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_NFS_V4_1
  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_DEF_SLOT_TABLE;
  58
  59 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  60 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  61
  62 #define XS_TCP_LINGER_TO        (15U * HZ)
  63 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  64
  65 /*
  66  * We can register our own files under /proc/sys/sunrpc by
  67  * calling register_sysctl_table() again.  The files in that
  68  * directory become the union of all files registered there.
  69  *
  70  * We simply need to make sure that we don't collide with
  71  * someone else's file names!
  72  */
  73
  74 #ifdef RPC_DEBUG
  75
  76 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  77 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  78 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  79 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  80
  81 static struct ctl_table_header *sunrpc_table_header;
  82
  83 /*
  84  * FIXME: changing the UDP slot table size should also resize the UDP
  85  *        socket buffers for existing UDP transports
  86  */
  87 static ctl_table xs_tunables_table[] = {
  88         {
  89                 .procname       = "udp_slot_table_entries",
  90                 .data           = &xprt_udp_slot_table_entries,
  91                 .maxlen         = sizeof(unsigned int),
  92                 .mode           = 0644,
  93                 .proc_handler   = proc_dointvec_minmax,
  94                 .extra1         = &min_slot_table_size,
  95                 .extra2         = &max_slot_table_size
  96         },
  97         {
  98                 .procname       = "tcp_slot_table_entries",
  99                 .data           = &xprt_tcp_slot_table_entries,
 100                 .maxlen         = sizeof(unsigned int),
 101                 .mode           = 0644,
 102                 .proc_handler   = proc_dointvec_minmax,
 103                 .extra1         = &min_slot_table_size,
 104                 .extra2         = &max_slot_table_size
 105         },
 106         {
 107                 .procname       = "min_resvport",
 108                 .data           = &xprt_min_resvport,
 109                 .maxlen         = sizeof(unsigned int),
 110                 .mode           = 0644,
 111                 .proc_handler   = proc_dointvec_minmax,
 112                 .extra1         = &xprt_min_resvport_limit,
 113                 .extra2         = &xprt_max_resvport_limit
 114         },
 115         {
 116                 .procname       = "max_resvport",
 117                 .data           = &xprt_max_resvport,
 118                 .maxlen         = sizeof(unsigned int),
 119                 .mode           = 0644,
 120                 .proc_handler   = proc_dointvec_minmax,
 121                 .extra1         = &xprt_min_resvport_limit,
 122                 .extra2         = &xprt_max_resvport_limit
 123         },
 124         {
 125                 .procname       = "tcp_fin_timeout",
 126                 .data           = &xs_tcp_fin_timeout,
 127                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 128                 .mode           = 0644,
 129                 .proc_handler   = proc_dointvec_jiffies,
 130         },
 131         { },
 132 };
 133
 134 static ctl_table sunrpc_table[] = {
 135         {
 136                 .procname       = "sunrpc",
 137                 .mode           = 0555,
 138                 .child          = xs_tunables_table
 139         },
 140         { },
 141 };
 142
 143 #endif
 144
 145 /*
 146  * Wait duration for a reply from the RPC portmapper.
 147  */
 148 #define XS_BIND_TO              (60U * HZ)
 149
 150 /*
 151  * Delay if a UDP socket connect error occurs.  This is most likely some
 152  * kind of resource problem on the local host.
 153  */
 154 #define XS_UDP_REEST_TO         (2U * HZ)
 155
 156 /*
 157  * The reestablish timeout allows clients to delay for a bit before attempting
 158  * to reconnect to a server that just dropped our connection.
 159  *
 160  * We implement an exponential backoff when trying to reestablish a TCP
 161  * transport connection with the server.  Some servers like to drop a TCP
 162  * connection when they are overworked, so we start with a short timeout and
 163  * increase over time if the server is down or not responding.
 164  */
 165 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 166 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 167
 168 /*
 169  * TCP idle timeout; client drops the transport socket if it is idle
 170  * for this long.  Note that we also timeout UDP sockets to prevent
 171  * holding port numbers when there is no RPC traffic.
 172  */
 173 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 174
 175 #ifdef RPC_DEBUG
 176 # undef  RPC_DEBUG_DATA
 177 # define RPCDBG_FACILITY        RPCDBG_TRANS
 178 #endif
 179
 180 #ifdef RPC_DEBUG_DATA
 181 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 182 {
 183         u8 *buf = (u8 *) packet;
 184         int j;
 185
 186         dprintk("RPC:       %s\n", msg);
 187         for (j = 0; j < count && j < 128; j += 4) {
 188                 if (!(j & 31)) {
 189                         if (j)
 190                                 dprintk("\n");
 191                         dprintk("0x%04x ", j);
 192                 }
 193                 dprintk("%02x%02x%02x%02x ",
 194                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 195         }
 196         dprintk("\n");
 197 }
 198 #else
 199 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 200 {
 201         /* NOP */
 202 }
 203 #endif
 204
 205 struct sock_xprt {
 206         struct rpc_xprt         xprt;
 207
 208         /*
 209          * Network layer
 210          */
 211         struct socket *         sock;
 212         struct sock *           inet;
 213
 214         /*
 215          * State of TCP reply receive
 216          */
 217         __be32                  tcp_fraghdr,
 218                                 tcp_xid,
 219                                 tcp_calldir;
 220
 221         u32                     tcp_offset,
 222                                 tcp_reclen;
 223
 224         unsigned long           tcp_copied,
 225                                 tcp_flags;
 226
 227         /*
 228          * Connection of transports
 229          */
 230         struct delayed_work     connect_worker;
 231         struct sockaddr_storage srcaddr;
 232         unsigned short          srcport;
 233
 234         /*
 235          * UDP socket buffer size parameters
 236          */
 237         size_t                  rcvsize,
 238                                 sndsize;
 239
 240         /*
 241          * Saved socket callback addresses
 242          */
 243         void                    (*old_data_ready)(struct sock *, int);
 244         void                    (*old_state_change)(struct sock *);
 245         void                    (*old_write_space)(struct sock *);
 246         void                    (*old_error_report)(struct sock *);
 247 };
 248
 249 /*
 250  * TCP receive state flags
 251  */
 252 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 253 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 254 #define TCP_RCV_COPY_XID        (1UL << 2)
 255 #define TCP_RCV_COPY_DATA       (1UL << 3)
 256 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 257 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 258
 259 /*
 260  * TCP RPC flags
 261  */
 262 #define TCP_RPC_REPLY           (1UL << 6)
 263
 264 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 265 {
 266         return (struct sockaddr *) &xprt->addr;
 267 }
 268
 269 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 270 {
 271         return (struct sockaddr_un *) &xprt->addr;
 272 }
 273
 274 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 275 {
 276         return (struct sockaddr_in *) &xprt->addr;
 277 }
 278
 279 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 280 {
 281         return (struct sockaddr_in6 *) &xprt->addr;
 282 }
 283
 284 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 285 {
 286         struct sockaddr *sap = xs_addr(xprt);
 287         struct sockaddr_in6 *sin6;
 288         struct sockaddr_in *sin;
 289         struct sockaddr_un *sun;
 290         char buf[128];
 291
 292         switch (sap->sa_family) {
 293         case AF_LOCAL:
 294                 sun = xs_addr_un(xprt);
 295                 strlcpy(buf, sun->sun_path, sizeof(buf));
 296                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 297                                                 kstrdup(buf, GFP_KERNEL);
 298                 break;
 299         case AF_INET:
 300                 (void)rpc_ntop(sap, buf, sizeof(buf));
 301                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 302                                                 kstrdup(buf, GFP_KERNEL);
 303                 sin = xs_addr_in(xprt);
 304                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 305                 break;
 306         case AF_INET6:
 307                 (void)rpc_ntop(sap, buf, sizeof(buf));
 308                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 309                                                 kstrdup(buf, GFP_KERNEL);
 310                 sin6 = xs_addr_in6(xprt);
 311                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 312                 break;
 313         default:
 314                 BUG();
 315         }
 316
 317         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 318 }
 319
 320 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 321 {
 322         struct sockaddr *sap = xs_addr(xprt);
 323         char buf[128];
 324
 325         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 326         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 327
 328         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 329         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 330 }
 331
 332 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 333                                      const char *protocol,
 334                                      const char *netid)
 335 {
 336         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 337         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 338         xs_format_common_peer_addresses(xprt);
 339         xs_format_common_peer_ports(xprt);
 340 }
 341
 342 static void xs_update_peer_port(struct rpc_xprt *xprt)
 343 {
 344         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 345         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 346
 347         xs_format_common_peer_ports(xprt);
 348 }
 349
 350 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 351 {
 352         unsigned int i;
 353
 354         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 355                 switch (i) {
 356                 case RPC_DISPLAY_PROTO:
 357                 case RPC_DISPLAY_NETID:
 358                         continue;
 359                 default:
 360                         kfree(xprt->address_strings[i]);
 361                 }
 362 }
 363
 364 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 365
 366 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 367 {
 368         struct msghdr msg = {
 369                 .msg_name       = addr,
 370                 .msg_namelen    = addrlen,
 371                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 372         };
 373         struct kvec iov = {
 374                 .iov_base       = vec->iov_base + base,
 375                 .iov_len        = vec->iov_len - base,
 376         };
 377
 378         if (iov.iov_len != 0)
 379                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 380         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 381 }
 382
 383 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 384 {
 385         struct page **ppage;
 386         unsigned int remainder;
 387         int err, sent = 0;
 388
 389         remainder = xdr->page_len - base;
 390         base += xdr->page_base;
 391         ppage = xdr->pages + (base >> PAGE_SHIFT);
 392         base &= ~PAGE_MASK;
 393         for(;;) {
 394                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 395                 int flags = XS_SENDMSG_FLAGS;
 396
 397                 remainder -= len;
 398                 if (remainder != 0 || more)
 399                         flags |= MSG_MORE;
 400                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 401                 if (remainder == 0 || err != len)
 402                         break;
 403                 sent += err;
 404                 ppage++;
 405                 base = 0;
 406         }
 407         if (sent == 0)
 408                 return err;
 409         if (err > 0)
 410                 sent += err;
 411         return sent;
 412 }
 413
 414 /**
 415  * xs_sendpages - write pages directly to a socket
 416  * @sock: socket to send on
 417  * @addr: UDP only -- address of destination
 418  * @addrlen: UDP only -- length of destination address
 419  * @xdr: buffer containing this request
 420  * @base: starting position in the buffer
 421  *
 422  */
 423 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 424 {
 425         unsigned int remainder = xdr->len - base;
 426         int err, sent = 0;
 427
 428         if (unlikely(!sock))
 429                 return -ENOTSOCK;
 430
 431         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 432         if (base != 0) {
 433                 addr = NULL;
 434                 addrlen = 0;
 435         }
 436
 437         if (base < xdr->head[0].iov_len || addr != NULL) {
 438                 unsigned int len = xdr->head[0].iov_len - base;
 439                 remainder -= len;
 440                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 441                 if (remainder == 0 || err != len)
 442                         goto out;
 443                 sent += err;
 444                 base = 0;
 445         } else
 446                 base -= xdr->head[0].iov_len;
 447
 448         if (base < xdr->page_len) {
 449                 unsigned int len = xdr->page_len - base;
 450                 remainder -= len;
 451                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 452                 if (remainder == 0 || err != len)
 453                         goto out;
 454                 sent += err;
 455                 base = 0;
 456         } else
 457                 base -= xdr->page_len;
 458
 459         if (base >= xdr->tail[0].iov_len)
 460                 return sent;
 461         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 462 out:
 463         if (sent == 0)
 464                 return err;
 465         if (err > 0)
 466                 sent += err;
 467         return sent;
 468 }
 469
 470 static void xs_nospace_callback(struct rpc_task *task)
 471 {
 472         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 473
 474         transport->inet->sk_write_pending--;
 475         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 476 }
 477
 478 /**
 479  * xs_nospace - place task on wait queue if transmit was incomplete
 480  * @task: task to put to sleep
 481  *
 482  */
 483 static int xs_nospace(struct rpc_task *task)
 484 {
 485         struct rpc_rqst *req = task->tk_rqstp;
 486         struct rpc_xprt *xprt = req->rq_xprt;
 487         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 488         int ret = 0;
 489
 490         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 491                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 492                         req->rq_slen);
 493
 494         /* Protect against races with write_space */
 495         spin_lock_bh(&xprt->transport_lock);
 496
 497         /* Don't race with disconnect */
 498         if (xprt_connected(xprt)) {
 499                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 500                         ret = -EAGAIN;
 501                         /*
 502                          * Notify TCP that we're limited by the application
 503                          * window size
 504                          */
 505                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 506                         transport->inet->sk_write_pending++;
 507                         /* ...and wait for more buffer space */
 508                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 509                 }
 510         } else {
 511                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 512                 ret = -ENOTCONN;
 513         }
 514
 515         spin_unlock_bh(&xprt->transport_lock);
 516         return ret;
 517 }
 518
 519 /*
 520  * Construct a stream transport record marker in @buf.
 521  */
 522 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 523 {
 524         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 525         rpc_fraghdr *base = buf->head[0].iov_base;
 526         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 527 }
 528
 529 /**
 530  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 531  * @task: RPC task that manages the state of an RPC request
 532  *
 533  * Return values:
 534  *        0:    The request has been sent
 535  *   EAGAIN:    The socket was blocked, please call again later to
 536  *              complete the request
 537  * ENOTCONN:    Caller needs to invoke connect logic then call again
 538  *    other:    Some other error occured, the request was not sent
 539  */
 540 static int xs_local_send_request(struct rpc_task *task)
 541 {
 542         struct rpc_rqst *req = task->tk_rqstp;
 543         struct rpc_xprt *xprt = req->rq_xprt;
 544         struct sock_xprt *transport =
 545                                 container_of(xprt, struct sock_xprt, xprt);
 546         struct xdr_buf *xdr = &req->rq_snd_buf;
 547         int status;
 548
 549         xs_encode_stream_record_marker(&req->rq_snd_buf);
 550
 551         xs_pktdump("packet data:",
 552                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 553
 554         status = xs_sendpages(transport->sock, NULL, 0,
 555                                                 xdr, req->rq_bytes_sent);
 556         dprintk("RPC:       %s(%u) = %d\n",
 557                         __func__, xdr->len - req->rq_bytes_sent, status);
 558         if (likely(status >= 0)) {
 559                 req->rq_bytes_sent += status;
 560                 req->rq_xmit_bytes_sent += status;
 561                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 562                         req->rq_bytes_sent = 0;
 563                         return 0;
 564                 }
 565                 status = -EAGAIN;
 566         }
 567
 568         switch (status) {
 569         case -EAGAIN:
 570                 status = xs_nospace(task);
 571                 break;
 572         default:
 573                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 574                         -status);
 575         case -EPIPE:
 576                 xs_close(xprt);
 577                 status = -ENOTCONN;
 578         }
 579
 580         return status;
 581 }
 582
 583 /**
 584  * xs_udp_send_request - write an RPC request to a UDP socket
 585  * @task: address of RPC task that manages the state of an RPC request
 586  *
 587  * Return values:
 588  *        0:    The request has been sent
 589  *   EAGAIN:    The socket was blocked, please call again later to
 590  *              complete the request
 591  * ENOTCONN:    Caller needs to invoke connect logic then call again
 592  *    other:    Some other error occurred, the request was not sent
 593  */
 594 static int xs_udp_send_request(struct rpc_task *task)
 595 {
 596         struct rpc_rqst *req = task->tk_rqstp;
 597         struct rpc_xprt *xprt = req->rq_xprt;
 598         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 599         struct xdr_buf *xdr = &req->rq_snd_buf;
 600         int status;
 601
 602         xs_pktdump("packet data:",
 603                                 req->rq_svec->iov_base,
 604                                 req->rq_svec->iov_len);
 605
 606         if (!xprt_bound(xprt))
 607                 return -ENOTCONN;
 608         status = xs_sendpages(transport->sock,
 609                               xs_addr(xprt),
 610                               xprt->addrlen, xdr,
 611                               req->rq_bytes_sent);
 612
 613         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 614                         xdr->len - req->rq_bytes_sent, status);
 615
 616         if (status >= 0) {
 617                 req->rq_xmit_bytes_sent += status;
 618                 if (status >= req->rq_slen)
 619                         return 0;
 620                 /* Still some bytes left; set up for a retry later. */
 621                 status = -EAGAIN;
 622         }
 623
 624         switch (status) {
 625         case -ENOTSOCK:
 626                 status = -ENOTCONN;
 627                 /* Should we call xs_close() here? */
 628                 break;
 629         case -EAGAIN:
 630                 status = xs_nospace(task);
 631                 break;
 632         default:
 633                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 634                         -status);
 635         case -ENETUNREACH:
 636         case -EPIPE:
 637         case -ECONNREFUSED:
 638                 /* When the server has died, an ICMP port unreachable message
 639                  * prompts ECONNREFUSED. */
 640                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 641         }
 642
 643         return status;
 644 }
 645
 646 /**
 647  * xs_tcp_shutdown - gracefully shut down a TCP socket
 648  * @xprt: transport
 649  *
 650  * Initiates a graceful shutdown of the TCP socket by calling the
 651  * equivalent of shutdown(SHUT_WR);
 652  */
 653 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 654 {
 655         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 656         struct socket *sock = transport->sock;
 657
 658         if (sock != NULL)
 659                 kernel_sock_shutdown(sock, SHUT_WR);
 660 }
 661
 662 /**
 663  * xs_tcp_send_request - write an RPC request to a TCP socket
 664  * @task: address of RPC task that manages the state of an RPC request
 665  *
 666  * Return values:
 667  *        0:    The request has been sent
 668  *   EAGAIN:    The socket was blocked, please call again later to
 669  *              complete the request
 670  * ENOTCONN:    Caller needs to invoke connect logic then call again
 671  *    other:    Some other error occurred, the request was not sent
 672  *
 673  * XXX: In the case of soft timeouts, should we eventually give up
 674  *      if sendmsg is not able to make progress?
 675  */
 676 static int xs_tcp_send_request(struct rpc_task *task)
 677 {
 678         struct rpc_rqst *req = task->tk_rqstp;
 679         struct rpc_xprt *xprt = req->rq_xprt;
 680         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 681         struct xdr_buf *xdr = &req->rq_snd_buf;
 682         int status;
 683
 684         xs_encode_stream_record_marker(&req->rq_snd_buf);
 685
 686         xs_pktdump("packet data:",
 687                                 req->rq_svec->iov_base,
 688                                 req->rq_svec->iov_len);
 689
 690         /* Continue transmitting the packet/record. We must be careful
 691          * to cope with writespace callbacks arriving _after_ we have
 692          * called sendmsg(). */
 693         while (1) {
 694                 status = xs_sendpages(transport->sock,
 695                                         NULL, 0, xdr, req->rq_bytes_sent);
 696
 697                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 698                                 xdr->len - req->rq_bytes_sent, status);
 699
 700                 if (unlikely(status < 0))
 701                         break;
 702
 703                 /* If we've sent the entire packet, immediately
 704                  * reset the count of bytes sent. */
 705                 req->rq_bytes_sent += status;
 706                 req->rq_xmit_bytes_sent += status;
 707                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 708                         req->rq_bytes_sent = 0;
 709                         return 0;
 710                 }
 711
 712                 if (status != 0)
 713                         continue;
 714                 status = -EAGAIN;
 715                 break;
 716         }
 717
 718         switch (status) {
 719         case -ENOTSOCK:
 720                 status = -ENOTCONN;
 721                 /* Should we call xs_close() here? */
 722                 break;
 723         case -EAGAIN:
 724                 status = xs_nospace(task);
 725                 break;
 726         default:
 727                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 728                         -status);
 729         case -ECONNRESET:
 730         case -EPIPE:
 731                 xs_tcp_shutdown(xprt);
 732         case -ECONNREFUSED:
 733         case -ENOTCONN:
 734                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 735         }
 736
 737         return status;
 738 }
 739
 740 /**
 741  * xs_tcp_release_xprt - clean up after a tcp transmission
 742  * @xprt: transport
 743  * @task: rpc task
 744  *
 745  * This cleans up if an error causes us to abort the transmission of a request.
 746  * In this case, the socket may need to be reset in order to avoid confusing
 747  * the server.
 748  */
 749 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 750 {
 751         struct rpc_rqst *req;
 752
 753         if (task != xprt->snd_task)
 754                 return;
 755         if (task == NULL)
 756                 goto out_release;
 757         req = task->tk_rqstp;
 758         if (req->rq_bytes_sent == 0)
 759                 goto out_release;
 760         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 761                 goto out_release;
 762         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 763 out_release:
 764         xprt_release_xprt(xprt, task);
 765 }
 766
 767 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 768 {
 769         transport->old_data_ready = sk->sk_data_ready;
 770         transport->old_state_change = sk->sk_state_change;
 771         transport->old_write_space = sk->sk_write_space;
 772         transport->old_error_report = sk->sk_error_report;
 773 }
 774
 775 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 776 {
 777         sk->sk_data_ready = transport->old_data_ready;
 778         sk->sk_state_change = transport->old_state_change;
 779         sk->sk_write_space = transport->old_write_space;
 780         sk->sk_error_report = transport->old_error_report;
 781 }
 782
 783 static void xs_reset_transport(struct sock_xprt *transport)
 784 {
 785         struct socket *sock = transport->sock;
 786         struct sock *sk = transport->inet;
 787
 788         if (sk == NULL)
 789                 return;
 790
 791         transport->srcport = 0;
 792
 793         write_lock_bh(&sk->sk_callback_lock);
 794         transport->inet = NULL;
 795         transport->sock = NULL;
 796
 797         sk->sk_user_data = NULL;
 798
 799         xs_restore_old_callbacks(transport, sk);
 800         write_unlock_bh(&sk->sk_callback_lock);
 801
 802         sk->sk_no_check = 0;
 803
 804         sock_release(sock);
 805 }
 806
 807 /**
 808  * xs_close - close a socket
 809  * @xprt: transport
 810  *
 811  * This is used when all requests are complete; ie, no DRC state remains
 812  * on the server we want to save.
 813  *
 814  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 815  * xs_reset_transport() zeroing the socket from underneath a writer.
 816  */
 817 static void xs_close(struct rpc_xprt *xprt)
 818 {
 819         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 820
 821         dprintk("RPC:       xs_close xprt %p\n", xprt);
 822
 823         xs_reset_transport(transport);
 824         xprt->reestablish_timeout = 0;
 825
 826         smp_mb__before_clear_bit();
 827         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 828         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 829         clear_bit(XPRT_CLOSING, &xprt->state);
 830         smp_mb__after_clear_bit();
 831         xprt_disconnect_done(xprt);
 832 }
 833
 834 static void xs_tcp_close(struct rpc_xprt *xprt)
 835 {
 836         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 837                 xs_close(xprt);
 838         else
 839                 xs_tcp_shutdown(xprt);
 840 }
 841
 842 /**
 843  * xs_destroy - prepare to shutdown a transport
 844  * @xprt: doomed transport
 845  *
 846  */
 847 static void xs_destroy(struct rpc_xprt *xprt)
 848 {
 849         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 850
 851         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 852
 853         cancel_delayed_work_sync(&transport->connect_worker);
 854
 855         xs_close(xprt);
 856         xs_free_peer_addresses(xprt);
 857         xprt_free(xprt);
 858         module_put(THIS_MODULE);
 859 }
 860
 861 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 862 {
 863         return (struct rpc_xprt *) sk->sk_user_data;
 864 }
 865
 866 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 867 {
 868         struct xdr_skb_reader desc = {
 869                 .skb            = skb,
 870                 .offset         = sizeof(rpc_fraghdr),
 871                 .count          = skb->len - sizeof(rpc_fraghdr),
 872         };
 873
 874         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 875                 return -1;
 876         if (desc.count)
 877                 return -1;
 878         return 0;
 879 }
 880
 881 /**
 882  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 883  * @sk: socket with data to read
 884  * @len: how much data to read
 885  *
 886  * Currently this assumes we can read the whole reply in a single gulp.
 887  */
 888 static void xs_local_data_ready(struct sock *sk, int len)
 889 {
 890         struct rpc_task *task;
 891         struct rpc_xprt *xprt;
 892         struct rpc_rqst *rovr;
 893         struct sk_buff *skb;
 894         int err, repsize, copied;
 895         u32 _xid;
 896         __be32 *xp;
 897
 898         read_lock_bh(&sk->sk_callback_lock);
 899         dprintk("RPC:       %s...\n", __func__);
 900         xprt = xprt_from_sock(sk);
 901         if (xprt == NULL)
 902                 goto out;
 903
 904         skb = skb_recv_datagram(sk, 0, 1, &err);
 905         if (skb == NULL)
 906                 goto out;
 907
 908         if (xprt->shutdown)
 909                 goto dropit;
 910
 911         repsize = skb->len - sizeof(rpc_fraghdr);
 912         if (repsize < 4) {
 913                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 914                 goto dropit;
 915         }
 916
 917         /* Copy the XID from the skb... */
 918         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 919         if (xp == NULL)
 920                 goto dropit;
 921
 922         /* Look up and lock the request corresponding to the given XID */
 923         spin_lock(&xprt->transport_lock);
 924         rovr = xprt_lookup_rqst(xprt, *xp);
 925         if (!rovr)
 926                 goto out_unlock;
 927         task = rovr->rq_task;
 928
 929         copied = rovr->rq_private_buf.buflen;
 930         if (copied > repsize)
 931                 copied = repsize;
 932
 933         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 934                 dprintk("RPC:       sk_buff copy failed\n");
 935                 goto out_unlock;
 936         }
 937
 938         xprt_complete_rqst(task, copied);
 939
 940  out_unlock:
 941         spin_unlock(&xprt->transport_lock);
 942  dropit:
 943         skb_free_datagram(sk, skb);
 944  out:
 945         read_unlock_bh(&sk->sk_callback_lock);
 946 }
 947
 948 /**
 949  * xs_udp_data_ready - "data ready" callback for UDP sockets
 950  * @sk: socket with data to read
 951  * @len: how much data to read
 952  *
 953  */
 954 static void xs_udp_data_ready(struct sock *sk, int len)
 955 {
 956         struct rpc_task *task;
 957         struct rpc_xprt *xprt;
 958         struct rpc_rqst *rovr;
 959         struct sk_buff *skb;
 960         int err, repsize, copied;
 961         u32 _xid;
 962         __be32 *xp;
 963
 964         read_lock_bh(&sk->sk_callback_lock);
 965         dprintk("RPC:       xs_udp_data_ready...\n");
 966         if (!(xprt = xprt_from_sock(sk)))
 967                 goto out;
 968
 969         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 970                 goto out;
 971
 972         if (xprt->shutdown)
 973                 goto dropit;
 974
 975         repsize = skb->len - sizeof(struct udphdr);
 976         if (repsize < 4) {
 977                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 978                 goto dropit;
 979         }
 980
 981         /* Copy the XID from the skb... */
 982         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 983                                 sizeof(_xid), &_xid);
 984         if (xp == NULL)
 985                 goto dropit;
 986
 987         /* Look up and lock the request corresponding to the given XID */
 988         spin_lock(&xprt->transport_lock);
 989         rovr = xprt_lookup_rqst(xprt, *xp);
 990         if (!rovr)
 991                 goto out_unlock;
 992         task = rovr->rq_task;
 993
 994         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 995                 copied = repsize;
 996
 997         /* Suck it into the iovec, verify checksum if not done by hw. */
 998         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 999                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1000                 goto out_unlock;
1001         }
1002
1003         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1004
1005         /* Something worked... */
1006         dst_confirm(skb_dst(skb));
1007
1008         xprt_adjust_cwnd(task, copied);
1009         xprt_complete_rqst(task, copied);
1010
1011  out_unlock:
1012         spin_unlock(&xprt->transport_lock);
1013  dropit:
1014         skb_free_datagram(sk, skb);
1015  out:
1016         read_unlock_bh(&sk->sk_callback_lock);
1017 }
1018
1019 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1020 {
1021         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1022         size_t len, used;
1023         char *p;
1024
1025         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1026         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1027         used = xdr_skb_read_bits(desc, p, len);
1028         transport->tcp_offset += used;
1029         if (used != len)
1030                 return;
1031
1032         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1033         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1034                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1035         else
1036                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1037         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1038
1039         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1040         transport->tcp_offset = 0;
1041
1042         /* Sanity check of the record length */
1043         if (unlikely(transport->tcp_reclen < 8)) {
1044                 dprintk("RPC:       invalid TCP record fragment length\n");
1045                 xprt_force_disconnect(xprt);
1046                 return;
1047         }
1048         dprintk("RPC:       reading TCP record fragment of length %d\n",
1049                         transport->tcp_reclen);
1050 }
1051
1052 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1053 {
1054         if (transport->tcp_offset == transport->tcp_reclen) {
1055                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1056                 transport->tcp_offset = 0;
1057                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1058                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1059                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1060                         transport->tcp_copied = 0;
1061                 }
1062         }
1063 }
1064
1065 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1066 {
1067         size_t len, used;
1068         char *p;
1069
1070         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1071         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1072         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1073         used = xdr_skb_read_bits(desc, p, len);
1074         transport->tcp_offset += used;
1075         if (used != len)
1076                 return;
1077         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1078         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1079         transport->tcp_copied = 4;
1080         dprintk("RPC:       reading %s XID %08x\n",
1081                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1082                                                               : "request with",
1083                         ntohl(transport->tcp_xid));
1084         xs_tcp_check_fraghdr(transport);
1085 }
1086
1087 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1088                                        struct xdr_skb_reader *desc)
1089 {
1090         size_t len, used;
1091         u32 offset;
1092         char *p;
1093
1094         /*
1095          * We want transport->tcp_offset to be 8 at the end of this routine
1096          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1097          * When this function is called for the first time,
1098          * transport->tcp_offset is 4 (after having already read the xid).
1099          */
1100         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1101         len = sizeof(transport->tcp_calldir) - offset;
1102         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1103         p = ((char *) &transport->tcp_calldir) + offset;
1104         used = xdr_skb_read_bits(desc, p, len);
1105         transport->tcp_offset += used;
1106         if (used != len)
1107                 return;
1108         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1109         /*
1110          * We don't yet have the XDR buffer, so we will write the calldir
1111          * out after we get the buffer from the 'struct rpc_rqst'
1112          */
1113         switch (ntohl(transport->tcp_calldir)) {
1114         case RPC_REPLY:
1115                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1116                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1117                 transport->tcp_flags |= TCP_RPC_REPLY;
1118                 break;
1119         case RPC_CALL:
1120                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1121                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1122                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1123                 break;
1124         default:
1125                 dprintk("RPC:       invalid request message type\n");
1126                 xprt_force_disconnect(&transport->xprt);
1127         }
1128         xs_tcp_check_fraghdr(transport);
1129 }
1130
1131 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1132                                      struct xdr_skb_reader *desc,
1133                                      struct rpc_rqst *req)
1134 {
1135         struct sock_xprt *transport =
1136                                 container_of(xprt, struct sock_xprt, xprt);
1137         struct xdr_buf *rcvbuf;
1138         size_t len;
1139         ssize_t r;
1140
1141         rcvbuf = &req->rq_private_buf;
1142
1143         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1144                 /*
1145                  * Save the RPC direction in the XDR buffer
1146                  */
1147                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1148                         &transport->tcp_calldir,
1149                         sizeof(transport->tcp_calldir));
1150                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1151                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1152         }
1153
1154         len = desc->count;
1155         if (len > transport->tcp_reclen - transport->tcp_offset) {
1156                 struct xdr_skb_reader my_desc;
1157
1158                 len = transport->tcp_reclen - transport->tcp_offset;
1159                 memcpy(&my_desc, desc, sizeof(my_desc));
1160                 my_desc.count = len;
1161                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1162                                           &my_desc, xdr_skb_read_bits);
1163                 desc->count -= r;
1164                 desc->offset += r;
1165         } else
1166                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1167                                           desc, xdr_skb_read_bits);
1168
1169         if (r > 0) {
1170                 transport->tcp_copied += r;
1171                 transport->tcp_offset += r;
1172         }
1173         if (r != len) {
1174                 /* Error when copying to the receive buffer,
1175                  * usually because we weren't able to allocate
1176                  * additional buffer pages. All we can do now
1177                  * is turn off TCP_RCV_COPY_DATA, so the request
1178                  * will not receive any additional updates,
1179                  * and time out.
1180                  * Any remaining data from this record will
1181                  * be discarded.
1182                  */
1183                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1184                 dprintk("RPC:       XID %08x truncated request\n",
1185                                 ntohl(transport->tcp_xid));
1186                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1187                                 "tcp_offset = %u, tcp_reclen = %u\n",
1188                                 xprt, transport->tcp_copied,
1189                                 transport->tcp_offset, transport->tcp_reclen);
1190                 return;
1191         }
1192
1193         dprintk("RPC:       XID %08x read %Zd bytes\n",
1194                         ntohl(transport->tcp_xid), r);
1195         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1196                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1197                         transport->tcp_offset, transport->tcp_reclen);
1198
1199         if (transport->tcp_copied == req->rq_private_buf.buflen)
1200                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1201         else if (transport->tcp_offset == transport->tcp_reclen) {
1202                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1203                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1204         }
1205 }
1206
1207 /*
1208  * Finds the request corresponding to the RPC xid and invokes the common
1209  * tcp read code to read the data.
1210  */
1211 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1212                                     struct xdr_skb_reader *desc)
1213 {
1214         struct sock_xprt *transport =
1215                                 container_of(xprt, struct sock_xprt, xprt);
1216         struct rpc_rqst *req;
1217
1218         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1219
1220         /* Find and lock the request corresponding to this xid */
1221         spin_lock(&xprt->transport_lock);
1222         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1223         if (!req) {
1224                 dprintk("RPC:       XID %08x request not found!\n",
1225                                 ntohl(transport->tcp_xid));
1226                 spin_unlock(&xprt->transport_lock);
1227                 return -1;
1228         }
1229
1230         xs_tcp_read_common(xprt, desc, req);
1231
1232         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1233                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1234
1235         spin_unlock(&xprt->transport_lock);
1236         return 0;
1237 }
1238
1239 #if defined(CONFIG_NFS_V4_1)
1240 /*
1241  * Obtains an rpc_rqst previously allocated and invokes the common
1242  * tcp read code to read the data.  The result is placed in the callback
1243  * queue.
1244  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1245  * connection and return -1.
1246  */
1247 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1248                                        struct xdr_skb_reader *desc)
1249 {
1250         struct sock_xprt *transport =
1251                                 container_of(xprt, struct sock_xprt, xprt);
1252         struct rpc_rqst *req;
1253
1254         req = xprt_alloc_bc_request(xprt);
1255         if (req == NULL) {
1256                 printk(KERN_WARNING "Callback slot table overflowed\n");
1257                 xprt_force_disconnect(xprt);
1258                 return -1;
1259         }
1260
1261         req->rq_xid = transport->tcp_xid;
1262         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1263         xs_tcp_read_common(xprt, desc, req);
1264
1265         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1266                 struct svc_serv *bc_serv = xprt->bc_serv;
1267
1268                 /*
1269                  * Add callback request to callback list.  The callback
1270                  * service sleeps on the sv_cb_waitq waiting for new
1271                  * requests.  Wake it up after adding enqueing the
1272                  * request.
1273                  */
1274                 dprintk("RPC:       add callback request to list\n");
1275                 spin_lock(&bc_serv->sv_cb_lock);
1276                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1277                 spin_unlock(&bc_serv->sv_cb_lock);
1278                 wake_up(&bc_serv->sv_cb_waitq);
1279         }
1280
1281         req->rq_private_buf.len = transport->tcp_copied;
1282
1283         return 0;
1284 }
1285
1286 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1287                                         struct xdr_skb_reader *desc)
1288 {
1289         struct sock_xprt *transport =
1290                                 container_of(xprt, struct sock_xprt, xprt);
1291
1292         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1293                 xs_tcp_read_reply(xprt, desc) :
1294                 xs_tcp_read_callback(xprt, desc);
1295 }
1296 #else
1297 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1298                                         struct xdr_skb_reader *desc)
1299 {
1300         return xs_tcp_read_reply(xprt, desc);
1301 }
1302 #endif /* CONFIG_NFS_V4_1 */
1303
1304 /*
1305  * Read data off the transport.  This can be either an RPC_CALL or an
1306  * RPC_REPLY.  Relay the processing to helper functions.
1307  */
1308 static void 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         if (_xs_tcp_read_data(xprt, desc) == 0)
1315                 xs_tcp_check_fraghdr(transport);
1316         else {
1317                 /*
1318                  * The transport_lock protects the request handling.
1319                  * There's no need to hold it to update the tcp_flags.
1320                  */
1321                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1322         }
1323 }
1324
1325 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1326 {
1327         size_t len;
1328
1329         len = transport->tcp_reclen - transport->tcp_offset;
1330         if (len > desc->count)
1331                 len = desc->count;
1332         desc->count -= len;
1333         desc->offset += len;
1334         transport->tcp_offset += len;
1335         dprintk("RPC:       discarded %Zu bytes\n", len);
1336         xs_tcp_check_fraghdr(transport);
1337 }
1338
1339 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1340 {
1341         struct rpc_xprt *xprt = rd_desc->arg.data;
1342         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1343         struct xdr_skb_reader desc = {
1344                 .skb    = skb,
1345                 .offset = offset,
1346                 .count  = len,
1347         };
1348
1349         dprintk("RPC:       xs_tcp_data_recv started\n");
1350         do {
1351                 /* Read in a new fragment marker if necessary */
1352                 /* Can we ever really expect to get completely empty fragments? */
1353                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1354                         xs_tcp_read_fraghdr(xprt, &desc);
1355                         continue;
1356                 }
1357                 /* Read in the xid if necessary */
1358                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1359                         xs_tcp_read_xid(transport, &desc);
1360                         continue;
1361                 }
1362                 /* Read in the call/reply flag */
1363                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1364                         xs_tcp_read_calldir(transport, &desc);
1365                         continue;
1366                 }
1367                 /* Read in the request data */
1368                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1369                         xs_tcp_read_data(xprt, &desc);
1370                         continue;
1371                 }
1372                 /* Skip over any trailing bytes on short reads */
1373                 xs_tcp_read_discard(transport, &desc);
1374         } while (desc.count);
1375         dprintk("RPC:       xs_tcp_data_recv done\n");
1376         return len - desc.count;
1377 }
1378
1379 /**
1380  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1381  * @sk: socket with data to read
1382  * @bytes: how much data to read
1383  *
1384  */
1385 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1386 {
1387         struct rpc_xprt *xprt;
1388         read_descriptor_t rd_desc;
1389         int read;
1390
1391         dprintk("RPC:       xs_tcp_data_ready...\n");
1392
1393         read_lock_bh(&sk->sk_callback_lock);
1394         if (!(xprt = xprt_from_sock(sk)))
1395                 goto out;
1396         if (xprt->shutdown)
1397                 goto out;
1398
1399         /* Any data means we had a useful conversation, so
1400          * the we don't need to delay the next reconnect
1401          */
1402         if (xprt->reestablish_timeout)
1403                 xprt->reestablish_timeout = 0;
1404
1405         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1406         rd_desc.arg.data = xprt;
1407         do {
1408                 rd_desc.count = 65536;
1409                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1410         } while (read > 0);
1411 out:
1412         read_unlock_bh(&sk->sk_callback_lock);
1413 }
1414
1415 /*
1416  * Do the equivalent of linger/linger2 handling for dealing with
1417  * broken servers that don't close the socket in a timely
1418  * fashion
1419  */
1420 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1421                 unsigned long timeout)
1422 {
1423         struct sock_xprt *transport;
1424
1425         if (xprt_test_and_set_connecting(xprt))
1426                 return;
1427         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1428         transport = container_of(xprt, struct sock_xprt, xprt);
1429         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1430                            timeout);
1431 }
1432
1433 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1434 {
1435         struct sock_xprt *transport;
1436
1437         transport = container_of(xprt, struct sock_xprt, xprt);
1438
1439         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1440             !cancel_delayed_work(&transport->connect_worker))
1441                 return;
1442         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1443         xprt_clear_connecting(xprt);
1444 }
1445
1446 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1447 {
1448         smp_mb__before_clear_bit();
1449         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1450         clear_bit(XPRT_CLOSING, &xprt->state);
1451         smp_mb__after_clear_bit();
1452         /* Mark transport as closed and wake up all pending tasks */
1453         xprt_disconnect_done(xprt);
1454 }
1455
1456 /**
1457  * xs_tcp_state_change - callback to handle TCP socket state changes
1458  * @sk: socket whose state has changed
1459  *
1460  */
1461 static void xs_tcp_state_change(struct sock *sk)
1462 {
1463         struct rpc_xprt *xprt;
1464
1465         read_lock_bh(&sk->sk_callback_lock);
1466         if (!(xprt = xprt_from_sock(sk)))
1467                 goto out;
1468         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1469         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1470                         sk->sk_state, xprt_connected(xprt),
1471                         sock_flag(sk, SOCK_DEAD),
1472                         sock_flag(sk, SOCK_ZAPPED),
1473                         sk->sk_shutdown);
1474
1475         switch (sk->sk_state) {
1476         case TCP_ESTABLISHED:
1477                 spin_lock(&xprt->transport_lock);
1478                 if (!xprt_test_and_set_connected(xprt)) {
1479                         struct sock_xprt *transport = container_of(xprt,
1480                                         struct sock_xprt, xprt);
1481
1482                         /* Reset TCP record info */
1483                         transport->tcp_offset = 0;
1484                         transport->tcp_reclen = 0;
1485                         transport->tcp_copied = 0;
1486                         transport->tcp_flags =
1487                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1488
1489                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1490                 }
1491                 spin_unlock(&xprt->transport_lock);
1492                 break;
1493         case TCP_FIN_WAIT1:
1494                 /* The client initiated a shutdown of the socket */
1495                 xprt->connect_cookie++;
1496                 xprt->reestablish_timeout = 0;
1497                 set_bit(XPRT_CLOSING, &xprt->state);
1498                 smp_mb__before_clear_bit();
1499                 clear_bit(XPRT_CONNECTED, &xprt->state);
1500                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1501                 smp_mb__after_clear_bit();
1502                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1503                 break;
1504         case TCP_CLOSE_WAIT:
1505                 /* The server initiated a shutdown of the socket */
1506                 xprt_force_disconnect(xprt);
1507                 xprt->connect_cookie++;
1508         case TCP_CLOSING:
1509                 /*
1510                  * If the server closed down the connection, make sure that
1511                  * we back off before reconnecting
1512                  */
1513                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1514                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1515                 break;
1516         case TCP_LAST_ACK:
1517                 set_bit(XPRT_CLOSING, &xprt->state);
1518                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1519                 smp_mb__before_clear_bit();
1520                 clear_bit(XPRT_CONNECTED, &xprt->state);
1521                 smp_mb__after_clear_bit();
1522                 break;
1523         case TCP_CLOSE:
1524                 xs_tcp_cancel_linger_timeout(xprt);
1525                 xs_sock_mark_closed(xprt);
1526         }
1527  out:
1528         read_unlock_bh(&sk->sk_callback_lock);
1529 }
1530
1531 /**
1532  * xs_error_report - callback mainly for catching socket errors
1533  * @sk: socket
1534  */
1535 static void xs_error_report(struct sock *sk)
1536 {
1537         struct rpc_xprt *xprt;
1538
1539         read_lock_bh(&sk->sk_callback_lock);
1540         if (!(xprt = xprt_from_sock(sk)))
1541                 goto out;
1542         dprintk("RPC:       %s client %p...\n"
1543                         "RPC:       error %d\n",
1544                         __func__, xprt, sk->sk_err);
1545         xprt_wake_pending_tasks(xprt, -EAGAIN);
1546 out:
1547         read_unlock_bh(&sk->sk_callback_lock);
1548 }
1549
1550 static void xs_write_space(struct sock *sk)
1551 {
1552         struct socket *sock;
1553         struct rpc_xprt *xprt;
1554
1555         if (unlikely(!(sock = sk->sk_socket)))
1556                 return;
1557         clear_bit(SOCK_NOSPACE, &sock->flags);
1558
1559         if (unlikely(!(xprt = xprt_from_sock(sk))))
1560                 return;
1561         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1562                 return;
1563
1564         xprt_write_space(xprt);
1565 }
1566
1567 /**
1568  * xs_udp_write_space - callback invoked when socket buffer space
1569  *                             becomes available
1570  * @sk: socket whose state has changed
1571  *
1572  * Called when more output buffer space is available for this socket.
1573  * We try not to wake our writers until they can make "significant"
1574  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1575  * with a bunch of small requests.
1576  */
1577 static void xs_udp_write_space(struct sock *sk)
1578 {
1579         read_lock_bh(&sk->sk_callback_lock);
1580
1581         /* from net/core/sock.c:sock_def_write_space */
1582         if (sock_writeable(sk))
1583                 xs_write_space(sk);
1584
1585         read_unlock_bh(&sk->sk_callback_lock);
1586 }
1587
1588 /**
1589  * xs_tcp_write_space - callback invoked when socket buffer space
1590  *                             becomes available
1591  * @sk: socket whose state has changed
1592  *
1593  * Called when more output buffer space is available for this socket.
1594  * We try not to wake our writers until they can make "significant"
1595  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1596  * with a bunch of small requests.
1597  */
1598 static void xs_tcp_write_space(struct sock *sk)
1599 {
1600         read_lock_bh(&sk->sk_callback_lock);
1601
1602         /* from net/core/stream.c:sk_stream_write_space */
1603         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1604                 xs_write_space(sk);
1605
1606         read_unlock_bh(&sk->sk_callback_lock);
1607 }
1608
1609 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1610 {
1611         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1612         struct sock *sk = transport->inet;
1613
1614         if (transport->rcvsize) {
1615                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1616                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1617         }
1618         if (transport->sndsize) {
1619                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1620                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1621                 sk->sk_write_space(sk);
1622         }
1623 }
1624
1625 /**
1626  * xs_udp_set_buffer_size - set send and receive limits
1627  * @xprt: generic transport
1628  * @sndsize: requested size of send buffer, in bytes
1629  * @rcvsize: requested size of receive buffer, in bytes
1630  *
1631  * Set socket send and receive buffer size limits.
1632  */
1633 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1634 {
1635         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1636
1637         transport->sndsize = 0;
1638         if (sndsize)
1639                 transport->sndsize = sndsize + 1024;
1640         transport->rcvsize = 0;
1641         if (rcvsize)
1642                 transport->rcvsize = rcvsize + 1024;
1643
1644         xs_udp_do_set_buffer_size(xprt);
1645 }
1646
1647 /**
1648  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1649  * @task: task that timed out
1650  *
1651  * Adjust the congestion window after a retransmit timeout has occurred.
1652  */
1653 static void xs_udp_timer(struct rpc_task *task)
1654 {
1655         xprt_adjust_cwnd(task, -ETIMEDOUT);
1656 }
1657
1658 static unsigned short xs_get_random_port(void)
1659 {
1660         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1661         unsigned short rand = (unsigned short) net_random() % range;
1662         return rand + xprt_min_resvport;
1663 }
1664
1665 /**
1666  * xs_set_port - reset the port number in the remote endpoint address
1667  * @xprt: generic transport
1668  * @port: new port number
1669  *
1670  */
1671 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1672 {
1673         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1674
1675         rpc_set_port(xs_addr(xprt), port);
1676         xs_update_peer_port(xprt);
1677 }
1678
1679 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1680 {
1681         unsigned short port = transport->srcport;
1682
1683         if (port == 0 && transport->xprt.resvport)
1684                 port = xs_get_random_port();
1685         return port;
1686 }
1687
1688 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1689 {
1690         if (transport->srcport != 0)
1691                 transport->srcport = 0;
1692         if (!transport->xprt.resvport)
1693                 return 0;
1694         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1695                 return xprt_max_resvport;
1696         return --port;
1697 }
1698 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1699 {
1700         struct sockaddr_storage myaddr;
1701         int err, nloop = 0;
1702         unsigned short port = xs_get_srcport(transport);
1703         unsigned short last;
1704
1705         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1706         do {
1707                 rpc_set_port((struct sockaddr *)&myaddr, port);
1708                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1709                                 transport->xprt.addrlen);
1710                 if (port == 0)
1711                         break;
1712                 if (err == 0) {
1713                         transport->srcport = port;
1714                         break;
1715                 }
1716                 last = port;
1717                 port = xs_next_srcport(transport, port);
1718                 if (port > last)
1719                         nloop++;
1720         } while (err == -EADDRINUSE && nloop != 2);
1721
1722         if (myaddr.ss_family == AF_INET)
1723                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1724                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1725                                 port, err ? "failed" : "ok", err);
1726         else
1727                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1728                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1729                                 port, err ? "failed" : "ok", err);
1730         return err;
1731 }
1732
1733 /*
1734  * We don't support autobind on AF_LOCAL sockets
1735  */
1736 static void xs_local_rpcbind(struct rpc_task *task)
1737 {
1738         xprt_set_bound(task->tk_xprt);
1739 }
1740
1741 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1742 {
1743 }
1744
1745 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1746 static struct lock_class_key xs_key[2];
1747 static struct lock_class_key xs_slock_key[2];
1748
1749 static inline void xs_reclassify_socketu(struct socket *sock)
1750 {
1751         struct sock *sk = sock->sk;
1752
1753         BUG_ON(sock_owned_by_user(sk));
1754         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1755                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1756 }
1757
1758 static inline void xs_reclassify_socket4(struct socket *sock)
1759 {
1760         struct sock *sk = sock->sk;
1761
1762         BUG_ON(sock_owned_by_user(sk));
1763         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1764                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1765 }
1766
1767 static inline void xs_reclassify_socket6(struct socket *sock)
1768 {
1769         struct sock *sk = sock->sk;
1770
1771         BUG_ON(sock_owned_by_user(sk));
1772         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1773                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1774 }
1775
1776 static inline void xs_reclassify_socket(int family, struct socket *sock)
1777 {
1778         switch (family) {
1779         case AF_LOCAL:
1780                 xs_reclassify_socketu(sock);
1781                 break;
1782         case AF_INET:
1783                 xs_reclassify_socket4(sock);
1784                 break;
1785         case AF_INET6:
1786                 xs_reclassify_socket6(sock);
1787                 break;
1788         }
1789 }
1790 #else
1791 static inline void xs_reclassify_socketu(struct socket *sock)
1792 {
1793 }
1794
1795 static inline void xs_reclassify_socket4(struct socket *sock)
1796 {
1797 }
1798
1799 static inline void xs_reclassify_socket6(struct socket *sock)
1800 {
1801 }
1802
1803 static inline void xs_reclassify_socket(int family, struct socket *sock)
1804 {
1805 }
1806 #endif
1807
1808 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1809                 struct sock_xprt *transport, int family, int type, int protocol)
1810 {
1811         struct socket *sock;
1812         int err;
1813
1814         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1815         if (err < 0) {
1816                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1817                                 protocol, -err);
1818                 goto out;
1819         }
1820         xs_reclassify_socket(family, sock);
1821
1822         err = xs_bind(transport, sock);
1823         if (err) {
1824                 sock_release(sock);
1825                 goto out;
1826         }
1827
1828         return sock;
1829 out:
1830         return ERR_PTR(err);
1831 }
1832
1833 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1834                                       struct socket *sock)
1835 {
1836         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1837                                                                         xprt);
1838
1839         if (!transport->inet) {
1840                 struct sock *sk = sock->sk;
1841
1842                 write_lock_bh(&sk->sk_callback_lock);
1843
1844                 xs_save_old_callbacks(transport, sk);
1845
1846                 sk->sk_user_data = xprt;
1847                 sk->sk_data_ready = xs_local_data_ready;
1848                 sk->sk_write_space = xs_udp_write_space;
1849                 sk->sk_error_report = xs_error_report;
1850                 sk->sk_allocation = GFP_ATOMIC;
1851
1852                 xprt_clear_connected(xprt);
1853
1854                 /* Reset to new socket */
1855                 transport->sock = sock;
1856                 transport->inet = sk;
1857
1858                 write_unlock_bh(&sk->sk_callback_lock);
1859         }
1860
1861         /* Tell the socket layer to start connecting... */
1862         xprt->stat.connect_count++;
1863         xprt->stat.connect_start = jiffies;
1864         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1865 }
1866
1867 /**
1868  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1869  * @xprt: RPC transport to connect
1870  * @transport: socket transport to connect
1871  * @create_sock: function to create a socket of the correct type
1872  *
1873  * Invoked by a work queue tasklet.
1874  */
1875 static void xs_local_setup_socket(struct work_struct *work)
1876 {
1877         struct sock_xprt *transport =
1878                 container_of(work, struct sock_xprt, connect_worker.work);
1879         struct rpc_xprt *xprt = &transport->xprt;
1880         struct socket *sock;
1881         int status = -EIO;
1882
1883         if (xprt->shutdown)
1884                 goto out;
1885
1886         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1887         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1888                                         SOCK_STREAM, 0, &sock, 1);
1889         if (status < 0) {
1890                 dprintk("RPC:       can't create AF_LOCAL "
1891                         "transport socket (%d).\n", -status);
1892                 goto out;
1893         }
1894         xs_reclassify_socketu(sock);
1895
1896         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1897                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1898
1899         status = xs_local_finish_connecting(xprt, sock);
1900         switch (status) {
1901         case 0:
1902                 dprintk("RPC:       xprt %p connected to %s\n",
1903                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1904                 xprt_set_connected(xprt);
1905                 break;
1906         case -ENOENT:
1907                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1908                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1909                 break;
1910         default:
1911                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1912                                 __func__, -status,
1913                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1914         }
1915
1916 out:
1917         xprt_clear_connecting(xprt);
1918         xprt_wake_pending_tasks(xprt, status);
1919 }
1920
1921 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1922 {
1923         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1924
1925         if (!transport->inet) {
1926                 struct sock *sk = sock->sk;
1927
1928                 write_lock_bh(&sk->sk_callback_lock);
1929
1930                 xs_save_old_callbacks(transport, sk);
1931
1932                 sk->sk_user_data = xprt;
1933                 sk->sk_data_ready = xs_udp_data_ready;
1934                 sk->sk_write_space = xs_udp_write_space;
1935                 sk->sk_error_report = xs_error_report;
1936                 sk->sk_no_check = UDP_CSUM_NORCV;
1937                 sk->sk_allocation = GFP_ATOMIC;
1938
1939                 xprt_set_connected(xprt);
1940
1941                 /* Reset to new socket */
1942                 transport->sock = sock;
1943                 transport->inet = sk;
1944
1945                 write_unlock_bh(&sk->sk_callback_lock);
1946         }
1947         xs_udp_do_set_buffer_size(xprt);
1948 }
1949
1950 static void xs_udp_setup_socket(struct work_struct *work)
1951 {
1952         struct sock_xprt *transport =
1953                 container_of(work, struct sock_xprt, connect_worker.work);
1954         struct rpc_xprt *xprt = &transport->xprt;
1955         struct socket *sock = transport->sock;
1956         int status = -EIO;
1957
1958         if (xprt->shutdown)
1959                 goto out;
1960
1961         /* Start by resetting any existing state */
1962         xs_reset_transport(transport);
1963         sock = xs_create_sock(xprt, transport,
1964                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
1965         if (IS_ERR(sock))
1966                 goto out;
1967
1968         dprintk("RPC:       worker connecting xprt %p via %s to "
1969                                 "%s (port %s)\n", xprt,
1970                         xprt->address_strings[RPC_DISPLAY_PROTO],
1971                         xprt->address_strings[RPC_DISPLAY_ADDR],
1972                         xprt->address_strings[RPC_DISPLAY_PORT]);
1973
1974         xs_udp_finish_connecting(xprt, sock);
1975         status = 0;
1976 out:
1977         xprt_clear_connecting(xprt);
1978         xprt_wake_pending_tasks(xprt, status);
1979 }
1980
1981 /*
1982  * We need to preserve the port number so the reply cache on the server can
1983  * find our cached RPC replies when we get around to reconnecting.
1984  */
1985 static void xs_abort_connection(struct sock_xprt *transport)
1986 {
1987         int result;
1988         struct sockaddr any;
1989
1990         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
1991
1992         /*
1993          * Disconnect the transport socket by doing a connect operation
1994          * with AF_UNSPEC.  This should return immediately...
1995          */
1996         memset(&any, 0, sizeof(any));
1997         any.sa_family = AF_UNSPEC;
1998         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1999         if (!result)
2000                 xs_sock_mark_closed(&transport->xprt);
2001         else
2002                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
2003                                 result);
2004 }
2005
2006 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2007 {
2008         unsigned int state = transport->inet->sk_state;
2009
2010         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2011                 /* we don't need to abort the connection if the socket
2012                  * hasn't undergone a shutdown
2013                  */
2014                 if (transport->inet->sk_shutdown == 0)
2015                         return;
2016                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2017                                 __func__, transport->inet->sk_shutdown);
2018         }
2019         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2020                 /* we don't need to abort the connection if the socket
2021                  * hasn't undergone a shutdown
2022                  */
2023                 if (transport->inet->sk_shutdown == 0)
2024                         return;
2025                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2026                                 "sk_shutdown set to %d\n",
2027                                 __func__, transport->inet->sk_shutdown);
2028         }
2029         xs_abort_connection(transport);
2030 }
2031
2032 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2033 {
2034         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2035         int ret = -ENOTCONN;
2036
2037         if (!transport->inet) {
2038                 struct sock *sk = sock->sk;
2039
2040                 write_lock_bh(&sk->sk_callback_lock);
2041
2042                 xs_save_old_callbacks(transport, sk);
2043
2044                 sk->sk_user_data = xprt;
2045                 sk->sk_data_ready = xs_tcp_data_ready;
2046                 sk->sk_state_change = xs_tcp_state_change;
2047                 sk->sk_write_space = xs_tcp_write_space;
2048                 sk->sk_error_report = xs_error_report;
2049                 sk->sk_allocation = GFP_ATOMIC;
2050
2051                 /* socket options */
2052                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2053                 sock_reset_flag(sk, SOCK_LINGER);
2054                 tcp_sk(sk)->linger2 = 0;
2055                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2056
2057                 xprt_clear_connected(xprt);
2058
2059                 /* Reset to new socket */
2060                 transport->sock = sock;
2061                 transport->inet = sk;
2062
2063                 write_unlock_bh(&sk->sk_callback_lock);
2064         }
2065
2066         if (!xprt_bound(xprt))
2067                 goto out;
2068
2069         /* Tell the socket layer to start connecting... */
2070         xprt->stat.connect_count++;
2071         xprt->stat.connect_start = jiffies;
2072         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2073         switch (ret) {
2074         case 0:
2075         case -EINPROGRESS:
2076                 /* SYN_SENT! */
2077                 xprt->connect_cookie++;
2078                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2079                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2080         }
2081 out:
2082         return ret;
2083 }
2084
2085 /**
2086  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2087  * @xprt: RPC transport to connect
2088  * @transport: socket transport to connect
2089  * @create_sock: function to create a socket of the correct type
2090  *
2091  * Invoked by a work queue tasklet.
2092  */
2093 static void xs_tcp_setup_socket(struct work_struct *work)
2094 {
2095         struct sock_xprt *transport =
2096                 container_of(work, struct sock_xprt, connect_worker.work);
2097         struct socket *sock = transport->sock;
2098         struct rpc_xprt *xprt = &transport->xprt;
2099         int status = -EIO;
2100
2101         if (xprt->shutdown)
2102                 goto out;
2103
2104         if (!sock) {
2105                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2106                 sock = xs_create_sock(xprt, transport,
2107                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2108                 if (IS_ERR(sock)) {
2109                         status = PTR_ERR(sock);
2110                         goto out;
2111                 }
2112         } else {
2113                 int abort_and_exit;
2114
2115                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2116                                 &xprt->state);
2117                 /* "close" the socket, preserving the local port */
2118                 xs_tcp_reuse_connection(transport);
2119
2120                 if (abort_and_exit)
2121                         goto out_eagain;
2122         }
2123
2124         dprintk("RPC:       worker connecting xprt %p via %s to "
2125                                 "%s (port %s)\n", xprt,
2126                         xprt->address_strings[RPC_DISPLAY_PROTO],
2127                         xprt->address_strings[RPC_DISPLAY_ADDR],
2128                         xprt->address_strings[RPC_DISPLAY_PORT]);
2129
2130         status = xs_tcp_finish_connecting(xprt, sock);
2131         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2132                         xprt, -status, xprt_connected(xprt),
2133                         sock->sk->sk_state);
2134         switch (status) {
2135         default:
2136                 printk("%s: connect returned unhandled error %d\n",
2137                         __func__, status);
2138         case -EADDRNOTAVAIL:
2139                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2140                  * and retry
2141                  */
2142                 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
2143                 xprt_force_disconnect(xprt);
2144                 break;
2145         case -ECONNREFUSED:
2146         case -ECONNRESET:
2147         case -ENETUNREACH:
2148                 /* retry with existing socket, after a delay */
2149         case 0:
2150         case -EINPROGRESS:
2151         case -EALREADY:
2152                 xprt_clear_connecting(xprt);
2153                 return;
2154         case -EINVAL:
2155                 /* Happens, for instance, if the user specified a link
2156                  * local IPv6 address without a scope-id.
2157                  */
2158                 goto out;
2159         }
2160 out_eagain:
2161         status = -EAGAIN;
2162 out:
2163         xprt_clear_connecting(xprt);
2164         xprt_wake_pending_tasks(xprt, status);
2165 }
2166
2167 /**
2168  * xs_connect - connect a socket to a remote endpoint
2169  * @task: address of RPC task that manages state of connect request
2170  *
2171  * TCP: If the remote end dropped the connection, delay reconnecting.
2172  *
2173  * UDP socket connects are synchronous, but we use a work queue anyway
2174  * to guarantee that even unprivileged user processes can set up a
2175  * socket on a privileged port.
2176  *
2177  * If a UDP socket connect fails, the delay behavior here prevents
2178  * retry floods (hard mounts).
2179  */
2180 static void xs_connect(struct rpc_task *task)
2181 {
2182         struct rpc_xprt *xprt = task->tk_xprt;
2183         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2184
2185         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2186                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2187                                 "seconds\n",
2188                                 xprt, xprt->reestablish_timeout / HZ);
2189                 queue_delayed_work(rpciod_workqueue,
2190                                    &transport->connect_worker,
2191                                    xprt->reestablish_timeout);
2192                 xprt->reestablish_timeout <<= 1;
2193                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2194                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2195                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2196                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2197         } else {
2198                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2199                 queue_delayed_work(rpciod_workqueue,
2200                                    &transport->connect_worker, 0);
2201         }
2202 }
2203
2204 /**
2205  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2206  * @xprt: rpc_xprt struct containing statistics
2207  * @seq: output file
2208  *
2209  */
2210 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2211 {
2212         long idle_time = 0;
2213
2214         if (xprt_connected(xprt))
2215                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2216
2217         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2218                         "%llu %llu\n",
2219                         xprt->stat.bind_count,
2220                         xprt->stat.connect_count,
2221                         xprt->stat.connect_time,
2222                         idle_time,
2223                         xprt->stat.sends,
2224                         xprt->stat.recvs,
2225                         xprt->stat.bad_xids,
2226                         xprt->stat.req_u,
2227                         xprt->stat.bklog_u);
2228 }
2229
2230 /**
2231  * xs_udp_print_stats - display UDP socket-specifc stats
2232  * @xprt: rpc_xprt struct containing statistics
2233  * @seq: output file
2234  *
2235  */
2236 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2237 {
2238         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2239
2240         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2241                         transport->srcport,
2242                         xprt->stat.bind_count,
2243                         xprt->stat.sends,
2244                         xprt->stat.recvs,
2245                         xprt->stat.bad_xids,
2246                         xprt->stat.req_u,
2247                         xprt->stat.bklog_u);
2248 }
2249
2250 /**
2251  * xs_tcp_print_stats - display TCP socket-specifc stats
2252  * @xprt: rpc_xprt struct containing statistics
2253  * @seq: output file
2254  *
2255  */
2256 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2257 {
2258         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2259         long idle_time = 0;
2260
2261         if (xprt_connected(xprt))
2262                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2263
2264         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2265                         transport->srcport,
2266                         xprt->stat.bind_count,
2267                         xprt->stat.connect_count,
2268                         xprt->stat.connect_time,
2269                         idle_time,
2270                         xprt->stat.sends,
2271                         xprt->stat.recvs,
2272                         xprt->stat.bad_xids,
2273                         xprt->stat.req_u,
2274                         xprt->stat.bklog_u);
2275 }
2276
2277 /*
2278  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2279  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2280  * to use the server side send routines.
2281  */
2282 static void *bc_malloc(struct rpc_task *task, size_t size)
2283 {
2284         struct page *page;
2285         struct rpc_buffer *buf;
2286
2287         BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2288         page = alloc_page(GFP_KERNEL);
2289
2290         if (!page)
2291                 return NULL;
2292
2293         buf = page_address(page);
2294         buf->len = PAGE_SIZE;
2295
2296         return buf->data;
2297 }
2298
2299 /*
2300  * Free the space allocated in the bc_alloc routine
2301  */
2302 static void bc_free(void *buffer)
2303 {
2304         struct rpc_buffer *buf;
2305
2306         if (!buffer)
2307                 return;
2308
2309         buf = container_of(buffer, struct rpc_buffer, data);
2310         free_page((unsigned long)buf);
2311 }
2312
2313 /*
2314  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2315  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2316  */
2317 static int bc_sendto(struct rpc_rqst *req)
2318 {
2319         int len;
2320         struct xdr_buf *xbufp = &req->rq_snd_buf;
2321         struct rpc_xprt *xprt = req->rq_xprt;
2322         struct sock_xprt *transport =
2323                                 container_of(xprt, struct sock_xprt, xprt);
2324         struct socket *sock = transport->sock;
2325         unsigned long headoff;
2326         unsigned long tailoff;
2327
2328         xs_encode_stream_record_marker(xbufp);
2329
2330         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2331         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2332         len = svc_send_common(sock, xbufp,
2333                               virt_to_page(xbufp->head[0].iov_base), headoff,
2334                               xbufp->tail[0].iov_base, tailoff);
2335
2336         if (len != xbufp->len) {
2337                 printk(KERN_NOTICE "Error sending entire callback!\n");
2338                 len = -EAGAIN;
2339         }
2340
2341         return len;
2342 }
2343
2344 /*
2345  * The send routine. Borrows from svc_send
2346  */
2347 static int bc_send_request(struct rpc_task *task)
2348 {
2349         struct rpc_rqst *req = task->tk_rqstp;
2350         struct svc_xprt *xprt;
2351         struct svc_sock         *svsk;
2352         u32                     len;
2353
2354         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2355         /*
2356          * Get the server socket associated with this callback xprt
2357          */
2358         xprt = req->rq_xprt->bc_xprt;
2359         svsk = container_of(xprt, struct svc_sock, sk_xprt);
2360
2361         /*
2362          * Grab the mutex to serialize data as the connection is shared
2363          * with the fore channel
2364          */
2365         if (!mutex_trylock(&xprt->xpt_mutex)) {
2366                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2367                 if (!mutex_trylock(&xprt->xpt_mutex))
2368                         return -EAGAIN;
2369                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2370         }
2371         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2372                 len = -ENOTCONN;
2373         else
2374                 len = bc_sendto(req);
2375         mutex_unlock(&xprt->xpt_mutex);
2376
2377         if (len > 0)
2378                 len = 0;
2379
2380         return len;
2381 }
2382
2383 /*
2384  * The close routine. Since this is client initiated, we do nothing
2385  */
2386
2387 static void bc_close(struct rpc_xprt *xprt)
2388 {
2389 }
2390
2391 /*
2392  * The xprt destroy routine. Again, because this connection is client
2393  * initiated, we do nothing
2394  */
2395
2396 static void bc_destroy(struct rpc_xprt *xprt)
2397 {
2398 }
2399
2400 static struct rpc_xprt_ops xs_local_ops = {
2401         .reserve_xprt           = xprt_reserve_xprt,
2402         .release_xprt           = xs_tcp_release_xprt,
2403         .rpcbind                = xs_local_rpcbind,
2404         .set_port               = xs_local_set_port,
2405         .connect                = xs_connect,
2406         .buf_alloc              = rpc_malloc,
2407         .buf_free               = rpc_free,
2408         .send_request           = xs_local_send_request,
2409         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2410         .close                  = xs_close,
2411         .destroy                = xs_destroy,
2412         .print_stats            = xs_local_print_stats,
2413 };
2414
2415 static struct rpc_xprt_ops xs_udp_ops = {
2416         .set_buffer_size        = xs_udp_set_buffer_size,
2417         .reserve_xprt           = xprt_reserve_xprt_cong,
2418         .release_xprt           = xprt_release_xprt_cong,
2419         .rpcbind                = rpcb_getport_async,
2420         .set_port               = xs_set_port,
2421         .connect                = xs_connect,
2422         .buf_alloc              = rpc_malloc,
2423         .buf_free               = rpc_free,
2424         .send_request           = xs_udp_send_request,
2425         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2426         .timer                  = xs_udp_timer,
2427         .release_request        = xprt_release_rqst_cong,
2428         .close                  = xs_close,
2429         .destroy                = xs_destroy,
2430         .print_stats            = xs_udp_print_stats,
2431 };
2432
2433 static struct rpc_xprt_ops xs_tcp_ops = {
2434         .reserve_xprt           = xprt_reserve_xprt,
2435         .release_xprt           = xs_tcp_release_xprt,
2436         .rpcbind                = rpcb_getport_async,
2437         .set_port               = xs_set_port,
2438         .connect                = xs_connect,
2439         .buf_alloc              = rpc_malloc,
2440         .buf_free               = rpc_free,
2441         .send_request           = xs_tcp_send_request,
2442         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2443         .close                  = xs_tcp_close,
2444         .destroy                = xs_destroy,
2445         .print_stats            = xs_tcp_print_stats,
2446 };
2447
2448 /*
2449  * The rpc_xprt_ops for the server backchannel
2450  */
2451
2452 static struct rpc_xprt_ops bc_tcp_ops = {
2453         .reserve_xprt           = xprt_reserve_xprt,
2454         .release_xprt           = xprt_release_xprt,
2455         .buf_alloc              = bc_malloc,
2456         .buf_free               = bc_free,
2457         .send_request           = bc_send_request,
2458         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2459         .close                  = bc_close,
2460         .destroy                = bc_destroy,
2461         .print_stats            = xs_tcp_print_stats,
2462 };
2463
2464 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2465 {
2466         static const struct sockaddr_in sin = {
2467                 .sin_family             = AF_INET,
2468                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2469         };
2470         static const struct sockaddr_in6 sin6 = {
2471                 .sin6_family            = AF_INET6,
2472                 .sin6_addr              = IN6ADDR_ANY_INIT,
2473         };
2474
2475         switch (family) {
2476         case AF_LOCAL:
2477                 break;
2478         case AF_INET:
2479                 memcpy(sap, &sin, sizeof(sin));
2480                 break;
2481         case AF_INET6:
2482                 memcpy(sap, &sin6, sizeof(sin6));
2483                 break;
2484         default:
2485                 dprintk("RPC:       %s: Bad address family\n", __func__);
2486                 return -EAFNOSUPPORT;
2487         }
2488         return 0;
2489 }
2490
2491 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2492                                       unsigned int slot_table_size)
2493 {
2494         struct rpc_xprt *xprt;
2495         struct sock_xprt *new;
2496
2497         if (args->addrlen > sizeof(xprt->addr)) {
2498                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2499                 return ERR_PTR(-EBADF);
2500         }
2501
2502         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size);
2503         if (xprt == NULL) {
2504                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2505                                 "rpc_xprt\n");
2506                 return ERR_PTR(-ENOMEM);
2507         }
2508
2509         new = container_of(xprt, struct sock_xprt, xprt);
2510         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2511         xprt->addrlen = args->addrlen;
2512         if (args->srcaddr)
2513                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2514         else {
2515                 int err;
2516                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2517                                         (struct sockaddr *)&new->srcaddr);
2518                 if (err != 0)
2519                         return ERR_PTR(err);
2520         }
2521
2522         return xprt;
2523 }
2524
2525 static const struct rpc_timeout xs_local_default_timeout = {
2526         .to_initval = 10 * HZ,
2527         .to_maxval = 10 * HZ,
2528         .to_retries = 2,
2529 };
2530
2531 /**
2532  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2533  * @args: rpc transport creation arguments
2534  *
2535  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2536  */
2537 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2538 {
2539         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2540         struct sock_xprt *transport;
2541         struct rpc_xprt *xprt;
2542         struct rpc_xprt *ret;
2543
2544         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2545         if (IS_ERR(xprt))
2546                 return xprt;
2547         transport = container_of(xprt, struct sock_xprt, xprt);
2548
2549         xprt->prot = 0;
2550         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2551         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2552
2553         xprt->bind_timeout = XS_BIND_TO;
2554         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2555         xprt->idle_timeout = XS_IDLE_DISC_TO;
2556
2557         xprt->ops = &xs_local_ops;
2558         xprt->timeout = &xs_local_default_timeout;
2559
2560         switch (sun->sun_family) {
2561         case AF_LOCAL:
2562                 if (sun->sun_path[0] != '/') {
2563                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2564                                         sun->sun_path);
2565                         ret = ERR_PTR(-EINVAL);
2566                         goto out_err;
2567                 }
2568                 xprt_set_bound(xprt);
2569                 INIT_DELAYED_WORK(&transport->connect_worker,
2570                                         xs_local_setup_socket);
2571                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2572                 break;
2573         default:
2574                 ret = ERR_PTR(-EAFNOSUPPORT);
2575                 goto out_err;
2576         }
2577
2578         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2579                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2580
2581         if (try_module_get(THIS_MODULE))
2582                 return xprt;
2583         ret = ERR_PTR(-EINVAL);
2584 out_err:
2585         xprt_free(xprt);
2586         return ret;
2587 }
2588
2589 static const struct rpc_timeout xs_udp_default_timeout = {
2590         .to_initval = 5 * HZ,
2591         .to_maxval = 30 * HZ,
2592         .to_increment = 5 * HZ,
2593         .to_retries = 5,
2594 };
2595
2596 /**
2597  * xs_setup_udp - Set up transport to use a UDP socket
2598  * @args: rpc transport creation arguments
2599  *
2600  */
2601 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2602 {
2603         struct sockaddr *addr = args->dstaddr;
2604         struct rpc_xprt *xprt;
2605         struct sock_xprt *transport;
2606         struct rpc_xprt *ret;
2607
2608         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2609         if (IS_ERR(xprt))
2610                 return xprt;
2611         transport = container_of(xprt, struct sock_xprt, xprt);
2612
2613         xprt->prot = IPPROTO_UDP;
2614         xprt->tsh_size = 0;
2615         /* XXX: header size can vary due to auth type, IPv6, etc. */
2616         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2617
2618         xprt->bind_timeout = XS_BIND_TO;
2619         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2620         xprt->idle_timeout = XS_IDLE_DISC_TO;
2621
2622         xprt->ops = &xs_udp_ops;
2623
2624         xprt->timeout = &xs_udp_default_timeout;
2625
2626         switch (addr->sa_family) {
2627         case AF_INET:
2628                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2629                         xprt_set_bound(xprt);
2630
2631                 INIT_DELAYED_WORK(&transport->connect_worker,
2632                                         xs_udp_setup_socket);
2633                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2634                 break;
2635         case AF_INET6:
2636                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2637                         xprt_set_bound(xprt);
2638
2639                 INIT_DELAYED_WORK(&transport->connect_worker,
2640                                         xs_udp_setup_socket);
2641                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2642                 break;
2643         default:
2644                 ret = ERR_PTR(-EAFNOSUPPORT);
2645                 goto out_err;
2646         }
2647
2648         if (xprt_bound(xprt))
2649                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2650                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2651                                 xprt->address_strings[RPC_DISPLAY_PORT],
2652                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2653         else
2654                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2655                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2656                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2657
2658         if (try_module_get(THIS_MODULE))
2659                 return xprt;
2660         ret = ERR_PTR(-EINVAL);
2661 out_err:
2662         xprt_free(xprt);
2663         return ret;
2664 }
2665
2666 static const struct rpc_timeout xs_tcp_default_timeout = {
2667         .to_initval = 60 * HZ,
2668         .to_maxval = 60 * HZ,
2669         .to_retries = 2,
2670 };
2671
2672 /**
2673  * xs_setup_tcp - Set up transport to use a TCP socket
2674  * @args: rpc transport creation arguments
2675  *
2676  */
2677 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2678 {
2679         struct sockaddr *addr = args->dstaddr;
2680         struct rpc_xprt *xprt;
2681         struct sock_xprt *transport;
2682         struct rpc_xprt *ret;
2683
2684         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2685         if (IS_ERR(xprt))
2686                 return xprt;
2687         transport = container_of(xprt, struct sock_xprt, xprt);
2688
2689         xprt->prot = IPPROTO_TCP;
2690         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2691         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2692
2693         xprt->bind_timeout = XS_BIND_TO;
2694         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2695         xprt->idle_timeout = XS_IDLE_DISC_TO;
2696
2697         xprt->ops = &xs_tcp_ops;
2698         xprt->timeout = &xs_tcp_default_timeout;
2699
2700         switch (addr->sa_family) {
2701         case AF_INET:
2702                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2703                         xprt_set_bound(xprt);
2704
2705                 INIT_DELAYED_WORK(&transport->connect_worker,
2706                                         xs_tcp_setup_socket);
2707                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2708                 break;
2709         case AF_INET6:
2710                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2711                         xprt_set_bound(xprt);
2712
2713                 INIT_DELAYED_WORK(&transport->connect_worker,
2714                                         xs_tcp_setup_socket);
2715                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2716                 break;
2717         default:
2718                 ret = ERR_PTR(-EAFNOSUPPORT);
2719                 goto out_err;
2720         }
2721
2722         if (xprt_bound(xprt))
2723                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2724                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2725                                 xprt->address_strings[RPC_DISPLAY_PORT],
2726                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2727         else
2728                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2729                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2730                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2731
2732
2733         if (try_module_get(THIS_MODULE))
2734                 return xprt;
2735         ret = ERR_PTR(-EINVAL);
2736 out_err:
2737         xprt_free(xprt);
2738         return ret;
2739 }
2740
2741 /**
2742  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2743  * @args: rpc transport creation arguments
2744  *
2745  */
2746 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2747 {
2748         struct sockaddr *addr = args->dstaddr;
2749         struct rpc_xprt *xprt;
2750         struct sock_xprt *transport;
2751         struct svc_sock *bc_sock;
2752         struct rpc_xprt *ret;
2753
2754         if (args->bc_xprt->xpt_bc_xprt) {
2755                 /*
2756                  * This server connection already has a backchannel
2757                  * export; we can't create a new one, as we wouldn't be
2758                  * able to match replies based on xid any more.  So,
2759                  * reuse the already-existing one:
2760                  */
2761                  return args->bc_xprt->xpt_bc_xprt;
2762         }
2763         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2764         if (IS_ERR(xprt))
2765                 return xprt;
2766         transport = container_of(xprt, struct sock_xprt, xprt);
2767
2768         xprt->prot = IPPROTO_TCP;
2769         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2770         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2771         xprt->timeout = &xs_tcp_default_timeout;
2772
2773         /* backchannel */
2774         xprt_set_bound(xprt);
2775         xprt->bind_timeout = 0;
2776         xprt->reestablish_timeout = 0;
2777         xprt->idle_timeout = 0;
2778
2779         xprt->ops = &bc_tcp_ops;
2780
2781         switch (addr->sa_family) {
2782         case AF_INET:
2783                 xs_format_peer_addresses(xprt, "tcp",
2784                                          RPCBIND_NETID_TCP);
2785                 break;
2786         case AF_INET6:
2787                 xs_format_peer_addresses(xprt, "tcp",
2788                                    RPCBIND_NETID_TCP6);
2789                 break;
2790         default:
2791                 ret = ERR_PTR(-EAFNOSUPPORT);
2792                 goto out_err;
2793         }
2794
2795         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2796                         xprt->address_strings[RPC_DISPLAY_ADDR],
2797                         xprt->address_strings[RPC_DISPLAY_PORT],
2798                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2799
2800         /*
2801          * Once we've associated a backchannel xprt with a connection,
2802          * we want to keep it around as long as long as the connection
2803          * lasts, in case we need to start using it for a backchannel
2804          * again; this reference won't be dropped until bc_xprt is
2805          * destroyed.
2806          */
2807         xprt_get(xprt);
2808         args->bc_xprt->xpt_bc_xprt = xprt;
2809         xprt->bc_xprt = args->bc_xprt;
2810         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2811         transport->sock = bc_sock->sk_sock;
2812         transport->inet = bc_sock->sk_sk;
2813
2814         /*
2815          * Since we don't want connections for the backchannel, we set
2816          * the xprt status to connected
2817          */
2818         xprt_set_connected(xprt);
2819
2820
2821         if (try_module_get(THIS_MODULE))
2822                 return xprt;
2823         xprt_put(xprt);
2824         ret = ERR_PTR(-EINVAL);
2825 out_err:
2826         xprt_free(xprt);
2827         return ret;
2828 }
2829
2830 static struct xprt_class        xs_local_transport = {
2831         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2832         .name           = "named UNIX socket",
2833         .owner          = THIS_MODULE,
2834         .ident          = XPRT_TRANSPORT_LOCAL,
2835         .setup          = xs_setup_local,
2836 };
2837
2838 static struct xprt_class        xs_udp_transport = {
2839         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2840         .name           = "udp",
2841         .owner          = THIS_MODULE,
2842         .ident          = XPRT_TRANSPORT_UDP,
2843         .setup          = xs_setup_udp,
2844 };
2845
2846 static struct xprt_class        xs_tcp_transport = {
2847         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2848         .name           = "tcp",
2849         .owner          = THIS_MODULE,
2850         .ident          = XPRT_TRANSPORT_TCP,
2851         .setup          = xs_setup_tcp,
2852 };
2853
2854 static struct xprt_class        xs_bc_tcp_transport = {
2855         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2856         .name           = "tcp NFSv4.1 backchannel",
2857         .owner          = THIS_MODULE,
2858         .ident          = XPRT_TRANSPORT_BC_TCP,
2859         .setup          = xs_setup_bc_tcp,
2860 };
2861
2862 /**
2863  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2864  *
2865  */
2866 int init_socket_xprt(void)
2867 {
2868 #ifdef RPC_DEBUG
2869         if (!sunrpc_table_header)
2870                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2871 #endif
2872
2873         xprt_register_transport(&xs_local_transport);
2874         xprt_register_transport(&xs_udp_transport);
2875         xprt_register_transport(&xs_tcp_transport);
2876         xprt_register_transport(&xs_bc_tcp_transport);
2877
2878         return 0;
2879 }
2880
2881 /**
2882  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2883  *
2884  */
2885 void cleanup_socket_xprt(void)
2886 {
2887 #ifdef RPC_DEBUG
2888         if (sunrpc_table_header) {
2889                 unregister_sysctl_table(sunrpc_table_header);
2890                 sunrpc_table_header = NULL;
2891         }
2892 #endif
2893
2894         xprt_unregister_transport(&xs_local_transport);
2895         xprt_unregister_transport(&xs_udp_transport);
2896         xprt_unregister_transport(&xs_tcp_transport);
2897         xprt_unregister_transport(&xs_bc_tcp_transport);
2898 }
2899
2900 static int param_set_uint_minmax(const char *val,
2901                 const struct kernel_param *kp,
2902                 unsigned int min, unsigned int max)
2903 {
2904         unsigned long num;
2905         int ret;
2906
2907         if (!val)
2908                 return -EINVAL;
2909         ret = strict_strtoul(val, 0, &num);
2910         if (ret == -EINVAL || num < min || num > max)
2911                 return -EINVAL;
2912         *((unsigned int *)kp->arg) = num;
2913         return 0;
2914 }
2915
2916 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2917 {
2918         return param_set_uint_minmax(val, kp,
2919                         RPC_MIN_RESVPORT,
2920                         RPC_MAX_RESVPORT);
2921 }
2922
2923 static struct kernel_param_ops param_ops_portnr = {
2924         .set = param_set_portnr,
2925         .get = param_get_uint,
2926 };
2927
2928 #define param_check_portnr(name, p) \
2929         __param_check(name, p, unsigned int);
2930
2931 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2932 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2933
2934 static int param_set_slot_table_size(const char *val,
2935                                      const struct kernel_param *kp)
2936 {
2937         return param_set_uint_minmax(val, kp,
2938                         RPC_MIN_SLOT_TABLE,
2939                         RPC_MAX_SLOT_TABLE);
2940 }
2941
2942 static struct kernel_param_ops param_ops_slot_table_size = {
2943         .set = param_set_slot_table_size,
2944         .get = param_get_uint,
2945 };
2946
2947 #define param_check_slot_table_size(name, p) \
2948         __param_check(name, p, unsigned int);
2949
2950 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2951                    slot_table_size, 0644);
2952 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2953                    slot_table_size, 0644);
2954