net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #define XS_TCP_LINGER_TO        (15U * HZ)
  67 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  68
  69 /*
  70  * We can register our own files under /proc/sys/sunrpc by
  71  * calling register_sysctl_table() again.  The files in that
  72  * directory become the union of all files registered there.
  73  *
  74  * We simply need to make sure that we don't collide with
  75  * someone else's file names!
  76  */
  77
  78 #ifdef RPC_DEBUG
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 181
 182 /*
 183  * TCP idle timeout; client drops the transport socket if it is idle
 184  * for this long.  Note that we also timeout UDP sockets to prevent
 185  * holding port numbers when there is no RPC traffic.
 186  */
 187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 188
 189 #ifdef RPC_DEBUG
 190 # undef  RPC_DEBUG_DATA
 191 # define RPCDBG_FACILITY        RPCDBG_TRANS
 192 #endif
 193
 194 #ifdef RPC_DEBUG_DATA
 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196 {
 197         u8 *buf = (u8 *) packet;
 198         int j;
 199
 200         dprintk("RPC:       %s\n", msg);
 201         for (j = 0; j < count && j < 128; j += 4) {
 202                 if (!(j & 31)) {
 203                         if (j)
 204                                 dprintk("\n");
 205                         dprintk("0x%04x ", j);
 206                 }
 207                 dprintk("%02x%02x%02x%02x ",
 208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209         }
 210         dprintk("\n");
 211 }
 212 #else
 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214 {
 215         /* NOP */
 216 }
 217 #endif
 218
 219 struct sock_xprt {
 220         struct rpc_xprt         xprt;
 221
 222         /*
 223          * Network layer
 224          */
 225         struct socket *         sock;
 226         struct sock *           inet;
 227
 228         /*
 229          * State of TCP reply receive
 230          */
 231         __be32                  tcp_fraghdr,
 232                                 tcp_xid,
 233                                 tcp_calldir;
 234
 235         u32                     tcp_offset,
 236                                 tcp_reclen;
 237
 238         unsigned long           tcp_copied,
 239                                 tcp_flags;
 240
 241         /*
 242          * Connection of transports
 243          */
 244         struct delayed_work     connect_worker;
 245         struct sockaddr_storage srcaddr;
 246         unsigned short          srcport;
 247
 248         /*
 249          * UDP socket buffer size parameters
 250          */
 251         size_t                  rcvsize,
 252                                 sndsize;
 253
 254         /*
 255          * Saved socket callback addresses
 256          */
 257         void                    (*old_data_ready)(struct sock *, int);
 258         void                    (*old_state_change)(struct sock *);
 259         void                    (*old_write_space)(struct sock *);
 260 };
 261
 262 /*
 263  * TCP receive state flags
 264  */
 265 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 266 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 267 #define TCP_RCV_COPY_XID        (1UL << 2)
 268 #define TCP_RCV_COPY_DATA       (1UL << 3)
 269 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 270 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 271
 272 /*
 273  * TCP RPC flags
 274  */
 275 #define TCP_RPC_REPLY           (1UL << 6)
 276
 277 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 278 {
 279         return (struct sockaddr *) &xprt->addr;
 280 }
 281
 282 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 283 {
 284         return (struct sockaddr_un *) &xprt->addr;
 285 }
 286
 287 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 288 {
 289         return (struct sockaddr_in *) &xprt->addr;
 290 }
 291
 292 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 293 {
 294         return (struct sockaddr_in6 *) &xprt->addr;
 295 }
 296
 297 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 298 {
 299         struct sockaddr *sap = xs_addr(xprt);
 300         struct sockaddr_in6 *sin6;
 301         struct sockaddr_in *sin;
 302         struct sockaddr_un *sun;
 303         char buf[128];
 304
 305         switch (sap->sa_family) {
 306         case AF_LOCAL:
 307                 sun = xs_addr_un(xprt);
 308                 strlcpy(buf, sun->sun_path, sizeof(buf));
 309                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 310                                                 kstrdup(buf, GFP_KERNEL);
 311                 break;
 312         case AF_INET:
 313                 (void)rpc_ntop(sap, buf, sizeof(buf));
 314                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 315                                                 kstrdup(buf, GFP_KERNEL);
 316                 sin = xs_addr_in(xprt);
 317                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 318                 break;
 319         case AF_INET6:
 320                 (void)rpc_ntop(sap, buf, sizeof(buf));
 321                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 322                                                 kstrdup(buf, GFP_KERNEL);
 323                 sin6 = xs_addr_in6(xprt);
 324                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 325                 break;
 326         default:
 327                 BUG();
 328         }
 329
 330         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 331 }
 332
 333 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 334 {
 335         struct sockaddr *sap = xs_addr(xprt);
 336         char buf[128];
 337
 338         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 339         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 340
 341         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 342         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 343 }
 344
 345 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 346                                      const char *protocol,
 347                                      const char *netid)
 348 {
 349         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 350         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 351         xs_format_common_peer_addresses(xprt);
 352         xs_format_common_peer_ports(xprt);
 353 }
 354
 355 static void xs_update_peer_port(struct rpc_xprt *xprt)
 356 {
 357         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 358         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 359
 360         xs_format_common_peer_ports(xprt);
 361 }
 362
 363 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 364 {
 365         unsigned int i;
 366
 367         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 368                 switch (i) {
 369                 case RPC_DISPLAY_PROTO:
 370                 case RPC_DISPLAY_NETID:
 371                         continue;
 372                 default:
 373                         kfree(xprt->address_strings[i]);
 374                 }
 375 }
 376
 377 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 378
 379 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 380 {
 381         struct msghdr msg = {
 382                 .msg_name       = addr,
 383                 .msg_namelen    = addrlen,
 384                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 385         };
 386         struct kvec iov = {
 387                 .iov_base       = vec->iov_base + base,
 388                 .iov_len        = vec->iov_len - base,
 389         };
 390
 391         if (iov.iov_len != 0)
 392                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 393         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 394 }
 395
 396 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 397 {
 398         struct page **ppage;
 399         unsigned int remainder;
 400         int err, sent = 0;
 401
 402         remainder = xdr->page_len - base;
 403         base += xdr->page_base;
 404         ppage = xdr->pages + (base >> PAGE_SHIFT);
 405         base &= ~PAGE_MASK;
 406         for(;;) {
 407                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 408                 int flags = XS_SENDMSG_FLAGS;
 409
 410                 remainder -= len;
 411                 if (remainder != 0 || more)
 412                         flags |= MSG_MORE;
 413                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 414                 if (remainder == 0 || err != len)
 415                         break;
 416                 sent += err;
 417                 ppage++;
 418                 base = 0;
 419         }
 420         if (sent == 0)
 421                 return err;
 422         if (err > 0)
 423                 sent += err;
 424         return sent;
 425 }
 426
 427 /**
 428  * xs_sendpages - write pages directly to a socket
 429  * @sock: socket to send on
 430  * @addr: UDP only -- address of destination
 431  * @addrlen: UDP only -- length of destination address
 432  * @xdr: buffer containing this request
 433  * @base: starting position in the buffer
 434  *
 435  */
 436 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 437 {
 438         unsigned int remainder = xdr->len - base;
 439         int err, sent = 0;
 440
 441         if (unlikely(!sock))
 442                 return -ENOTSOCK;
 443
 444         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 445         if (base != 0) {
 446                 addr = NULL;
 447                 addrlen = 0;
 448         }
 449
 450         if (base < xdr->head[0].iov_len || addr != NULL) {
 451                 unsigned int len = xdr->head[0].iov_len - base;
 452                 remainder -= len;
 453                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 454                 if (remainder == 0 || err != len)
 455                         goto out;
 456                 sent += err;
 457                 base = 0;
 458         } else
 459                 base -= xdr->head[0].iov_len;
 460
 461         if (base < xdr->page_len) {
 462                 unsigned int len = xdr->page_len - base;
 463                 remainder -= len;
 464                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 465                 if (remainder == 0 || err != len)
 466                         goto out;
 467                 sent += err;
 468                 base = 0;
 469         } else
 470                 base -= xdr->page_len;
 471
 472         if (base >= xdr->tail[0].iov_len)
 473                 return sent;
 474         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 475 out:
 476         if (sent == 0)
 477                 return err;
 478         if (err > 0)
 479                 sent += err;
 480         return sent;
 481 }
 482
 483 static void xs_nospace_callback(struct rpc_task *task)
 484 {
 485         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 486
 487         transport->inet->sk_write_pending--;
 488         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 489 }
 490
 491 /**
 492  * xs_nospace - place task on wait queue if transmit was incomplete
 493  * @task: task to put to sleep
 494  *
 495  */
 496 static int xs_nospace(struct rpc_task *task)
 497 {
 498         struct rpc_rqst *req = task->tk_rqstp;
 499         struct rpc_xprt *xprt = req->rq_xprt;
 500         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 501         int ret = -EAGAIN;
 502
 503         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 504                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 505                         req->rq_slen);
 506
 507         /* Protect against races with write_space */
 508         spin_lock_bh(&xprt->transport_lock);
 509
 510         /* Don't race with disconnect */
 511         if (xprt_connected(xprt)) {
 512                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 513                         /*
 514                          * Notify TCP that we're limited by the application
 515                          * window size
 516                          */
 517                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 518                         transport->inet->sk_write_pending++;
 519                         /* ...and wait for more buffer space */
 520                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 521                 }
 522         } else {
 523                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 524                 ret = -ENOTCONN;
 525         }
 526
 527         spin_unlock_bh(&xprt->transport_lock);
 528         return ret;
 529 }
 530
 531 /*
 532  * Construct a stream transport record marker in @buf.
 533  */
 534 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 535 {
 536         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 537         rpc_fraghdr *base = buf->head[0].iov_base;
 538         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 539 }
 540
 541 /**
 542  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 543  * @task: RPC task that manages the state of an RPC request
 544  *
 545  * Return values:
 546  *        0:    The request has been sent
 547  *   EAGAIN:    The socket was blocked, please call again later to
 548  *              complete the request
 549  * ENOTCONN:    Caller needs to invoke connect logic then call again
 550  *    other:    Some other error occured, the request was not sent
 551  */
 552 static int xs_local_send_request(struct rpc_task *task)
 553 {
 554         struct rpc_rqst *req = task->tk_rqstp;
 555         struct rpc_xprt *xprt = req->rq_xprt;
 556         struct sock_xprt *transport =
 557                                 container_of(xprt, struct sock_xprt, xprt);
 558         struct xdr_buf *xdr = &req->rq_snd_buf;
 559         int status;
 560
 561         xs_encode_stream_record_marker(&req->rq_snd_buf);
 562
 563         xs_pktdump("packet data:",
 564                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 565
 566         status = xs_sendpages(transport->sock, NULL, 0,
 567                                                 xdr, req->rq_bytes_sent);
 568         dprintk("RPC:       %s(%u) = %d\n",
 569                         __func__, xdr->len - req->rq_bytes_sent, status);
 570         if (likely(status >= 0)) {
 571                 req->rq_bytes_sent += status;
 572                 req->rq_xmit_bytes_sent += status;
 573                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 574                         req->rq_bytes_sent = 0;
 575                         return 0;
 576                 }
 577                 status = -EAGAIN;
 578         }
 579
 580         switch (status) {
 581         case -EAGAIN:
 582                 status = xs_nospace(task);
 583                 break;
 584         default:
 585                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 586                         -status);
 587         case -EPIPE:
 588                 xs_close(xprt);
 589                 status = -ENOTCONN;
 590         }
 591
 592         return status;
 593 }
 594
 595 /**
 596  * xs_udp_send_request - write an RPC request to a UDP socket
 597  * @task: address of RPC task that manages the state of an RPC request
 598  *
 599  * Return values:
 600  *        0:    The request has been sent
 601  *   EAGAIN:    The socket was blocked, please call again later to
 602  *              complete the request
 603  * ENOTCONN:    Caller needs to invoke connect logic then call again
 604  *    other:    Some other error occurred, the request was not sent
 605  */
 606 static int xs_udp_send_request(struct rpc_task *task)
 607 {
 608         struct rpc_rqst *req = task->tk_rqstp;
 609         struct rpc_xprt *xprt = req->rq_xprt;
 610         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 611         struct xdr_buf *xdr = &req->rq_snd_buf;
 612         int status;
 613
 614         xs_pktdump("packet data:",
 615                                 req->rq_svec->iov_base,
 616                                 req->rq_svec->iov_len);
 617
 618         if (!xprt_bound(xprt))
 619                 return -ENOTCONN;
 620         status = xs_sendpages(transport->sock,
 621                               xs_addr(xprt),
 622                               xprt->addrlen, xdr,
 623                               req->rq_bytes_sent);
 624
 625         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 626                         xdr->len - req->rq_bytes_sent, status);
 627
 628         if (status >= 0) {
 629                 req->rq_xmit_bytes_sent += status;
 630                 if (status >= req->rq_slen)
 631                         return 0;
 632                 /* Still some bytes left; set up for a retry later. */
 633                 status = -EAGAIN;
 634         }
 635
 636         switch (status) {
 637         case -ENOTSOCK:
 638                 status = -ENOTCONN;
 639                 /* Should we call xs_close() here? */
 640                 break;
 641         case -EAGAIN:
 642                 status = xs_nospace(task);
 643                 break;
 644         default:
 645                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 646                         -status);
 647         case -ENETUNREACH:
 648         case -EPIPE:
 649         case -ECONNREFUSED:
 650                 /* When the server has died, an ICMP port unreachable message
 651                  * prompts ECONNREFUSED. */
 652                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 653         }
 654
 655         return status;
 656 }
 657
 658 /**
 659  * xs_tcp_shutdown - gracefully shut down a TCP socket
 660  * @xprt: transport
 661  *
 662  * Initiates a graceful shutdown of the TCP socket by calling the
 663  * equivalent of shutdown(SHUT_WR);
 664  */
 665 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 666 {
 667         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 668         struct socket *sock = transport->sock;
 669
 670         if (sock != NULL) {
 671                 kernel_sock_shutdown(sock, SHUT_WR);
 672                 trace_rpc_socket_shutdown(xprt, sock);
 673         }
 674 }
 675
 676 /**
 677  * xs_tcp_send_request - write an RPC request to a TCP socket
 678  * @task: address of RPC task that manages the state of an RPC request
 679  *
 680  * Return values:
 681  *        0:    The request has been sent
 682  *   EAGAIN:    The socket was blocked, please call again later to
 683  *              complete the request
 684  * ENOTCONN:    Caller needs to invoke connect logic then call again
 685  *    other:    Some other error occurred, the request was not sent
 686  *
 687  * XXX: In the case of soft timeouts, should we eventually give up
 688  *      if sendmsg is not able to make progress?
 689  */
 690 static int xs_tcp_send_request(struct rpc_task *task)
 691 {
 692         struct rpc_rqst *req = task->tk_rqstp;
 693         struct rpc_xprt *xprt = req->rq_xprt;
 694         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 695         struct xdr_buf *xdr = &req->rq_snd_buf;
 696         int status;
 697
 698         xs_encode_stream_record_marker(&req->rq_snd_buf);
 699
 700         xs_pktdump("packet data:",
 701                                 req->rq_svec->iov_base,
 702                                 req->rq_svec->iov_len);
 703
 704         /* Continue transmitting the packet/record. We must be careful
 705          * to cope with writespace callbacks arriving _after_ we have
 706          * called sendmsg(). */
 707         while (1) {
 708                 status = xs_sendpages(transport->sock,
 709                                         NULL, 0, xdr, req->rq_bytes_sent);
 710
 711                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 712                                 xdr->len - req->rq_bytes_sent, status);
 713
 714                 if (unlikely(status < 0))
 715                         break;
 716
 717                 /* If we've sent the entire packet, immediately
 718                  * reset the count of bytes sent. */
 719                 req->rq_bytes_sent += status;
 720                 req->rq_xmit_bytes_sent += status;
 721                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 722                         req->rq_bytes_sent = 0;
 723                         return 0;
 724                 }
 725
 726                 if (status != 0)
 727                         continue;
 728                 status = -EAGAIN;
 729                 break;
 730         }
 731
 732         switch (status) {
 733         case -ENOTSOCK:
 734                 status = -ENOTCONN;
 735                 /* Should we call xs_close() here? */
 736                 break;
 737         case -EAGAIN:
 738                 status = xs_nospace(task);
 739                 break;
 740         default:
 741                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 742                         -status);
 743         case -ECONNRESET:
 744                 xs_tcp_shutdown(xprt);
 745         case -ECONNREFUSED:
 746         case -ENOTCONN:
 747         case -EPIPE:
 748                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 749         }
 750
 751         return status;
 752 }
 753
 754 /**
 755  * xs_tcp_release_xprt - clean up after a tcp transmission
 756  * @xprt: transport
 757  * @task: rpc task
 758  *
 759  * This cleans up if an error causes us to abort the transmission of a request.
 760  * In this case, the socket may need to be reset in order to avoid confusing
 761  * the server.
 762  */
 763 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 764 {
 765         struct rpc_rqst *req;
 766
 767         if (task != xprt->snd_task)
 768                 return;
 769         if (task == NULL)
 770                 goto out_release;
 771         req = task->tk_rqstp;
 772         if (req == NULL)
 773                 goto out_release;
 774         if (req->rq_bytes_sent == 0)
 775                 goto out_release;
 776         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 777                 goto out_release;
 778         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 779 out_release:
 780         xprt_release_xprt(xprt, task);
 781 }
 782
 783 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 784 {
 785         transport->old_data_ready = sk->sk_data_ready;
 786         transport->old_state_change = sk->sk_state_change;
 787         transport->old_write_space = sk->sk_write_space;
 788 }
 789
 790 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 791 {
 792         sk->sk_data_ready = transport->old_data_ready;
 793         sk->sk_state_change = transport->old_state_change;
 794         sk->sk_write_space = transport->old_write_space;
 795 }
 796
 797 static void xs_reset_transport(struct sock_xprt *transport)
 798 {
 799         struct socket *sock = transport->sock;
 800         struct sock *sk = transport->inet;
 801
 802         if (sk == NULL)
 803                 return;
 804
 805         transport->srcport = 0;
 806
 807         write_lock_bh(&sk->sk_callback_lock);
 808         transport->inet = NULL;
 809         transport->sock = NULL;
 810
 811         sk->sk_user_data = NULL;
 812
 813         xs_restore_old_callbacks(transport, sk);
 814         write_unlock_bh(&sk->sk_callback_lock);
 815
 816         sk->sk_no_check = 0;
 817
 818         trace_rpc_socket_close(&transport->xprt, sock);
 819         sock_release(sock);
 820 }
 821
 822 /**
 823  * xs_close - close a socket
 824  * @xprt: transport
 825  *
 826  * This is used when all requests are complete; ie, no DRC state remains
 827  * on the server we want to save.
 828  *
 829  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 830  * xs_reset_transport() zeroing the socket from underneath a writer.
 831  */
 832 static void xs_close(struct rpc_xprt *xprt)
 833 {
 834         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 835
 836         dprintk("RPC:       xs_close xprt %p\n", xprt);
 837
 838         xs_reset_transport(transport);
 839         xprt->reestablish_timeout = 0;
 840
 841         smp_mb__before_clear_bit();
 842         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 843         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 844         clear_bit(XPRT_CLOSING, &xprt->state);
 845         smp_mb__after_clear_bit();
 846         xprt_disconnect_done(xprt);
 847 }
 848
 849 static void xs_tcp_close(struct rpc_xprt *xprt)
 850 {
 851         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 852                 xs_close(xprt);
 853         else
 854                 xs_tcp_shutdown(xprt);
 855 }
 856
 857 static void xs_local_destroy(struct rpc_xprt *xprt)
 858 {
 859         xs_close(xprt);
 860         xs_free_peer_addresses(xprt);
 861         xprt_free(xprt);
 862         module_put(THIS_MODULE);
 863 }
 864
 865 /**
 866  * xs_destroy - prepare to shutdown a transport
 867  * @xprt: doomed transport
 868  *
 869  */
 870 static void xs_destroy(struct rpc_xprt *xprt)
 871 {
 872         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 873
 874         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 875
 876         cancel_delayed_work_sync(&transport->connect_worker);
 877
 878         xs_local_destroy(xprt);
 879 }
 880
 881 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 882 {
 883         return (struct rpc_xprt *) sk->sk_user_data;
 884 }
 885
 886 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 887 {
 888         struct xdr_skb_reader desc = {
 889                 .skb            = skb,
 890                 .offset         = sizeof(rpc_fraghdr),
 891                 .count          = skb->len - sizeof(rpc_fraghdr),
 892         };
 893
 894         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 895                 return -1;
 896         if (desc.count)
 897                 return -1;
 898         return 0;
 899 }
 900
 901 /**
 902  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 903  * @sk: socket with data to read
 904  * @len: how much data to read
 905  *
 906  * Currently this assumes we can read the whole reply in a single gulp.
 907  */
 908 static void xs_local_data_ready(struct sock *sk, int len)
 909 {
 910         struct rpc_task *task;
 911         struct rpc_xprt *xprt;
 912         struct rpc_rqst *rovr;
 913         struct sk_buff *skb;
 914         int err, repsize, copied;
 915         u32 _xid;
 916         __be32 *xp;
 917
 918         read_lock_bh(&sk->sk_callback_lock);
 919         dprintk("RPC:       %s...\n", __func__);
 920         xprt = xprt_from_sock(sk);
 921         if (xprt == NULL)
 922                 goto out;
 923
 924         skb = skb_recv_datagram(sk, 0, 1, &err);
 925         if (skb == NULL)
 926                 goto out;
 927
 928         repsize = skb->len - sizeof(rpc_fraghdr);
 929         if (repsize < 4) {
 930                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 931                 goto dropit;
 932         }
 933
 934         /* Copy the XID from the skb... */
 935         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 936         if (xp == NULL)
 937                 goto dropit;
 938
 939         /* Look up and lock the request corresponding to the given XID */
 940         spin_lock(&xprt->transport_lock);
 941         rovr = xprt_lookup_rqst(xprt, *xp);
 942         if (!rovr)
 943                 goto out_unlock;
 944         task = rovr->rq_task;
 945
 946         copied = rovr->rq_private_buf.buflen;
 947         if (copied > repsize)
 948                 copied = repsize;
 949
 950         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 951                 dprintk("RPC:       sk_buff copy failed\n");
 952                 goto out_unlock;
 953         }
 954
 955         xprt_complete_rqst(task, copied);
 956
 957  out_unlock:
 958         spin_unlock(&xprt->transport_lock);
 959  dropit:
 960         skb_free_datagram(sk, skb);
 961  out:
 962         read_unlock_bh(&sk->sk_callback_lock);
 963 }
 964
 965 /**
 966  * xs_udp_data_ready - "data ready" callback for UDP sockets
 967  * @sk: socket with data to read
 968  * @len: how much data to read
 969  *
 970  */
 971 static void xs_udp_data_ready(struct sock *sk, int len)
 972 {
 973         struct rpc_task *task;
 974         struct rpc_xprt *xprt;
 975         struct rpc_rqst *rovr;
 976         struct sk_buff *skb;
 977         int err, repsize, copied;
 978         u32 _xid;
 979         __be32 *xp;
 980
 981         read_lock_bh(&sk->sk_callback_lock);
 982         dprintk("RPC:       xs_udp_data_ready...\n");
 983         if (!(xprt = xprt_from_sock(sk)))
 984                 goto out;
 985
 986         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 987                 goto out;
 988
 989         repsize = skb->len - sizeof(struct udphdr);
 990         if (repsize < 4) {
 991                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 992                 goto dropit;
 993         }
 994
 995         /* Copy the XID from the skb... */
 996         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 997                                 sizeof(_xid), &_xid);
 998         if (xp == NULL)
 999                 goto dropit;
1000
1001         /* Look up and lock the request corresponding to the given XID */
1002         spin_lock(&xprt->transport_lock);
1003         rovr = xprt_lookup_rqst(xprt, *xp);
1004         if (!rovr)
1005                 goto out_unlock;
1006         task = rovr->rq_task;
1007
1008         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1009                 copied = repsize;
1010
1011         /* Suck it into the iovec, verify checksum if not done by hw. */
1012         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1013                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1014                 goto out_unlock;
1015         }
1016
1017         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1018
1019         xprt_adjust_cwnd(xprt, task, copied);
1020         xprt_complete_rqst(task, copied);
1021
1022  out_unlock:
1023         spin_unlock(&xprt->transport_lock);
1024  dropit:
1025         skb_free_datagram(sk, skb);
1026  out:
1027         read_unlock_bh(&sk->sk_callback_lock);
1028 }
1029
1030 /*
1031  * Helper function to force a TCP close if the server is sending
1032  * junk and/or it has put us in CLOSE_WAIT
1033  */
1034 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1035 {
1036         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1037         xprt_force_disconnect(xprt);
1038 }
1039
1040 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1041 {
1042         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1043         size_t len, used;
1044         char *p;
1045
1046         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1047         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1048         used = xdr_skb_read_bits(desc, p, len);
1049         transport->tcp_offset += used;
1050         if (used != len)
1051                 return;
1052
1053         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1054         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1055                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1056         else
1057                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1058         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1059
1060         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1061         transport->tcp_offset = 0;
1062
1063         /* Sanity check of the record length */
1064         if (unlikely(transport->tcp_reclen < 8)) {
1065                 dprintk("RPC:       invalid TCP record fragment length\n");
1066                 xs_tcp_force_close(xprt);
1067                 return;
1068         }
1069         dprintk("RPC:       reading TCP record fragment of length %d\n",
1070                         transport->tcp_reclen);
1071 }
1072
1073 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1074 {
1075         if (transport->tcp_offset == transport->tcp_reclen) {
1076                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1077                 transport->tcp_offset = 0;
1078                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1079                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1080                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1081                         transport->tcp_copied = 0;
1082                 }
1083         }
1084 }
1085
1086 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1087 {
1088         size_t len, used;
1089         char *p;
1090
1091         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1092         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1093         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1094         used = xdr_skb_read_bits(desc, p, len);
1095         transport->tcp_offset += used;
1096         if (used != len)
1097                 return;
1098         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1099         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1100         transport->tcp_copied = 4;
1101         dprintk("RPC:       reading %s XID %08x\n",
1102                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1103                                                               : "request with",
1104                         ntohl(transport->tcp_xid));
1105         xs_tcp_check_fraghdr(transport);
1106 }
1107
1108 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1109                                        struct xdr_skb_reader *desc)
1110 {
1111         size_t len, used;
1112         u32 offset;
1113         char *p;
1114
1115         /*
1116          * We want transport->tcp_offset to be 8 at the end of this routine
1117          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1118          * When this function is called for the first time,
1119          * transport->tcp_offset is 4 (after having already read the xid).
1120          */
1121         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1122         len = sizeof(transport->tcp_calldir) - offset;
1123         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1124         p = ((char *) &transport->tcp_calldir) + offset;
1125         used = xdr_skb_read_bits(desc, p, len);
1126         transport->tcp_offset += used;
1127         if (used != len)
1128                 return;
1129         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1130         /*
1131          * We don't yet have the XDR buffer, so we will write the calldir
1132          * out after we get the buffer from the 'struct rpc_rqst'
1133          */
1134         switch (ntohl(transport->tcp_calldir)) {
1135         case RPC_REPLY:
1136                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1137                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1138                 transport->tcp_flags |= TCP_RPC_REPLY;
1139                 break;
1140         case RPC_CALL:
1141                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1142                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1143                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1144                 break;
1145         default:
1146                 dprintk("RPC:       invalid request message type\n");
1147                 xs_tcp_force_close(&transport->xprt);
1148         }
1149         xs_tcp_check_fraghdr(transport);
1150 }
1151
1152 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1153                                      struct xdr_skb_reader *desc,
1154                                      struct rpc_rqst *req)
1155 {
1156         struct sock_xprt *transport =
1157                                 container_of(xprt, struct sock_xprt, xprt);
1158         struct xdr_buf *rcvbuf;
1159         size_t len;
1160         ssize_t r;
1161
1162         rcvbuf = &req->rq_private_buf;
1163
1164         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1165                 /*
1166                  * Save the RPC direction in the XDR buffer
1167                  */
1168                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1169                         &transport->tcp_calldir,
1170                         sizeof(transport->tcp_calldir));
1171                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1172                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1173         }
1174
1175         len = desc->count;
1176         if (len > transport->tcp_reclen - transport->tcp_offset) {
1177                 struct xdr_skb_reader my_desc;
1178
1179                 len = transport->tcp_reclen - transport->tcp_offset;
1180                 memcpy(&my_desc, desc, sizeof(my_desc));
1181                 my_desc.count = len;
1182                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1183                                           &my_desc, xdr_skb_read_bits);
1184                 desc->count -= r;
1185                 desc->offset += r;
1186         } else
1187                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1188                                           desc, xdr_skb_read_bits);
1189
1190         if (r > 0) {
1191                 transport->tcp_copied += r;
1192                 transport->tcp_offset += r;
1193         }
1194         if (r != len) {
1195                 /* Error when copying to the receive buffer,
1196                  * usually because we weren't able to allocate
1197                  * additional buffer pages. All we can do now
1198                  * is turn off TCP_RCV_COPY_DATA, so the request
1199                  * will not receive any additional updates,
1200                  * and time out.
1201                  * Any remaining data from this record will
1202                  * be discarded.
1203                  */
1204                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1205                 dprintk("RPC:       XID %08x truncated request\n",
1206                                 ntohl(transport->tcp_xid));
1207                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1208                                 "tcp_offset = %u, tcp_reclen = %u\n",
1209                                 xprt, transport->tcp_copied,
1210                                 transport->tcp_offset, transport->tcp_reclen);
1211                 return;
1212         }
1213
1214         dprintk("RPC:       XID %08x read %Zd bytes\n",
1215                         ntohl(transport->tcp_xid), r);
1216         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1217                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1218                         transport->tcp_offset, transport->tcp_reclen);
1219
1220         if (transport->tcp_copied == req->rq_private_buf.buflen)
1221                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1222         else if (transport->tcp_offset == transport->tcp_reclen) {
1223                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1224                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1225         }
1226 }
1227
1228 /*
1229  * Finds the request corresponding to the RPC xid and invokes the common
1230  * tcp read code to read the data.
1231  */
1232 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1233                                     struct xdr_skb_reader *desc)
1234 {
1235         struct sock_xprt *transport =
1236                                 container_of(xprt, struct sock_xprt, xprt);
1237         struct rpc_rqst *req;
1238
1239         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1240
1241         /* Find and lock the request corresponding to this xid */
1242         spin_lock(&xprt->transport_lock);
1243         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1244         if (!req) {
1245                 dprintk("RPC:       XID %08x request not found!\n",
1246                                 ntohl(transport->tcp_xid));
1247                 spin_unlock(&xprt->transport_lock);
1248                 return -1;
1249         }
1250
1251         xs_tcp_read_common(xprt, desc, req);
1252
1253         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1254                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1255
1256         spin_unlock(&xprt->transport_lock);
1257         return 0;
1258 }
1259
1260 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1261 /*
1262  * Obtains an rpc_rqst previously allocated and invokes the common
1263  * tcp read code to read the data.  The result is placed in the callback
1264  * queue.
1265  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1266  * connection and return -1.
1267  */
1268 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1269                                        struct xdr_skb_reader *desc)
1270 {
1271         struct sock_xprt *transport =
1272                                 container_of(xprt, struct sock_xprt, xprt);
1273         struct rpc_rqst *req;
1274
1275         req = xprt_alloc_bc_request(xprt);
1276         if (req == NULL) {
1277                 printk(KERN_WARNING "Callback slot table overflowed\n");
1278                 xprt_force_disconnect(xprt);
1279                 return -1;
1280         }
1281
1282         req->rq_xid = transport->tcp_xid;
1283         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1284         xs_tcp_read_common(xprt, desc, req);
1285
1286         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1287                 struct svc_serv *bc_serv = xprt->bc_serv;
1288
1289                 /*
1290                  * Add callback request to callback list.  The callback
1291                  * service sleeps on the sv_cb_waitq waiting for new
1292                  * requests.  Wake it up after adding enqueing the
1293                  * request.
1294                  */
1295                 dprintk("RPC:       add callback request to list\n");
1296                 spin_lock(&bc_serv->sv_cb_lock);
1297                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1298                 spin_unlock(&bc_serv->sv_cb_lock);
1299                 wake_up(&bc_serv->sv_cb_waitq);
1300         }
1301
1302         req->rq_private_buf.len = transport->tcp_copied;
1303
1304         return 0;
1305 }
1306
1307 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1308                                         struct xdr_skb_reader *desc)
1309 {
1310         struct sock_xprt *transport =
1311                                 container_of(xprt, struct sock_xprt, xprt);
1312
1313         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1314                 xs_tcp_read_reply(xprt, desc) :
1315                 xs_tcp_read_callback(xprt, desc);
1316 }
1317 #else
1318 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1319                                         struct xdr_skb_reader *desc)
1320 {
1321         return xs_tcp_read_reply(xprt, desc);
1322 }
1323 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1324
1325 /*
1326  * Read data off the transport.  This can be either an RPC_CALL or an
1327  * RPC_REPLY.  Relay the processing to helper functions.
1328  */
1329 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1330                                     struct xdr_skb_reader *desc)
1331 {
1332         struct sock_xprt *transport =
1333                                 container_of(xprt, struct sock_xprt, xprt);
1334
1335         if (_xs_tcp_read_data(xprt, desc) == 0)
1336                 xs_tcp_check_fraghdr(transport);
1337         else {
1338                 /*
1339                  * The transport_lock protects the request handling.
1340                  * There's no need to hold it to update the tcp_flags.
1341                  */
1342                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1343         }
1344 }
1345
1346 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1347 {
1348         size_t len;
1349
1350         len = transport->tcp_reclen - transport->tcp_offset;
1351         if (len > desc->count)
1352                 len = desc->count;
1353         desc->count -= len;
1354         desc->offset += len;
1355         transport->tcp_offset += len;
1356         dprintk("RPC:       discarded %Zu bytes\n", len);
1357         xs_tcp_check_fraghdr(transport);
1358 }
1359
1360 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1361 {
1362         struct rpc_xprt *xprt = rd_desc->arg.data;
1363         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1364         struct xdr_skb_reader desc = {
1365                 .skb    = skb,
1366                 .offset = offset,
1367                 .count  = len,
1368         };
1369
1370         dprintk("RPC:       xs_tcp_data_recv started\n");
1371         do {
1372                 /* Read in a new fragment marker if necessary */
1373                 /* Can we ever really expect to get completely empty fragments? */
1374                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1375                         xs_tcp_read_fraghdr(xprt, &desc);
1376                         continue;
1377                 }
1378                 /* Read in the xid if necessary */
1379                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1380                         xs_tcp_read_xid(transport, &desc);
1381                         continue;
1382                 }
1383                 /* Read in the call/reply flag */
1384                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1385                         xs_tcp_read_calldir(transport, &desc);
1386                         continue;
1387                 }
1388                 /* Read in the request data */
1389                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1390                         xs_tcp_read_data(xprt, &desc);
1391                         continue;
1392                 }
1393                 /* Skip over any trailing bytes on short reads */
1394                 xs_tcp_read_discard(transport, &desc);
1395         } while (desc.count);
1396         dprintk("RPC:       xs_tcp_data_recv done\n");
1397         return len - desc.count;
1398 }
1399
1400 /**
1401  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1402  * @sk: socket with data to read
1403  * @bytes: how much data to read
1404  *
1405  */
1406 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1407 {
1408         struct rpc_xprt *xprt;
1409         read_descriptor_t rd_desc;
1410         int read;
1411
1412         dprintk("RPC:       xs_tcp_data_ready...\n");
1413
1414         read_lock_bh(&sk->sk_callback_lock);
1415         if (!(xprt = xprt_from_sock(sk)))
1416                 goto out;
1417         /* Any data means we had a useful conversation, so
1418          * the we don't need to delay the next reconnect
1419          */
1420         if (xprt->reestablish_timeout)
1421                 xprt->reestablish_timeout = 0;
1422
1423         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1424         rd_desc.arg.data = xprt;
1425         do {
1426                 rd_desc.count = 65536;
1427                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1428         } while (read > 0);
1429 out:
1430         read_unlock_bh(&sk->sk_callback_lock);
1431 }
1432
1433 /*
1434  * Do the equivalent of linger/linger2 handling for dealing with
1435  * broken servers that don't close the socket in a timely
1436  * fashion
1437  */
1438 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1439                 unsigned long timeout)
1440 {
1441         struct sock_xprt *transport;
1442
1443         if (xprt_test_and_set_connecting(xprt))
1444                 return;
1445         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1446         transport = container_of(xprt, struct sock_xprt, xprt);
1447         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1448                            timeout);
1449 }
1450
1451 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1452 {
1453         struct sock_xprt *transport;
1454
1455         transport = container_of(xprt, struct sock_xprt, xprt);
1456
1457         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1458             !cancel_delayed_work(&transport->connect_worker))
1459                 return;
1460         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1461         xprt_clear_connecting(xprt);
1462 }
1463
1464 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1465 {
1466         smp_mb__before_clear_bit();
1467         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1468         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1469         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1470         clear_bit(XPRT_CLOSING, &xprt->state);
1471         smp_mb__after_clear_bit();
1472 }
1473
1474 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1475 {
1476         xs_sock_reset_connection_flags(xprt);
1477         /* Mark transport as closed and wake up all pending tasks */
1478         xprt_disconnect_done(xprt);
1479 }
1480
1481 /**
1482  * xs_tcp_state_change - callback to handle TCP socket state changes
1483  * @sk: socket whose state has changed
1484  *
1485  */
1486 static void xs_tcp_state_change(struct sock *sk)
1487 {
1488         struct rpc_xprt *xprt;
1489
1490         read_lock_bh(&sk->sk_callback_lock);
1491         if (!(xprt = xprt_from_sock(sk)))
1492                 goto out;
1493         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1494         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1495                         sk->sk_state, xprt_connected(xprt),
1496                         sock_flag(sk, SOCK_DEAD),
1497                         sock_flag(sk, SOCK_ZAPPED),
1498                         sk->sk_shutdown);
1499
1500         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1501         switch (sk->sk_state) {
1502         case TCP_ESTABLISHED:
1503                 spin_lock(&xprt->transport_lock);
1504                 if (!xprt_test_and_set_connected(xprt)) {
1505                         struct sock_xprt *transport = container_of(xprt,
1506                                         struct sock_xprt, xprt);
1507
1508                         /* Reset TCP record info */
1509                         transport->tcp_offset = 0;
1510                         transport->tcp_reclen = 0;
1511                         transport->tcp_copied = 0;
1512                         transport->tcp_flags =
1513                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1514
1515                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1516                 }
1517                 spin_unlock(&xprt->transport_lock);
1518                 break;
1519         case TCP_FIN_WAIT1:
1520                 /* The client initiated a shutdown of the socket */
1521                 xprt->connect_cookie++;
1522                 xprt->reestablish_timeout = 0;
1523                 set_bit(XPRT_CLOSING, &xprt->state);
1524                 smp_mb__before_clear_bit();
1525                 clear_bit(XPRT_CONNECTED, &xprt->state);
1526                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1527                 smp_mb__after_clear_bit();
1528                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1529                 break;
1530         case TCP_CLOSE_WAIT:
1531                 /* The server initiated a shutdown of the socket */
1532                 xprt->connect_cookie++;
1533                 clear_bit(XPRT_CONNECTED, &xprt->state);
1534                 xs_tcp_force_close(xprt);
1535         case TCP_CLOSING:
1536                 /*
1537                  * If the server closed down the connection, make sure that
1538                  * we back off before reconnecting
1539                  */
1540                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1541                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1542                 break;
1543         case TCP_LAST_ACK:
1544                 set_bit(XPRT_CLOSING, &xprt->state);
1545                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1546                 smp_mb__before_clear_bit();
1547                 clear_bit(XPRT_CONNECTED, &xprt->state);
1548                 smp_mb__after_clear_bit();
1549                 break;
1550         case TCP_CLOSE:
1551                 xs_tcp_cancel_linger_timeout(xprt);
1552                 xs_sock_mark_closed(xprt);
1553         }
1554  out:
1555         read_unlock_bh(&sk->sk_callback_lock);
1556 }
1557
1558 static void xs_write_space(struct sock *sk)
1559 {
1560         struct socket *sock;
1561         struct rpc_xprt *xprt;
1562
1563         if (unlikely(!(sock = sk->sk_socket)))
1564                 return;
1565         clear_bit(SOCK_NOSPACE, &sock->flags);
1566
1567         if (unlikely(!(xprt = xprt_from_sock(sk))))
1568                 return;
1569         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1570                 return;
1571
1572         xprt_write_space(xprt);
1573 }
1574
1575 /**
1576  * xs_udp_write_space - callback invoked when socket buffer space
1577  *                             becomes available
1578  * @sk: socket whose state has changed
1579  *
1580  * Called when more output buffer space is available for this socket.
1581  * We try not to wake our writers until they can make "significant"
1582  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1583  * with a bunch of small requests.
1584  */
1585 static void xs_udp_write_space(struct sock *sk)
1586 {
1587         read_lock_bh(&sk->sk_callback_lock);
1588
1589         /* from net/core/sock.c:sock_def_write_space */
1590         if (sock_writeable(sk))
1591                 xs_write_space(sk);
1592
1593         read_unlock_bh(&sk->sk_callback_lock);
1594 }
1595
1596 /**
1597  * xs_tcp_write_space - callback invoked when socket buffer space
1598  *                             becomes available
1599  * @sk: socket whose state has changed
1600  *
1601  * Called when more output buffer space is available for this socket.
1602  * We try not to wake our writers until they can make "significant"
1603  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1604  * with a bunch of small requests.
1605  */
1606 static void xs_tcp_write_space(struct sock *sk)
1607 {
1608         read_lock_bh(&sk->sk_callback_lock);
1609
1610         /* from net/core/stream.c:sk_stream_write_space */
1611         if (sk_stream_is_writeable(sk))
1612                 xs_write_space(sk);
1613
1614         read_unlock_bh(&sk->sk_callback_lock);
1615 }
1616
1617 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1618 {
1619         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1620         struct sock *sk = transport->inet;
1621
1622         if (transport->rcvsize) {
1623                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1624                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1625         }
1626         if (transport->sndsize) {
1627                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1628                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1629                 sk->sk_write_space(sk);
1630         }
1631 }
1632
1633 /**
1634  * xs_udp_set_buffer_size - set send and receive limits
1635  * @xprt: generic transport
1636  * @sndsize: requested size of send buffer, in bytes
1637  * @rcvsize: requested size of receive buffer, in bytes
1638  *
1639  * Set socket send and receive buffer size limits.
1640  */
1641 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1642 {
1643         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1644
1645         transport->sndsize = 0;
1646         if (sndsize)
1647                 transport->sndsize = sndsize + 1024;
1648         transport->rcvsize = 0;
1649         if (rcvsize)
1650                 transport->rcvsize = rcvsize + 1024;
1651
1652         xs_udp_do_set_buffer_size(xprt);
1653 }
1654
1655 /**
1656  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1657  * @task: task that timed out
1658  *
1659  * Adjust the congestion window after a retransmit timeout has occurred.
1660  */
1661 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1662 {
1663         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1664 }
1665
1666 static unsigned short xs_get_random_port(void)
1667 {
1668         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1669         unsigned short rand = (unsigned short) net_random() % range;
1670         return rand + xprt_min_resvport;
1671 }
1672
1673 /**
1674  * xs_set_port - reset the port number in the remote endpoint address
1675  * @xprt: generic transport
1676  * @port: new port number
1677  *
1678  */
1679 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1680 {
1681         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1682
1683         rpc_set_port(xs_addr(xprt), port);
1684         xs_update_peer_port(xprt);
1685 }
1686
1687 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1688 {
1689         unsigned short port = transport->srcport;
1690
1691         if (port == 0 && transport->xprt.resvport)
1692                 port = xs_get_random_port();
1693         return port;
1694 }
1695
1696 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1697 {
1698         if (transport->srcport != 0)
1699                 transport->srcport = 0;
1700         if (!transport->xprt.resvport)
1701                 return 0;
1702         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1703                 return xprt_max_resvport;
1704         return --port;
1705 }
1706 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1707 {
1708         struct sockaddr_storage myaddr;
1709         int err, nloop = 0;
1710         unsigned short port = xs_get_srcport(transport);
1711         unsigned short last;
1712
1713         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1714         do {
1715                 rpc_set_port((struct sockaddr *)&myaddr, port);
1716                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1717                                 transport->xprt.addrlen);
1718                 if (port == 0)
1719                         break;
1720                 if (err == 0) {
1721                         transport->srcport = port;
1722                         break;
1723                 }
1724                 last = port;
1725                 port = xs_next_srcport(transport, port);
1726                 if (port > last)
1727                         nloop++;
1728         } while (err == -EADDRINUSE && nloop != 2);
1729
1730         if (myaddr.ss_family == AF_INET)
1731                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1732                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1733                                 port, err ? "failed" : "ok", err);
1734         else
1735                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1736                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1737                                 port, err ? "failed" : "ok", err);
1738         return err;
1739 }
1740
1741 /*
1742  * We don't support autobind on AF_LOCAL sockets
1743  */
1744 static void xs_local_rpcbind(struct rpc_task *task)
1745 {
1746         rcu_read_lock();
1747         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1748         rcu_read_unlock();
1749 }
1750
1751 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1752 {
1753 }
1754
1755 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1756 static struct lock_class_key xs_key[2];
1757 static struct lock_class_key xs_slock_key[2];
1758
1759 static inline void xs_reclassify_socketu(struct socket *sock)
1760 {
1761         struct sock *sk = sock->sk;
1762
1763         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1764                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1765 }
1766
1767 static inline void xs_reclassify_socket4(struct socket *sock)
1768 {
1769         struct sock *sk = sock->sk;
1770
1771         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1772                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1773 }
1774
1775 static inline void xs_reclassify_socket6(struct socket *sock)
1776 {
1777         struct sock *sk = sock->sk;
1778
1779         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1780                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1781 }
1782
1783 static inline void xs_reclassify_socket(int family, struct socket *sock)
1784 {
1785         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1786         if (sock_owned_by_user(sock->sk))
1787                 return;
1788
1789         switch (family) {
1790         case AF_LOCAL:
1791                 xs_reclassify_socketu(sock);
1792                 break;
1793         case AF_INET:
1794                 xs_reclassify_socket4(sock);
1795                 break;
1796         case AF_INET6:
1797                 xs_reclassify_socket6(sock);
1798                 break;
1799         }
1800 }
1801 #else
1802 static inline void xs_reclassify_socketu(struct socket *sock)
1803 {
1804 }
1805
1806 static inline void xs_reclassify_socket4(struct socket *sock)
1807 {
1808 }
1809
1810 static inline void xs_reclassify_socket6(struct socket *sock)
1811 {
1812 }
1813
1814 static inline void xs_reclassify_socket(int family, struct socket *sock)
1815 {
1816 }
1817 #endif
1818
1819 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1820                 struct sock_xprt *transport, int family, int type, int protocol)
1821 {
1822         struct socket *sock;
1823         int err;
1824
1825         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1826         if (err < 0) {
1827                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1828                                 protocol, -err);
1829                 goto out;
1830         }
1831         xs_reclassify_socket(family, sock);
1832
1833         err = xs_bind(transport, sock);
1834         if (err) {
1835                 sock_release(sock);
1836                 goto out;
1837         }
1838
1839         return sock;
1840 out:
1841         return ERR_PTR(err);
1842 }
1843
1844 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1845                                       struct socket *sock)
1846 {
1847         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1848                                                                         xprt);
1849
1850         if (!transport->inet) {
1851                 struct sock *sk = sock->sk;
1852
1853                 write_lock_bh(&sk->sk_callback_lock);
1854
1855                 xs_save_old_callbacks(transport, sk);
1856
1857                 sk->sk_user_data = xprt;
1858                 sk->sk_data_ready = xs_local_data_ready;
1859                 sk->sk_write_space = xs_udp_write_space;
1860                 sk->sk_allocation = GFP_ATOMIC;
1861
1862                 xprt_clear_connected(xprt);
1863
1864                 /* Reset to new socket */
1865                 transport->sock = sock;
1866                 transport->inet = sk;
1867
1868                 write_unlock_bh(&sk->sk_callback_lock);
1869         }
1870
1871         /* Tell the socket layer to start connecting... */
1872         xprt->stat.connect_count++;
1873         xprt->stat.connect_start = jiffies;
1874         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1875 }
1876
1877 /**
1878  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1879  * @xprt: RPC transport to connect
1880  * @transport: socket transport to connect
1881  * @create_sock: function to create a socket of the correct type
1882  */
1883 static int xs_local_setup_socket(struct sock_xprt *transport)
1884 {
1885         struct rpc_xprt *xprt = &transport->xprt;
1886         struct socket *sock;
1887         int status = -EIO;
1888
1889         current->flags |= PF_FSTRANS;
1890
1891         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1892         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1893                                         SOCK_STREAM, 0, &sock, 1);
1894         if (status < 0) {
1895                 dprintk("RPC:       can't create AF_LOCAL "
1896                         "transport socket (%d).\n", -status);
1897                 goto out;
1898         }
1899         xs_reclassify_socketu(sock);
1900
1901         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1902                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1903
1904         status = xs_local_finish_connecting(xprt, sock);
1905         trace_rpc_socket_connect(xprt, sock, status);
1906         switch (status) {
1907         case 0:
1908                 dprintk("RPC:       xprt %p connected to %s\n",
1909                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1910                 xprt_set_connected(xprt);
1911                 break;
1912         case -ENOENT:
1913                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1914                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915                 break;
1916         case -ECONNREFUSED:
1917                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1918                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1919                 break;
1920         default:
1921                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1922                                 __func__, -status,
1923                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1924         }
1925
1926 out:
1927         xprt_clear_connecting(xprt);
1928         xprt_wake_pending_tasks(xprt, status);
1929         current->flags &= ~PF_FSTRANS;
1930         return status;
1931 }
1932
1933 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1934 {
1935         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1936         int ret;
1937
1938          if (RPC_IS_ASYNC(task)) {
1939                 /*
1940                  * We want the AF_LOCAL connect to be resolved in the
1941                  * filesystem namespace of the process making the rpc
1942                  * call.  Thus we connect synchronously.
1943                  *
1944                  * If we want to support asynchronous AF_LOCAL calls,
1945                  * we'll need to figure out how to pass a namespace to
1946                  * connect.
1947                  */
1948                 rpc_exit(task, -ENOTCONN);
1949                 return;
1950         }
1951         ret = xs_local_setup_socket(transport);
1952         if (ret && !RPC_IS_SOFTCONN(task))
1953                 msleep_interruptible(15000);
1954 }
1955
1956 #ifdef CONFIG_SUNRPC_SWAP
1957 static void xs_set_memalloc(struct rpc_xprt *xprt)
1958 {
1959         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1960                         xprt);
1961
1962         if (xprt->swapper)
1963                 sk_set_memalloc(transport->inet);
1964 }
1965
1966 /**
1967  * xs_swapper - Tag this transport as being used for swap.
1968  * @xprt: transport to tag
1969  * @enable: enable/disable
1970  *
1971  */
1972 int xs_swapper(struct rpc_xprt *xprt, int enable)
1973 {
1974         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1975                         xprt);
1976         int err = 0;
1977
1978         if (enable) {
1979                 xprt->swapper++;
1980                 xs_set_memalloc(xprt);
1981         } else if (xprt->swapper) {
1982                 xprt->swapper--;
1983                 sk_clear_memalloc(transport->inet);
1984         }
1985
1986         return err;
1987 }
1988 EXPORT_SYMBOL_GPL(xs_swapper);
1989 #else
1990 static void xs_set_memalloc(struct rpc_xprt *xprt)
1991 {
1992 }
1993 #endif
1994
1995 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1996 {
1997         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1998
1999         if (!transport->inet) {
2000                 struct sock *sk = sock->sk;
2001
2002                 write_lock_bh(&sk->sk_callback_lock);
2003
2004                 xs_save_old_callbacks(transport, sk);
2005
2006                 sk->sk_user_data = xprt;
2007                 sk->sk_data_ready = xs_udp_data_ready;
2008                 sk->sk_write_space = xs_udp_write_space;
2009                 sk->sk_no_check = UDP_CSUM_NORCV;
2010                 sk->sk_allocation = GFP_ATOMIC;
2011
2012                 xprt_set_connected(xprt);
2013
2014                 /* Reset to new socket */
2015                 transport->sock = sock;
2016                 transport->inet = sk;
2017
2018                 xs_set_memalloc(xprt);
2019
2020                 write_unlock_bh(&sk->sk_callback_lock);
2021         }
2022         xs_udp_do_set_buffer_size(xprt);
2023 }
2024
2025 static void xs_udp_setup_socket(struct work_struct *work)
2026 {
2027         struct sock_xprt *transport =
2028                 container_of(work, struct sock_xprt, connect_worker.work);
2029         struct rpc_xprt *xprt = &transport->xprt;
2030         struct socket *sock = transport->sock;
2031         int status = -EIO;
2032
2033         current->flags |= PF_FSTRANS;
2034
2035         /* Start by resetting any existing state */
2036         xs_reset_transport(transport);
2037         sock = xs_create_sock(xprt, transport,
2038                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2039         if (IS_ERR(sock))
2040                 goto out;
2041
2042         dprintk("RPC:       worker connecting xprt %p via %s to "
2043                                 "%s (port %s)\n", xprt,
2044                         xprt->address_strings[RPC_DISPLAY_PROTO],
2045                         xprt->address_strings[RPC_DISPLAY_ADDR],
2046                         xprt->address_strings[RPC_DISPLAY_PORT]);
2047
2048         xs_udp_finish_connecting(xprt, sock);
2049         trace_rpc_socket_connect(xprt, sock, 0);
2050         status = 0;
2051 out:
2052         xprt_clear_connecting(xprt);
2053         xprt_wake_pending_tasks(xprt, status);
2054         current->flags &= ~PF_FSTRANS;
2055 }
2056
2057 /*
2058  * We need to preserve the port number so the reply cache on the server can
2059  * find our cached RPC replies when we get around to reconnecting.
2060  */
2061 static void xs_abort_connection(struct sock_xprt *transport)
2062 {
2063         int result;
2064         struct sockaddr any;
2065
2066         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2067
2068         /*
2069          * Disconnect the transport socket by doing a connect operation
2070          * with AF_UNSPEC.  This should return immediately...
2071          */
2072         memset(&any, 0, sizeof(any));
2073         any.sa_family = AF_UNSPEC;
2074         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2075         trace_rpc_socket_reset_connection(&transport->xprt,
2076                         transport->sock, result);
2077         if (!result)
2078                 xs_sock_reset_connection_flags(&transport->xprt);
2079         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2080 }
2081
2082 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2083 {
2084         unsigned int state = transport->inet->sk_state;
2085
2086         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2087                 /* we don't need to abort the connection if the socket
2088                  * hasn't undergone a shutdown
2089                  */
2090                 if (transport->inet->sk_shutdown == 0)
2091                         return;
2092                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2093                                 __func__, transport->inet->sk_shutdown);
2094         }
2095         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2096                 /* we don't need to abort the connection if the socket
2097                  * hasn't undergone a shutdown
2098                  */
2099                 if (transport->inet->sk_shutdown == 0)
2100                         return;
2101                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2102                                 "sk_shutdown set to %d\n",
2103                                 __func__, transport->inet->sk_shutdown);
2104         }
2105         xs_abort_connection(transport);
2106 }
2107
2108 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2109 {
2110         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2111         int ret = -ENOTCONN;
2112
2113         if (!transport->inet) {
2114                 struct sock *sk = sock->sk;
2115
2116                 write_lock_bh(&sk->sk_callback_lock);
2117
2118                 xs_save_old_callbacks(transport, sk);
2119
2120                 sk->sk_user_data = xprt;
2121                 sk->sk_data_ready = xs_tcp_data_ready;
2122                 sk->sk_state_change = xs_tcp_state_change;
2123                 sk->sk_write_space = xs_tcp_write_space;
2124                 sk->sk_allocation = GFP_ATOMIC;
2125
2126                 /* socket options */
2127                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2128                 sock_reset_flag(sk, SOCK_LINGER);
2129                 tcp_sk(sk)->linger2 = 0;
2130                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2131
2132                 xprt_clear_connected(xprt);
2133
2134                 /* Reset to new socket */
2135                 transport->sock = sock;
2136                 transport->inet = sk;
2137
2138                 write_unlock_bh(&sk->sk_callback_lock);
2139         }
2140
2141         if (!xprt_bound(xprt))
2142                 goto out;
2143
2144         xs_set_memalloc(xprt);
2145
2146         /* Tell the socket layer to start connecting... */
2147         xprt->stat.connect_count++;
2148         xprt->stat.connect_start = jiffies;
2149         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2150         switch (ret) {
2151         case 0:
2152         case -EINPROGRESS:
2153                 /* SYN_SENT! */
2154                 xprt->connect_cookie++;
2155                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2156                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2157         }
2158 out:
2159         return ret;
2160 }
2161
2162 /**
2163  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2164  * @xprt: RPC transport to connect
2165  * @transport: socket transport to connect
2166  * @create_sock: function to create a socket of the correct type
2167  *
2168  * Invoked by a work queue tasklet.
2169  */
2170 static void xs_tcp_setup_socket(struct work_struct *work)
2171 {
2172         struct sock_xprt *transport =
2173                 container_of(work, struct sock_xprt, connect_worker.work);
2174         struct socket *sock = transport->sock;
2175         struct rpc_xprt *xprt = &transport->xprt;
2176         int status = -EIO;
2177
2178         current->flags |= PF_FSTRANS;
2179
2180         if (!sock) {
2181                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2182                 sock = xs_create_sock(xprt, transport,
2183                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2184                 if (IS_ERR(sock)) {
2185                         status = PTR_ERR(sock);
2186                         goto out;
2187                 }
2188         } else {
2189                 int abort_and_exit;
2190
2191                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2192                                 &xprt->state);
2193                 /* "close" the socket, preserving the local port */
2194                 xs_tcp_reuse_connection(transport);
2195
2196                 if (abort_and_exit)
2197                         goto out_eagain;
2198         }
2199
2200         dprintk("RPC:       worker connecting xprt %p via %s to "
2201                                 "%s (port %s)\n", xprt,
2202                         xprt->address_strings[RPC_DISPLAY_PROTO],
2203                         xprt->address_strings[RPC_DISPLAY_ADDR],
2204                         xprt->address_strings[RPC_DISPLAY_PORT]);
2205
2206         status = xs_tcp_finish_connecting(xprt, sock);
2207         trace_rpc_socket_connect(xprt, sock, status);
2208         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2209                         xprt, -status, xprt_connected(xprt),
2210                         sock->sk->sk_state);
2211         switch (status) {
2212         default:
2213                 printk("%s: connect returned unhandled error %d\n",
2214                         __func__, status);
2215         case -EADDRNOTAVAIL:
2216                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2217                  * and retry
2218                  */
2219                 xs_tcp_force_close(xprt);
2220                 break;
2221         case 0:
2222         case -EINPROGRESS:
2223         case -EALREADY:
2224                 xprt_clear_connecting(xprt);
2225                 current->flags &= ~PF_FSTRANS;
2226                 return;
2227         case -EINVAL:
2228                 /* Happens, for instance, if the user specified a link
2229                  * local IPv6 address without a scope-id.
2230                  */
2231         case -ECONNREFUSED:
2232         case -ECONNRESET:
2233         case -ENETUNREACH:
2234                 /* retry with existing socket, after a delay */
2235                 goto out;
2236         }
2237 out_eagain:
2238         status = -EAGAIN;
2239 out:
2240         xprt_clear_connecting(xprt);
2241         xprt_wake_pending_tasks(xprt, status);
2242         current->flags &= ~PF_FSTRANS;
2243 }
2244
2245 /**
2246  * xs_connect - connect a socket to a remote endpoint
2247  * @xprt: pointer to transport structure
2248  * @task: address of RPC task that manages state of connect request
2249  *
2250  * TCP: If the remote end dropped the connection, delay reconnecting.
2251  *
2252  * UDP socket connects are synchronous, but we use a work queue anyway
2253  * to guarantee that even unprivileged user processes can set up a
2254  * socket on a privileged port.
2255  *
2256  * If a UDP socket connect fails, the delay behavior here prevents
2257  * retry floods (hard mounts).
2258  */
2259 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2260 {
2261         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2262
2263         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2264                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2265                                 "seconds\n",
2266                                 xprt, xprt->reestablish_timeout / HZ);
2267                 queue_delayed_work(rpciod_workqueue,
2268                                    &transport->connect_worker,
2269                                    xprt->reestablish_timeout);
2270                 xprt->reestablish_timeout <<= 1;
2271                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2272                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2273                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2274                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2275         } else {
2276                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2277                 queue_delayed_work(rpciod_workqueue,
2278                                    &transport->connect_worker, 0);
2279         }
2280 }
2281
2282 /**
2283  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2284  * @xprt: rpc_xprt struct containing statistics
2285  * @seq: output file
2286  *
2287  */
2288 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2289 {
2290         long idle_time = 0;
2291
2292         if (xprt_connected(xprt))
2293                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2294
2295         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2296                         "%llu %llu %lu %llu %llu\n",
2297                         xprt->stat.bind_count,
2298                         xprt->stat.connect_count,
2299                         xprt->stat.connect_time,
2300                         idle_time,
2301                         xprt->stat.sends,
2302                         xprt->stat.recvs,
2303                         xprt->stat.bad_xids,
2304                         xprt->stat.req_u,
2305                         xprt->stat.bklog_u,
2306                         xprt->stat.max_slots,
2307                         xprt->stat.sending_u,
2308                         xprt->stat.pending_u);
2309 }
2310
2311 /**
2312  * xs_udp_print_stats - display UDP socket-specifc stats
2313  * @xprt: rpc_xprt struct containing statistics
2314  * @seq: output file
2315  *
2316  */
2317 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2318 {
2319         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2320
2321         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2322                         "%lu %llu %llu\n",
2323                         transport->srcport,
2324                         xprt->stat.bind_count,
2325                         xprt->stat.sends,
2326                         xprt->stat.recvs,
2327                         xprt->stat.bad_xids,
2328                         xprt->stat.req_u,
2329                         xprt->stat.bklog_u,
2330                         xprt->stat.max_slots,
2331                         xprt->stat.sending_u,
2332                         xprt->stat.pending_u);
2333 }
2334
2335 /**
2336  * xs_tcp_print_stats - display TCP socket-specifc stats
2337  * @xprt: rpc_xprt struct containing statistics
2338  * @seq: output file
2339  *
2340  */
2341 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2342 {
2343         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2344         long idle_time = 0;
2345
2346         if (xprt_connected(xprt))
2347                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2348
2349         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2350                         "%llu %llu %lu %llu %llu\n",
2351                         transport->srcport,
2352                         xprt->stat.bind_count,
2353                         xprt->stat.connect_count,
2354                         xprt->stat.connect_time,
2355                         idle_time,
2356                         xprt->stat.sends,
2357                         xprt->stat.recvs,
2358                         xprt->stat.bad_xids,
2359                         xprt->stat.req_u,
2360                         xprt->stat.bklog_u,
2361                         xprt->stat.max_slots,
2362                         xprt->stat.sending_u,
2363                         xprt->stat.pending_u);
2364 }
2365
2366 /*
2367  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2368  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2369  * to use the server side send routines.
2370  */
2371 static void *bc_malloc(struct rpc_task *task, size_t size)
2372 {
2373         struct page *page;
2374         struct rpc_buffer *buf;
2375
2376         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2377         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2378                 return NULL;
2379
2380         page = alloc_page(GFP_KERNEL);
2381         if (!page)
2382                 return NULL;
2383
2384         buf = page_address(page);
2385         buf->len = PAGE_SIZE;
2386
2387         return buf->data;
2388 }
2389
2390 /*
2391  * Free the space allocated in the bc_alloc routine
2392  */
2393 static void bc_free(void *buffer)
2394 {
2395         struct rpc_buffer *buf;
2396
2397         if (!buffer)
2398                 return;
2399
2400         buf = container_of(buffer, struct rpc_buffer, data);
2401         free_page((unsigned long)buf);
2402 }
2403
2404 /*
2405  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2406  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2407  */
2408 static int bc_sendto(struct rpc_rqst *req)
2409 {
2410         int len;
2411         struct xdr_buf *xbufp = &req->rq_snd_buf;
2412         struct rpc_xprt *xprt = req->rq_xprt;
2413         struct sock_xprt *transport =
2414                                 container_of(xprt, struct sock_xprt, xprt);
2415         struct socket *sock = transport->sock;
2416         unsigned long headoff;
2417         unsigned long tailoff;
2418
2419         xs_encode_stream_record_marker(xbufp);
2420
2421         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2422         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2423         len = svc_send_common(sock, xbufp,
2424                               virt_to_page(xbufp->head[0].iov_base), headoff,
2425                               xbufp->tail[0].iov_base, tailoff);
2426
2427         if (len != xbufp->len) {
2428                 printk(KERN_NOTICE "Error sending entire callback!\n");
2429                 len = -EAGAIN;
2430         }
2431
2432         return len;
2433 }
2434
2435 /*
2436  * The send routine. Borrows from svc_send
2437  */
2438 static int bc_send_request(struct rpc_task *task)
2439 {
2440         struct rpc_rqst *req = task->tk_rqstp;
2441         struct svc_xprt *xprt;
2442         u32                     len;
2443
2444         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2445         /*
2446          * Get the server socket associated with this callback xprt
2447          */
2448         xprt = req->rq_xprt->bc_xprt;
2449
2450         /*
2451          * Grab the mutex to serialize data as the connection is shared
2452          * with the fore channel
2453          */
2454         if (!mutex_trylock(&xprt->xpt_mutex)) {
2455                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2456                 if (!mutex_trylock(&xprt->xpt_mutex))
2457                         return -EAGAIN;
2458                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2459         }
2460         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2461                 len = -ENOTCONN;
2462         else
2463                 len = bc_sendto(req);
2464         mutex_unlock(&xprt->xpt_mutex);
2465
2466         if (len > 0)
2467                 len = 0;
2468
2469         return len;
2470 }
2471
2472 /*
2473  * The close routine. Since this is client initiated, we do nothing
2474  */
2475
2476 static void bc_close(struct rpc_xprt *xprt)
2477 {
2478 }
2479
2480 /*
2481  * The xprt destroy routine. Again, because this connection is client
2482  * initiated, we do nothing
2483  */
2484
2485 static void bc_destroy(struct rpc_xprt *xprt)
2486 {
2487 }
2488
2489 static struct rpc_xprt_ops xs_local_ops = {
2490         .reserve_xprt           = xprt_reserve_xprt,
2491         .release_xprt           = xs_tcp_release_xprt,
2492         .alloc_slot             = xprt_alloc_slot,
2493         .rpcbind                = xs_local_rpcbind,
2494         .set_port               = xs_local_set_port,
2495         .connect                = xs_local_connect,
2496         .buf_alloc              = rpc_malloc,
2497         .buf_free               = rpc_free,
2498         .send_request           = xs_local_send_request,
2499         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2500         .close                  = xs_close,
2501         .destroy                = xs_local_destroy,
2502         .print_stats            = xs_local_print_stats,
2503 };
2504
2505 static struct rpc_xprt_ops xs_udp_ops = {
2506         .set_buffer_size        = xs_udp_set_buffer_size,
2507         .reserve_xprt           = xprt_reserve_xprt_cong,
2508         .release_xprt           = xprt_release_xprt_cong,
2509         .alloc_slot             = xprt_alloc_slot,
2510         .rpcbind                = rpcb_getport_async,
2511         .set_port               = xs_set_port,
2512         .connect                = xs_connect,
2513         .buf_alloc              = rpc_malloc,
2514         .buf_free               = rpc_free,
2515         .send_request           = xs_udp_send_request,
2516         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2517         .timer                  = xs_udp_timer,
2518         .release_request        = xprt_release_rqst_cong,
2519         .close                  = xs_close,
2520         .destroy                = xs_destroy,
2521         .print_stats            = xs_udp_print_stats,
2522 };
2523
2524 static struct rpc_xprt_ops xs_tcp_ops = {
2525         .reserve_xprt           = xprt_reserve_xprt,
2526         .release_xprt           = xs_tcp_release_xprt,
2527         .alloc_slot             = xprt_lock_and_alloc_slot,
2528         .rpcbind                = rpcb_getport_async,
2529         .set_port               = xs_set_port,
2530         .connect                = xs_connect,
2531         .buf_alloc              = rpc_malloc,
2532         .buf_free               = rpc_free,
2533         .send_request           = xs_tcp_send_request,
2534         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2535         .close                  = xs_tcp_close,
2536         .destroy                = xs_destroy,
2537         .print_stats            = xs_tcp_print_stats,
2538 };
2539
2540 /*
2541  * The rpc_xprt_ops for the server backchannel
2542  */
2543
2544 static struct rpc_xprt_ops bc_tcp_ops = {
2545         .reserve_xprt           = xprt_reserve_xprt,
2546         .release_xprt           = xprt_release_xprt,
2547         .alloc_slot             = xprt_alloc_slot,
2548         .buf_alloc              = bc_malloc,
2549         .buf_free               = bc_free,
2550         .send_request           = bc_send_request,
2551         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2552         .close                  = bc_close,
2553         .destroy                = bc_destroy,
2554         .print_stats            = xs_tcp_print_stats,
2555 };
2556
2557 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2558 {
2559         static const struct sockaddr_in sin = {
2560                 .sin_family             = AF_INET,
2561                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2562         };
2563         static const struct sockaddr_in6 sin6 = {
2564                 .sin6_family            = AF_INET6,
2565                 .sin6_addr              = IN6ADDR_ANY_INIT,
2566         };
2567
2568         switch (family) {
2569         case AF_LOCAL:
2570                 break;
2571         case AF_INET:
2572                 memcpy(sap, &sin, sizeof(sin));
2573                 break;
2574         case AF_INET6:
2575                 memcpy(sap, &sin6, sizeof(sin6));
2576                 break;
2577         default:
2578                 dprintk("RPC:       %s: Bad address family\n", __func__);
2579                 return -EAFNOSUPPORT;
2580         }
2581         return 0;
2582 }
2583
2584 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2585                                       unsigned int slot_table_size,
2586                                       unsigned int max_slot_table_size)
2587 {
2588         struct rpc_xprt *xprt;
2589         struct sock_xprt *new;
2590
2591         if (args->addrlen > sizeof(xprt->addr)) {
2592                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2593                 return ERR_PTR(-EBADF);
2594         }
2595
2596         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2597                         max_slot_table_size);
2598         if (xprt == NULL) {
2599                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2600                                 "rpc_xprt\n");
2601                 return ERR_PTR(-ENOMEM);
2602         }
2603
2604         new = container_of(xprt, struct sock_xprt, xprt);
2605         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2606         xprt->addrlen = args->addrlen;
2607         if (args->srcaddr)
2608                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2609         else {
2610                 int err;
2611                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2612                                         (struct sockaddr *)&new->srcaddr);
2613                 if (err != 0) {
2614                         xprt_free(xprt);
2615                         return ERR_PTR(err);
2616                 }
2617         }
2618
2619         return xprt;
2620 }
2621
2622 static const struct rpc_timeout xs_local_default_timeout = {
2623         .to_initval = 10 * HZ,
2624         .to_maxval = 10 * HZ,
2625         .to_retries = 2,
2626 };
2627
2628 /**
2629  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2630  * @args: rpc transport creation arguments
2631  *
2632  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2633  */
2634 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2635 {
2636         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2637         struct sock_xprt *transport;
2638         struct rpc_xprt *xprt;
2639         struct rpc_xprt *ret;
2640
2641         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2642                         xprt_max_tcp_slot_table_entries);
2643         if (IS_ERR(xprt))
2644                 return xprt;
2645         transport = container_of(xprt, struct sock_xprt, xprt);
2646
2647         xprt->prot = 0;
2648         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2649         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2650
2651         xprt->bind_timeout = XS_BIND_TO;
2652         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2653         xprt->idle_timeout = XS_IDLE_DISC_TO;
2654
2655         xprt->ops = &xs_local_ops;
2656         xprt->timeout = &xs_local_default_timeout;
2657
2658         switch (sun->sun_family) {
2659         case AF_LOCAL:
2660                 if (sun->sun_path[0] != '/') {
2661                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2662                                         sun->sun_path);
2663                         ret = ERR_PTR(-EINVAL);
2664                         goto out_err;
2665                 }
2666                 xprt_set_bound(xprt);
2667                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2668                 ret = ERR_PTR(xs_local_setup_socket(transport));
2669                 if (ret)
2670                         goto out_err;
2671                 break;
2672         default:
2673                 ret = ERR_PTR(-EAFNOSUPPORT);
2674                 goto out_err;
2675         }
2676
2677         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2678                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2679
2680         if (try_module_get(THIS_MODULE))
2681                 return xprt;
2682         ret = ERR_PTR(-EINVAL);
2683 out_err:
2684         xprt_free(xprt);
2685         return ret;
2686 }
2687
2688 static const struct rpc_timeout xs_udp_default_timeout = {
2689         .to_initval = 5 * HZ,
2690         .to_maxval = 30 * HZ,
2691         .to_increment = 5 * HZ,
2692         .to_retries = 5,
2693 };
2694
2695 /**
2696  * xs_setup_udp - Set up transport to use a UDP socket
2697  * @args: rpc transport creation arguments
2698  *
2699  */
2700 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2701 {
2702         struct sockaddr *addr = args->dstaddr;
2703         struct rpc_xprt *xprt;
2704         struct sock_xprt *transport;
2705         struct rpc_xprt *ret;
2706
2707         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2708                         xprt_udp_slot_table_entries);
2709         if (IS_ERR(xprt))
2710                 return xprt;
2711         transport = container_of(xprt, struct sock_xprt, xprt);
2712
2713         xprt->prot = IPPROTO_UDP;
2714         xprt->tsh_size = 0;
2715         /* XXX: header size can vary due to auth type, IPv6, etc. */
2716         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2717
2718         xprt->bind_timeout = XS_BIND_TO;
2719         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2720         xprt->idle_timeout = XS_IDLE_DISC_TO;
2721
2722         xprt->ops = &xs_udp_ops;
2723
2724         xprt->timeout = &xs_udp_default_timeout;
2725
2726         switch (addr->sa_family) {
2727         case AF_INET:
2728                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2729                         xprt_set_bound(xprt);
2730
2731                 INIT_DELAYED_WORK(&transport->connect_worker,
2732                                         xs_udp_setup_socket);
2733                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2734                 break;
2735         case AF_INET6:
2736                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2737                         xprt_set_bound(xprt);
2738
2739                 INIT_DELAYED_WORK(&transport->connect_worker,
2740                                         xs_udp_setup_socket);
2741                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2742                 break;
2743         default:
2744                 ret = ERR_PTR(-EAFNOSUPPORT);
2745                 goto out_err;
2746         }
2747
2748         if (xprt_bound(xprt))
2749                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2750                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2751                                 xprt->address_strings[RPC_DISPLAY_PORT],
2752                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2753         else
2754                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2755                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2756                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2757
2758         if (try_module_get(THIS_MODULE))
2759                 return xprt;
2760         ret = ERR_PTR(-EINVAL);
2761 out_err:
2762         xprt_free(xprt);
2763         return ret;
2764 }
2765
2766 static const struct rpc_timeout xs_tcp_default_timeout = {
2767         .to_initval = 60 * HZ,
2768         .to_maxval = 60 * HZ,
2769         .to_retries = 2,
2770 };
2771
2772 /**
2773  * xs_setup_tcp - Set up transport to use a TCP socket
2774  * @args: rpc transport creation arguments
2775  *
2776  */
2777 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2778 {
2779         struct sockaddr *addr = args->dstaddr;
2780         struct rpc_xprt *xprt;
2781         struct sock_xprt *transport;
2782         struct rpc_xprt *ret;
2783         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2784
2785         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2786                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2787
2788         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2789                         max_slot_table_size);
2790         if (IS_ERR(xprt))
2791                 return xprt;
2792         transport = container_of(xprt, struct sock_xprt, xprt);
2793
2794         xprt->prot = IPPROTO_TCP;
2795         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2796         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2797
2798         xprt->bind_timeout = XS_BIND_TO;
2799         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2800         xprt->idle_timeout = XS_IDLE_DISC_TO;
2801
2802         xprt->ops = &xs_tcp_ops;
2803         xprt->timeout = &xs_tcp_default_timeout;
2804
2805         switch (addr->sa_family) {
2806         case AF_INET:
2807                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2808                         xprt_set_bound(xprt);
2809
2810                 INIT_DELAYED_WORK(&transport->connect_worker,
2811                                         xs_tcp_setup_socket);
2812                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2813                 break;
2814         case AF_INET6:
2815                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2816                         xprt_set_bound(xprt);
2817
2818                 INIT_DELAYED_WORK(&transport->connect_worker,
2819                                         xs_tcp_setup_socket);
2820                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2821                 break;
2822         default:
2823                 ret = ERR_PTR(-EAFNOSUPPORT);
2824                 goto out_err;
2825         }
2826
2827         if (xprt_bound(xprt))
2828                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2829                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2830                                 xprt->address_strings[RPC_DISPLAY_PORT],
2831                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2832         else
2833                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2834                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2835                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2836
2837
2838         if (try_module_get(THIS_MODULE))
2839                 return xprt;
2840         ret = ERR_PTR(-EINVAL);
2841 out_err:
2842         xprt_free(xprt);
2843         return ret;
2844 }
2845
2846 /**
2847  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2848  * @args: rpc transport creation arguments
2849  *
2850  */
2851 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2852 {
2853         struct sockaddr *addr = args->dstaddr;
2854         struct rpc_xprt *xprt;
2855         struct sock_xprt *transport;
2856         struct svc_sock *bc_sock;
2857         struct rpc_xprt *ret;
2858
2859         if (args->bc_xprt->xpt_bc_xprt) {
2860                 /*
2861                  * This server connection already has a backchannel
2862                  * export; we can't create a new one, as we wouldn't be
2863                  * able to match replies based on xid any more.  So,
2864                  * reuse the already-existing one:
2865                  */
2866                  return args->bc_xprt->xpt_bc_xprt;
2867         }
2868         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2869                         xprt_tcp_slot_table_entries);
2870         if (IS_ERR(xprt))
2871                 return xprt;
2872         transport = container_of(xprt, struct sock_xprt, xprt);
2873
2874         xprt->prot = IPPROTO_TCP;
2875         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2876         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2877         xprt->timeout = &xs_tcp_default_timeout;
2878
2879         /* backchannel */
2880         xprt_set_bound(xprt);
2881         xprt->bind_timeout = 0;
2882         xprt->reestablish_timeout = 0;
2883         xprt->idle_timeout = 0;
2884
2885         xprt->ops = &bc_tcp_ops;
2886
2887         switch (addr->sa_family) {
2888         case AF_INET:
2889                 xs_format_peer_addresses(xprt, "tcp",
2890                                          RPCBIND_NETID_TCP);
2891                 break;
2892         case AF_INET6:
2893                 xs_format_peer_addresses(xprt, "tcp",
2894                                    RPCBIND_NETID_TCP6);
2895                 break;
2896         default:
2897                 ret = ERR_PTR(-EAFNOSUPPORT);
2898                 goto out_err;
2899         }
2900
2901         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2902                         xprt->address_strings[RPC_DISPLAY_ADDR],
2903                         xprt->address_strings[RPC_DISPLAY_PORT],
2904                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2905
2906         /*
2907          * Once we've associated a backchannel xprt with a connection,
2908          * we want to keep it around as long as long as the connection
2909          * lasts, in case we need to start using it for a backchannel
2910          * again; this reference won't be dropped until bc_xprt is
2911          * destroyed.
2912          */
2913         xprt_get(xprt);
2914         args->bc_xprt->xpt_bc_xprt = xprt;
2915         xprt->bc_xprt = args->bc_xprt;
2916         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2917         transport->sock = bc_sock->sk_sock;
2918         transport->inet = bc_sock->sk_sk;
2919
2920         /*
2921          * Since we don't want connections for the backchannel, we set
2922          * the xprt status to connected
2923          */
2924         xprt_set_connected(xprt);
2925
2926
2927         if (try_module_get(THIS_MODULE))
2928                 return xprt;
2929         xprt_put(xprt);
2930         ret = ERR_PTR(-EINVAL);
2931 out_err:
2932         xprt_free(xprt);
2933         return ret;
2934 }
2935
2936 static struct xprt_class        xs_local_transport = {
2937         .list           = LIST_HEAD_INIT(xs_local_transport.list),
2938         .name           = "named UNIX socket",
2939         .owner          = THIS_MODULE,
2940         .ident          = XPRT_TRANSPORT_LOCAL,
2941         .setup          = xs_setup_local,
2942 };
2943
2944 static struct xprt_class        xs_udp_transport = {
2945         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2946         .name           = "udp",
2947         .owner          = THIS_MODULE,
2948         .ident          = XPRT_TRANSPORT_UDP,
2949         .setup          = xs_setup_udp,
2950 };
2951
2952 static struct xprt_class        xs_tcp_transport = {
2953         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2954         .name           = "tcp",
2955         .owner          = THIS_MODULE,
2956         .ident          = XPRT_TRANSPORT_TCP,
2957         .setup          = xs_setup_tcp,
2958 };
2959
2960 static struct xprt_class        xs_bc_tcp_transport = {
2961         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2962         .name           = "tcp NFSv4.1 backchannel",
2963         .owner          = THIS_MODULE,
2964         .ident          = XPRT_TRANSPORT_BC_TCP,
2965         .setup          = xs_setup_bc_tcp,
2966 };
2967
2968 /**
2969  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2970  *
2971  */
2972 int init_socket_xprt(void)
2973 {
2974 #ifdef RPC_DEBUG
2975         if (!sunrpc_table_header)
2976                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2977 #endif
2978
2979         xprt_register_transport(&xs_local_transport);
2980         xprt_register_transport(&xs_udp_transport);
2981         xprt_register_transport(&xs_tcp_transport);
2982         xprt_register_transport(&xs_bc_tcp_transport);
2983
2984         return 0;
2985 }
2986
2987 /**
2988  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2989  *
2990  */
2991 void cleanup_socket_xprt(void)
2992 {
2993 #ifdef RPC_DEBUG
2994         if (sunrpc_table_header) {
2995                 unregister_sysctl_table(sunrpc_table_header);
2996                 sunrpc_table_header = NULL;
2997         }
2998 #endif
2999
3000         xprt_unregister_transport(&xs_local_transport);
3001         xprt_unregister_transport(&xs_udp_transport);
3002         xprt_unregister_transport(&xs_tcp_transport);
3003         xprt_unregister_transport(&xs_bc_tcp_transport);
3004 }
3005
3006 static int param_set_uint_minmax(const char *val,
3007                 const struct kernel_param *kp,
3008                 unsigned int min, unsigned int max)
3009 {
3010         unsigned long num;
3011         int ret;
3012
3013         if (!val)
3014                 return -EINVAL;
3015         ret = strict_strtoul(val, 0, &num);
3016         if (ret == -EINVAL || num < min || num > max)
3017                 return -EINVAL;
3018         *((unsigned int *)kp->arg) = num;
3019         return 0;
3020 }
3021
3022 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3023 {
3024         return param_set_uint_minmax(val, kp,
3025                         RPC_MIN_RESVPORT,
3026                         RPC_MAX_RESVPORT);
3027 }
3028
3029 static struct kernel_param_ops param_ops_portnr = {
3030         .set = param_set_portnr,
3031         .get = param_get_uint,
3032 };
3033
3034 #define param_check_portnr(name, p) \
3035         __param_check(name, p, unsigned int);
3036
3037 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3038 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3039
3040 static int param_set_slot_table_size(const char *val,
3041                                      const struct kernel_param *kp)
3042 {
3043         return param_set_uint_minmax(val, kp,
3044                         RPC_MIN_SLOT_TABLE,
3045                         RPC_MAX_SLOT_TABLE);
3046 }
3047
3048 static struct kernel_param_ops param_ops_slot_table_size = {
3049         .set = param_set_slot_table_size,
3050         .get = param_get_uint,
3051 };
3052
3053 #define param_check_slot_table_size(name, p) \
3054         __param_check(name, p, unsigned int);
3055
3056 static int param_set_max_slot_table_size(const char *val,
3057                                      const struct kernel_param *kp)
3058 {
3059         return param_set_uint_minmax(val, kp,
3060                         RPC_MIN_SLOT_TABLE,
3061                         RPC_MAX_SLOT_TABLE_LIMIT);
3062 }
3063
3064 static struct kernel_param_ops param_ops_max_slot_table_size = {
3065         .set = param_set_max_slot_table_size,
3066         .get = param_get_uint,
3067 };
3068
3069 #define param_check_max_slot_table_size(name, p) \
3070         __param_check(name, p, unsigned int);
3071
3072 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3073                    slot_table_size, 0644);
3074 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3075                    max_slot_table_size, 0644);
3076 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3077                    slot_table_size, 0644);
3078