchronously.
1954                  *
1955                  * If we want to support asynchronous AF_LOCAL calls,
1956                  * we'll need to figure out how to pass a namespace to
1957                  * connect.
1958                  */
1959                 rpc_exit(task, -ENOTCONN);
1960                 return;
1961         }
1962         ret = xs_local_setup_socket(transport);
1963         if (ret && !RPC_IS_SOFTCONN(task))
1964                 msleep_interruptible(15000);
1965 }
1966
1967 #ifdef CONFIG_SUNRPC_SWAP
1968 static void xs_set_memalloc(struct rpc_xprt *xprt)
1969 {
1970         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1971                         xprt);
1972
1973         if (xprt->swapper)
1974                 sk_set_memalloc(transport->inet);
1975 }
1976
1977 /**
1978  * xs_swapper - Tag this transport as being used for swap.
1979  * @xprt: transport to tag
1980  * @enable: enable/disable
1981  *
1982  */
1983 int xs_swapper(struct rpc_xprt *xprt, int enable)
1984 {
1985         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1986                         xprt);
1987         int err = 0;
1988
1989         if (enable) {
1990                 xprt->swapper++;
1991                 xs_set_memalloc(xprt);
1992         } else if (xprt->swapper) {
1993                 xprt->swapper--;
1994                 sk_clear_memalloc(transport->inet);
1995         }
1996
1997         return err;
1998 }
1999 EXPORT_SYMBOL_GPL(xs_swapper);
2000 #else
2001 static void xs_set_memalloc(struct rpc_xprt *xprt)
2002 {
2003 }
2004 #endif
2005
2006 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2007 {
2008         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2009
2010         if (!transport->inet) {
2011                 struct sock *sk = sock->sk;
2012
2013                 write_lock_bh(&sk->sk_callback_lock);
2014
2015                 xs_save_old_callbacks(transport, sk);
2016
2017                 sk->sk_user_data = xprt;
2018                 sk->sk_data_ready = xs_udp_data_ready;
2019                 sk->sk_write_space = xs_udp_write_space;
2020                 sk->sk_no_check = UDP_CSUM_NORCV;
2021                 sk->sk_allocation = GFP_ATOMIC;
2022
2023                 xprt_set_connected(xprt);
2024
2025                 /* Reset to new socket */
2026                 transport->sock = sock;
2027                 transport->inet = sk;
2028
2029                 xs_set_memalloc(xprt);
2030
2031                 write_unlock_bh(&sk->sk_callback_lock);
2032         }
2033         xs_udp_do_set_buffer_size(xprt);
2034 }
2035
2036 static void xs_udp_setup_socket(struct work_struct *work)
2037 {
2038         struct sock_xprt *transport =
2039                 container_of(work, struct sock_xprt, connect_worker.work);
2040         struct rpc_xprt *xprt = &transport->xprt;
2041         struct socket *sock = transport->sock;
2042         int status = -EIO;
2043
2044         current->flags |= PF_FSTRANS;
2045
2046         /* Start by resetting any existing state */
2047         xs_reset_transport(transport);
2048         sock = xs_create_sock(xprt, transport,
2049                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2050         if (IS_ERR(sock))
2051                 goto out;
2052
2053         dprintk("RPC:       worker connecting xprt %p via %s to "
2054                                 "%s (port %s)\n", xprt,
2055                         xprt->address_strings[RPC_DISPLAY_PROTO],
2056                         xprt->address_strings[RPC_DISPLAY_ADDR],
2057                         xprt->address_strings[RPC_DISPLAY_PORT]);
2058
2059         xs_udp_finish_connecting(xprt, sock);
2060         status = 0;
2061 out:
2062         xprt_clear_connecting(xprt);
2063         xprt_wake_pending_tasks(xprt, status);
2064         current->flags &= ~PF_FSTRANS;
2065 }
2066
2067 /*
2068  * We need to preserve the port number so the reply cache on the server can
2069  * find our cached RPC replies when we get around to reconnecting.
2070  */
2071 static void xs_abort_connection(struct sock_xprt *transport)
2072 {
2073         int result;
2074         struct sockaddr any;
2075
2076         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2077
2078         /*
2079          * Disconnect the transport socket by doing a connect operation
2080          * with AF_UNSPEC.  This should return immediately...
2081          */
2082         memset(&any, 0, sizeof(any));
2083         any.sa_family = AF_UNSPEC;
2084         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2085         if (!result)
2086                 xs_sock_reset_connection_flags(&transport->xprt);
2087         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2088 }
2089
2090 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2091 {
2092         unsigned int state = transport->inet->sk_state;
2093
2094         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2095                 /* we don't need to abort the connection if the socket
2096                  * hasn't undergone a shutdown
2097                  */
2098                 if (transport->inet->sk_shutdown == 0)
2099                         return;
2100                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2101                                 __func__, transport->inet->sk_shutdown);
2102         }
2103         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2104                 /* we don't need to abort the connection if the socket
2105                  * hasn't undergone a shutdown
2106                  */
2107                 if (transport->inet->sk_shutdown == 0)
2108                         return;
2109                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2110                                 "sk_shutdown set to %d\n",
2111                                 __func__, transport->inet->sk_shutdown);
2112         }
2113         xs_abort_connection(transport);
2114 }
2115
2116 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2117 {
2118         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2119         int ret = -ENOTCONN;
2120
2121         if (!transport->inet) {
2122                 struct sock *sk = sock->sk;
2123
2124                 write_lock_bh(&sk->sk_callback_lock);
2125
2126                 xs_save_old_callbacks(transport, sk);
2127
2128                 sk->sk_user_data = xprt;
2129                 sk->sk_data_ready = xs_tcp_data_ready;
2130                 sk->sk_state_change = xs_tcp_state_change;
2131                 sk->sk_write_space = xs_tcp_write_space;
2132                 sk->sk_allocation = GFP_ATOMIC;
2133
2134                 /* socket options */
2135                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2136                 sock_reset_flag(sk, SOCK_LINGER);
2137                 tcp_sk(sk)->linger2 = 0;
2138                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2139
2140                 xprt_clear_connected(xprt);
2141
2142                 /* Reset to new socket */
2143                 transport->sock = sock;
2144                 transport->inet = sk;
2145
2146                 write_unlock_bh(&sk->sk_callback_lock);
2147         }
2148
2149         if (!xprt_bound(xprt))
2150                 goto out;
2151
2152         xs_set_memalloc(xprt);
2153
2154         /* Tell the socket layer to start connecting... */
2155         xprt->stat.connect_count++;
2156         xprt->stat.connect_start = jiffies;
2157         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2158         switch (ret) {
2159         case 0:
2160         case -EINPROGRESS:
2161                 /* SYN_SENT! */
2162                 xprt->connect_cookie++;
2163                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2164                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2165         }
2166 out:
2167         return ret;
2168 }
2169
2170 /**
2171  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2172  * @xprt: RPC transport to connect
2173  * @transport: socket transport to connect
2174  * @create_sock: function to create a socket of the correct type
2175  *
2176  * Invoked by a work queue tasklet.
2177  */
2178 static void xs_tcp_setup_socket(struct work_struct *work)
2179 {
2180         struct sock_xprt *transport =
2181                 container_of(work, struct sock_xprt, connect_worker.work);
2182         struct socket *sock = transport->sock;
2183         struct rpc_xprt *xprt = &transport->xprt;
2184         int status = -EIO;
2185
2186         current->flags |= PF_FSTRANS;
2187
2188         if (!sock) {
2189                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2190                 sock = xs_create_sock(xprt, transport,
2191                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2192                 if (IS_ERR(sock)) {
2193                         status = PTR_ERR(sock);
2194                         goto out;
2195                 }
2196         } else {
2197                 int abort_and_exit;
2198
2199                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2200                                 &xprt->state);
2201                 /* "close" the socket, preserving the local port */
2202                 xs_tcp_reuse_connection(transport);
2203
2204                 if (abort_and_exit)
2205                         goto out_eagain;
2206         }
2207
2208         dprintk("RPC:       worker connecting xprt %p via %s to "
2209                                 "%s (port %s)\n", xprt,
2210                         xprt->address_strings[RPC_DISPLAY_PROTO],
2211                         xprt->address_strings[RPC_DISPLAY_ADDR],
2212                         xprt->address_strings[RPC_DISPLAY_PORT]);
2213
2214         status = xs_tcp_finish_connecting(xprt, sock);
2215         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2216                         xprt, -status, xprt_connected(xprt),
2217                         sock->sk->sk_state);
2218         switch (status) {
2219         default:
2220                 printk("%s: connect returned unhandled error %d\n",
2221                         __func__, status);
2222         case -EADDRNOTAVAIL:
2223                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2224                  * and retry
2225                  */
2226                 xs_tcp_force_close(xprt);
2227                 break;
2228         case 0:
2229         case -EINPROGRESS:
2230         case -EALREADY:
2231                 xprt_clear_connecting(xprt);
2232                 current->flags &= ~PF_FSTRANS;
2233                 return;
2234         case -EINVAL:
2235                 /* Happens, for instance, if the user specified a link
2236                  * local IPv6 address without a scope-id.
2237                  */
2238         case -ECONNREFUSED:
2239         case -ECONNRESET:
2240         case -ENETUNREACH:
2241                 /* retry with existing socket, after a delay */
2242                 goto out;
2243         }
2244 out_eagain:
2245         status = -EAGAIN;
2246 out:
2247         xprt_clear_connecting(xprt);
2248         xprt_wake_pending_tasks(xprt, status);
2249         current->flags &= ~PF_FSTRANS;
2250 }
2251
2252 /**
2253  * xs_connect - connect a socket to a remote endpoint
2254  * @xprt: pointer to transport structure
2255  * @task: address of RPC task that manages state of connect request
2256  *
2257  * TCP: If the remote end dropped the connection, delay reconnecting.
2258  *
2259  * UDP socket connects are synchronous, but we use a work queue anyway
2260  * to guarantee that even unprivileged user processes can set up a
2261  * socket on a privileged port.
2262  *
2263  * If a UDP socket connect fails, the delay behavior here prevents
2264  * retry floods (hard mounts).
2265  */
2266 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2267 {
2268         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2269
2270         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2271                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2272                                 "seconds\n",
2273                                 xprt, xprt->reestablish_timeout / HZ);
2274                 queue_delayed_work(rpciod_workqueue,
2275                                    &transport->connect_worker,
2276                                    xprt->reestablish_timeout);
2277                 xprt->reestablish_timeout <<= 1;
2278                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2279                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2280                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2281                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2282         } else {
2283                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2284                 queue_delayed_work(rpciod_workqueue,
2285                                    &transport->connect_worker, 0);
2286         }
2287 }
2288
2289 /**
2290  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2291  * @xprt: rpc_xprt struct containing statistics
2292  * @seq: output file
2293  *
2294  */
2295 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2296 {
2297         long idle_time = 0;
2298
2299         if (xprt_connected(xprt))
2300                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2301
2302         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2303                         "%llu %llu %lu %llu %llu\n",
2304                         xprt->stat.bind_count,
2305                         xprt->stat.connect_count,
2306                         xprt->stat.connect_time,
2307                         idle_time,
2308                         xprt->stat.sends,
2309                         xprt->stat.recvs,
2310                         xprt->stat.bad_xids,
2311                         xprt->stat.req_u,
2312                         xprt->stat.bklog_u,
2313                         xprt->stat.max_slots,
2314                         xprt->stat.sending_u,
2315                         xprt->stat.pending_u);
2316 }
2317
2318 /**
2319  * xs_udp_print_stats - display UDP socket-specifc stats
2320  * @xprt: rpc_xprt struct containing statistics
2321  * @seq: output file
2322  *
2323  */
2324 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2325 {
2326         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2327
2328         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2329                         "%lu %llu %llu\n",
2330                         transport->srcport,
2331                         xprt->stat.bind_count,
2332                         xprt->stat.sends,
2333                         xprt->stat.recvs,
2334                         xprt->stat.bad_xids,
2335                         xprt->stat.req_u,
2336                         xprt->stat.bklog_u,
2337                         xprt->stat.max_slots,
2338                         xprt->stat.sending_u,
2339                         xprt->stat.pending_u);
2340 }
2341
2342 /**
2343  * xs_tcp_print_stats - display TCP socket-specifc stats
2344  * @xprt: rpc_xprt struct containing statistics
2345  * @seq: output file
2346  *
2347  */
2348 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2349 {
2350         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2351         long idle_time = 0;
2352
2353         if (xprt_connected(xprt))
2354                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2355
2356         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2357                         "%llu %llu %lu %llu %llu\n",
2358                         transport->srcport,
2359                         xprt->stat.bind_count,
2360                         xprt->stat.connect_count,
2361                         xprt->stat.connect_time,
2362                         idle_time,
2363                         xprt->stat.sends,
2364                         xprt->stat.recvs,
2365                         xprt->stat.bad_xids,
2366                         xprt->stat.req_u,
2367                         xprt->stat.bklog_u,
2368                         xprt->stat.max_slots,
2369                         xprt->stat.sending_u,
2370                         xprt->stat.pending_u);
2371 }
2372
2373 /*
2374  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2375  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2376  * to use the server side send routines.
2377  */
2378 static void *bc_malloc(struct rpc_task *task, size_t size)
2379 {
2380         struct page *page;
2381         struct rpc_buffer *buf;
2382
2383         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2384         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2385                 return NULL;
2386
2387         page = alloc_page(GFP_KERNEL);
2388         if (!page)
2389                 return NULL;
2390
2391         buf = page_address(page);
2392         buf->len = PAGE_SIZE;
2393
2394         return buf->data;
2395 }
2396
2397 /*
2398  * Free the space allocated in the bc_alloc routine
2399  */
2400 static void bc_free(void *buffer)
2401 {
2402         struct rpc_buffer *buf;
2403
2404         if (!buffer)
2405                 return;
2406
2407         buf = container_of(buffer, struct rpc_buffer, data);
2408         free_page((unsigned long)buf);
2409 }
2410
2411 /*
2412  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2413  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2414  */
2415 static int bc_sendto(struct rpc_rqst *req)
2416 {
2417         int len;
2418         struct xdr_buf *xbufp = &req->rq_snd_buf;
2419         struct rpc_xprt *xprt = req->rq_xprt;
2420         struct sock_xprt *transport =
2421                                 container_of(xprt, struct sock_xprt, xprt);
2422         struct socket *sock = transport->sock;
2423         unsigned long headoff;
2424         unsigned long tailoff;
2425
2426         xs_encode_stream_record_marker(xbufp);
2427
2428         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2429         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2430         len = svc_send_common(sock, xbufp,
2431                               virt_to_page(xbufp->head[0].iov_base), headoff,
2432                               xbufp->tail[0].iov_base, tailoff);
2433
2434         if (len != xbufp->len) {
2435                 printk(KERN_NOTICE "Error sending entire callback!\n");
2436                 len = -EAGAIN;
2437         }
2438
2439         return len;
2440 }
2441
2442 /*
2443  * The send routine. Borrows from svc_send
2444  */
2445 static int bc_send_request(struct rpc_task *task)
2446 {
2447         struct rpc_rqst *req = task->tk_rqstp;
2448         struct svc_xprt *xprt;
2449         u32                     len;
2450
2451         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2452         /*
2453          * Get the server socket associated with this callback xprt
2454          */
2455         xprt = req->rq_xprt->bc_xprt;
2456
2457         /*
2458          * Grab the mutex to serialize data as the connection is shared
2459          * with the fore channel
2460          */
2461         if (!mutex_trylock(&xprt->xpt_mutex)) {
2462                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2463                 if (!mutex_trylock(&xprt->xpt_mutex))
2464                         return -EAGAIN;
2465                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2466         }
2467         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2468                 len = -ENOTCONN;
2469         else
2470                 len = bc_sendto(req);
2471         mutex_unlock(&xprt->xpt_mutex);
2472
2473         if (len > 0)
2474                 len = 0;
2475
2476         return len;
2477 }
2478
2479 /*
2480  * The close routine. Since this is client initiated, we do nothing
2481  */
2482
2483 static void bc_close(struct rpc_xprt *xprt)
2484 {
2485 }
2486
2487 /*
2488  * The xprt destroy routine. Again, because this connection is client
2489  * initiated, we do nothing
2490  */
2491
2492 static void bc_destroy(struct rpc_xprt *xprt)
2493 {
2494 }
2495
2496 static struct rpc_xprt_ops xs_local_ops = {
2497         .reserve_xprt           = xprt_reserve_xprt,
2498         .release_xprt           = xs_tcp_release_xprt,
2499         .alloc_slot             = xprt_alloc_slot,
2500         .rpcbind                = xs_local_rpcbind,
2501         .set_port               = xs_local_set_port,
2502         .connect                = xs_local_connect,
2503         .buf_alloc              = rpc_malloc,
2504         .buf_free               = rpc_free,
2505         .send_request           = xs_local_send_request,
2506         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2507         .close                  = xs_close,
2508         .destroy                = xs_local_destroy,
2509         .print_stats            = xs_local_print_stats,
2510 };
2511
2512 static struct rpc_xprt_ops xs_udp_ops = {
2513         .set_buffer_size        = xs_udp_set_buffer_size,
2514         .reserve_xprt           = xprt_reserve_xprt_cong,
2515         .release_xprt           = xprt_release_xprt_cong,
2516         .alloc_slot             = xprt_alloc_slot,
2517         .rpcbind                = rpcb_getport_async,
2518         .set_port               = xs_set_port,
2519         .connect                = xs_connect,
2520         .buf_alloc              = rpc_malloc,
2521         .buf_free               = rpc_free,
2522         .send_request           = xs_udp_send_request,
2523         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2524         .timer                  = xs_udp_timer,
2525         .release_request        = xprt_release_rqst_cong,
2526         .close                  = xs_close,
2527         .destroy                = xs_destroy,
2528         .print_stats            = xs_udp_print_stats,
2529 };
2530
2531 static struct rpc_xprt_ops xs_tcp_ops = {
2532         .reserve_xprt           = xprt_reserve_xprt,
2533         .release_xprt           = xs_tcp_release_xprt,
2534         .alloc_slot             = xprt_lock_and_alloc_slot,
2535         .rpcbind                = rpcb_getport_async,
2536         .set_port               = xs_set_port,
2537         .connect                = xs_connect,
2538         .buf_alloc              = rpc_malloc,
2539         .buf_free               = rpc_free,
2540         .send_request           = xs_tcp_send_request,
2541         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2542         .close                  = xs_tcp_close,
2543         .destroy                = xs_destroy,
2544         .print_stats            = xs_tcp_print_stats,
2545 };
2546
2547 /*
2548  * The rpc_xprt_ops for the server backchannel
2549  */
2550
2551 static struct rpc_xprt_ops bc_tcp_ops = {
2552         .reserve_xprt           = xprt_reserve_xprt,
2553         .release_xprt           = xprt_release_xprt,
2554         .alloc_slot             = xprt_alloc_slot,
2555         .rpcbind                = xs_local_rpcbind,
2556         .buf_alloc              = bc_malloc,
2557         .buf_free               = bc_free,
2558         .send_request           = bc_send_request,
2559         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2560         .close                  = bc_close,
2561         .destroy                = bc_destroy,
2562         .print_stats            = xs_tcp_print_stats,
2563 };
2564
2565 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2566 {
2567         static const struct sockaddr_in sin = {
2568                 .sin_family             = AF_INET,
2569                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2570         };
2571         static const struct sockaddr_in6 sin6 = {
2572                 .sin6_family            = AF_INET6,
2573                 .sin6_addr              = IN6ADDR_ANY_INIT,
2574         };
2575
2576         switch (family) {
2577         case AF_LOCAL:
2578                 break;
2579         case AF_INET:
2580                 memcpy(sap, &sin, sizeof(sin));
2581                 break;
2582         case AF_INET6:
2583                 memcpy(sap, &sin6, sizeof(sin6));
2584                 break;
2585         default:
2586                 dprintk("RPC:       %s: Bad address family\n", __func__);
2587                 return -EAFNOSUPPORT;
2588         }
2589         return 0;
2590 }
2591
2592 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2593                                       unsigned int slot_table_size,
2594                                       unsigned int max_slot_table_size)
2595 {
2596         struct rpc_xprt *xprt;
2597         struct sock_xprt *new;
2598
2599         if (args->addrlen > sizeof(xprt->addr)) {
2600                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2601                 return ERR_PTR(-EBADF);
2602         }
2603
2604         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2605                         max_slot_table_size);
2606         if (xprt == NULL) {
2607                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2608                                 "rpc_xprt\n");
2609                 return ERR_PTR(-ENOMEM);
2610         }
2611
2612         new = container_of(xprt, struct sock_xprt, xprt);
2613         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2614         xprt->addrlen = args->addrlen;
2615         if (args->srcaddr)
2616                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2617         else {
2618                 int err;
2619                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2620                                         (struct sockaddr *)&new->srcaddr);
2621                 if (err != 0) {
2622                         xprt_free(xprt);
2623                         return ERR_PTR(err);
2624                 }
2625         }
2626
2627         return xprt;
2628 }
2629
2630 static const struct rpc_timeout xs_local_default_timeout = {
2631         .to_initval = 10 * HZ,
2632         .to_maxval = 10 * HZ,
2633         .to_retries = 2,
2634 };
2635
2636 /**
2637  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2638  * @args: rpc transport creation arguments
2639  *
2640  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2641  */
2642 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2643 {
2644         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2645         struct sock_xprt *transport;
2646         struct rpc_xprt *xprt;
2647         struct rpc_xprt *ret;
2648
2649         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2650                         xprt_max_tcp_slot_table_entries);
2651         if (IS_ERR(xprt))
2652                 return xprt;
2653         transport = container_of(xprt, struct sock_xprt, xprt);
2654
2655         xprt->prot = 0;
2656         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2657         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2658
2659         xprt->bind_timeout = XS_BIND_TO;
2660         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2661         xprt->idle_timeout = XS_IDLE_DISC_TO;
2662
2663         xprt->ops = &xs_local_ops;
2664         xprt->timeout = &xs_local_default_timeout;
2665
2666         switch (sun->sun_family) {
2667         case AF_LOCAL:
2668                 if (sun->sun_path[0] != '/') {
2669                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2670                                         sun->sun_path);
2671                         ret = ERR_PTR(-EINVAL);
2672                         goto out_err;
2673                 }
2674                 xprt_set_bound(xprt);
2675                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2676                 ret = ERR_PTR(xs_local_setup_socket(transport));
2677                 if (ret)
2678                         goto out_err;
2679                 break;
2680         default:
2681                 ret = ERR_PTR(-EAFNOSUPPORT);
2682                 goto out_err;
2683         }
2684
2685         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2686                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2687
2688         if (try_module_get(THIS_MODULE))
2689                 return xprt;
2690         ret = ERR_PTR(-EINVAL);
2691 out_err:
2692         xprt_free(xprt);
2693         return ret;
2694 }
2695
2696 static const struct rpc_timeout xs_udp_default_timeout = {
2697         .to_initval = 5 * HZ,
2698         .to_maxval = 30 * HZ,
2699         .to_increment = 5 * HZ,
2700         .to_retries = 5,
2701 };
2702
2703 /**
2704  * xs_setup_udp - Set up transport to use a UDP socket
2705  * @args: rpc transport creation arguments
2706  *
2707  */
2708 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2709 {
2710         struct sockaddr *addr = args->dstaddr;
2711         struct rpc_xprt *xprt;
2712         struct sock_xprt *transport;
2713         struct rpc_xprt *ret;
2714
2715         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2716                         xprt_udp_slot_table_entries);
2717         if (IS_ERR(xprt))
2718                 return xprt;
2719         transport = container_of(xprt, struct sock_xprt, xprt);
2720
2721         xprt->prot = IPPROTO_UDP;
2722         xprt->tsh_size = 0;
2723         /* XXX: header size can vary due to auth type, IPv6, etc. */
2724         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2725
2726         xprt->bind_timeout = XS_BIND_TO;
2727         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2728         xprt->idle_timeout = XS_IDLE_DISC_TO;
2729
2730         xprt->ops = &xs_udp_ops;
2731
2732         xprt->timeout = &xs_udp_default_timeout;
2733
2734         switch (addr->sa_family) {
2735         case AF_INET:
2736                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2737                         xprt_set_bound(xprt);
2738
2739                 INIT_DELAYED_WORK(&transport->connect_worker,
2740                                         xs_udp_setup_socket);
2741                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2742                 break;
2743         case AF_INET6:
2744                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2745                         xprt_set_bound(xprt);
2746
2747                 INIT_DELAYED_WORK(&transport->connect_worker,
2748                                         xs_udp_setup_socket);
2749                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2750                 break;
2751         default:
2752                 ret = ERR_PTR(-EAFNOSUPPORT);
2753                 goto out_err;
2754         }
2755
2756         if (xprt_bound(xprt))
2757                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2758                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2759                                 xprt->address_strings[RPC_DISPLAY_PORT],
2760                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2761         else
2762                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2763                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2764                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2765
2766         if (try_module_get(THIS_MODULE))
2767                 return xprt;
2768         ret = ERR_PTR(-EINVAL);
2769 out_err:
2770         xprt_free(xprt);
2771         return ret;
2772 }
2773
2774 static const struct rpc_timeout xs_tcp_default_timeout = {
2775         .to_initval = 60 * HZ,
2776         .to_maxval = 60 * HZ,
2777         .to_retries = 2,
2778 };
2779
2780 /**
2781  * xs_setup_tcp - Set up transport to use a TCP socket
2782  * @args: rpc transport creation arguments
2783  *
2784  */
2785 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2786 {
2787         struct sockaddr *addr = args->dstaddr;
2788         struct rpc_xprt *xprt;
2789         struct sock_xprt *transport;
2790         struct rpc_xprt *ret;
2791         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2792
2793         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2794                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2795
2796         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2797                         max_slot_table_size);
2798         if (IS_ERR(xprt))
2799                 return xprt;
2800         transport = container_of(xprt, struct sock_xprt, xprt);
2801
2802         xprt->prot = IPPROTO_TCP;
2803         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2804         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2805
2806         xprt->bind_timeout = XS_BIND_TO;
2807         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2808         xprt->idle_timeout = XS_IDLE_DISC_TO;
2809
2810         xprt->ops = &xs_tcp_ops;
2811         xprt->timeout = &xs_tcp_default_timeout;
2812
2813         switch (addr->sa_family) {
2814         case AF_INET:
2815                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2816                         xprt_set_bound(xprt);
2817
2818                 INIT_DELAYED_WORK(&transport->connect_worker,
2819                                         xs_tcp_setup_socket);
2820                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2821                 break;
2822         case AF_INET6:
2823                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2824                         xprt_set_bound(xprt);
2825
2826                 INIT_DELAYED_WORK(&transport->connect_worker,
2827                                         xs_tcp_setup_socket);
2828                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2829                 break;
2830         default:
2831                 ret = ERR_PTR(-EAFNOSUPPORT);
2832                 goto out_err;
2833         }
2834
2835         if (xprt_bound(xprt))
2836                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2837                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2838                                 xprt->address_strings[RPC_DISPLAY_PORT],
2839                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2840         else
2841                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2842                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2843                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2844
2845
2846         if (try_module_get(THIS_MODULE))
2847                 return xprt;
2848         ret = ERR_PTR(-EINVAL);
2849 out_err:
2850         xprt_free(xprt);
2851         return ret;
2852 }
2853
2854 /**
2855  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2856  * @args: rpc transport creation arguments
2857  *
2858  */
2859 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2860 {
2861         struct sockaddr *addr = args->dstaddr;
2862         struct rpc_xprt *xprt;
2863         struct sock_xprt *transport;
2864         struct svc_sock *bc_sock;
2865         struct rpc_xprt *ret;
2866
2867         if (args->bc_xprt->xpt_bc_xprt) {
2868                 /*
2869                  * This server connection already has a backchannel
2870                  * export; we can't create a new one, as we wouldn't be
2871                  * able to match replies based on xid any more.  So,
2872                  * reuse the already-existing one:
2873                  */
2874                  return args->bc_xprt->xpt_bc_xprt;
2875         }
2876         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2877                         xprt_tcp_slot_table_entries);
2878         if (IS_ERR(xprt))
2879                 return xprt;
2880         transport = container_of(xprt, struct sock_xprt, xprt);
2881
2882         xprt->prot = IPPROTO_TCP;
2883         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2884         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2885         xprt->timeout = &xs_tcp_default_timeout;
2886
2887         /* backchannel */
2888         xprt_set_bound(xprt);
2889         xprt->bind_timeout = 0;
2890         xprt->reestablish_timeout = 0;
2891         xprt->idle_timeout = 0;
2892
2893         xprt->ops = &bc_tcp_ops;
2894
2895         switch (addr->sa_family) {
2896         case AF_INET:
2897                 xs_format_peer_addresses(xprt, "tcp",
2898                                          RPCBIND_NETID_TCP);
2899                 break;
2900         case AF_INET6:
2901                 xs_format_peer_addresses(xprt, "tcp",
2902                                    RPCBIND_NETID_TCP6);
2903                 break;
2904         default:
2905                 ret = ERR_PTR(-EAFNOSUPPORT);
2906                 goto out_err;
2907         }
2908
2909         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2910                         xprt->address_strings[RPC_DISPLAY_ADDR],
2911                         xprt->address_strings[RPC_DISPLAY_PORT],
2912                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2913
2914         /*
2915          * Once we've associated a backchannel xprt with a connection,
2916          * we want to keep it around as long as long as the connection
2917          * lasts, in case we need to start using it for a backchannel
2918          * again; this reference won't be dropped until bc_xprt is
2919          * destroyed.
2920          */
2921         xprt_get(xprt);
2922         args->bc_xprt->xpt_bc_xprt = xprt;
2923         xprt->bc_xprt = args->bc_xprt;
2924         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2925         transport->sock = bc_sock->sk_sock;
2926         transport->inet = bc_sock->sk_sk;
2927
2928         /*
2929          * Since we don't want connections for the backchannel, we set
2930          * the xprt status to connected
2931          */
2932         xprt_set_connected(xprt);
2933
2934
2935         if (try_module_get(THIS_MODULE))
2936                 return xprt;
2937         xprt_put(xprt);
2938         ret = ERR_PTR(-EINVAL);
2939 out_err:
2940         xprt_free(xprt);
2941         return ret;
2942 }
2943
2944 static struct xprt_class        xs_local_transport = {
2945         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2946         .name           = "named UNIX socket",
2947         .owner          = THIS_MODULE,
2948         .ident          = XPRT_TRANSPORT_LOCAL,
2949         .setup          = xs_setup_local,
2950 };
2951
2952 static struct xprt_class        xs_udp_transport = {
2953         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2954         .name           = "udp",
2955         .owner          = THIS_MODULE,
2956         .ident          = XPRT_TRANSPORT_UDP,
2957         .setup          = xs_setup_udp,
2958 };
2959
2960 static struct xprt_class        xs_tcp_transport = {
2961         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2962         .name           = "tcp",
2963         .owner          = THIS_MODULE,
2964         .ident          = XPRT_TRANSPORT_TCP,
2965         .setup          = xs_setup_tcp,
2966 };
2967
2968 static struct xprt_class        xs_bc_tcp_transport = {
2969         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2970         .name           = "tcp NFSv4.1 backchannel",
2971         .owner          = THIS_MODULE,
2972         .ident          = XPRT_TRANSPORT_BC_TCP,
2973         .setup          = xs_setup_bc_tcp,
2974 };
2975
2976 /**
2977  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2978  *
2979  */
2980 int init_socket_xprt(void)
2981 {
2982 #ifdef RPC_DEBUG
2983         if (!sunrpc_table_header)
2984                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2985 #endif
2986
2987         xprt_register_transport(&xs_local_transport);
2988         xprt_register_transport(&xs_udp_transport);
2989         xprt_register_transport(&xs_tcp_transport);
2990         xprt_register_transport(&xs_bc_tcp_transport);
2991
2992         return 0;
2993 }
2994
2995 /**
2996  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2997  *
2998  */
2999 void cleanup_socket_xprt(void)
3000 {
3001 #ifdef RPC_DEBUG
3002         if (sunrpc_table_header) {
3003                 unregister_sysctl_table(sunrpc_table_header);
3004                 sunrpc_table_header = NULL;
3005         }
3006 #endif
3007
3008         xprt_unregister_transport(&xs_local_transport);
3009         xprt_unregister_transport(&xs_udp_transport);
3010         xprt_unregister_transport(&xs_tcp_transport);
3011         xprt_unregister_transport(&xs_bc_tcp_transport);
3012 }
3013
3014 static int param_set_uint_minmax(const char *val,
3015                 const struct kernel_param *kp,
3016                 unsigned int min, unsigned int max)
3017 {
3018         unsigned long num;
3019         int ret;
3020
3021         if (!val)
3022                 return -EINVAL;
3023         ret = strict_strtoul(val, 0, &num);
3024         if (ret == -EINVAL || num < min || num > max)
3025                 return -EINVAL;
3026         *((unsigned int *)kp->arg) = num;
3027         return 0;
3028 }
3029
3030 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3031 {
3032         return param_set_uint_minmax(val, kp,
3033                         RPC_MIN_RESVPORT,
3034                         RPC_MAX_RESVPORT);
3035 }
3036
3037 static struct kernel_param_ops param_ops_portnr = {
3038         .set = param_set_portnr,
3039         .get = param_get_uint,
3040 };
3041
3042 #define param_check_portnr(name, p) \
3043         __param_check(name, p, unsigned int);
3044
3045 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3046 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3047
3048 static int param_set_slot_table_size(const char *val,
3049                                      const struct kernel_param *kp)
3050 {
3051         return param_set_uint_minmax(val, kp,
3052                         RPC_MIN_SLOT_TABLE,
3053                         RPC_MAX_SLOT_TABLE);
3054 }
3055
3056 static struct kernel_param_ops param_ops_slot_table_size = {
3057         .set = param_set_slot_table_size,
3058         .get = param_get_uint,
3059 };
3060
3061 #define param_check_slot_table_size(name, p) \
3062         __param_check(name, p, unsigned int);
3063
3064 static int param_set_max_slot_table_size(const char *val,
3065                                      const struct kernel_param *kp)
3066 {
3067         return param_set_uint_minmax(val, kp,
3068                         RPC_MIN_SLOT_TABLE,
3069                         RPC_MAX_SLOT_TABLE_LIMIT);
3070 }
3071
3072 static struct kernel_param_ops param_ops_max_slot_table_size = {
3073         .set = param_set_max_slot_table_size,
3074         .get = param_get_uint,
3075 };
3076
3077 #define param_check_max_slot_table_size(name, p) \
3078         __param_check(name, p, unsigned int);
3079
3080 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3081                    slot_table_size, 0644);
3082 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3083                    max_slot_table_size, 0644);
3084 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3085                    slot_table_size, 0644);
3086