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