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/sched.h>
  37 #include <linux/sunrpc/svcsock.h>
  38 #include <linux/sunrpc/xprtsock.h>
  39 #include <linux/file.h>
  40 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  41 #include <linux/sunrpc/bc_xprt.h>
  42 #endif
  43
  44 #include <net/sock.h>
  45 #include <net/checksum.h>
  46 #include <net/udp.h>
  47 #include <net/tcp.h>
  48
  49 #include "sunrpc.h"
  50
  51 static void xs_close(struct rpc_xprt *xprt);
  52
  53 /*
  54  * xprtsock tunables
  55  */
  56 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  57 unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  58 unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  59
  60 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  61 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  62
  63 #define XS_TCP_LINGER_TO        (15U * HZ)
  64 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  65
  66 /*
  67  * We can register our own files under /proc/sys/sunrpc by
  68  * calling register_sysctl_table() again.  The files in that
  69  * directory become the union of all files registered there.
  70  *
  71  * We simply need to make sure that we don't collide with
  72  * someone else's file names!
  73  */
  74
  75 #ifdef RPC_DEBUG
  76
  77 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  78 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  79 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  80 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  81 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  82
  83 static struct ctl_table_header *sunrpc_table_header;
  84
  85 /*
  86  * FIXME: changing the UDP slot table size should also resize the UDP
  87  *        socket buffers for existing UDP transports
  88  */
  89 static ctl_table xs_tunables_table[] = {
  90         {
  91                 .procname       = "udp_slot_table_entries",
  92                 .data           = &xprt_udp_slot_table_entries,
  93                 .maxlen         = sizeof(unsigned int),
  94                 .mode           = 0644,
  95                 .proc_handler   = proc_dointvec_minmax,
  96                 .extra1         = &min_slot_table_size,
  97                 .extra2         = &max_slot_table_size
  98         },
  99         {
 100                 .procname       = "tcp_slot_table_entries",
 101                 .data           = &xprt_tcp_slot_table_entries,
 102                 .maxlen         = sizeof(unsigned int),
 103                 .mode           = 0644,
 104                 .proc_handler   = proc_dointvec_minmax,
 105                 .extra1         = &min_slot_table_size,
 106                 .extra2         = &max_slot_table_size
 107         },
 108         {
 109                 .procname       = "tcp_max_slot_table_entries",
 110                 .data           = &xprt_max_tcp_slot_table_entries,
 111                 .maxlen         = sizeof(unsigned int),
 112                 .mode           = 0644,
 113                 .proc_handler   = proc_dointvec_minmax,
 114                 .extra1         = &min_slot_table_size,
 115                 .extra2         = &max_tcp_slot_table_limit
 116         },
 117         {
 118                 .procname       = "min_resvport",
 119                 .data           = &xprt_min_resvport,
 120                 .maxlen         = sizeof(unsigned int),
 121                 .mode           = 0644,
 122                 .proc_handler   = proc_dointvec_minmax,
 123                 .extra1         = &xprt_min_resvport_limit,
 124                 .extra2         = &xprt_max_resvport_limit
 125         },
 126         {
 127                 .procname       = "max_resvport",
 128                 .data           = &xprt_max_resvport,
 129                 .maxlen         = sizeof(unsigned int),
 130                 .mode           = 0644,
 131                 .proc_handler   = proc_dointvec_minmax,
 132                 .extra1         = &xprt_min_resvport_limit,
 133                 .extra2         = &xprt_max_resvport_limit
 134         },
 135         {
 136                 .procname       = "tcp_fin_timeout",
 137                 .data           = &xs_tcp_fin_timeout,
 138                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 139                 .mode           = 0644,
 140                 .proc_handler   = proc_dointvec_jiffies,
 141         },
 142         { },
 143 };
 144
 145 static ctl_table sunrpc_table[] = {
 146         {
 147                 .procname       = "sunrpc",
 148                 .mode           = 0555,
 149                 .child          = xs_tunables_table
 150         },
 151         { },
 152 };
 153
 154 #endif
 155
 156 /*
 157  * Wait duration for a reply from the RPC portmapper.
 158  */
 159 #define XS_BIND_TO              (60U * HZ)
 160
 161 /*
 162  * Delay if a UDP socket connect error occurs.  This is most likely some
 163  * kind of resource problem on the local host.
 164  */
 165 #define XS_UDP_REEST_TO         (2U * HZ)
 166
 167 /*
 168  * The reestablish timeout allows clients to delay for a bit before attempting
 169  * to reconnect to a server that just dropped our connection.
 170  *
 171  * We implement an exponential backoff when trying to reestablish a TCP
 172  * transport connection with the server.  Some servers like to drop a TCP
 173  * connection when they are overworked, so we start with a short timeout and
 174  * increase over time if the server is down or not responding.
 175  */
 176 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 177 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 178
 179 /*
 180  * TCP idle timeout; client drops the transport socket if it is idle
 181  * for this long.  Note that we also timeout UDP sockets to prevent
 182  * holding port numbers when there is no RPC traffic.
 183  */
 184 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 185
 186 #ifdef RPC_DEBUG
 187 # undef  RPC_DEBUG_DATA
 188 # define RPCDBG_FACILITY        RPCDBG_TRANS
 189 #endif
 190
 191 #ifdef RPC_DEBUG_DATA
 192 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 193 {
 194         u8 *buf = (u8 *) packet;
 195         int j;
 196
 197         dprintk("RPC:       %s\n", msg);
 198         for (j = 0; j < count && j < 128; j += 4) {
 199                 if (!(j & 31)) {
 200                         if (j)
 201                                 dprintk("\n");
 202                         dprintk("0x%04x ", j);
 203                 }
 204                 dprintk("%02x%02x%02x%02x ",
 205                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 206         }
 207         dprintk("\n");
 208 }
 209 #else
 210 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 211 {
 212         /* NOP */
 213 }
 214 #endif
 215
 216 struct sock_xprt {
 217         struct rpc_xprt         xprt;
 218
 219         /*
 220          * Network layer
 221          */
 222         struct socket *         sock;
 223         struct sock *           inet;
 224
 225         /*
 226          * State of TCP reply receive
 227          */
 228         __be32                  tcp_fraghdr,
 229                                 tcp_xid,
 230                                 tcp_calldir;
 231
 232         u32                     tcp_offset,
 233                                 tcp_reclen;
 234
 235         unsigned long           tcp_copied,
 236                                 tcp_flags;
 237
 238         /*
 239          * Connection of transports
 240          */
 241         struct delayed_work     connect_worker;
 242         struct sockaddr_storage srcaddr;
 243         unsigned short          srcport;
 244
 245         /*
 246          * UDP socket buffer size parameters
 247          */
 248         size_t                  rcvsize,
 249                                 sndsize;
 250
 251         /*
 252          * Saved socket callback addresses
 253          */
 254         void                    (*old_data_ready)(struct sock *, int);
 255         void                    (*old_state_change)(struct sock *);
 256         void                    (*old_write_space)(struct sock *);
 257 };
 258
 259 /*
 260  * TCP receive state flags
 261  */
 262 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 263 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 264 #define TCP_RCV_COPY_XID        (1UL << 2)
 265 #define TCP_RCV_COPY_DATA       (1UL << 3)
 266 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 267 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 268
 269 /*
 270  * TCP RPC flags
 271  */
 272 #define TCP_RPC_REPLY           (1UL << 6)
 273
 274 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 275 {
 276         return (struct sockaddr *) &xprt->addr;
 277 }
 278
 279 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 280 {
 281         return (struct sockaddr_un *) &xprt->addr;
 282 }
 283
 284 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 285 {
 286         return (struct sockaddr_in *) &xprt->addr;
 287 }
 288
 289 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 290 {
 291         return (struct sockaddr_in6 *) &xprt->addr;
 292 }
 293
 294 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 295 {
 296         struct sockaddr *sap = xs_addr(xprt);
 297         struct sockaddr_in6 *sin6;
 298         struct sockaddr_in *sin;
 299         struct sockaddr_un *sun;
 300         char buf[128];
 301
 302         switch (sap->sa_family) {
 303         case AF_LOCAL:
 304                 sun = xs_addr_un(xprt);
 305                 strlcpy(buf, sun->sun_path, sizeof(buf));
 306                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 307                                                 kstrdup(buf, GFP_KERNEL);
 308                 break;
 309         case AF_INET:
 310                 (void)rpc_ntop(sap, buf, sizeof(buf));
 311                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 312                                                 kstrdup(buf, GFP_KERNEL);
 313                 sin = xs_addr_in(xprt);
 314                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 315                 break;
 316         case AF_INET6:
 317                 (void)rpc_ntop(sap, buf, sizeof(buf));
 318                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 319                                                 kstrdup(buf, GFP_KERNEL);
 320                 sin6 = xs_addr_in6(xprt);
 321                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 322                 break;
 323         default:
 324                 BUG();
 325         }
 326
 327         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 328 }
 329
 330 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 331 {
 332         struct sockaddr *sap = xs_addr(xprt);
 333         char buf[128];
 334
 335         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 336         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 337
 338         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 339         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 340 }
 341
 342 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 343                                      const char *protocol,
 344                                      const char *netid)
 345 {
 346         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 347         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 348         xs_format_common_peer_addresses(xprt);
 349         xs_format_common_peer_ports(xprt);
 350 }
 351
 352 static void xs_update_peer_port(struct rpc_xprt *xprt)
 353 {
 354         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 355         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 356
 357         xs_format_common_peer_ports(xprt);
 358 }
 359
 360 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 361 {
 362         unsigned int i;
 363
 364         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 365                 switch (i) {
 366                 case RPC_DISPLAY_PROTO:
 367                 case RPC_DISPLAY_NETID:
 368                         continue;
 369                 default:
 370                         kfree(xprt->address_strings[i]);
 371                 }
 372 }
 373
 374 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 375
 376 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 377 {
 378         struct msghdr msg = {
 379                 .msg_name       = addr,
 380                 .msg_namelen    = addrlen,
 381                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 382         };
 383         struct kvec iov = {
 384                 .iov_base       = vec->iov_base + base,
 385                 .iov_len        = vec->iov_len - base,
 386         };
 387
 388         if (iov.iov_len != 0)
 389                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 390         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 391 }
 392
 393 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
 394 {
 395         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 396                         int offset, size_t size, int flags);
 397         struct page **ppage;
 398         unsigned int remainder;
 399         int err, sent = 0;
 400
 401         remainder = xdr->page_len - base;
 402         base += xdr->page_base;
 403         ppage = xdr->pages + (base >> PAGE_SHIFT);
 404         base &= ~PAGE_MASK;
 405         do_sendpage = sock->ops->sendpage;
 406         if (!zerocopy)
 407                 do_sendpage = sock_no_sendpage;
 408         for(;;) {
 409                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 410                 int flags = XS_SENDMSG_FLAGS;
 411
 412                 remainder -= len;
 413                 if (remainder != 0 || more)
 414                         flags |= MSG_MORE;
 415                 err = do_sendpage(sock, *ppage, base, len, flags);
 416                 if (remainder == 0 || err != len)
 417                         break;
 418                 sent += err;
 419                 ppage++;
 420                 base = 0;
 421         }
 422         if (sent == 0)
 423                 return err;
 424         if (err > 0)
 425                 sent += err;
 426         return sent;
 427 }
 428
 429 /**
 430  * xs_sendpages - write pages directly to a socket
 431  * @sock: socket to send on
 432  * @addr: UDP only -- address of destination
 433  * @addrlen: UDP only -- length of destination address
 434  * @xdr: buffer containing this request
 435  * @base: starting position in the buffer
 436  * @zerocopy: true if it is safe to use sendpage()
 437  *
 438  */
 439 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
 440 {
 441         unsigned int remainder = xdr->len - base;
 442         int err, sent = 0;
 443
 444         if (unlikely(!sock))
 445                 return -ENOTSOCK;
 446
 447         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 448         if (base != 0) {
 449                 addr = NULL;
 450                 addrlen = 0;
 451         }
 452
 453         if (base < xdr->head[0].iov_len || addr != NULL) {
 454                 unsigned int len = xdr->head[0].iov_len - base;
 455                 remainder -= len;
 456                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 457                 if (remainder == 0 || err != len)
 458                         goto out;
 459                 sent += err;
 460                 base = 0;
 461         } else
 462                 base -= xdr->head[0].iov_len;
 463
 464         if (base < xdr->page_len) {
 465                 unsigned int len = xdr->page_len - base;
 466                 remainder -= len;
 467                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
 468                 if (remainder == 0 || err != len)
 469                         goto out;
 470                 sent += err;
 471                 base = 0;
 472         } else
 473                 base -= xdr->page_len;
 474
 475         if (base >= xdr->tail[0].iov_len)
 476                 return sent;
 477         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 478 out:
 479         if (sent == 0)
 480                 return err;
 481         if (err > 0)
 482                 sent += err;
 483         return sent;
 484 }
 485
 486 static void xs_nospace_callback(struct rpc_task *task)
 487 {
 488         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 489
 490         transport->inet->sk_write_pending--;
 491         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 492 }
 493
 494 /**
 495  * xs_nospace - place task on wait queue if transmit was incomplete
 496  * @task: task to put to sleep
 497  *
 498  */
 499 static int xs_nospace(struct rpc_task *task)
 500 {
 501         struct rpc_rqst *req = task->tk_rqstp;
 502         struct rpc_xprt *xprt = req->rq_xprt;
 503         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 504         struct sock *sk = transport->inet;
 505         int ret = -EAGAIN;
 506
 507         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 508                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 509                         req->rq_slen);
 510
 511         /* Protect against races with write_space */
 512         spin_lock_bh(&xprt->transport_lock);
 513
 514         /* Don't race with disconnect */
 515         if (xprt_connected(xprt)) {
 516                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 517                         /*
 518                          * Notify TCP that we're limited by the application
 519                          * window size
 520                          */
 521                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 522                         sk->sk_write_pending++;
 523                         /* ...and wait for more buffer space */
 524                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 525                 }
 526         } else {
 527                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 528                 ret = -ENOTCONN;
 529         }
 530
 531         spin_unlock_bh(&xprt->transport_lock);
 532
 533         /* Race breaker in case memory is freed before above code is called */
 534         sk->sk_write_space(sk);
 535         return ret;
 536 }
 537
 538 /*
 539  * Construct a stream transport record marker in @buf.
 540  */
 541 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 542 {
 543         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 544         rpc_fraghdr *base = buf->head[0].iov_base;
 545         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 546 }
 547
 548 /**
 549  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 550  * @task: RPC task that manages the state of an RPC request
 551  *
 552  * Return values:
 553  *        0:    The request has been sent
 554  *   EAGAIN:    The socket was blocked, please call again later to
 555  *              complete the request
 556  * ENOTCONN:    Caller needs to invoke connect logic then call again
 557  *    other:    Some other error occured, the request was not sent
 558  */
 559 static int xs_local_send_request(struct rpc_task *task)
 560 {
 561         struct rpc_rqst *req = task->tk_rqstp;
 562         struct rpc_xprt *xprt = req->rq_xprt;
 563         struct sock_xprt *transport =
 564                                 container_of(xprt, struct sock_xprt, xprt);
 565         struct xdr_buf *xdr = &req->rq_snd_buf;
 566         int status;
 567
 568         xs_encode_stream_record_marker(&req->rq_snd_buf);
 569
 570         xs_pktdump("packet data:",
 571                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 572
 573         status = xs_sendpages(transport->sock, NULL, 0,
 574                                                 xdr, req->rq_bytes_sent, true);
 575         dprintk("RPC:       %s(%u) = %d\n",
 576                         __func__, xdr->len - req->rq_bytes_sent, status);
 577         if (likely(status >= 0)) {
 578                 req->rq_bytes_sent += status;
 579                 req->rq_xmit_bytes_sent += status;
 580                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 581                         req->rq_bytes_sent = 0;
 582                         return 0;
 583                 }
 584                 status = -EAGAIN;
 585         }
 586
 587         switch (status) {
 588         case -EAGAIN:
 589                 status = xs_nospace(task);
 590                 break;
 591         default:
 592                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 593                         -status);
 594         case -EPIPE:
 595                 xs_close(xprt);
 596                 status = -ENOTCONN;
 597         }
 598
 599         return status;
 600 }
 601
 602 /**
 603  * xs_udp_send_request - write an RPC request to a UDP socket
 604  * @task: address of RPC task that manages the state of an RPC request
 605  *
 606  * Return values:
 607  *        0:    The request has been sent
 608  *   EAGAIN:    The socket was blocked, please call again later to
 609  *              complete the request
 610  * ENOTCONN:    Caller needs to invoke connect logic then call again
 611  *    other:    Some other error occurred, the request was not sent
 612  */
 613 static int xs_udp_send_request(struct rpc_task *task)
 614 {
 615         struct rpc_rqst *req = task->tk_rqstp;
 616         struct rpc_xprt *xprt = req->rq_xprt;
 617         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 618         struct xdr_buf *xdr = &req->rq_snd_buf;
 619         int status;
 620
 621         xs_pktdump("packet data:",
 622                                 req->rq_svec->iov_base,
 623                                 req->rq_svec->iov_len);
 624
 625         if (!xprt_bound(xprt))
 626                 return -ENOTCONN;
 627         status = xs_sendpages(transport->sock,
 628                               xs_addr(xprt),
 629                               xprt->addrlen, xdr,
 630                               req->rq_bytes_sent, true);
 631
 632         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 633                         xdr->len - req->rq_bytes_sent, status);
 634
 635         if (status >= 0) {
 636                 req->rq_xmit_bytes_sent += status;
 637                 if (status >= req->rq_slen)
 638                         return 0;
 639                 /* Still some bytes left; set up for a retry later. */
 640                 status = -EAGAIN;
 641         }
 642
 643         switch (status) {
 644         case -ENOTSOCK:
 645                 status = -ENOTCONN;
 646                 /* Should we call xs_close() here? */
 647                 break;
 648         case -EAGAIN:
 649                 status = xs_nospace(task);
 650                 break;
 651         default:
 652                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 653                         -status);
 654         case -ENETUNREACH:
 655         case -EPIPE:
 656         case -ECONNREFUSED:
 657                 /* When the server has died, an ICMP port unreachable message
 658                  * prompts ECONNREFUSED. */
 659                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 660         }
 661
 662         return status;
 663 }
 664
 665 /**
 666  * xs_tcp_shutdown - gracefully shut down a TCP socket
 667  * @xprt: transport
 668  *
 669  * Initiates a graceful shutdown of the TCP socket by calling the
 670  * equivalent of shutdown(SHUT_WR);
 671  */
 672 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 673 {
 674         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 675         struct socket *sock = transport->sock;
 676
 677         if (sock != NULL)
 678                 kernel_sock_shutdown(sock, SHUT_WR);
 679 }
 680
 681 /**
 682  * xs_tcp_send_request - write an RPC request to a TCP socket
 683  * @task: address of RPC task that manages the state of an RPC request
 684  *
 685  * Return values:
 686  *        0:    The request has been sent
 687  *   EAGAIN:    The socket was blocked, please call again later to
 688  *              complete the request
 689  * ENOTCONN:    Caller needs to invoke connect logic then call again
 690  *    other:    Some other error occurred, the request was not sent
 691  *
 692  * XXX: In the case of soft timeouts, should we eventually give up
 693  *      if sendmsg is not able to make progress?
 694  */
 695 static int xs_tcp_send_request(struct rpc_task *task)
 696 {
 697         struct rpc_rqst *req = task->tk_rqstp;
 698         struct rpc_xprt *xprt = req->rq_xprt;
 699         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 700         struct xdr_buf *xdr = &req->rq_snd_buf;
 701         bool zerocopy = true;
 702         int status;
 703
 704         xs_encode_stream_record_marker(&req->rq_snd_buf);
 705
 706         xs_pktdump("packet data:",
 707                                 req->rq_svec->iov_base,
 708                                 req->rq_svec->iov_len);
 709         /* Don't use zero copy if this is a resend. If the RPC call
 710          * completes while the socket holds a reference to the pages,
 711          * then we may end up resending corrupted data.
 712          */
 713         if (task->tk_flags & RPC_TASK_SENT)
 714                 zerocopy = false;
 715
 716         /* Continue transmitting the packet/record. We must be careful
 717          * to cope with writespace callbacks arriving _after_ we have
 718          * called sendmsg(). */
 719         while (1) {
 720                 status = xs_sendpages(transport->sock,
 721                                         NULL, 0, xdr, req->rq_bytes_sent,
 722                                         zerocopy);
 723
 724                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 725                                 xdr->len - req->rq_bytes_sent, status);
 726
 727                 if (unlikely(status < 0))
 728                         break;
 729
 730                 /* If we've sent the entire packet, immediately
 731                  * reset the count of bytes sent. */
 732                 req->rq_bytes_sent += status;
 733                 req->rq_xmit_bytes_sent += status;
 734                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 735                         req->rq_bytes_sent = 0;
 736                         return 0;
 737                 }
 738
 739                 if (status != 0)
 740                         continue;
 741                 status = -EAGAIN;
 742                 break;
 743         }
 744
 745         switch (status) {
 746         case -ENOTSOCK:
 747                 status = -ENOTCONN;
 748                 /* Should we call xs_close() here? */
 749                 break;
 750         case -EAGAIN:
 751                 status = xs_nospace(task);
 752                 break;
 753         default:
 754                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 755                         -status);
 756         case -ECONNRESET:
 757                 xs_tcp_shutdown(xprt);
 758         case -ECONNREFUSED:
 759         case -ENOTCONN:
 760         case -EPIPE:
 761                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 762         }
 763
 764         return status;
 765 }
 766
 767 /**
 768  * xs_tcp_release_xprt - clean up after a tcp transmission
 769  * @xprt: transport
 770  * @task: rpc task
 771  *
 772  * This cleans up if an error causes us to abort the transmission of a request.
 773  * In this case, the socket may need to be reset in order to avoid confusing
 774  * the server.
 775  */
 776 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 777 {
 778         struct rpc_rqst *req;
 779
 780         if (task != xprt->snd_task)
 781                 return;
 782         if (task == NULL)
 783                 goto out_release;
 784         req = task->tk_rqstp;
 785         if (req == NULL)
 786                 goto out_release;
 787         if (req->rq_bytes_sent == 0)
 788                 goto out_release;
 789         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 790                 goto out_release;
 791         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 792 out_release:
 793         xprt_release_xprt(xprt, task);
 794 }
 795
 796 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 797 {
 798         transport->old_data_ready = sk->sk_data_ready;
 799         transport->old_state_change = sk->sk_state_change;
 800         transport->old_write_space = sk->sk_write_space;
 801 }
 802
 803 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 804 {
 805         sk->sk_data_ready = transport->old_data_ready;
 806         sk->sk_state_change = transport->old_state_change;
 807         sk->sk_write_space = transport->old_write_space;
 808 }
 809
 810 static void xs_reset_transport(struct sock_xprt *transport)
 811 {
 812         struct socket *sock = transport->sock;
 813         struct sock *sk = transport->inet;
 814
 815         if (sk == NULL)
 816                 return;
 817
 818         transport->srcport = 0;
 819
 820         write_lock_bh(&sk->sk_callback_lock);
 821         transport->inet = NULL;
 822         transport->sock = NULL;
 823
 824         sk->sk_user_data = NULL;
 825
 826         xs_restore_old_callbacks(transport, sk);
 827         write_unlock_bh(&sk->sk_callback_lock);
 828
 829         sk->sk_no_check = 0;
 830
 831         sock_release(sock);
 832 }
 833
 834 /**
 835  * xs_close - close a socket
 836  * @xprt: transport
 837  *
 838  * This is used when all requests are complete; ie, no DRC state remains
 839  * on the server we want to save.
 840  *
 841  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 842  * xs_reset_transport() zeroing the socket from underneath a writer.
 843  */
 844 static void xs_close(struct rpc_xprt *xprt)
 845 {
 846         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 847
 848         dprintk("RPC:       xs_close xprt %p\n", xprt);
 849
 850         xs_reset_transport(transport);
 851         xprt->reestablish_timeout = 0;
 852
 853         smp_mb__before_clear_bit();
 854         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 855         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 856         clear_bit(XPRT_CLOSING, &xprt->state);
 857         smp_mb__after_clear_bit();
 858         xprt_disconnect_done(xprt);
 859 }
 860
 861 static void xs_tcp_close(struct rpc_xprt *xprt)
 862 {
 863         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 864                 xs_close(xprt);
 865         else
 866                 xs_tcp_shutdown(xprt);
 867 }
 868
 869 /**
 870  * xs_destroy - prepare to shutdown a transport
 871  * @xprt: doomed transport
 872  *
 873  */
 874 static void xs_destroy(struct rpc_xprt *xprt)
 875 {
 876         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 877
 878         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 879
 880         cancel_delayed_work_sync(&transport->connect_worker);
 881
 882         xs_close(xprt);
 883         xs_free_peer_addresses(xprt);
 884         xprt_free(xprt);
 885         module_put(THIS_MODULE);
 886 }
 887
 888 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 889 {
 890         return (struct rpc_xprt *) sk->sk_user_data;
 891 }
 892
 893 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 894 {
 895         struct xdr_skb_reader desc = {
 896                 .skb            = skb,
 897                 .offset         = sizeof(rpc_fraghdr),
 898                 .count          = skb->len - sizeof(rpc_fraghdr),
 899         };
 900
 901         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 902                 return -1;
 903         if (desc.count)
 904                 return -1;
 905         return 0;
 906 }
 907
 908 /**
 909  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 910  * @sk: socket with data to read
 911  * @len: how much data to read
 912  *
 913  * Currently this assumes we can read the whole reply in a single gulp.
 914  */
 915 static void xs_local_data_ready(struct sock *sk, int len)
 916 {
 917         struct rpc_task *task;
 918         struct rpc_xprt *xprt;
 919         struct rpc_rqst *rovr;
 920         struct sk_buff *skb;
 921         int err, repsize, copied;
 922         u32 _xid;
 923         __be32 *xp;
 924
 925         read_lock_bh(&sk->sk_callback_lock);
 926         dprintk("RPC:       %s...\n", __func__);
 927         xprt = xprt_from_sock(sk);
 928         if (xprt == NULL)
 929                 goto out;
 930
 931         skb = skb_recv_datagram(sk, 0, 1, &err);
 932         if (skb == NULL)
 933                 goto out;
 934
 935         if (xprt->shutdown)
 936                 goto dropit;
 937
 938         repsize = skb->len - sizeof(rpc_fraghdr);
 939         if (repsize < 4) {
 940                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 941                 goto dropit;
 942         }
 943
 944         /* Copy the XID from the skb... */
 945         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 946         if (xp == NULL)
 947                 goto dropit;
 948
 949         /* Look up and lock the request corresponding to the given XID */
 950         spin_lock(&xprt->transport_lock);
 951         rovr = xprt_lookup_rqst(xprt, *xp);
 952         if (!rovr)
 953                 goto out_unlock;
 954         task = rovr->rq_task;
 955
 956         copied = rovr->rq_private_buf.buflen;
 957         if (copied > repsize)
 958                 copied = repsize;
 959
 960         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 961                 dprintk("RPC:       sk_buff copy failed\n");
 962                 goto out_unlock;
 963         }
 964
 965         xprt_complete_rqst(task, copied);
 966
 967  out_unlock:
 968         spin_unlock(&xprt->transport_lock);
 969  dropit:
 970         skb_free_datagram(sk, skb);
 971  out:
 972         read_unlock_bh(&sk->sk_callback_lock);
 973 }
 974
 975 /**
 976  * xs_udp_data_ready - "data ready" callback for UDP sockets
 977  * @sk: socket with data to read
 978  * @len: how much data to read
 979  *
 980  */
 981 static void xs_udp_data_ready(struct sock *sk, int len)
 982 {
 983         struct rpc_task *task;
 984         struct rpc_xprt *xprt;
 985         struct rpc_rqst *rovr;
 986         struct sk_buff *skb;
 987         int err, repsize, copied;
 988         u32 _xid;
 989         __be32 *xp;
 990
 991         read_lock_bh(&sk->sk_callback_lock);
 992         dprintk("RPC:       xs_udp_data_ready...\n");
 993         if (!(xprt = xprt_from_sock(sk)))
 994                 goto out;
 995
 996         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 997                 goto out;
 998
 999         if (xprt->shutdown)
1000                 goto dropit;
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         /* Something worked... */
1033         dst_confirm(skb_dst(skb));
1034
1035         xprt_adjust_cwnd(task, copied);
1036         xprt_complete_rqst(task, copied);
1037
1038  out_unlock:
1039         spin_unlock(&xprt->transport_lock);
1040  dropit:
1041         skb_free_datagram(sk, skb);
1042  out:
1043         read_unlock_bh(&sk->sk_callback_lock);
1044 }
1045
1046 /*
1047  * Helper function to force a TCP close if the server is sending
1048  * junk and/or it has put us in CLOSE_WAIT
1049  */
1050 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1051 {
1052         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1053         xprt_force_disconnect(xprt);
1054 }
1055
1056 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1057 {
1058         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1059         size_t len, used;
1060         char *p;
1061
1062         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1063         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1064         used = xdr_skb_read_bits(desc, p, len);
1065         transport->tcp_offset += used;
1066         if (used != len)
1067                 return;
1068
1069         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1070         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1071                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1072         else
1073                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1074         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1075
1076         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1077         transport->tcp_offset = 0;
1078
1079         /* Sanity check of the record length */
1080         if (unlikely(transport->tcp_reclen < 8)) {
1081                 dprintk("RPC:       invalid TCP record fragment length\n");
1082                 xs_tcp_force_close(xprt);
1083                 return;
1084         }
1085         dprintk("RPC:       reading TCP record fragment of length %d\n",
1086                         transport->tcp_reclen);
1087 }
1088
1089 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1090 {
1091         if (transport->tcp_offset == transport->tcp_reclen) {
1092                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1093                 transport->tcp_offset = 0;
1094                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1095                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1096                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1097                         transport->tcp_copied = 0;
1098                 }
1099         }
1100 }
1101
1102 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1103 {
1104         size_t len, used;
1105         char *p;
1106
1107         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1108         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1109         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1110         used = xdr_skb_read_bits(desc, p, len);
1111         transport->tcp_offset += used;
1112         if (used != len)
1113                 return;
1114         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1115         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1116         transport->tcp_copied = 4;
1117         dprintk("RPC:       reading %s XID %08x\n",
1118                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1119                                                               : "request with",
1120                         ntohl(transport->tcp_xid));
1121         xs_tcp_check_fraghdr(transport);
1122 }
1123
1124 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1125                                        struct xdr_skb_reader *desc)
1126 {
1127         size_t len, used;
1128         u32 offset;
1129         char *p;
1130
1131         /*
1132          * We want transport->tcp_offset to be 8 at the end of this routine
1133          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1134          * When this function is called for the first time,
1135          * transport->tcp_offset is 4 (after having already read the xid).
1136          */
1137         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1138         len = sizeof(transport->tcp_calldir) - offset;
1139         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1140         p = ((char *) &transport->tcp_calldir) + offset;
1141         used = xdr_skb_read_bits(desc, p, len);
1142         transport->tcp_offset += used;
1143         if (used != len)
1144                 return;
1145         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1146         /*
1147          * We don't yet have the XDR buffer, so we will write the calldir
1148          * out after we get the buffer from the 'struct rpc_rqst'
1149          */
1150         switch (ntohl(transport->tcp_calldir)) {
1151         case RPC_REPLY:
1152                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1153                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1154                 transport->tcp_flags |= TCP_RPC_REPLY;
1155                 break;
1156         case RPC_CALL:
1157                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1158                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1159                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1160                 break;
1161         default:
1162                 dprintk("RPC:       invalid request message type\n");
1163                 xs_tcp_force_close(&transport->xprt);
1164         }
1165         xs_tcp_check_fraghdr(transport);
1166 }
1167
1168 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1169                                      struct xdr_skb_reader *desc,
1170                                      struct rpc_rqst *req)
1171 {
1172         struct sock_xprt *transport =
1173                                 container_of(xprt, struct sock_xprt, xprt);
1174         struct xdr_buf *rcvbuf;
1175         size_t len;
1176         ssize_t r;
1177
1178         rcvbuf = &req->rq_private_buf;
1179
1180         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1181                 /*
1182                  * Save the RPC direction in the XDR buffer
1183                  */
1184                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1185                         &transport->tcp_calldir,
1186                         sizeof(transport->tcp_calldir));
1187                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1188                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1189         }
1190
1191         len = desc->count;
1192         if (len > transport->tcp_reclen - transport->tcp_offset) {
1193                 struct xdr_skb_reader my_desc;
1194
1195                 len = transport->tcp_reclen - transport->tcp_offset;
1196                 memcpy(&my_desc, desc, sizeof(my_desc));
1197                 my_desc.count = len;
1198                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1199                                           &my_desc, xdr_skb_read_bits);
1200                 desc->count -= r;
1201                 desc->offset += r;
1202         } else
1203                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1204                                           desc, xdr_skb_read_bits);
1205
1206         if (r > 0) {
1207                 transport->tcp_copied += r;
1208                 transport->tcp_offset += r;
1209         }
1210         if (r != len) {
1211                 /* Error when copying to the receive buffer,
1212                  * usually because we weren't able to allocate
1213                  * additional buffer pages. All we can do now
1214                  * is turn off TCP_RCV_COPY_DATA, so the request
1215                  * will not receive any additional updates,
1216                  * and time out.
1217                  * Any remaining data from this record will
1218                  * be discarded.
1219                  */
1220                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1221                 dprintk("RPC:       XID %08x truncated request\n",
1222                                 ntohl(transport->tcp_xid));
1223                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1224                                 "tcp_offset = %u, tcp_reclen = %u\n",
1225                                 xprt, transport->tcp_copied,
1226                                 transport->tcp_offset, transport->tcp_reclen);
1227                 return;
1228         }
1229
1230         dprintk("RPC:       XID %08x read %Zd bytes\n",
1231                         ntohl(transport->tcp_xid), r);
1232         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1233                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1234                         transport->tcp_offset, transport->tcp_reclen);
1235
1236         if (transport->tcp_copied == req->rq_private_buf.buflen)
1237                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1238         else if (transport->tcp_offset == transport->tcp_reclen) {
1239                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1240                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1241         }
1242 }
1243
1244 /*
1245  * Finds the request corresponding to the RPC xid and invokes the common
1246  * tcp read code to read the data.
1247  */
1248 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1249                                     struct xdr_skb_reader *desc)
1250 {
1251         struct sock_xprt *transport =
1252                                 container_of(xprt, struct sock_xprt, xprt);
1253         struct rpc_rqst *req;
1254
1255         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1256
1257         /* Find and lock the request corresponding to this xid */
1258         spin_lock(&xprt->transport_lock);
1259         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1260         if (!req) {
1261                 dprintk("RPC:       XID %08x request not found!\n",
1262                                 ntohl(transport->tcp_xid));
1263                 spin_unlock(&xprt->transport_lock);
1264                 return -1;
1265         }
1266
1267         xs_tcp_read_common(xprt, desc, req);
1268
1269         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1270                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1271
1272         spin_unlock(&xprt->transport_lock);
1273         return 0;
1274 }
1275
1276 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1277 /*
1278  * Obtains an rpc_rqst previously allocated and invokes the common
1279  * tcp read code to read the data.  The result is placed in the callback
1280  * queue.
1281  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1282  * connection and return -1.
1283  */
1284 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1285                                        struct xdr_skb_reader *desc)
1286 {
1287         struct sock_xprt *transport =
1288                                 container_of(xprt, struct sock_xprt, xprt);
1289         struct rpc_rqst *req;
1290
1291         req = xprt_alloc_bc_request(xprt);
1292         if (req == NULL) {
1293                 printk(KERN_WARNING "Callback slot table overflowed\n");
1294                 xprt_force_disconnect(xprt);
1295                 return -1;
1296         }
1297
1298         req->rq_xid = transport->tcp_xid;
1299         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1300         xs_tcp_read_common(xprt, desc, req);
1301
1302         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1303                 struct svc_serv *bc_serv = xprt->bc_serv;
1304
1305                 /*
1306                  * Add callback request to callback list.  The callback
1307                  * service sleeps on the sv_cb_waitq waiting for new
1308                  * requests.  Wake it up after adding enqueing the
1309                  * request.
1310                  */
1311                 dprintk("RPC:       add callback request to list\n");
1312                 spin_lock(&bc_serv->sv_cb_lock);
1313                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1314                 spin_unlock(&bc_serv->sv_cb_lock);
1315                 wake_up(&bc_serv->sv_cb_waitq);
1316         }
1317
1318         req->rq_private_buf.len = transport->tcp_copied;
1319
1320         return 0;
1321 }
1322
1323 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1324                                         struct xdr_skb_reader *desc)
1325 {
1326         struct sock_xprt *transport =
1327                                 container_of(xprt, struct sock_xprt, xprt);
1328
1329         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1330                 xs_tcp_read_reply(xprt, desc) :
1331                 xs_tcp_read_callback(xprt, desc);
1332 }
1333 #else
1334 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1335                                         struct xdr_skb_reader *desc)
1336 {
1337         return xs_tcp_read_reply(xprt, desc);
1338 }
1339 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1340
1341 /*
1342  * Read data off the transport.  This can be either an RPC_CALL or an
1343  * RPC_REPLY.  Relay the processing to helper functions.
1344  */
1345 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1346                                     struct xdr_skb_reader *desc)
1347 {
1348         struct sock_xprt *transport =
1349                                 container_of(xprt, struct sock_xprt, xprt);
1350
1351         if (_xs_tcp_read_data(xprt, desc) == 0)
1352                 xs_tcp_check_fraghdr(transport);
1353         else {
1354                 /*
1355                  * The transport_lock protects the request handling.
1356                  * There's no need to hold it to update the tcp_flags.
1357                  */
1358                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1359         }
1360 }
1361
1362 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1363 {
1364         size_t len;
1365
1366         len = transport->tcp_reclen - transport->tcp_offset;
1367         if (len > desc->count)
1368                 len = desc->count;
1369         desc->count -= len;
1370         desc->offset += len;
1371         transport->tcp_offset += len;
1372         dprintk("RPC:       discarded %Zu bytes\n", len);
1373         xs_tcp_check_fraghdr(transport);
1374 }
1375
1376 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1377 {
1378         struct rpc_xprt *xprt = rd_desc->arg.data;
1379         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1380         struct xdr_skb_reader desc = {
1381                 .skb    = skb,
1382                 .offset = offset,
1383                 .count  = len,
1384         };
1385
1386         dprintk("RPC:       xs_tcp_data_recv started\n");
1387         do {
1388                 /* Read in a new fragment marker if necessary */
1389                 /* Can we ever really expect to get completely empty fragments? */
1390                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1391                         xs_tcp_read_fraghdr(xprt, &desc);
1392                         continue;
1393                 }
1394                 /* Read in the xid if necessary */
1395                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1396                         xs_tcp_read_xid(transport, &desc);
1397                         continue;
1398                 }
1399                 /* Read in the call/reply flag */
1400                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1401                         xs_tcp_read_calldir(transport, &desc);
1402                         continue;
1403                 }
1404                 /* Read in the request data */
1405                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1406                         xs_tcp_read_data(xprt, &desc);
1407                         continue;
1408                 }
1409                 /* Skip over any trailing bytes on short reads */
1410                 xs_tcp_read_discard(transport, &desc);
1411         } while (desc.count);
1412         dprintk("RPC:       xs_tcp_data_recv done\n");
1413         return len - desc.count;
1414 }
1415
1416 /**
1417  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1418  * @sk: socket with data to read
1419  * @bytes: how much data to read
1420  *
1421  */
1422 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1423 {
1424         struct rpc_xprt *xprt;
1425         read_descriptor_t rd_desc;
1426         int read;
1427
1428         dprintk("RPC:       xs_tcp_data_ready...\n");
1429
1430         read_lock_bh(&sk->sk_callback_lock);
1431         if (!(xprt = xprt_from_sock(sk)))
1432                 goto out;
1433         if (xprt->shutdown)
1434                 goto out;
1435
1436         /* Any data means we had a useful conversation, so
1437          * the we don't need to delay the next reconnect
1438          */
1439         if (xprt->reestablish_timeout)
1440                 xprt->reestablish_timeout = 0;
1441
1442         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1443         rd_desc.arg.data = xprt;
1444         do {
1445                 rd_desc.count = 65536;
1446                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1447         } while (read > 0);
1448 out:
1449         read_unlock_bh(&sk->sk_callback_lock);
1450 }
1451
1452 /*
1453  * Do the equivalent of linger/linger2 handling for dealing with
1454  * broken servers that don't close the socket in a timely
1455  * fashion
1456  */
1457 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1458                 unsigned long timeout)
1459 {
1460         struct sock_xprt *transport;
1461
1462         if (xprt_test_and_set_connecting(xprt))
1463                 return;
1464         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1465         transport = container_of(xprt, struct sock_xprt, xprt);
1466         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1467                            timeout);
1468 }
1469
1470 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1471 {
1472         struct sock_xprt *transport;
1473
1474         transport = container_of(xprt, struct sock_xprt, xprt);
1475
1476         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1477             !cancel_delayed_work(&transport->connect_worker))
1478                 return;
1479         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1480         xprt_clear_connecting(xprt);
1481 }
1482
1483 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1484 {
1485         smp_mb__before_clear_bit();
1486         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1487         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1488         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1489         clear_bit(XPRT_CLOSING, &xprt->state);
1490         smp_mb__after_clear_bit();
1491 }
1492
1493 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1494 {
1495         xs_sock_reset_connection_flags(xprt);
1496         /* Mark transport as closed and wake up all pending tasks */
1497         xprt_disconnect_done(xprt);
1498 }
1499
1500 /**
1501  * xs_tcp_state_change - callback to handle TCP socket state changes
1502  * @sk: socket whose state has changed
1503  *
1504  */
1505 static void xs_tcp_state_change(struct sock *sk)
1506 {
1507         struct rpc_xprt *xprt;
1508
1509         read_lock_bh(&sk->sk_callback_lock);
1510         if (!(xprt = xprt_from_sock(sk)))
1511                 goto out;
1512         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1513         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1514                         sk->sk_state, xprt_connected(xprt),
1515                         sock_flag(sk, SOCK_DEAD),
1516                         sock_flag(sk, SOCK_ZAPPED),
1517                         sk->sk_shutdown);
1518
1519         switch (sk->sk_state) {
1520         case TCP_ESTABLISHED:
1521                 spin_lock(&xprt->transport_lock);
1522                 if (!xprt_test_and_set_connected(xprt)) {
1523                         struct sock_xprt *transport = container_of(xprt,
1524                                         struct sock_xprt, xprt);
1525
1526                         /* Reset TCP record info */
1527                         transport->tcp_offset = 0;
1528                         transport->tcp_reclen = 0;
1529                         transport->tcp_copied = 0;
1530                         transport->tcp_flags =
1531                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1532
1533                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1534                 }
1535                 spin_unlock(&xprt->transport_lock);
1536                 break;
1537         case TCP_FIN_WAIT1:
1538                 /* The client initiated a shutdown of the socket */
1539                 xprt->connect_cookie++;
1540                 xprt->reestablish_timeout = 0;
1541                 set_bit(XPRT_CLOSING, &xprt->state);
1542                 smp_mb__before_clear_bit();
1543                 clear_bit(XPRT_CONNECTED, &xprt->state);
1544                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1545                 smp_mb__after_clear_bit();
1546                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1547                 break;
1548         case TCP_CLOSE_WAIT:
1549                 /* The server initiated a shutdown of the socket */
1550                 xprt->connect_cookie++;
1551                 clear_bit(XPRT_CONNECTED, &xprt->state);
1552                 xs_tcp_force_close(xprt);
1553         case TCP_CLOSING:
1554                 /*
1555                  * If the server closed down the connection, make sure that
1556                  * we back off before reconnecting
1557                  */
1558                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1559                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1560                 break;
1561         case TCP_LAST_ACK:
1562                 set_bit(XPRT_CLOSING, &xprt->state);
1563                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1564                 smp_mb__before_clear_bit();
1565                 clear_bit(XPRT_CONNECTED, &xprt->state);
1566                 smp_mb__after_clear_bit();
1567                 break;
1568         case TCP_CLOSE:
1569                 xs_tcp_cancel_linger_timeout(xprt);
1570                 xs_sock_mark_closed(xprt);
1571         }
1572  out:
1573         read_unlock_bh(&sk->sk_callback_lock);
1574 }
1575
1576 static void xs_write_space(struct sock *sk)
1577 {
1578         struct socket *sock;
1579         struct rpc_xprt *xprt;
1580
1581         if (unlikely(!(sock = sk->sk_socket)))
1582                 return;
1583         clear_bit(SOCK_NOSPACE, &sock->flags);
1584
1585         if (unlikely(!(xprt = xprt_from_sock(sk))))
1586                 return;
1587         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1588                 return;
1589
1590         xprt_write_space(xprt);
1591 }
1592
1593 /**
1594  * xs_udp_write_space - callback invoked when socket buffer space
1595  *                             becomes available
1596  * @sk: socket whose state has changed
1597  *
1598  * Called when more output buffer space is available for this socket.
1599  * We try not to wake our writers until they can make "significant"
1600  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1601  * with a bunch of small requests.
1602  */
1603 static void xs_udp_write_space(struct sock *sk)
1604 {
1605         read_lock_bh(&sk->sk_callback_lock);
1606
1607         /* from net/core/sock.c:sock_def_write_space */
1608         if (sock_writeable(sk))
1609                 xs_write_space(sk);
1610
1611         read_unlock_bh(&sk->sk_callback_lock);
1612 }
1613
1614 /**
1615  * xs_tcp_write_space - callback invoked when socket buffer space
1616  *                             becomes available
1617  * @sk: socket whose state has changed
1618  *
1619  * Called when more output buffer space is available for this socket.
1620  * We try not to wake our writers until they can make "significant"
1621  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1622  * with a bunch of small requests.
1623  */
1624 static void xs_tcp_write_space(struct sock *sk)
1625 {
1626         read_lock_bh(&sk->sk_callback_lock);
1627
1628         /* from net/core/stream.c:sk_stream_write_space */
1629         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1630                 xs_write_space(sk);
1631
1632         read_unlock_bh(&sk->sk_callback_lock);
1633 }
1634
1635 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1636 {
1637         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1638         struct sock *sk = transport->inet;
1639
1640         if (transport->rcvsize) {
1641                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1642                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1643         }
1644         if (transport->sndsize) {
1645                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1646                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1647                 sk->sk_write_space(sk);
1648         }
1649 }
1650
1651 /**
1652  * xs_udp_set_buffer_size - set send and receive limits
1653  * @xprt: generic transport
1654  * @sndsize: requested size of send buffer, in bytes
1655  * @rcvsize: requested size of receive buffer, in bytes
1656  *
1657  * Set socket send and receive buffer size limits.
1658  */
1659 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1660 {
1661         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1662
1663         transport->sndsize = 0;
1664         if (sndsize)
1665                 transport->sndsize = sndsize + 1024;
1666         transport->rcvsize = 0;
1667         if (rcvsize)
1668                 transport->rcvsize = rcvsize + 1024;
1669
1670         xs_udp_do_set_buffer_size(xprt);
1671 }
1672
1673 /**
1674  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1675  * @task: task that timed out
1676  *
1677  * Adjust the congestion window after a retransmit timeout has occurred.
1678  */
1679 static void xs_udp_timer(struct rpc_task *task)
1680 {
1681         xprt_adjust_cwnd(task, -ETIMEDOUT);
1682 }
1683
1684 static unsigned short xs_get_random_port(void)
1685 {
1686         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1687         unsigned short rand = (unsigned short) net_random() % range;
1688         return rand + xprt_min_resvport;
1689 }
1690
1691 /**
1692  * xs_set_port - reset the port number in the remote endpoint address
1693  * @xprt: generic transport
1694  * @port: new port number
1695  *
1696  */
1697 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1698 {
1699         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1700
1701         rpc_set_port(xs_addr(xprt), port);
1702         xs_update_peer_port(xprt);
1703 }
1704
1705 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1706 {
1707         unsigned short port = transport->srcport;
1708
1709         if (port == 0 && transport->xprt.resvport)
1710                 port = xs_get_random_port();
1711         return port;
1712 }
1713
1714 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1715 {
1716         if (transport->srcport != 0)
1717                 transport->srcport = 0;
1718         if (!transport->xprt.resvport)
1719                 return 0;
1720         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1721                 return xprt_max_resvport;
1722         return --port;
1723 }
1724 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1725 {
1726         struct sockaddr_storage myaddr;
1727         int err, nloop = 0;
1728         unsigned short port = xs_get_srcport(transport);
1729         unsigned short last;
1730
1731         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1732         do {
1733                 rpc_set_port((struct sockaddr *)&myaddr, port);
1734                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1735                                 transport->xprt.addrlen);
1736                 if (port == 0)
1737                         break;
1738                 if (err == 0) {
1739                         transport->srcport = port;
1740                         break;
1741                 }
1742                 last = port;
1743                 port = xs_next_srcport(transport, port);
1744                 if (port > last)
1745                         nloop++;
1746         } while (err == -EADDRINUSE && nloop != 2);
1747
1748         if (myaddr.ss_family == AF_INET)
1749                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1750                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1751                                 port, err ? "failed" : "ok", err);
1752         else
1753                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1754                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1755                                 port, err ? "failed" : "ok", err);
1756         return err;
1757 }
1758
1759 /*
1760  * We don't support autobind on AF_LOCAL sockets
1761  */
1762 static void xs_local_rpcbind(struct rpc_task *task)
1763 {
1764         xprt_set_bound(task->tk_xprt);
1765 }
1766
1767 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1768 {
1769 }
1770
1771 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1772 static struct lock_class_key xs_key[2];
1773 static struct lock_class_key xs_slock_key[2];
1774
1775 static inline void xs_reclassify_socketu(struct socket *sock)
1776 {
1777         struct sock *sk = sock->sk;
1778
1779         BUG_ON(sock_owned_by_user(sk));
1780         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1781                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1782 }
1783
1784 static inline void xs_reclassify_socket4(struct socket *sock)
1785 {
1786         struct sock *sk = sock->sk;
1787
1788         BUG_ON(sock_owned_by_user(sk));
1789         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1790                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1791 }
1792
1793 static inline void xs_reclassify_socket6(struct socket *sock)
1794 {
1795         struct sock *sk = sock->sk;
1796
1797         BUG_ON(sock_owned_by_user(sk));
1798         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1799                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1800 }
1801
1802 static inline void xs_reclassify_socket(int family, struct socket *sock)
1803 {
1804         switch (family) {
1805         case AF_LOCAL:
1806                 xs_reclassify_socketu(sock);
1807                 break;
1808         case AF_INET:
1809                 xs_reclassify_socket4(sock);
1810                 break;
1811         case AF_INET6:
1812                 xs_reclassify_socket6(sock);
1813                 break;
1814         }
1815 }
1816 #else
1817 static inline void xs_reclassify_socketu(struct socket *sock)
1818 {
1819 }
1820
1821 static inline void xs_reclassify_socket4(struct socket *sock)
1822 {
1823 }
1824
1825 static inline void xs_reclassify_socket6(struct socket *sock)
1826 {
1827 }
1828
1829 static inline void xs_reclassify_socket(int family, struct socket *sock)
1830 {
1831 }
1832 #endif
1833
1834 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1835                 struct sock_xprt *transport, int family, int type, int protocol)
1836 {
1837         struct socket *sock;
1838         int err;
1839
1840         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1841         if (err < 0) {
1842                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1843                                 protocol, -err);
1844                 goto out;
1845         }
1846         xs_reclassify_socket(family, sock);
1847
1848         err = xs_bind(transport, sock);
1849         if (err) {
1850                 sock_release(sock);
1851                 goto out;
1852         }
1853
1854         return sock;
1855 out:
1856         return ERR_PTR(err);
1857 }
1858
1859 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1860                                       struct socket *sock)
1861 {
1862         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1863                                                                         xprt);
1864
1865         if (!transport->inet) {
1866                 struct sock *sk = sock->sk;
1867
1868                 write_lock_bh(&sk->sk_callback_lock);
1869
1870                 xs_save_old_callbacks(transport, sk);
1871
1872                 sk->sk_user_data = xprt;
1873                 sk->sk_data_ready = xs_local_data_ready;
1874                 sk->sk_write_space = xs_udp_write_space;
1875                 sk->sk_allocation = GFP_ATOMIC;
1876
1877                 xprt_clear_connected(xprt);
1878
1879                 /* Reset to new socket */
1880                 transport->sock = sock;
1881                 transport->inet = sk;
1882
1883                 write_unlock_bh(&sk->sk_callback_lock);
1884         }
1885
1886         /* Tell the socket layer to start connecting... */
1887         xprt->stat.connect_count++;
1888         xprt->stat.connect_start = jiffies;
1889         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1890 }
1891
1892 /**
1893  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1894  * @xprt: RPC transport to connect
1895  * @transport: socket transport to connect
1896  * @create_sock: function to create a socket of the correct type
1897  *
1898  * Invoked by a work queue tasklet.
1899  */
1900 static void xs_local_setup_socket(struct work_struct *work)
1901 {
1902         struct sock_xprt *transport =
1903                 container_of(work, struct sock_xprt, connect_worker.work);
1904         struct rpc_xprt *xprt = &transport->xprt;
1905         struct socket *sock;
1906         int status = -EIO;
1907
1908         if (xprt->shutdown)
1909                 goto out;
1910
1911         current->flags |= PF_FSTRANS;
1912
1913         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1914         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1915                                         SOCK_STREAM, 0, &sock, 1);
1916         if (status < 0) {
1917                 dprintk("RPC:       can't create AF_LOCAL "
1918                         "transport socket (%d).\n", -status);
1919                 goto out;
1920         }
1921         xs_reclassify_socketu(sock);
1922
1923         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1924                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1925
1926         status = xs_local_finish_connecting(xprt, sock);
1927         switch (status) {
1928         case 0:
1929                 dprintk("RPC:       xprt %p connected to %s\n",
1930                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1931                 xprt_set_connected(xprt);
1932                 break;
1933         case -ENOENT:
1934                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1935                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1936                 break;
1937         default:
1938                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1939                                 __func__, -status,
1940                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1941         }
1942
1943 out:
1944         xprt_clear_connecting(xprt);
1945         xprt_wake_pending_tasks(xprt, status);
1946         current->flags &= ~PF_FSTRANS;
1947 }
1948
1949 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1950 {
1951         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1952
1953         if (!transport->inet) {
1954                 struct sock *sk = sock->sk;
1955
1956                 write_lock_bh(&sk->sk_callback_lock);
1957
1958                 xs_save_old_callbacks(transport, sk);
1959
1960                 sk->sk_user_data = xprt;
1961                 sk->sk_data_ready = xs_udp_data_ready;
1962                 sk->sk_write_space = xs_udp_write_space;
1963                 sk->sk_no_check = UDP_CSUM_NORCV;
1964                 sk->sk_allocation = GFP_ATOMIC;
1965
1966                 xprt_set_connected(xprt);
1967
1968                 /* Reset to new socket */
1969                 transport->sock = sock;
1970                 transport->inet = sk;
1971
1972                 write_unlock_bh(&sk->sk_callback_lock);
1973         }
1974         xs_udp_do_set_buffer_size(xprt);
1975 }
1976
1977 static void xs_udp_setup_socket(struct work_struct *work)
1978 {
1979         struct sock_xprt *transport =
1980                 container_of(work, struct sock_xprt, connect_worker.work);
1981         struct rpc_xprt *xprt = &transport->xprt;
1982         struct socket *sock = transport->sock;
1983         int status = -EIO;
1984
1985         if (xprt->shutdown)
1986                 goto out;
1987
1988         current->flags |= PF_FSTRANS;
1989
1990         /* Start by resetting any existing state */
1991         xs_reset_transport(transport);
1992         sock = xs_create_sock(xprt, transport,
1993                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
1994         if (IS_ERR(sock))
1995                 goto out;
1996
1997         dprintk("RPC:       worker connecting xprt %p via %s to "
1998                                 "%s (port %s)\n", xprt,
1999                         xprt->address_strings[RPC_DISPLAY_PROTO],
2000                         xprt->address_strings[RPC_DISPLAY_ADDR],
2001                         xprt->address_strings[RPC_DISPLAY_PORT]);
2002
2003         xs_udp_finish_connecting(xprt, sock);
2004         status = 0;
2005 out:
2006         xprt_clear_connecting(xprt);
2007         xprt_wake_pending_tasks(xprt, status);
2008         current->flags &= ~PF_FSTRANS;
2009 }
2010
2011 /*
2012  * We need to preserve the port number so the reply cache on the server can
2013  * find our cached RPC replies when we get around to reconnecting.
2014  */
2015 static void xs_abort_connection(struct sock_xprt *transport)
2016 {
2017         int result;
2018         struct sockaddr any;
2019
2020         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2021
2022         /*
2023          * Disconnect the transport socket by doing a connect operation
2024          * with AF_UNSPEC.  This should return immediately...
2025          */
2026         memset(&any, 0, sizeof(any));
2027         any.sa_family = AF_UNSPEC;
2028         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2029         if (!result)
2030                 xs_sock_reset_connection_flags(&transport->xprt);
2031         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2032 }
2033
2034 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2035 {
2036         unsigned int state = transport->inet->sk_state;
2037
2038         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2039                 /* we don't need to abort the connection if the socket
2040                  * hasn't undergone a shutdown
2041                  */
2042                 if (transport->inet->sk_shutdown == 0)
2043                         return;
2044                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2045                                 __func__, transport->inet->sk_shutdown);
2046         }
2047         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2048                 /* we don't need to abort the connection if the socket
2049                  * hasn't undergone a shutdown
2050                  */
2051                 if (transport->inet->sk_shutdown == 0)
2052                         return;
2053                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2054                                 "sk_shutdown set to %d\n",
2055                                 __func__, transport->inet->sk_shutdown);
2056         }
2057         xs_abort_connection(transport);
2058 }
2059
2060 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2061 {
2062         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2063         int ret = -ENOTCONN;
2064
2065         if (!transport->inet) {
2066                 struct sock *sk = sock->sk;
2067
2068                 write_lock_bh(&sk->sk_callback_lock);
2069
2070                 xs_save_old_callbacks(transport, sk);
2071
2072                 sk->sk_user_data = xprt;
2073                 sk->sk_data_ready = xs_tcp_data_ready;
2074                 sk->sk_state_change = xs_tcp_state_change;
2075                 sk->sk_write_space = xs_tcp_write_space;
2076                 sk->sk_allocation = GFP_ATOMIC;
2077
2078                 /* socket options */
2079                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2080                 sock_reset_flag(sk, SOCK_LINGER);
2081                 tcp_sk(sk)->linger2 = 0;
2082                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2083
2084                 xprt_clear_connected(xprt);
2085
2086                 /* Reset to new socket */
2087                 transport->sock = sock;
2088                 transport->inet = sk;
2089
2090                 write_unlock_bh(&sk->sk_callback_lock);
2091         }
2092
2093         if (!xprt_bound(xprt))
2094                 goto out;
2095
2096         /* Tell the socket layer to start connecting... */
2097         xprt->stat.connect_count++;
2098         xprt->stat.connect_start = jiffies;
2099         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2100         switch (ret) {
2101         case 0:
2102         case -EINPROGRESS:
2103                 /* SYN_SENT! */
2104                 xprt->connect_cookie++;
2105                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2106                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2107         }
2108 out:
2109         return ret;
2110 }
2111
2112 /**
2113  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2114  * @xprt: RPC transport to connect
2115  * @transport: socket transport to connect
2116  * @create_sock: function to create a socket of the correct type
2117  *
2118  * Invoked by a work queue tasklet.
2119  */
2120 static void xs_tcp_setup_socket(struct work_struct *work)
2121 {
2122         struct sock_xprt *transport =
2123                 container_of(work, struct sock_xprt, connect_worker.work);
2124         struct socket *sock = transport->sock;
2125         struct rpc_xprt *xprt = &transport->xprt;
2126         int status = -EIO;
2127
2128         if (xprt->shutdown)
2129                 goto out;
2130
2131         current->flags |= PF_FSTRANS;
2132
2133         if (!sock) {
2134                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2135                 sock = xs_create_sock(xprt, transport,
2136                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2137                 if (IS_ERR(sock)) {
2138                         status = PTR_ERR(sock);
2139                         goto out;
2140                 }
2141         } else {
2142                 int abort_and_exit;
2143
2144                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2145                                 &xprt->state);
2146                 /* "close" the socket, preserving the local port */
2147                 xs_tcp_reuse_connection(transport);
2148
2149                 if (abort_and_exit)
2150                         goto out_eagain;
2151         }
2152
2153         dprintk("RPC:       worker connecting xprt %p via %s to "
2154                                 "%s (port %s)\n", xprt,
2155                         xprt->address_strings[RPC_DISPLAY_PROTO],
2156                         xprt->address_strings[RPC_DISPLAY_ADDR],
2157                         xprt->address_strings[RPC_DISPLAY_PORT]);
2158
2159         status = xs_tcp_finish_connecting(xprt, sock);
2160         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2161                         xprt, -status, xprt_connected(xprt),
2162                         sock->sk->sk_state);
2163         switch (status) {
2164         default:
2165                 printk("%s: connect returned unhandled error %d\n",
2166                         __func__, status);
2167         case -EADDRNOTAVAIL:
2168                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2169                  * and retry
2170                  */
2171                 xs_tcp_force_close(xprt);
2172                 break;
2173         case -ECONNREFUSED:
2174         case -ECONNRESET:
2175         case -ENETUNREACH:
2176                 /* retry with existing socket, after a delay */
2177         case 0:
2178         case -EINPROGRESS:
2179         case -EALREADY:
2180                 xprt_clear_connecting(xprt);
2181                 current->flags &= ~PF_FSTRANS;
2182                 return;
2183         case -EINVAL:
2184                 /* Happens, for instance, if the user specified a link
2185                  * local IPv6 address without a scope-id.
2186                  */
2187                 goto out;
2188         }
2189 out_eagain:
2190         status = -EAGAIN;
2191 out:
2192         xprt_clear_connecting(xprt);
2193         xprt_wake_pending_tasks(xprt, status);
2194         current->flags &= ~PF_FSTRANS;
2195 }
2196
2197 /**
2198  * xs_connect - connect a socket to a remote endpoint
2199  * @task: address of RPC task that manages state of connect request
2200  *
2201  * TCP: If the remote end dropped the connection, delay reconnecting.
2202  *
2203  * UDP socket connects are synchronous, but we use a work queue anyway
2204  * to guarantee that even unprivileged user processes can set up a
2205  * socket on a privileged port.
2206  *
2207  * If a UDP socket connect fails, the delay behavior here prevents
2208  * retry floods (hard mounts).
2209  */
2210 static void xs_connect(struct rpc_task *task)
2211 {
2212         struct rpc_xprt *xprt = task->tk_xprt;
2213         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2214
2215         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2216                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2217                                 "seconds\n",
2218                                 xprt, xprt->reestablish_timeout / HZ);
2219                 queue_delayed_work(rpciod_workqueue,
2220                                    &transport->connect_worker,
2221                                    xprt->reestablish_timeout);
2222                 xprt->reestablish_timeout <<= 1;
2223                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2224                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2225                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2226                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2227         } else {
2228                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2229                 queue_delayed_work(rpciod_workqueue,
2230                                    &transport->connect_worker, 0);
2231         }
2232 }
2233
2234 /**
2235  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2236  * @xprt: rpc_xprt struct containing statistics
2237  * @seq: output file
2238  *
2239  */
2240 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2241 {
2242         long idle_time = 0;
2243
2244         if (xprt_connected(xprt))
2245                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2246
2247         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2248                         "%llu %llu\n",
2249                         xprt->stat.bind_count,
2250                         xprt->stat.connect_count,
2251                         xprt->stat.connect_time,
2252                         idle_time,
2253                         xprt->stat.sends,
2254                         xprt->stat.recvs,
2255                         xprt->stat.bad_xids,
2256                         xprt->stat.req_u,
2257                         xprt->stat.bklog_u);
2258 }
2259
2260 /**
2261  * xs_udp_print_stats - display UDP socket-specifc stats
2262  * @xprt: rpc_xprt struct containing statistics
2263  * @seq: output file
2264  *
2265  */
2266 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2267 {
2268         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2269
2270         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2271                         transport->srcport,
2272                         xprt->stat.bind_count,
2273                         xprt->stat.sends,
2274                         xprt->stat.recvs,
2275                         xprt->stat.bad_xids,
2276                         xprt->stat.req_u,
2277                         xprt->stat.bklog_u);
2278 }
2279
2280 /**
2281  * xs_tcp_print_stats - display TCP socket-specifc stats
2282  * @xprt: rpc_xprt struct containing statistics
2283  * @seq: output file
2284  *
2285  */
2286 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2287 {
2288         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2289         long idle_time = 0;
2290
2291         if (xprt_connected(xprt))
2292                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2293
2294         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2295                         transport->srcport,
2296                         xprt->stat.bind_count,
2297                         xprt->stat.connect_count,
2298                         xprt->stat.connect_time,
2299                         idle_time,
2300                         xprt->stat.sends,
2301                         xprt->stat.recvs,
2302                         xprt->stat.bad_xids,
2303                         xprt->stat.req_u,
2304                         xprt->stat.bklog_u);
2305 }
2306
2307 /*
2308  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2309  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2310  * to use the server side send routines.
2311  */
2312 static void *bc_malloc(struct rpc_task *task, size_t size)
2313 {
2314         struct page *page;
2315         struct rpc_buffer *buf;
2316
2317         BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2318         page = alloc_page(GFP_KERNEL);
2319
2320         if (!page)
2321                 return NULL;
2322
2323         buf = page_address(page);
2324         buf->len = PAGE_SIZE;
2325
2326         return buf->data;
2327 }
2328
2329 /*
2330  * Free the space allocated in the bc_alloc routine
2331  */
2332 static void bc_free(void *buffer)
2333 {
2334         struct rpc_buffer *buf;
2335
2336         if (!buffer)
2337                 return;
2338
2339         buf = container_of(buffer, struct rpc_buffer, data);
2340         free_page((unsigned long)buf);
2341 }
2342
2343 /*
2344  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2345  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2346  */
2347 static int bc_sendto(struct rpc_rqst *req)
2348 {
2349         int len;
2350         struct xdr_buf *xbufp = &req->rq_snd_buf;
2351         struct rpc_xprt *xprt = req->rq_xprt;
2352         struct sock_xprt *transport =
2353                                 container_of(xprt, struct sock_xprt, xprt);
2354         struct socket *sock = transport->sock;
2355         unsigned long headoff;
2356         unsigned long tailoff;
2357
2358         xs_encode_stream_record_marker(xbufp);
2359
2360         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2361         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2362         len = svc_send_common(sock, xbufp,
2363                               virt_to_page(xbufp->head[0].iov_base), headoff,
2364                               xbufp->tail[0].iov_base, tailoff);
2365
2366         if (len != xbufp->len) {
2367                 printk(KERN_NOTICE "Error sending entire callback!\n");
2368                 len = -EAGAIN;
2369         }
2370
2371         return len;
2372 }
2373
2374 /*
2375  * The send routine. Borrows from svc_send
2376  */
2377 static int bc_send_request(struct rpc_task *task)
2378 {
2379         struct rpc_rqst *req = task->tk_rqstp;
2380         struct svc_xprt *xprt;
2381         struct svc_sock         *svsk;
2382         u32                     len;
2383
2384         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2385         /*
2386          * Get the server socket associated with this callback xprt
2387          */
2388         xprt = req->rq_xprt->bc_xprt;
2389         svsk = container_of(xprt, struct svc_sock, sk_xprt);
2390
2391         /*
2392          * Grab the mutex to serialize data as the connection is shared
2393          * with the fore channel
2394          */
2395         if (!mutex_trylock(&xprt->xpt_mutex)) {
2396                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2397                 if (!mutex_trylock(&xprt->xpt_mutex))
2398                         return -EAGAIN;
2399                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2400         }
2401         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2402                 len = -ENOTCONN;
2403         else
2404                 len = bc_sendto(req);
2405         mutex_unlock(&xprt->xpt_mutex);
2406
2407         if (len > 0)
2408                 len = 0;
2409
2410         return len;
2411 }
2412
2413 /*
2414  * The close routine. Since this is client initiated, we do nothing
2415  */
2416
2417 static void bc_close(struct rpc_xprt *xprt)
2418 {
2419 }
2420
2421 /*
2422  * The xprt destroy routine. Again, because this connection is client
2423  * initiated, we do nothing
2424  */
2425
2426 static void bc_destroy(struct rpc_xprt *xprt)
2427 {
2428 }
2429
2430 static struct rpc_xprt_ops xs_local_ops = {
2431         .reserve_xprt           = xprt_reserve_xprt,
2432         .release_xprt           = xs_tcp_release_xprt,
2433         .alloc_slot             = xprt_alloc_slot,
2434         .rpcbind                = xs_local_rpcbind,
2435         .set_port               = xs_local_set_port,
2436         .connect                = xs_connect,
2437         .buf_alloc              = rpc_malloc,
2438         .buf_free               = rpc_free,
2439         .send_request           = xs_local_send_request,
2440         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2441         .close                  = xs_close,
2442         .destroy                = xs_destroy,
2443         .print_stats            = xs_local_print_stats,
2444 };
2445
2446 static struct rpc_xprt_ops xs_udp_ops = {
2447         .set_buffer_size        = xs_udp_set_buffer_size,
2448         .reserve_xprt           = xprt_reserve_xprt_cong,
2449         .release_xprt           = xprt_release_xprt_cong,
2450         .alloc_slot             = xprt_alloc_slot,
2451         .rpcbind                = rpcb_getport_async,
2452         .set_port               = xs_set_port,
2453         .connect                = xs_connect,
2454         .buf_alloc              = rpc_malloc,
2455         .buf_free               = rpc_free,
2456         .send_request           = xs_udp_send_request,
2457         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2458         .timer                  = xs_udp_timer,
2459         .release_request        = xprt_release_rqst_cong,
2460         .close                  = xs_close,
2461         .destroy                = xs_destroy,
2462         .print_stats            = xs_udp_print_stats,
2463 };
2464
2465 static struct rpc_xprt_ops xs_tcp_ops = {
2466         .reserve_xprt           = xprt_reserve_xprt,
2467         .release_xprt           = xs_tcp_release_xprt,
2468         .alloc_slot             = xprt_lock_and_alloc_slot,
2469         .rpcbind                = rpcb_getport_async,
2470         .set_port               = xs_set_port,
2471         .connect                = xs_connect,
2472         .buf_alloc              = rpc_malloc,
2473         .buf_free               = rpc_free,
2474         .send_request           = xs_tcp_send_request,
2475         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2476         .close                  = xs_tcp_close,
2477         .destroy                = xs_destroy,
2478         .print_stats            = xs_tcp_print_stats,
2479 };
2480
2481 /*
2482  * The rpc_xprt_ops for the server backchannel
2483  */
2484
2485 static struct rpc_xprt_ops bc_tcp_ops = {
2486         .reserve_xprt           = xprt_reserve_xprt,
2487         .release_xprt           = xprt_release_xprt,
2488         .alloc_slot             = xprt_alloc_slot,
2489         .buf_alloc              = bc_malloc,
2490         .buf_free               = bc_free,
2491         .send_request           = bc_send_request,
2492         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2493         .close                  = bc_close,
2494         .destroy                = bc_destroy,
2495         .print_stats            = xs_tcp_print_stats,
2496 };
2497
2498 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2499 {
2500         static const struct sockaddr_in sin = {
2501                 .sin_family             = AF_INET,
2502                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2503         };
2504         static const struct sockaddr_in6 sin6 = {
2505                 .sin6_family            = AF_INET6,
2506                 .sin6_addr              = IN6ADDR_ANY_INIT,
2507         };
2508
2509         switch (family) {
2510         case AF_LOCAL:
2511                 break;
2512         case AF_INET:
2513                 memcpy(sap, &sin, sizeof(sin));
2514                 break;
2515         case AF_INET6:
2516                 memcpy(sap, &sin6, sizeof(sin6));
2517                 break;
2518         default:
2519                 dprintk("RPC:       %s: Bad address family\n", __func__);
2520                 return -EAFNOSUPPORT;
2521         }
2522         return 0;
2523 }
2524
2525 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2526                                       unsigned int slot_table_size,
2527                                       unsigned int max_slot_table_size)
2528 {
2529         struct rpc_xprt *xprt;
2530         struct sock_xprt *new;
2531
2532         if (args->addrlen > sizeof(xprt->addr)) {
2533                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2534                 return ERR_PTR(-EBADF);
2535         }
2536
2537         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2538                         max_slot_table_size);
2539         if (xprt == NULL) {
2540                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2541                                 "rpc_xprt\n");
2542                 return ERR_PTR(-ENOMEM);
2543         }
2544
2545         new = container_of(xprt, struct sock_xprt, xprt);
2546         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2547         xprt->addrlen = args->addrlen;
2548         if (args->srcaddr)
2549                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2550         else {
2551                 int err;
2552                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2553                                         (struct sockaddr *)&new->srcaddr);
2554                 if (err != 0) {
2555                         xprt_free(xprt);
2556                         return ERR_PTR(err);
2557                 }
2558         }
2559
2560         return xprt;
2561 }
2562
2563 static const struct rpc_timeout xs_local_default_timeout = {
2564         .to_initval = 10 * HZ,
2565         .to_maxval = 10 * HZ,
2566         .to_retries = 2,
2567 };
2568
2569 /**
2570  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2571  * @args: rpc transport creation arguments
2572  *
2573  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2574  */
2575 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2576 {
2577         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2578         struct sock_xprt *transport;
2579         struct rpc_xprt *xprt;
2580         struct rpc_xprt *ret;
2581
2582         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2583                         xprt_max_tcp_slot_table_entries);
2584         if (IS_ERR(xprt))
2585                 return xprt;
2586         transport = container_of(xprt, struct sock_xprt, xprt);
2587
2588         xprt->prot = 0;
2589         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2590         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2591
2592         xprt->bind_timeout = XS_BIND_TO;
2593         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2594         xprt->idle_timeout = XS_IDLE_DISC_TO;
2595
2596         xprt->ops = &xs_local_ops;
2597         xprt->timeout = &xs_local_default_timeout;
2598
2599         switch (sun->sun_family) {
2600         case AF_LOCAL:
2601                 if (sun->sun_path[0] != '/') {
2602                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2603                                         sun->sun_path);
2604                         ret = ERR_PTR(-EINVAL);
2605                         goto out_err;
2606                 }
2607                 xprt_set_bound(xprt);
2608                 INIT_DELAYED_WORK(&transport->connect_worker,
2609                                         xs_local_setup_socket);
2610                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2611                 break;
2612         default:
2613                 ret = ERR_PTR(-EAFNOSUPPORT);
2614                 goto out_err;
2615         }
2616
2617         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2618                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2619
2620         if (try_module_get(THIS_MODULE))
2621                 return xprt;
2622         ret = ERR_PTR(-EINVAL);
2623 out_err:
2624         xprt_free(xprt);
2625         return ret;
2626 }
2627
2628 static const struct rpc_timeout xs_udp_default_timeout = {
2629         .to_initval = 5 * HZ,
2630         .to_maxval = 30 * HZ,
2631         .to_increment = 5 * HZ,
2632         .to_retries = 5,
2633 };
2634
2635 /**
2636  * xs_setup_udp - Set up transport to use a UDP socket
2637  * @args: rpc transport creation arguments
2638  *
2639  */
2640 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2641 {
2642         struct sockaddr *addr = args->dstaddr;
2643         struct rpc_xprt *xprt;
2644         struct sock_xprt *transport;
2645         struct rpc_xprt *ret;
2646
2647         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2648                         xprt_udp_slot_table_entries);
2649         if (IS_ERR(xprt))
2650                 return xprt;
2651         transport = container_of(xprt, struct sock_xprt, xprt);
2652
2653         xprt->prot = IPPROTO_UDP;
2654         xprt->tsh_size = 0;
2655         /* XXX: header size can vary due to auth type, IPv6, etc. */
2656         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2657
2658         xprt->bind_timeout = XS_BIND_TO;
2659         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2660         xprt->idle_timeout = XS_IDLE_DISC_TO;
2661
2662         xprt->ops = &xs_udp_ops;
2663
2664         xprt->timeout = &xs_udp_default_timeout;
2665
2666         switch (addr->sa_family) {
2667         case AF_INET:
2668                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2669                         xprt_set_bound(xprt);
2670
2671                 INIT_DELAYED_WORK(&transport->connect_worker,
2672                                         xs_udp_setup_socket);
2673                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2674                 break;
2675         case AF_INET6:
2676                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2677                         xprt_set_bound(xprt);
2678
2679                 INIT_DELAYED_WORK(&transport->connect_worker,
2680                                         xs_udp_setup_socket);
2681                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2682                 break;
2683         default:
2684                 ret = ERR_PTR(-EAFNOSUPPORT);
2685                 goto out_err;
2686         }
2687
2688         if (xprt_bound(xprt))
2689                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2690                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2691                                 xprt->address_strings[RPC_DISPLAY_PORT],
2692                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2693         else
2694                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2695                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2696                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2697
2698         if (try_module_get(THIS_MODULE))
2699                 return xprt;
2700         ret = ERR_PTR(-EINVAL);
2701 out_err:
2702         xprt_free(xprt);
2703         return ret;
2704 }
2705
2706 static const struct rpc_timeout xs_tcp_default_timeout = {
2707         .to_initval = 60 * HZ,
2708         .to_maxval = 60 * HZ,
2709         .to_retries = 2,
2710 };
2711
2712 /**
2713  * xs_setup_tcp - Set up transport to use a TCP socket
2714  * @args: rpc transport creation arguments
2715  *
2716  */
2717 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2718 {
2719         struct sockaddr *addr = args->dstaddr;
2720         struct rpc_xprt *xprt;
2721         struct sock_xprt *transport;
2722         struct rpc_xprt *ret;
2723
2724         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2725                         xprt_max_tcp_slot_table_entries);
2726         if (IS_ERR(xprt))
2727                 return xprt;
2728         transport = container_of(xprt, struct sock_xprt, xprt);
2729
2730         xprt->prot = IPPROTO_TCP;
2731         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2732         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2733
2734         xprt->bind_timeout = XS_BIND_TO;
2735         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2736         xprt->idle_timeout = XS_IDLE_DISC_TO;
2737
2738         xprt->ops = &xs_tcp_ops;
2739         xprt->timeout = &xs_tcp_default_timeout;
2740
2741         switch (addr->sa_family) {
2742         case AF_INET:
2743                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2744                         xprt_set_bound(xprt);
2745
2746                 INIT_DELAYED_WORK(&transport->connect_worker,
2747                                         xs_tcp_setup_socket);
2748                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2749                 break;
2750         case AF_INET6:
2751                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2752                         xprt_set_bound(xprt);
2753
2754                 INIT_DELAYED_WORK(&transport->connect_worker,
2755                                         xs_tcp_setup_socket);
2756                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2757                 break;
2758         default:
2759                 ret = ERR_PTR(-EAFNOSUPPORT);
2760                 goto out_err;
2761         }
2762
2763         if (xprt_bound(xprt))
2764                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2765                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2766                                 xprt->address_strings[RPC_DISPLAY_PORT],
2767                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2768         else
2769                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2770                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2771                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2772
2773
2774         if (try_module_get(THIS_MODULE))
2775                 return xprt;
2776         ret = ERR_PTR(-EINVAL);
2777 out_err:
2778         xprt_free(xprt);
2779         return ret;
2780 }
2781
2782 /**
2783  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2784  * @args: rpc transport creation arguments
2785  *
2786  */
2787 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2788 {
2789         struct sockaddr *addr = args->dstaddr;
2790         struct rpc_xprt *xprt;
2791         struct sock_xprt *transport;
2792         struct svc_sock *bc_sock;
2793         struct rpc_xprt *ret;
2794
2795         if (args->bc_xprt->xpt_bc_xprt) {
2796                 /*
2797                  * This server connection already has a backchannel
2798                  * export; we can't create a new one, as we wouldn't be
2799                  * able to match replies based on xid any more.  So,
2800                  * reuse the already-existing one:
2801                  */
2802                  return args->bc_xprt->xpt_bc_xprt;
2803         }
2804         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2805                         xprt_tcp_slot_table_entries);
2806         if (IS_ERR(xprt))
2807                 return xprt;
2808         transport = container_of(xprt, struct sock_xprt, xprt);
2809
2810         xprt->prot = IPPROTO_TCP;
2811         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2812         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2813         xprt->timeout = &xs_tcp_default_timeout;
2814
2815         /* backchannel */
2816         xprt_set_bound(xprt);
2817         xprt->bind_timeout = 0;
2818         xprt->reestablish_timeout = 0;
2819         xprt->idle_timeout = 0;
2820
2821         xprt->ops = &bc_tcp_ops;
2822
2823         switch (addr->sa_family) {
2824         case AF_INET:
2825                 xs_format_peer_addresses(xprt, "tcp",
2826                                          RPCBIND_NETID_TCP);
2827                 break;
2828         case AF_INET6:
2829                 xs_format_peer_addresses(xprt, "tcp",
2830                                    RPCBIND_NETID_TCP6);
2831                 break;
2832         default:
2833                 ret = ERR_PTR(-EAFNOSUPPORT);
2834                 goto out_err;
2835         }
2836
2837         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2838                         xprt->address_strings[RPC_DISPLAY_ADDR],
2839                         xprt->address_strings[RPC_DISPLAY_PORT],
2840                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2841
2842         /*
2843          * Once we've associated a backchannel xprt with a connection,
2844          * we want to keep it around as long as long as the connection
2845          * lasts, in case we need to start using it for a backchannel
2846          * again; this reference won't be dropped until bc_xprt is
2847          * destroyed.
2848          */
2849         xprt_get(xprt);
2850         args->bc_xprt->xpt_bc_xprt = xprt;
2851         xprt->bc_xprt = args->bc_xprt;
2852         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2853         transport->sock = bc_sock->sk_sock;
2854         transport->inet = bc_sock->sk_sk;
2855
2856         /*
2857          * Since we don't want connections for the backchannel, we set
2858          * the xprt status to connected
2859          */
2860         xprt_set_connected(xprt);
2861
2862
2863         if (try_module_get(THIS_MODULE))
2864                 return xprt;
2865         xprt_put(xprt);
2866         ret = ERR_PTR(-EINVAL);
2867 out_err:
2868         xprt_free(xprt);
2869         return ret;
2870 }
2871
2872 static struct xprt_class        xs_local_transport = {
2873         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2874         .name           = "named UNIX socket",
2875         .owner          = THIS_MODULE,
2876         .ident          = XPRT_TRANSPORT_LOCAL,
2877         .setup          = xs_setup_local,
2878 };
2879
2880 static struct xprt_class        xs_udp_transport = {
2881         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2882         .name           = "udp",
2883         .owner          = THIS_MODULE,
2884         .ident          = XPRT_TRANSPORT_UDP,
2885         .setup          = xs_setup_udp,
2886 };
2887
2888 static struct xprt_class        xs_tcp_transport = {
2889         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2890         .name           = "tcp",
2891         .owner          = THIS_MODULE,
2892         .ident          = XPRT_TRANSPORT_TCP,
2893         .setup          = xs_setup_tcp,
2894 };
2895
2896 static struct xprt_class        xs_bc_tcp_transport = {
2897         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2898         .name           = "tcp NFSv4.1 backchannel",
2899         .owner          = THIS_MODULE,
2900         .ident          = XPRT_TRANSPORT_BC_TCP,
2901         .setup          = xs_setup_bc_tcp,
2902 };
2903
2904 /**
2905  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2906  *
2907  */
2908 int init_socket_xprt(void)
2909 {
2910 #ifdef RPC_DEBUG
2911         if (!sunrpc_table_header)
2912                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2913 #endif
2914
2915         xprt_register_transport(&xs_local_transport);
2916         xprt_register_transport(&xs_udp_transport);
2917         xprt_register_transport(&xs_tcp_transport);
2918         xprt_register_transport(&xs_bc_tcp_transport);
2919
2920         return 0;
2921 }
2922
2923 /**
2924  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2925  *
2926  */
2927 void cleanup_socket_xprt(void)
2928 {
2929 #ifdef RPC_DEBUG
2930         if (sunrpc_table_header) {
2931                 unregister_sysctl_table(sunrpc_table_header);
2932                 sunrpc_table_header = NULL;
2933         }
2934 #endif
2935
2936         xprt_unregister_transport(&xs_local_transport);
2937         xprt_unregister_transport(&xs_udp_transport);
2938         xprt_unregister_transport(&xs_tcp_transport);
2939         xprt_unregister_transport(&xs_bc_tcp_transport);
2940 }
2941
2942 static int param_set_uint_minmax(const char *val,
2943                 const struct kernel_param *kp,
2944                 unsigned int min, unsigned int max)
2945 {
2946         unsigned long num;
2947         int ret;
2948
2949         if (!val)
2950                 return -EINVAL;
2951         ret = strict_strtoul(val, 0, &num);
2952         if (ret == -EINVAL || num < min || num > max)
2953                 return -EINVAL;
2954         *((unsigned int *)kp->arg) = num;
2955         return 0;
2956 }
2957
2958 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2959 {
2960         return param_set_uint_minmax(val, kp,
2961                         RPC_MIN_RESVPORT,
2962                         RPC_MAX_RESVPORT);
2963 }
2964
2965 static struct kernel_param_ops param_ops_portnr = {
2966         .set = param_set_portnr,
2967         .get = param_get_uint,
2968 };
2969
2970 #define param_check_portnr(name, p) \
2971         __param_check(name, p, unsigned int);
2972
2973 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2974 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2975
2976 static int param_set_slot_table_size(const char *val,
2977                                      const struct kernel_param *kp)
2978 {
2979         return param_set_uint_minmax(val, kp,
2980                         RPC_MIN_SLOT_TABLE,
2981                         RPC_MAX_SLOT_TABLE);
2982 }
2983
2984 static struct kernel_param_ops param_ops_slot_table_size = {
2985         .set = param_set_slot_table_size,
2986         .get = param_get_uint,
2987 };
2988
2989 #define param_check_slot_table_size(name, p) \
2990         __param_check(name, p, unsigned int);
2991
2992 static int param_set_max_slot_table_size(const char *val,
2993                                      const struct kernel_param *kp)
2994 {
2995         return param_set_uint_minmax(val, kp,
2996                         RPC_MIN_SLOT_TABLE,
2997                         RPC_MAX_SLOT_TABLE_LIMIT);
2998 }
2999
3000 static struct kernel_param_ops param_ops_max_slot_table_size = {
3001         .set = param_set_max_slot_table_size,
3002         .get = param_get_uint,
3003 };
3004
3005 #define param_check_max_slot_table_size(name, p) \
3006         __param_check(name, p, unsigned int);
3007
3008 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3009                    slot_table_size, 0644);
3010 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3011                    max_slot_table_size, 0644);
3012 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3013                    slot_table_size, 0644);
3014