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 Software <alan@redhat.com>
   7  * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
   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
  18 #include <linux/types.h>
  19 #include <linux/slab.h>
  20 #include <linux/capability.h>
  21 #include <linux/sched.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/errno.h>
  24 #include <linux/socket.h>
  25 #include <linux/in.h>
  26 #include <linux/net.h>
  27 #include <linux/mm.h>
  28 #include <linux/udp.h>
  29 #include <linux/tcp.h>
  30 #include <linux/sunrpc/clnt.h>
  31 #include <linux/sunrpc/sched.h>
  32 #include <linux/file.h>
  33
  34 #include <net/sock.h>
  35 #include <net/checksum.h>
  36 #include <net/udp.h>
  37 #include <net/tcp.h>
  38
  39 /*
  40  * xprtsock tunables
  41  */
  42 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  43 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  44
  45 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  46 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  47
  48 /*
  49  * How many times to try sending a request on a socket before waiting
  50  * for the socket buffer to clear.
  51  */
  52 #define XS_SENDMSG_RETRY        (10U)
  53
  54 /*
  55  * Time out for an RPC UDP socket connect.  UDP socket connects are
  56  * synchronous, but we set a timeout anyway in case of resource
  57  * exhaustion on the local host.
  58  */
  59 #define XS_UDP_CONN_TO          (5U * HZ)
  60
  61 /*
  62  * Wait duration for an RPC TCP connection to be established.  Solaris
  63  * NFS over TCP uses 60 seconds, for example, which is in line with how
  64  * long a server takes to reboot.
  65  */
  66 #define XS_TCP_CONN_TO          (60U * HZ)
  67
  68 /*
  69  * Wait duration for a reply from the RPC portmapper.
  70  */
  71 #define XS_BIND_TO              (60U * HZ)
  72
  73 /*
  74  * Delay if a UDP socket connect error occurs.  This is most likely some
  75  * kind of resource problem on the local host.
  76  */
  77 #define XS_UDP_REEST_TO         (2U * HZ)
  78
  79 /*
  80  * The reestablish timeout allows clients to delay for a bit before attempting
  81  * to reconnect to a server that just dropped our connection.
  82  *
  83  * We implement an exponential backoff when trying to reestablish a TCP
  84  * transport connection with the server.  Some servers like to drop a TCP
  85  * connection when they are overworked, so we start with a short timeout and
  86  * increase over time if the server is down or not responding.
  87  */
  88 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
  89 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
  90
  91 /*
  92  * TCP idle timeout; client drops the transport socket if it is idle
  93  * for this long.  Note that we also timeout UDP sockets to prevent
  94  * holding port numbers when there is no RPC traffic.
  95  */
  96 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
  97
  98 #ifdef RPC_DEBUG
  99 # undef  RPC_DEBUG_DATA
 100 # define RPCDBG_FACILITY        RPCDBG_TRANS
 101 #endif
 102
 103 #ifdef RPC_DEBUG_DATA
 104 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 105 {
 106         u8 *buf = (u8 *) packet;
 107         int j;
 108
 109         dprintk("RPC:      %s\n", msg);
 110         for (j = 0; j < count && j < 128; j += 4) {
 111                 if (!(j & 31)) {
 112                         if (j)
 113                                 dprintk("\n");
 114                         dprintk("0x%04x ", j);
 115                 }
 116                 dprintk("%02x%02x%02x%02x ",
 117                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 118         }
 119         dprintk("\n");
 120 }
 121 #else
 122 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 123 {
 124         /* NOP */
 125 }
 126 #endif
 127
 128 struct sock_xprt {
 129         struct rpc_xprt         xprt;
 130
 131         /*
 132          * Network layer
 133          */
 134         struct socket *         sock;
 135         struct sock *           inet;
 136
 137         /*
 138          * State of TCP reply receive
 139          */
 140         __be32                  tcp_fraghdr,
 141                                 tcp_xid;
 142
 143         u32                     tcp_offset,
 144                                 tcp_reclen;
 145
 146         unsigned long           tcp_copied,
 147                                 tcp_flags;
 148
 149         /*
 150          * Connection of transports
 151          */
 152         struct work_struct      connect_worker;
 153         unsigned short          port;
 154
 155         /*
 156          * UDP socket buffer size parameters
 157          */
 158         size_t                  rcvsize,
 159                                 sndsize;
 160
 161         /*
 162          * Saved socket callback addresses
 163          */
 164         void                    (*old_data_ready)(struct sock *, int);
 165         void                    (*old_state_change)(struct sock *);
 166         void                    (*old_write_space)(struct sock *);
 167 };
 168
 169 /*
 170  * TCP receive state flags
 171  */
 172 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 173 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 174 #define TCP_RCV_COPY_XID        (1UL << 2)
 175 #define TCP_RCV_COPY_DATA       (1UL << 3)
 176
 177 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
 178 {
 179         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
 180         char *buf;
 181
 182         buf = kzalloc(20, GFP_KERNEL);
 183         if (buf) {
 184                 snprintf(buf, 20, "%u.%u.%u.%u",
 185                                 NIPQUAD(addr->sin_addr.s_addr));
 186         }
 187         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 188
 189         buf = kzalloc(8, GFP_KERNEL);
 190         if (buf) {
 191                 snprintf(buf, 8, "%u",
 192                                 ntohs(addr->sin_port));
 193         }
 194         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 195
 196         if (xprt->prot == IPPROTO_UDP)
 197                 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
 198         else
 199                 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
 200
 201         buf = kzalloc(48, GFP_KERNEL);
 202         if (buf) {
 203                 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
 204                         NIPQUAD(addr->sin_addr.s_addr),
 205                         ntohs(addr->sin_port),
 206                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 207         }
 208         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 209 }
 210
 211 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 212 {
 213         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
 214         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 215         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
 216 }
 217
 218 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 219
 220 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 221 {
 222         struct msghdr msg = {
 223                 .msg_name       = addr,
 224                 .msg_namelen    = addrlen,
 225                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 226         };
 227         struct kvec iov = {
 228                 .iov_base       = vec->iov_base + base,
 229                 .iov_len        = vec->iov_len - base,
 230         };
 231
 232         if (iov.iov_len != 0)
 233                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 234         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 235 }
 236
 237 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 238 {
 239         struct page **ppage;
 240         unsigned int remainder;
 241         int err, sent = 0;
 242
 243         remainder = xdr->page_len - base;
 244         base += xdr->page_base;
 245         ppage = xdr->pages + (base >> PAGE_SHIFT);
 246         base &= ~PAGE_MASK;
 247         for(;;) {
 248                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 249                 int flags = XS_SENDMSG_FLAGS;
 250
 251                 remainder -= len;
 252                 if (remainder != 0 || more)
 253                         flags |= MSG_MORE;
 254                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 255                 if (remainder == 0 || err != len)
 256                         break;
 257                 sent += err;
 258                 ppage++;
 259                 base = 0;
 260         }
 261         if (sent == 0)
 262                 return err;
 263         if (err > 0)
 264                 sent += err;
 265         return sent;
 266 }
 267
 268 /**
 269  * xs_sendpages - write pages directly to a socket
 270  * @sock: socket to send on
 271  * @addr: UDP only -- address of destination
 272  * @addrlen: UDP only -- length of destination address
 273  * @xdr: buffer containing this request
 274  * @base: starting position in the buffer
 275  *
 276  */
 277 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 278 {
 279         unsigned int remainder = xdr->len - base;
 280         int err, sent = 0;
 281
 282         if (unlikely(!sock))
 283                 return -ENOTCONN;
 284
 285         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 286         if (base != 0) {
 287                 addr = NULL;
 288                 addrlen = 0;
 289         }
 290
 291         if (base < xdr->head[0].iov_len || addr != NULL) {
 292                 unsigned int len = xdr->head[0].iov_len - base;
 293                 remainder -= len;
 294                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 295                 if (remainder == 0 || err != len)
 296                         goto out;
 297                 sent += err;
 298                 base = 0;
 299         } else
 300                 base -= xdr->head[0].iov_len;
 301
 302         if (base < xdr->page_len) {
 303                 unsigned int len = xdr->page_len - base;
 304                 remainder -= len;
 305                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 306                 if (remainder == 0 || err != len)
 307                         goto out;
 308                 sent += err;
 309                 base = 0;
 310         } else
 311                 base -= xdr->page_len;
 312
 313         if (base >= xdr->tail[0].iov_len)
 314                 return sent;
 315         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 316 out:
 317         if (sent == 0)
 318                 return err;
 319         if (err > 0)
 320                 sent += err;
 321         return sent;
 322 }
 323
 324 /**
 325  * xs_nospace - place task on wait queue if transmit was incomplete
 326  * @task: task to put to sleep
 327  *
 328  */
 329 static void xs_nospace(struct rpc_task *task)
 330 {
 331         struct rpc_rqst *req = task->tk_rqstp;
 332         struct rpc_xprt *xprt = req->rq_xprt;
 333         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 334
 335         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 336                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 337                         req->rq_slen);
 338
 339         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 340                 /* Protect against races with write_space */
 341                 spin_lock_bh(&xprt->transport_lock);
 342
 343                 /* Don't race with disconnect */
 344                 if (!xprt_connected(xprt))
 345                         task->tk_status = -ENOTCONN;
 346                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 347                         xprt_wait_for_buffer_space(task);
 348
 349                 spin_unlock_bh(&xprt->transport_lock);
 350         } else
 351                 /* Keep holding the socket if it is blocked */
 352                 rpc_delay(task, HZ>>4);
 353 }
 354
 355 /**
 356  * xs_udp_send_request - write an RPC request to a UDP socket
 357  * @task: address of RPC task that manages the state of an RPC request
 358  *
 359  * Return values:
 360  *        0:    The request has been sent
 361  *   EAGAIN:    The socket was blocked, please call again later to
 362  *              complete the request
 363  * ENOTCONN:    Caller needs to invoke connect logic then call again
 364  *    other:    Some other error occured, the request was not sent
 365  */
 366 static int xs_udp_send_request(struct rpc_task *task)
 367 {
 368         struct rpc_rqst *req = task->tk_rqstp;
 369         struct rpc_xprt *xprt = req->rq_xprt;
 370         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 371         struct xdr_buf *xdr = &req->rq_snd_buf;
 372         int status;
 373
 374         xs_pktdump("packet data:",
 375                                 req->rq_svec->iov_base,
 376                                 req->rq_svec->iov_len);
 377
 378         req->rq_xtime = jiffies;
 379         status = xs_sendpages(transport->sock,
 380                               (struct sockaddr *) &xprt->addr,
 381                               xprt->addrlen, xdr,
 382                               req->rq_bytes_sent);
 383
 384         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 385                         xdr->len - req->rq_bytes_sent, status);
 386
 387         if (likely(status >= (int) req->rq_slen))
 388                 return 0;
 389
 390         /* Still some bytes left; set up for a retry later. */
 391         if (status > 0)
 392                 status = -EAGAIN;
 393
 394         switch (status) {
 395         case -ENETUNREACH:
 396         case -EPIPE:
 397         case -ECONNREFUSED:
 398                 /* When the server has died, an ICMP port unreachable message
 399                  * prompts ECONNREFUSED. */
 400                 break;
 401         case -EAGAIN:
 402                 xs_nospace(task);
 403                 break;
 404         default:
 405                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 406                         -status);
 407                 break;
 408         }
 409
 410         return status;
 411 }
 412
 413 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 414 {
 415         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 416         rpc_fraghdr *base = buf->head[0].iov_base;
 417         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 418 }
 419
 420 /**
 421  * xs_tcp_send_request - write an RPC request to a TCP socket
 422  * @task: address of RPC task that manages the state of an RPC request
 423  *
 424  * Return values:
 425  *        0:    The request has been sent
 426  *   EAGAIN:    The socket was blocked, please call again later to
 427  *              complete the request
 428  * ENOTCONN:    Caller needs to invoke connect logic then call again
 429  *    other:    Some other error occured, the request was not sent
 430  *
 431  * XXX: In the case of soft timeouts, should we eventually give up
 432  *      if sendmsg is not able to make progress?
 433  */
 434 static int xs_tcp_send_request(struct rpc_task *task)
 435 {
 436         struct rpc_rqst *req = task->tk_rqstp;
 437         struct rpc_xprt *xprt = req->rq_xprt;
 438         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 439         struct xdr_buf *xdr = &req->rq_snd_buf;
 440         int status, retry = 0;
 441
 442         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 443
 444         xs_pktdump("packet data:",
 445                                 req->rq_svec->iov_base,
 446                                 req->rq_svec->iov_len);
 447
 448         /* Continue transmitting the packet/record. We must be careful
 449          * to cope with writespace callbacks arriving _after_ we have
 450          * called sendmsg(). */
 451         while (1) {
 452                 req->rq_xtime = jiffies;
 453                 status = xs_sendpages(transport->sock,
 454                                         NULL, 0, xdr, req->rq_bytes_sent);
 455
 456                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 457                                 xdr->len - req->rq_bytes_sent, status);
 458
 459                 if (unlikely(status < 0))
 460                         break;
 461
 462                 /* If we've sent the entire packet, immediately
 463                  * reset the count of bytes sent. */
 464                 req->rq_bytes_sent += status;
 465                 task->tk_bytes_sent += status;
 466                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 467                         req->rq_bytes_sent = 0;
 468                         return 0;
 469                 }
 470
 471                 status = -EAGAIN;
 472                 if (retry++ > XS_SENDMSG_RETRY)
 473                         break;
 474         }
 475
 476         switch (status) {
 477         case -EAGAIN:
 478                 xs_nospace(task);
 479                 break;
 480         case -ECONNREFUSED:
 481         case -ECONNRESET:
 482         case -ENOTCONN:
 483         case -EPIPE:
 484                 status = -ENOTCONN;
 485                 break;
 486         default:
 487                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 488                         -status);
 489                 xprt_disconnect(xprt);
 490                 break;
 491         }
 492
 493         return status;
 494 }
 495
 496 /**
 497  * xs_tcp_release_xprt - clean up after a tcp transmission
 498  * @xprt: transport
 499  * @task: rpc task
 500  *
 501  * This cleans up if an error causes us to abort the transmission of a request.
 502  * In this case, the socket may need to be reset in order to avoid confusing
 503  * the server.
 504  */
 505 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 506 {
 507         struct rpc_rqst *req;
 508
 509         if (task != xprt->snd_task)
 510                 return;
 511         if (task == NULL)
 512                 goto out_release;
 513         req = task->tk_rqstp;
 514         if (req->rq_bytes_sent == 0)
 515                 goto out_release;
 516         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 517                 goto out_release;
 518         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 519 out_release:
 520         xprt_release_xprt(xprt, task);
 521 }
 522
 523 /**
 524  * xs_close - close a socket
 525  * @xprt: transport
 526  *
 527  * This is used when all requests are complete; ie, no DRC state remains
 528  * on the server we want to save.
 529  */
 530 static void xs_close(struct rpc_xprt *xprt)
 531 {
 532         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 533         struct socket *sock = transport->sock;
 534         struct sock *sk = transport->inet;
 535
 536         if (!sk)
 537                 goto clear_close_wait;
 538
 539         dprintk("RPC:      xs_close xprt %p\n", xprt);
 540
 541         write_lock_bh(&sk->sk_callback_lock);
 542         transport->inet = NULL;
 543         transport->sock = NULL;
 544
 545         sk->sk_user_data = NULL;
 546         sk->sk_data_ready = transport->old_data_ready;
 547         sk->sk_state_change = transport->old_state_change;
 548         sk->sk_write_space = transport->old_write_space;
 549         write_unlock_bh(&sk->sk_callback_lock);
 550
 551         sk->sk_no_check = 0;
 552
 553         sock_release(sock);
 554 clear_close_wait:
 555         smp_mb__before_clear_bit();
 556         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 557         smp_mb__after_clear_bit();
 558 }
 559
 560 /**
 561  * xs_destroy - prepare to shutdown a transport
 562  * @xprt: doomed transport
 563  *
 564  */
 565 static void xs_destroy(struct rpc_xprt *xprt)
 566 {
 567         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 568
 569         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 570
 571         cancel_delayed_work(&transport->connect_worker);
 572         flush_scheduled_work();
 573
 574         xprt_disconnect(xprt);
 575         xs_close(xprt);
 576         xs_free_peer_addresses(xprt);
 577         kfree(xprt->slot);
 578         kfree(xprt);
 579 }
 580
 581 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 582 {
 583         return (struct rpc_xprt *) sk->sk_user_data;
 584 }
 585
 586 /**
 587  * xs_udp_data_ready - "data ready" callback for UDP sockets
 588  * @sk: socket with data to read
 589  * @len: how much data to read
 590  *
 591  */
 592 static void xs_udp_data_ready(struct sock *sk, int len)
 593 {
 594         struct rpc_task *task;
 595         struct rpc_xprt *xprt;
 596         struct rpc_rqst *rovr;
 597         struct sk_buff *skb;
 598         int err, repsize, copied;
 599         u32 _xid;
 600         __be32 *xp;
 601
 602         read_lock(&sk->sk_callback_lock);
 603         dprintk("RPC:      xs_udp_data_ready...\n");
 604         if (!(xprt = xprt_from_sock(sk)))
 605                 goto out;
 606
 607         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 608                 goto out;
 609
 610         if (xprt->shutdown)
 611                 goto dropit;
 612
 613         repsize = skb->len - sizeof(struct udphdr);
 614         if (repsize < 4) {
 615                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 616                 goto dropit;
 617         }
 618
 619         /* Copy the XID from the skb... */
 620         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 621                                 sizeof(_xid), &_xid);
 622         if (xp == NULL)
 623                 goto dropit;
 624
 625         /* Look up and lock the request corresponding to the given XID */
 626         spin_lock(&xprt->transport_lock);
 627         rovr = xprt_lookup_rqst(xprt, *xp);
 628         if (!rovr)
 629                 goto out_unlock;
 630         task = rovr->rq_task;
 631
 632         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 633                 copied = repsize;
 634
 635         /* Suck it into the iovec, verify checksum if not done by hw. */
 636         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 637                 goto out_unlock;
 638
 639         /* Something worked... */
 640         dst_confirm(skb->dst);
 641
 642         xprt_adjust_cwnd(task, copied);
 643         xprt_update_rtt(task);
 644         xprt_complete_rqst(task, copied);
 645
 646  out_unlock:
 647         spin_unlock(&xprt->transport_lock);
 648  dropit:
 649         skb_free_datagram(sk, skb);
 650  out:
 651         read_unlock(&sk->sk_callback_lock);
 652 }
 653
 654 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 655 {
 656         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 657         size_t len, used;
 658         char *p;
 659
 660         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 661         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 662         used = xdr_skb_read_bits(desc, p, len);
 663         transport->tcp_offset += used;
 664         if (used != len)
 665                 return;
 666
 667         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 668         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 669                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 670         else
 671                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 672         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 673
 674         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 675         transport->tcp_offset = 0;
 676
 677         /* Sanity check of the record length */
 678         if (unlikely(transport->tcp_reclen < 4)) {
 679                 dprintk("RPC:      invalid TCP record fragment length\n");
 680                 xprt_disconnect(xprt);
 681                 return;
 682         }
 683         dprintk("RPC:      reading TCP record fragment of length %d\n",
 684                         transport->tcp_reclen);
 685 }
 686
 687 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 688 {
 689         if (transport->tcp_offset == transport->tcp_reclen) {
 690                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 691                 transport->tcp_offset = 0;
 692                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 693                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 694                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 695                         transport->tcp_copied = 0;
 696                 }
 697         }
 698 }
 699
 700 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 701 {
 702         size_t len, used;
 703         char *p;
 704
 705         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 706         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 707         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 708         used = xdr_skb_read_bits(desc, p, len);
 709         transport->tcp_offset += used;
 710         if (used != len)
 711                 return;
 712         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 713         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 714         transport->tcp_copied = 4;
 715         dprintk("RPC:      reading reply for XID %08x\n",
 716                         ntohl(transport->tcp_xid));
 717         xs_tcp_check_fraghdr(transport);
 718 }
 719
 720 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 721 {
 722         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 723         struct rpc_rqst *req;
 724         struct xdr_buf *rcvbuf;
 725         size_t len;
 726         ssize_t r;
 727
 728         /* Find and lock the request corresponding to this xid */
 729         spin_lock(&xprt->transport_lock);
 730         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 731         if (!req) {
 732                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 733                 dprintk("RPC:      XID %08x request not found!\n",
 734                                 ntohl(transport->tcp_xid));
 735                 spin_unlock(&xprt->transport_lock);
 736                 return;
 737         }
 738
 739         rcvbuf = &req->rq_private_buf;
 740         len = desc->count;
 741         if (len > transport->tcp_reclen - transport->tcp_offset) {
 742                 struct xdr_skb_reader my_desc;
 743
 744                 len = transport->tcp_reclen - transport->tcp_offset;
 745                 memcpy(&my_desc, desc, sizeof(my_desc));
 746                 my_desc.count = len;
 747                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 748                                           &my_desc, xdr_skb_read_bits);
 749                 desc->count -= r;
 750                 desc->offset += r;
 751         } else
 752                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 753                                           desc, xdr_skb_read_bits);
 754
 755         if (r > 0) {
 756                 transport->tcp_copied += r;
 757                 transport->tcp_offset += r;
 758         }
 759         if (r != len) {
 760                 /* Error when copying to the receive buffer,
 761                  * usually because we weren't able to allocate
 762                  * additional buffer pages. All we can do now
 763                  * is turn off TCP_RCV_COPY_DATA, so the request
 764                  * will not receive any additional updates,
 765                  * and time out.
 766                  * Any remaining data from this record will
 767                  * be discarded.
 768                  */
 769                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 770                 dprintk("RPC:      XID %08x truncated request\n",
 771                                 ntohl(transport->tcp_xid));
 772                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 773                                 xprt, transport->tcp_copied, transport->tcp_offset,
 774                                         transport->tcp_reclen);
 775                 goto out;
 776         }
 777
 778         dprintk("RPC:      XID %08x read %Zd bytes\n",
 779                         ntohl(transport->tcp_xid), r);
 780         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 781                         xprt, transport->tcp_copied, transport->tcp_offset,
 782                                 transport->tcp_reclen);
 783
 784         if (transport->tcp_copied == req->rq_private_buf.buflen)
 785                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 786         else if (transport->tcp_offset == transport->tcp_reclen) {
 787                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
 788                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 789         }
 790
 791 out:
 792         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
 793                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
 794         spin_unlock(&xprt->transport_lock);
 795         xs_tcp_check_fraghdr(transport);
 796 }
 797
 798 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 799 {
 800         size_t len;
 801
 802         len = transport->tcp_reclen - transport->tcp_offset;
 803         if (len > desc->count)
 804                 len = desc->count;
 805         desc->count -= len;
 806         desc->offset += len;
 807         transport->tcp_offset += len;
 808         dprintk("RPC:      discarded %Zu bytes\n", len);
 809         xs_tcp_check_fraghdr(transport);
 810 }
 811
 812 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 813 {
 814         struct rpc_xprt *xprt = rd_desc->arg.data;
 815         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 816         struct xdr_skb_reader desc = {
 817                 .skb    = skb,
 818                 .offset = offset,
 819                 .count  = len,
 820         };
 821
 822         dprintk("RPC:      xs_tcp_data_recv started\n");
 823         do {
 824                 /* Read in a new fragment marker if necessary */
 825                 /* Can we ever really expect to get completely empty fragments? */
 826                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
 827                         xs_tcp_read_fraghdr(xprt, &desc);
 828                         continue;
 829                 }
 830                 /* Read in the xid if necessary */
 831                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
 832                         xs_tcp_read_xid(transport, &desc);
 833                         continue;
 834                 }
 835                 /* Read in the request data */
 836                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
 837                         xs_tcp_read_request(xprt, &desc);
 838                         continue;
 839                 }
 840                 /* Skip over any trailing bytes on short reads */
 841                 xs_tcp_read_discard(transport, &desc);
 842         } while (desc.count);
 843         dprintk("RPC:      xs_tcp_data_recv done\n");
 844         return len - desc.count;
 845 }
 846
 847 /**
 848  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 849  * @sk: socket with data to read
 850  * @bytes: how much data to read
 851  *
 852  */
 853 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 854 {
 855         struct rpc_xprt *xprt;
 856         read_descriptor_t rd_desc;
 857
 858         read_lock(&sk->sk_callback_lock);
 859         dprintk("RPC:      xs_tcp_data_ready...\n");
 860         if (!(xprt = xprt_from_sock(sk)))
 861                 goto out;
 862         if (xprt->shutdown)
 863                 goto out;
 864
 865         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 866         rd_desc.arg.data = xprt;
 867         rd_desc.count = 65536;
 868         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 869 out:
 870         read_unlock(&sk->sk_callback_lock);
 871 }
 872
 873 /**
 874  * xs_tcp_state_change - callback to handle TCP socket state changes
 875  * @sk: socket whose state has changed
 876  *
 877  */
 878 static void xs_tcp_state_change(struct sock *sk)
 879 {
 880         struct rpc_xprt *xprt;
 881
 882         read_lock(&sk->sk_callback_lock);
 883         if (!(xprt = xprt_from_sock(sk)))
 884                 goto out;
 885         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 886         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 887                                 sk->sk_state, xprt_connected(xprt),
 888                                 sock_flag(sk, SOCK_DEAD),
 889                                 sock_flag(sk, SOCK_ZAPPED));
 890
 891         switch (sk->sk_state) {
 892         case TCP_ESTABLISHED:
 893                 spin_lock_bh(&xprt->transport_lock);
 894                 if (!xprt_test_and_set_connected(xprt)) {
 895                         struct sock_xprt *transport = container_of(xprt,
 896                                         struct sock_xprt, xprt);
 897
 898                         /* Reset TCP record info */
 899                         transport->tcp_offset = 0;
 900                         transport->tcp_reclen = 0;
 901                         transport->tcp_copied = 0;
 902                         transport->tcp_flags =
 903                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
 904
 905                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 906                         xprt_wake_pending_tasks(xprt, 0);
 907                 }
 908                 spin_unlock_bh(&xprt->transport_lock);
 909                 break;
 910         case TCP_SYN_SENT:
 911         case TCP_SYN_RECV:
 912                 break;
 913         case TCP_CLOSE_WAIT:
 914                 /* Try to schedule an autoclose RPC calls */
 915                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 916                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 917                         schedule_work(&xprt->task_cleanup);
 918         default:
 919                 xprt_disconnect(xprt);
 920         }
 921  out:
 922         read_unlock(&sk->sk_callback_lock);
 923 }
 924
 925 /**
 926  * xs_udp_write_space - callback invoked when socket buffer space
 927  *                             becomes available
 928  * @sk: socket whose state has changed
 929  *
 930  * Called when more output buffer space is available for this socket.
 931  * We try not to wake our writers until they can make "significant"
 932  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 933  * with a bunch of small requests.
 934  */
 935 static void xs_udp_write_space(struct sock *sk)
 936 {
 937         read_lock(&sk->sk_callback_lock);
 938
 939         /* from net/core/sock.c:sock_def_write_space */
 940         if (sock_writeable(sk)) {
 941                 struct socket *sock;
 942                 struct rpc_xprt *xprt;
 943
 944                 if (unlikely(!(sock = sk->sk_socket)))
 945                         goto out;
 946                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 947                         goto out;
 948                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 949                         goto out;
 950
 951                 xprt_write_space(xprt);
 952         }
 953
 954  out:
 955         read_unlock(&sk->sk_callback_lock);
 956 }
 957
 958 /**
 959  * xs_tcp_write_space - callback invoked when socket buffer space
 960  *                             becomes available
 961  * @sk: socket whose state has changed
 962  *
 963  * Called when more output buffer space is available for this socket.
 964  * We try not to wake our writers until they can make "significant"
 965  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 966  * with a bunch of small requests.
 967  */
 968 static void xs_tcp_write_space(struct sock *sk)
 969 {
 970         read_lock(&sk->sk_callback_lock);
 971
 972         /* from net/core/stream.c:sk_stream_write_space */
 973         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 974                 struct socket *sock;
 975                 struct rpc_xprt *xprt;
 976
 977                 if (unlikely(!(sock = sk->sk_socket)))
 978                         goto out;
 979                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 980                         goto out;
 981                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 982                         goto out;
 983
 984                 xprt_write_space(xprt);
 985         }
 986
 987  out:
 988         read_unlock(&sk->sk_callback_lock);
 989 }
 990
 991 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 992 {
 993         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 994         struct sock *sk = transport->inet;
 995
 996         if (transport->rcvsize) {
 997                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 998                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
 999         }
1000         if (transport->sndsize) {
1001                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1002                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1003                 sk->sk_write_space(sk);
1004         }
1005 }
1006
1007 /**
1008  * xs_udp_set_buffer_size - set send and receive limits
1009  * @xprt: generic transport
1010  * @sndsize: requested size of send buffer, in bytes
1011  * @rcvsize: requested size of receive buffer, in bytes
1012  *
1013  * Set socket send and receive buffer size limits.
1014  */
1015 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1016 {
1017         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1018
1019         transport->sndsize = 0;
1020         if (sndsize)
1021                 transport->sndsize = sndsize + 1024;
1022         transport->rcvsize = 0;
1023         if (rcvsize)
1024                 transport->rcvsize = rcvsize + 1024;
1025
1026         xs_udp_do_set_buffer_size(xprt);
1027 }
1028
1029 /**
1030  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1031  * @task: task that timed out
1032  *
1033  * Adjust the congestion window after a retransmit timeout has occurred.
1034  */
1035 static void xs_udp_timer(struct rpc_task *task)
1036 {
1037         xprt_adjust_cwnd(task, -ETIMEDOUT);
1038 }
1039
1040 static unsigned short xs_get_random_port(void)
1041 {
1042         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1043         unsigned short rand = (unsigned short) net_random() % range;
1044         return rand + xprt_min_resvport;
1045 }
1046
1047 /**
1048  * xs_set_port - reset the port number in the remote endpoint address
1049  * @xprt: generic transport
1050  * @port: new port number
1051  *
1052  */
1053 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1054 {
1055         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1056
1057         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
1058
1059         sap->sin_port = htons(port);
1060 }
1061
1062 static int xs_bindresvport(struct sock_xprt *transport, struct socket *sock)
1063 {
1064         struct sockaddr_in myaddr = {
1065                 .sin_family = AF_INET,
1066         };
1067         int err;
1068         unsigned short port = transport->port;
1069
1070         do {
1071                 myaddr.sin_port = htons(port);
1072                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1073                                                 sizeof(myaddr));
1074                 if (err == 0) {
1075                         transport->port = port;
1076                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
1077                                         port);
1078                         return 0;
1079                 }
1080                 if (port <= xprt_min_resvport)
1081                         port = xprt_max_resvport;
1082                 else
1083                         port--;
1084         } while (err == -EADDRINUSE && port != transport->port);
1085
1086         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
1087         return err;
1088 }
1089
1090 /**
1091  * xs_udp_connect_worker - set up a UDP socket
1092  * @args: RPC transport to connect
1093  *
1094  * Invoked by a work queue tasklet.
1095  */
1096 static void xs_udp_connect_worker(void *args)
1097 {
1098         struct sock_xprt *transport = (struct sock_xprt *)args;
1099         struct rpc_xprt *xprt = &transport->xprt;
1100         struct socket *sock = transport->sock;
1101         int err, status = -EIO;
1102
1103         if (xprt->shutdown || !xprt_bound(xprt))
1104                 goto out;
1105
1106         /* Start by resetting any existing state */
1107         xs_close(xprt);
1108
1109         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1110                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1111                 goto out;
1112         }
1113
1114         if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1115                 sock_release(sock);
1116                 goto out;
1117         }
1118
1119         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1120                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1121
1122         if (!transport->inet) {
1123                 struct sock *sk = sock->sk;
1124
1125                 write_lock_bh(&sk->sk_callback_lock);
1126
1127                 sk->sk_user_data = xprt;
1128                 transport->old_data_ready = sk->sk_data_ready;
1129                 transport->old_state_change = sk->sk_state_change;
1130                 transport->old_write_space = sk->sk_write_space;
1131                 sk->sk_data_ready = xs_udp_data_ready;
1132                 sk->sk_write_space = xs_udp_write_space;
1133                 sk->sk_no_check = UDP_CSUM_NORCV;
1134                 sk->sk_allocation = GFP_ATOMIC;
1135
1136                 xprt_set_connected(xprt);
1137
1138                 /* Reset to new socket */
1139                 transport->sock = sock;
1140                 transport->inet = sk;
1141
1142                 write_unlock_bh(&sk->sk_callback_lock);
1143         }
1144         xs_udp_do_set_buffer_size(xprt);
1145         status = 0;
1146 out:
1147         xprt_wake_pending_tasks(xprt, status);
1148         xprt_clear_connecting(xprt);
1149 }
1150
1151 /*
1152  * We need to preserve the port number so the reply cache on the server can
1153  * find our cached RPC replies when we get around to reconnecting.
1154  */
1155 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1156 {
1157         int result;
1158         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1159         struct sockaddr any;
1160
1161         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1162
1163         /*
1164          * Disconnect the transport socket by doing a connect operation
1165          * with AF_UNSPEC.  This should return immediately...
1166          */
1167         memset(&any, 0, sizeof(any));
1168         any.sa_family = AF_UNSPEC;
1169         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1170         if (result)
1171                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1172                                 result);
1173 }
1174
1175 /**
1176  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1177  * @args: RPC transport to connect
1178  *
1179  * Invoked by a work queue tasklet.
1180  */
1181 static void xs_tcp_connect_worker(void *args)
1182 {
1183         struct sock_xprt *transport = (struct sock_xprt *)args;
1184         struct rpc_xprt *xprt = &transport->xprt;
1185         struct socket *sock = transport->sock;
1186         int err, status = -EIO;
1187
1188         if (xprt->shutdown || !xprt_bound(xprt))
1189                 goto out;
1190
1191         if (!sock) {
1192                 /* start from scratch */
1193                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1194                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1195                         goto out;
1196                 }
1197
1198                 if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1199                         sock_release(sock);
1200                         goto out;
1201                 }
1202         } else
1203                 /* "close" the socket, preserving the local port */
1204                 xs_tcp_reuse_connection(xprt);
1205
1206         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1207                         xprt, xprt->address_strings[RPC_DISPLAY_ALL]);
1208
1209         if (!transport->inet) {
1210                 struct sock *sk = sock->sk;
1211
1212                 write_lock_bh(&sk->sk_callback_lock);
1213
1214                 sk->sk_user_data = xprt;
1215                 transport->old_data_ready = sk->sk_data_ready;
1216                 transport->old_state_change = sk->sk_state_change;
1217                 transport->old_write_space = sk->sk_write_space;
1218                 sk->sk_data_ready = xs_tcp_data_ready;
1219                 sk->sk_state_change = xs_tcp_state_change;
1220                 sk->sk_write_space = xs_tcp_write_space;
1221                 sk->sk_allocation = GFP_ATOMIC;
1222
1223                 /* socket options */
1224                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1225                 sock_reset_flag(sk, SOCK_LINGER);
1226                 tcp_sk(sk)->linger2 = 0;
1227                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1228
1229                 xprt_clear_connected(xprt);
1230
1231                 /* Reset to new socket */
1232                 transport->sock = sock;
1233                 transport->inet = sk;
1234
1235                 write_unlock_bh(&sk->sk_callback_lock);
1236         }
1237
1238         /* Tell the socket layer to start connecting... */
1239         xprt->stat.connect_count++;
1240         xprt->stat.connect_start = jiffies;
1241         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1242                         xprt->addrlen, O_NONBLOCK);
1243         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1244                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1245         if (status < 0) {
1246                 switch (status) {
1247                         case -EINPROGRESS:
1248                         case -EALREADY:
1249                                 goto out_clear;
1250                         case -ECONNREFUSED:
1251                         case -ECONNRESET:
1252                                 /* retry with existing socket, after a delay */
1253                                 break;
1254                         default:
1255                                 /* get rid of existing socket, and retry */
1256                                 xs_close(xprt);
1257                                 break;
1258                 }
1259         }
1260 out:
1261         xprt_wake_pending_tasks(xprt, status);
1262 out_clear:
1263         xprt_clear_connecting(xprt);
1264 }
1265
1266 /**
1267  * xs_connect - connect a socket to a remote endpoint
1268  * @task: address of RPC task that manages state of connect request
1269  *
1270  * TCP: If the remote end dropped the connection, delay reconnecting.
1271  *
1272  * UDP socket connects are synchronous, but we use a work queue anyway
1273  * to guarantee that even unprivileged user processes can set up a
1274  * socket on a privileged port.
1275  *
1276  * If a UDP socket connect fails, the delay behavior here prevents
1277  * retry floods (hard mounts).
1278  */
1279 static void xs_connect(struct rpc_task *task)
1280 {
1281         struct rpc_xprt *xprt = task->tk_xprt;
1282         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1283
1284         if (xprt_test_and_set_connecting(xprt))
1285                 return;
1286
1287         if (transport->sock != NULL) {
1288                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1289                                 xprt, xprt->reestablish_timeout / HZ);
1290                 schedule_delayed_work(&transport->connect_worker,
1291                                         xprt->reestablish_timeout);
1292                 xprt->reestablish_timeout <<= 1;
1293                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1294                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1295         } else {
1296                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1297                 schedule_work(&transport->connect_worker);
1298
1299                 /* flush_scheduled_work can sleep... */
1300                 if (!RPC_IS_ASYNC(task))
1301                         flush_scheduled_work();
1302         }
1303 }
1304
1305 /**
1306  * xs_udp_print_stats - display UDP socket-specifc stats
1307  * @xprt: rpc_xprt struct containing statistics
1308  * @seq: output file
1309  *
1310  */
1311 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1312 {
1313         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1314
1315         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1316                         transport->port,
1317                         xprt->stat.bind_count,
1318                         xprt->stat.sends,
1319                         xprt->stat.recvs,
1320                         xprt->stat.bad_xids,
1321                         xprt->stat.req_u,
1322                         xprt->stat.bklog_u);
1323 }
1324
1325 /**
1326  * xs_tcp_print_stats - display TCP socket-specifc stats
1327  * @xprt: rpc_xprt struct containing statistics
1328  * @seq: output file
1329  *
1330  */
1331 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1332 {
1333         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1334         long idle_time = 0;
1335
1336         if (xprt_connected(xprt))
1337                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1338
1339         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1340                         transport->port,
1341                         xprt->stat.bind_count,
1342                         xprt->stat.connect_count,
1343                         xprt->stat.connect_time,
1344                         idle_time,
1345                         xprt->stat.sends,
1346                         xprt->stat.recvs,
1347                         xprt->stat.bad_xids,
1348                         xprt->stat.req_u,
1349                         xprt->stat.bklog_u);
1350 }
1351
1352 static struct rpc_xprt_ops xs_udp_ops = {
1353         .set_buffer_size        = xs_udp_set_buffer_size,
1354         .reserve_xprt           = xprt_reserve_xprt_cong,
1355         .release_xprt           = xprt_release_xprt_cong,
1356         .rpcbind                = rpc_getport,
1357         .set_port               = xs_set_port,
1358         .connect                = xs_connect,
1359         .buf_alloc              = rpc_malloc,
1360         .buf_free               = rpc_free,
1361         .send_request           = xs_udp_send_request,
1362         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1363         .timer                  = xs_udp_timer,
1364         .release_request        = xprt_release_rqst_cong,
1365         .close                  = xs_close,
1366         .destroy                = xs_destroy,
1367         .print_stats            = xs_udp_print_stats,
1368 };
1369
1370 static struct rpc_xprt_ops xs_tcp_ops = {
1371         .reserve_xprt           = xprt_reserve_xprt,
1372         .release_xprt           = xs_tcp_release_xprt,
1373         .rpcbind                = rpc_getport,
1374         .set_port               = xs_set_port,
1375         .connect                = xs_connect,
1376         .buf_alloc              = rpc_malloc,
1377         .buf_free               = rpc_free,
1378         .send_request           = xs_tcp_send_request,
1379         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1380         .close                  = xs_close,
1381         .destroy                = xs_destroy,
1382         .print_stats            = xs_tcp_print_stats,
1383 };
1384
1385 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1386 {
1387         struct rpc_xprt *xprt;
1388         struct sock_xprt *new;
1389
1390         if (addrlen > sizeof(xprt->addr)) {
1391                 dprintk("RPC:      xs_setup_xprt: address too large\n");
1392                 return ERR_PTR(-EBADF);
1393         }
1394
1395         new = kzalloc(sizeof(*new), GFP_KERNEL);
1396         if (new == NULL) {
1397                 dprintk("RPC:      xs_setup_xprt: couldn't allocate rpc_xprt\n");
1398                 return ERR_PTR(-ENOMEM);
1399         }
1400         xprt = &new->xprt;
1401
1402         xprt->max_reqs = slot_table_size;
1403         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1404         if (xprt->slot == NULL) {
1405                 kfree(xprt);
1406                 dprintk("RPC:      xs_setup_xprt: couldn't allocate slot table\n");
1407                 return ERR_PTR(-ENOMEM);
1408         }
1409
1410         memcpy(&xprt->addr, addr, addrlen);
1411         xprt->addrlen = addrlen;
1412         new->port = xs_get_random_port();
1413
1414         return xprt;
1415 }
1416
1417 /**
1418  * xs_setup_udp - Set up transport to use a UDP socket
1419  * @addr: address of remote server
1420  * @addrlen: length of address in bytes
1421  * @to:   timeout parameters
1422  *
1423  */
1424 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1425 {
1426         struct rpc_xprt *xprt;
1427         struct sock_xprt *transport;
1428
1429         xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1430         if (IS_ERR(xprt))
1431                 return xprt;
1432         transport = container_of(xprt, struct sock_xprt, xprt);
1433
1434         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1435                 xprt_set_bound(xprt);
1436
1437         xprt->prot = IPPROTO_UDP;
1438         xprt->tsh_size = 0;
1439         /* XXX: header size can vary due to auth type, IPv6, etc. */
1440         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1441
1442         INIT_WORK(&transport->connect_worker, xs_udp_connect_worker, transport);
1443         xprt->bind_timeout = XS_BIND_TO;
1444         xprt->connect_timeout = XS_UDP_CONN_TO;
1445         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1446         xprt->idle_timeout = XS_IDLE_DISC_TO;
1447
1448         xprt->ops = &xs_udp_ops;
1449
1450         if (to)
1451                 xprt->timeout = *to;
1452         else
1453                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1454
1455         xs_format_peer_addresses(xprt);
1456         dprintk("RPC:      set up transport to address %s\n",
1457                         xprt->address_strings[RPC_DISPLAY_ALL]);
1458
1459         return xprt;
1460 }
1461
1462 /**
1463  * xs_setup_tcp - Set up transport to use a TCP socket
1464  * @addr: address of remote server
1465  * @addrlen: length of address in bytes
1466  * @to: timeout parameters
1467  *
1468  */
1469 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1470 {
1471         struct rpc_xprt *xprt;
1472         struct sock_xprt *transport;
1473
1474         xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1475         if (IS_ERR(xprt))
1476                 return xprt;
1477         transport = container_of(xprt, struct sock_xprt, xprt);
1478
1479         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1480                 xprt_set_bound(xprt);
1481
1482         xprt->prot = IPPROTO_TCP;
1483         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1484         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1485
1486         INIT_WORK(&transport->connect_worker, xs_tcp_connect_worker, transport);
1487         xprt->bind_timeout = XS_BIND_TO;
1488         xprt->connect_timeout = XS_TCP_CONN_TO;
1489         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1490         xprt->idle_timeout = XS_IDLE_DISC_TO;
1491
1492         xprt->ops = &xs_tcp_ops;
1493
1494         if (to)
1495                 xprt->timeout = *to;
1496         else
1497                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1498
1499         xs_format_peer_addresses(xprt);
1500         dprintk("RPC:      set up transport to address %s\n",
1501                         xprt->address_strings[RPC_DISPLAY_ALL]);
1502
1503         return xprt;
1504 }
1505
1506 /**
1507  * init_socket_xprt - stub
1508  *
1509  */
1510 int init_socket_xprt(void)
1511 {
1512         return 0;
1513 }
1514
1515 /**
1516  * cleanup_socket_xprt - stub
1517  *
1518  */
1519 void cleanup_socket_xprt(void)
1520 {
1521         return;
1522 }