SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/file.h>
  32
  33 #include <net/sock.h>
  34 #include <net/checksum.h>
  35 #include <net/udp.h>
  36 #include <net/tcp.h>
  37
  38 /*
  39  * Maximum port number to use when requesting a reserved port.
  40  */
  41 #define XS_MAX_RESVPORT         (800U)
  42
  43 /*
  44  * How many times to try sending a request on a socket before waiting
  45  * for the socket buffer to clear.
  46  */
  47 #define XS_SENDMSG_RETRY        (10U)
  48
  49 #ifdef RPC_DEBUG
  50 # undef  RPC_DEBUG_DATA
  51 # define RPCDBG_FACILITY        RPCDBG_TRANS
  52 #endif
  53
  54 #ifdef RPC_DEBUG_DATA
  55 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
  56 {
  57         u8 *buf = (u8 *) packet;
  58         int j;
  59
  60         dprintk("RPC:      %s\n", msg);
  61         for (j = 0; j < count && j < 128; j += 4) {
  62                 if (!(j & 31)) {
  63                         if (j)
  64                                 dprintk("\n");
  65                         dprintk("0x%04x ", j);
  66                 }
  67                 dprintk("%02x%02x%02x%02x ",
  68                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
  69         }
  70         dprintk("\n");
  71 }
  72 #else
  73 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
  74 {
  75         /* NOP */
  76 }
  77 #endif
  78
  79 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
  80
  81 static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
  82 {
  83         struct kvec iov = {
  84                 .iov_base       = xdr->head[0].iov_base + base,
  85                 .iov_len        = len - base,
  86         };
  87         struct msghdr msg = {
  88                 .msg_name       = addr,
  89                 .msg_namelen    = addrlen,
  90                 .msg_flags      = XS_SENDMSG_FLAGS,
  91         };
  92
  93         if (xdr->len > len)
  94                 msg.msg_flags |= MSG_MORE;
  95
  96         if (likely(iov.iov_len))
  97                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
  98         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
  99 }
 100
 101 static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
 102 {
 103         struct kvec iov = {
 104                 .iov_base       = xdr->tail[0].iov_base + base,
 105                 .iov_len        = len - base,
 106         };
 107         struct msghdr msg = {
 108                 .msg_flags      = XS_SENDMSG_FLAGS,
 109         };
 110
 111         return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 112 }
 113
 114 /**
 115  * xs_sendpages - write pages directly to a socket
 116  * @sock: socket to send on
 117  * @addr: UDP only -- address of destination
 118  * @addrlen: UDP only -- length of destination address
 119  * @xdr: buffer containing this request
 120  * @base: starting position in the buffer
 121  *
 122  */
 123 static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 124 {
 125         struct page **ppage = xdr->pages;
 126         unsigned int len, pglen = xdr->page_len;
 127         int err, ret = 0;
 128         ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
 129
 130         if (unlikely(!sock))
 131                 return -ENOTCONN;
 132
 133         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 134
 135         len = xdr->head[0].iov_len;
 136         if (base < len || (addr != NULL && base == 0)) {
 137                 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
 138                 if (ret == 0)
 139                         ret = err;
 140                 else if (err > 0)
 141                         ret += err;
 142                 if (err != (len - base))
 143                         goto out;
 144                 base = 0;
 145         } else
 146                 base -= len;
 147
 148         if (unlikely(pglen == 0))
 149                 goto copy_tail;
 150         if (unlikely(base >= pglen)) {
 151                 base -= pglen;
 152                 goto copy_tail;
 153         }
 154         if (base || xdr->page_base) {
 155                 pglen -= base;
 156                 base += xdr->page_base;
 157                 ppage += base >> PAGE_CACHE_SHIFT;
 158                 base &= ~PAGE_CACHE_MASK;
 159         }
 160
 161         sendpage = sock->ops->sendpage ? : sock_no_sendpage;
 162         do {
 163                 int flags = XS_SENDMSG_FLAGS;
 164
 165                 len = PAGE_CACHE_SIZE;
 166                 if (base)
 167                         len -= base;
 168                 if (pglen < len)
 169                         len = pglen;
 170
 171                 if (pglen != len || xdr->tail[0].iov_len != 0)
 172                         flags |= MSG_MORE;
 173
 174                 /* Hmm... We might be dealing with highmem pages */
 175                 if (PageHighMem(*ppage))
 176                         sendpage = sock_no_sendpage;
 177                 err = sendpage(sock, *ppage, base, len, flags);
 178                 if (ret == 0)
 179                         ret = err;
 180                 else if (err > 0)
 181                         ret += err;
 182                 if (err != len)
 183                         goto out;
 184                 base = 0;
 185                 ppage++;
 186         } while ((pglen -= len) != 0);
 187 copy_tail:
 188         len = xdr->tail[0].iov_len;
 189         if (base < len) {
 190                 err = xs_send_tail(sock, xdr, base, len);
 191                 if (ret == 0)
 192                         ret = err;
 193                 else if (err > 0)
 194                         ret += err;
 195         }
 196 out:
 197         return ret;
 198 }
 199
 200 /**
 201  * xs_nospace - place task on wait queue if transmit was incomplete
 202  * @task: task to put to sleep
 203  *
 204  */
 205 static void xs_nospace(struct rpc_task *task)
 206 {
 207         struct rpc_rqst *req = task->tk_rqstp;
 208         struct rpc_xprt *xprt = req->rq_xprt;
 209
 210         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 211                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 212                         req->rq_slen);
 213
 214         if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
 215                 /* Protect against races with write_space */
 216                 spin_lock_bh(&xprt->transport_lock);
 217
 218                 /* Don't race with disconnect */
 219                 if (!xprt_connected(xprt))
 220                         task->tk_status = -ENOTCONN;
 221                 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
 222                         xprt_wait_for_buffer_space(task);
 223
 224                 spin_unlock_bh(&xprt->transport_lock);
 225         } else
 226                 /* Keep holding the socket if it is blocked */
 227                 rpc_delay(task, HZ>>4);
 228 }
 229
 230 /**
 231  * xs_udp_send_request - write an RPC request to a UDP socket
 232  * @task: address of RPC task that manages the state of an RPC request
 233  *
 234  * Return values:
 235  *        0:    The request has been sent
 236  *   EAGAIN:    The socket was blocked, please call again later to
 237  *              complete the request
 238  * ENOTCONN:    Caller needs to invoke connect logic then call again
 239  *    other:    Some other error occured, the request was not sent
 240  */
 241 static int xs_udp_send_request(struct rpc_task *task)
 242 {
 243         struct rpc_rqst *req = task->tk_rqstp;
 244         struct rpc_xprt *xprt = req->rq_xprt;
 245         struct xdr_buf *xdr = &req->rq_snd_buf;
 246         int status;
 247
 248         xs_pktdump("packet data:",
 249                                 req->rq_svec->iov_base,
 250                                 req->rq_svec->iov_len);
 251
 252         req->rq_xtime = jiffies;
 253         status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
 254                                 sizeof(xprt->addr), xdr, req->rq_bytes_sent);
 255
 256         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 257                         xdr->len - req->rq_bytes_sent, status);
 258
 259         if (likely(status >= (int) req->rq_slen))
 260                 return 0;
 261
 262         /* Still some bytes left; set up for a retry later. */
 263         if (status > 0)
 264                 status = -EAGAIN;
 265
 266         switch (status) {
 267         case -ENETUNREACH:
 268         case -EPIPE:
 269         case -ECONNREFUSED:
 270                 /* When the server has died, an ICMP port unreachable message
 271                  * prompts ECONNREFUSED. */
 272                 break;
 273         case -EAGAIN:
 274                 xs_nospace(task);
 275                 break;
 276         default:
 277                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 278                         -status);
 279                 break;
 280         }
 281
 282         return status;
 283 }
 284
 285 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 286 {
 287         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 288         rpc_fraghdr *base = buf->head[0].iov_base;
 289         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 290 }
 291
 292 /**
 293  * xs_tcp_send_request - write an RPC request to a TCP socket
 294  * @task: address of RPC task that manages the state of an RPC request
 295  *
 296  * Return values:
 297  *        0:    The request has been sent
 298  *   EAGAIN:    The socket was blocked, please call again later to
 299  *              complete the request
 300  * ENOTCONN:    Caller needs to invoke connect logic then call again
 301  *    other:    Some other error occured, the request was not sent
 302  *
 303  * XXX: In the case of soft timeouts, should we eventually give up
 304  *      if sendmsg is not able to make progress?
 305  */
 306 static int xs_tcp_send_request(struct rpc_task *task)
 307 {
 308         struct rpc_rqst *req = task->tk_rqstp;
 309         struct rpc_xprt *xprt = req->rq_xprt;
 310         struct xdr_buf *xdr = &req->rq_snd_buf;
 311         int status, retry = 0;
 312
 313         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 314
 315         xs_pktdump("packet data:",
 316                                 req->rq_svec->iov_base,
 317                                 req->rq_svec->iov_len);
 318
 319         /* Continue transmitting the packet/record. We must be careful
 320          * to cope with writespace callbacks arriving _after_ we have
 321          * called sendmsg(). */
 322         while (1) {
 323                 req->rq_xtime = jiffies;
 324                 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
 325                                                 req->rq_bytes_sent);
 326
 327                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 328                                 xdr->len - req->rq_bytes_sent, status);
 329
 330                 if (unlikely(status < 0))
 331                         break;
 332
 333                 /* If we've sent the entire packet, immediately
 334                  * reset the count of bytes sent. */
 335                 req->rq_bytes_sent += status;
 336                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 337                         req->rq_bytes_sent = 0;
 338                         return 0;
 339                 }
 340
 341                 status = -EAGAIN;
 342                 if (retry++ > XS_SENDMSG_RETRY)
 343                         break;
 344         }
 345
 346         switch (status) {
 347         case -EAGAIN:
 348                 xs_nospace(task);
 349                 break;
 350         case -ECONNREFUSED:
 351         case -ECONNRESET:
 352         case -ENOTCONN:
 353         case -EPIPE:
 354                 status = -ENOTCONN;
 355                 break;
 356         default:
 357                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 358                         -status);
 359                 xprt_disconnect(xprt);
 360                 break;
 361         }
 362
 363         return status;
 364 }
 365
 366 /**
 367  * xs_close - close a socket
 368  * @xprt: transport
 369  *
 370  */
 371 static void xs_close(struct rpc_xprt *xprt)
 372 {
 373         struct socket *sock = xprt->sock;
 374         struct sock *sk = xprt->inet;
 375
 376         if (!sk)
 377                 return;
 378
 379         dprintk("RPC:      xs_close xprt %p\n", xprt);
 380
 381         write_lock_bh(&sk->sk_callback_lock);
 382         xprt->inet = NULL;
 383         xprt->sock = NULL;
 384
 385         sk->sk_user_data = NULL;
 386         sk->sk_data_ready = xprt->old_data_ready;
 387         sk->sk_state_change = xprt->old_state_change;
 388         sk->sk_write_space = xprt->old_write_space;
 389         write_unlock_bh(&sk->sk_callback_lock);
 390
 391         sk->sk_no_check = 0;
 392
 393         sock_release(sock);
 394 }
 395
 396 /**
 397  * xs_destroy - prepare to shutdown a transport
 398  * @xprt: doomed transport
 399  *
 400  */
 401 static void xs_destroy(struct rpc_xprt *xprt)
 402 {
 403         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 404
 405         cancel_delayed_work(&xprt->connect_worker);
 406         flush_scheduled_work();
 407
 408         xprt_disconnect(xprt);
 409         xs_close(xprt);
 410         kfree(xprt->slot);
 411 }
 412
 413 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 414 {
 415         return (struct rpc_xprt *) sk->sk_user_data;
 416 }
 417
 418 /**
 419  * xs_udp_data_ready - "data ready" callback for UDP sockets
 420  * @sk: socket with data to read
 421  * @len: how much data to read
 422  *
 423  */
 424 static void xs_udp_data_ready(struct sock *sk, int len)
 425 {
 426         struct rpc_task *task;
 427         struct rpc_xprt *xprt;
 428         struct rpc_rqst *rovr;
 429         struct sk_buff *skb;
 430         int err, repsize, copied;
 431         u32 _xid, *xp;
 432
 433         read_lock(&sk->sk_callback_lock);
 434         dprintk("RPC:      xs_udp_data_ready...\n");
 435         if (!(xprt = xprt_from_sock(sk)))
 436                 goto out;
 437
 438         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 439                 goto out;
 440
 441         if (xprt->shutdown)
 442                 goto dropit;
 443
 444         repsize = skb->len - sizeof(struct udphdr);
 445         if (repsize < 4) {
 446                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 447                 goto dropit;
 448         }
 449
 450         /* Copy the XID from the skb... */
 451         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 452                                 sizeof(_xid), &_xid);
 453         if (xp == NULL)
 454                 goto dropit;
 455
 456         /* Look up and lock the request corresponding to the given XID */
 457         spin_lock(&xprt->transport_lock);
 458         rovr = xprt_lookup_rqst(xprt, *xp);
 459         if (!rovr)
 460                 goto out_unlock;
 461         task = rovr->rq_task;
 462
 463         dprintk("RPC: %4d received reply\n", task->tk_pid);
 464
 465         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 466                 copied = repsize;
 467
 468         /* Suck it into the iovec, verify checksum if not done by hw. */
 469         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 470                 goto out_unlock;
 471
 472         /* Something worked... */
 473         dst_confirm(skb->dst);
 474
 475         xprt_complete_rqst(xprt, rovr, copied);
 476
 477  out_unlock:
 478         spin_unlock(&xprt->transport_lock);
 479  dropit:
 480         skb_free_datagram(sk, skb);
 481  out:
 482         read_unlock(&sk->sk_callback_lock);
 483 }
 484
 485 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 486 {
 487         if (len > desc->count)
 488                 len = desc->count;
 489         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 490                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 491                                 len, desc->count);
 492                 return 0;
 493         }
 494         desc->offset += len;
 495         desc->count -= len;
 496         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 497                         len, desc->count);
 498         return len;
 499 }
 500
 501 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 502 {
 503         size_t len, used;
 504         char *p;
 505
 506         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 507         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 508         used = xs_tcp_copy_data(desc, p, len);
 509         xprt->tcp_offset += used;
 510         if (used != len)
 511                 return;
 512
 513         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 514         if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 515                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 516         else
 517                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 518         xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 519
 520         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 521         xprt->tcp_offset = 0;
 522
 523         /* Sanity check of the record length */
 524         if (unlikely(xprt->tcp_reclen < 4)) {
 525                 dprintk("RPC:      invalid TCP record fragment length\n");
 526                 xprt_disconnect(xprt);
 527                 return;
 528         }
 529         dprintk("RPC:      reading TCP record fragment of length %d\n",
 530                         xprt->tcp_reclen);
 531 }
 532
 533 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 534 {
 535         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 536                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 537         if (xprt->tcp_offset == xprt->tcp_reclen) {
 538                 xprt->tcp_flags |= XPRT_COPY_RECM;
 539                 xprt->tcp_offset = 0;
 540                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 541                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 542                         xprt->tcp_flags |= XPRT_COPY_XID;
 543                         xprt->tcp_copied = 0;
 544                 }
 545         }
 546 }
 547
 548 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 549 {
 550         size_t len, used;
 551         char *p;
 552
 553         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 554         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 555         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 556         used = xs_tcp_copy_data(desc, p, len);
 557         xprt->tcp_offset += used;
 558         if (used != len)
 559                 return;
 560         xprt->tcp_flags &= ~XPRT_COPY_XID;
 561         xprt->tcp_flags |= XPRT_COPY_DATA;
 562         xprt->tcp_copied = 4;
 563         dprintk("RPC:      reading reply for XID %08x\n",
 564                                                 ntohl(xprt->tcp_xid));
 565         xs_tcp_check_recm(xprt);
 566 }
 567
 568 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 569 {
 570         struct rpc_rqst *req;
 571         struct xdr_buf *rcvbuf;
 572         size_t len;
 573         ssize_t r;
 574
 575         /* Find and lock the request corresponding to this xid */
 576         spin_lock(&xprt->transport_lock);
 577         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 578         if (!req) {
 579                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 580                 dprintk("RPC:      XID %08x request not found!\n",
 581                                 ntohl(xprt->tcp_xid));
 582                 spin_unlock(&xprt->transport_lock);
 583                 return;
 584         }
 585
 586         rcvbuf = &req->rq_private_buf;
 587         len = desc->count;
 588         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 589                 skb_reader_t my_desc;
 590
 591                 len = xprt->tcp_reclen - xprt->tcp_offset;
 592                 memcpy(&my_desc, desc, sizeof(my_desc));
 593                 my_desc.count = len;
 594                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 595                                           &my_desc, xs_tcp_copy_data);
 596                 desc->count -= r;
 597                 desc->offset += r;
 598         } else
 599                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 600                                           desc, xs_tcp_copy_data);
 601
 602         if (r > 0) {
 603                 xprt->tcp_copied += r;
 604                 xprt->tcp_offset += r;
 605         }
 606         if (r != len) {
 607                 /* Error when copying to the receive buffer,
 608                  * usually because we weren't able to allocate
 609                  * additional buffer pages. All we can do now
 610                  * is turn off XPRT_COPY_DATA, so the request
 611                  * will not receive any additional updates,
 612                  * and time out.
 613                  * Any remaining data from this record will
 614                  * be discarded.
 615                  */
 616                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 617                 dprintk("RPC:      XID %08x truncated request\n",
 618                                 ntohl(xprt->tcp_xid));
 619                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 620                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 621                 goto out;
 622         }
 623
 624         dprintk("RPC:      XID %08x read %Zd bytes\n",
 625                         ntohl(xprt->tcp_xid), r);
 626         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 627                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 628
 629         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 630                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 631         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 632                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 633                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 634         }
 635
 636 out:
 637         if (!(xprt->tcp_flags & XPRT_COPY_DATA)) {
 638                 dprintk("RPC: %4d received reply complete\n",
 639                                 req->rq_task->tk_pid);
 640                 xprt_complete_rqst(xprt, req, xprt->tcp_copied);
 641         }
 642         spin_unlock(&xprt->transport_lock);
 643         xs_tcp_check_recm(xprt);
 644 }
 645
 646 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 647 {
 648         size_t len;
 649
 650         len = xprt->tcp_reclen - xprt->tcp_offset;
 651         if (len > desc->count)
 652                 len = desc->count;
 653         desc->count -= len;
 654         desc->offset += len;
 655         xprt->tcp_offset += len;
 656         dprintk("RPC:      discarded %Zu bytes\n", len);
 657         xs_tcp_check_recm(xprt);
 658 }
 659
 660 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 661 {
 662         struct rpc_xprt *xprt = rd_desc->arg.data;
 663         skb_reader_t desc = {
 664                 .skb    = skb,
 665                 .offset = offset,
 666                 .count  = len,
 667                 .csum   = 0
 668         };
 669
 670         dprintk("RPC:      xs_tcp_data_recv started\n");
 671         do {
 672                 /* Read in a new fragment marker if necessary */
 673                 /* Can we ever really expect to get completely empty fragments? */
 674                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 675                         xs_tcp_read_fraghdr(xprt, &desc);
 676                         continue;
 677                 }
 678                 /* Read in the xid if necessary */
 679                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 680                         xs_tcp_read_xid(xprt, &desc);
 681                         continue;
 682                 }
 683                 /* Read in the request data */
 684                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 685                         xs_tcp_read_request(xprt, &desc);
 686                         continue;
 687                 }
 688                 /* Skip over any trailing bytes on short reads */
 689                 xs_tcp_read_discard(xprt, &desc);
 690         } while (desc.count);
 691         dprintk("RPC:      xs_tcp_data_recv done\n");
 692         return len - desc.count;
 693 }
 694
 695 /**
 696  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 697  * @sk: socket with data to read
 698  * @bytes: how much data to read
 699  *
 700  */
 701 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 702 {
 703         struct rpc_xprt *xprt;
 704         read_descriptor_t rd_desc;
 705
 706         read_lock(&sk->sk_callback_lock);
 707         dprintk("RPC:      xs_tcp_data_ready...\n");
 708         if (!(xprt = xprt_from_sock(sk)))
 709                 goto out;
 710         if (xprt->shutdown)
 711                 goto out;
 712
 713         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 714         rd_desc.arg.data = xprt;
 715         rd_desc.count = 65536;
 716         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 717 out:
 718         read_unlock(&sk->sk_callback_lock);
 719 }
 720
 721 /**
 722  * xs_tcp_state_change - callback to handle TCP socket state changes
 723  * @sk: socket whose state has changed
 724  *
 725  */
 726 static void xs_tcp_state_change(struct sock *sk)
 727 {
 728         struct rpc_xprt *xprt;
 729
 730         read_lock(&sk->sk_callback_lock);
 731         if (!(xprt = xprt_from_sock(sk)))
 732                 goto out;
 733         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 734         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 735                                 sk->sk_state, xprt_connected(xprt),
 736                                 sock_flag(sk, SOCK_DEAD),
 737                                 sock_flag(sk, SOCK_ZAPPED));
 738
 739         switch (sk->sk_state) {
 740         case TCP_ESTABLISHED:
 741                 spin_lock_bh(&xprt->transport_lock);
 742                 if (!xprt_test_and_set_connected(xprt)) {
 743                         /* Reset TCP record info */
 744                         xprt->tcp_offset = 0;
 745                         xprt->tcp_reclen = 0;
 746                         xprt->tcp_copied = 0;
 747                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 748                         xprt_wake_pending_tasks(xprt, 0);
 749                 }
 750                 spin_unlock_bh(&xprt->transport_lock);
 751                 break;
 752         case TCP_SYN_SENT:
 753         case TCP_SYN_RECV:
 754                 break;
 755         default:
 756                 xprt_disconnect(xprt);
 757                 break;
 758         }
 759  out:
 760         read_unlock(&sk->sk_callback_lock);
 761 }
 762
 763 /**
 764  * xs_udp_write_space - callback invoked when socket buffer space
 765  *                             becomes available
 766  * @sk: socket whose state has changed
 767  *
 768  * Called when more output buffer space is available for this socket.
 769  * We try not to wake our writers until they can make "significant"
 770  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 771  * with a bunch of small requests.
 772  */
 773 static void xs_udp_write_space(struct sock *sk)
 774 {
 775         read_lock(&sk->sk_callback_lock);
 776
 777         /* from net/core/sock.c:sock_def_write_space */
 778         if (sock_writeable(sk)) {
 779                 struct socket *sock;
 780                 struct rpc_xprt *xprt;
 781
 782                 if (unlikely(!(sock = sk->sk_socket)))
 783                         goto out;
 784                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 785                         goto out;
 786                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 787                         goto out;
 788
 789                 xprt_write_space(xprt);
 790         }
 791
 792  out:
 793         read_unlock(&sk->sk_callback_lock);
 794 }
 795
 796 /**
 797  * xs_tcp_write_space - callback invoked when socket buffer space
 798  *                             becomes available
 799  * @sk: socket whose state has changed
 800  *
 801  * Called when more output buffer space is available for this socket.
 802  * We try not to wake our writers until they can make "significant"
 803  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 804  * with a bunch of small requests.
 805  */
 806 static void xs_tcp_write_space(struct sock *sk)
 807 {
 808         read_lock(&sk->sk_callback_lock);
 809
 810         /* from net/core/stream.c:sk_stream_write_space */
 811         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 812                 struct socket *sock;
 813                 struct rpc_xprt *xprt;
 814
 815                 if (unlikely(!(sock = sk->sk_socket)))
 816                         goto out;
 817                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 818                         goto out;
 819                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 820                         goto out;
 821
 822                 xprt_write_space(xprt);
 823         }
 824
 825  out:
 826         read_unlock(&sk->sk_callback_lock);
 827 }
 828
 829 /**
 830  * xs_udp_set_buffer_size - set send and receive limits
 831  * @xprt: generic transport
 832  *
 833  * Set socket send and receive limits based on the
 834  * sndsize and rcvsize fields in the generic transport
 835  * structure.
 836  */
 837 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt)
 838 {
 839         struct sock *sk = xprt->inet;
 840
 841         if (xprt->rcvsize) {
 842                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 843                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 844         }
 845         if (xprt->sndsize) {
 846                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 847                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 848                 sk->sk_write_space(sk);
 849         }
 850 }
 851
 852 /**
 853  * xs_tcp_set_buffer_size - set send and receive limits
 854  * @xprt: generic transport
 855  *
 856  * Nothing to do for TCP.
 857  */
 858 static void xs_tcp_set_buffer_size(struct rpc_xprt *xprt)
 859 {
 860         return;
 861 }
 862
 863 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
 864 {
 865         struct sockaddr_in myaddr = {
 866                 .sin_family = AF_INET,
 867         };
 868         int err, port;
 869
 870         /* Were we already bound to a given port? Try to reuse it */
 871         port = xprt->port;
 872         do {
 873                 myaddr.sin_port = htons(port);
 874                 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
 875                                                 sizeof(myaddr));
 876                 if (err == 0) {
 877                         xprt->port = port;
 878                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
 879                                         port);
 880                         return 0;
 881                 }
 882                 if (--port == 0)
 883                         port = XS_MAX_RESVPORT;
 884         } while (err == -EADDRINUSE && port != xprt->port);
 885
 886         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
 887         return err;
 888 }
 889
 890 /**
 891  * xs_udp_connect_worker - set up a UDP socket
 892  * @args: RPC transport to connect
 893  *
 894  * Invoked by a work queue tasklet.
 895  */
 896 static void xs_udp_connect_worker(void *args)
 897 {
 898         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
 899         struct socket *sock = xprt->sock;
 900         int err, status = -EIO;
 901
 902         if (xprt->shutdown || xprt->addr.sin_port == 0)
 903                 goto out;
 904
 905         dprintk("RPC:      xs_udp_connect_worker for xprt %p\n", xprt);
 906
 907         /* Start by resetting any existing state */
 908         xs_close(xprt);
 909
 910         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
 911                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
 912                 goto out;
 913         }
 914
 915         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
 916                 sock_release(sock);
 917                 goto out;
 918         }
 919
 920         if (!xprt->inet) {
 921                 struct sock *sk = sock->sk;
 922
 923                 write_lock_bh(&sk->sk_callback_lock);
 924
 925                 sk->sk_user_data = xprt;
 926                 xprt->old_data_ready = sk->sk_data_ready;
 927                 xprt->old_state_change = sk->sk_state_change;
 928                 xprt->old_write_space = sk->sk_write_space;
 929                 sk->sk_data_ready = xs_udp_data_ready;
 930                 sk->sk_write_space = xs_udp_write_space;
 931                 sk->sk_no_check = UDP_CSUM_NORCV;
 932
 933                 xprt_set_connected(xprt);
 934
 935                 /* Reset to new socket */
 936                 xprt->sock = sock;
 937                 xprt->inet = sk;
 938
 939                 write_unlock_bh(&sk->sk_callback_lock);
 940         }
 941         xs_udp_set_buffer_size(xprt);
 942         status = 0;
 943 out:
 944         xprt_wake_pending_tasks(xprt, status);
 945         xprt_clear_connecting(xprt);
 946 }
 947
 948 /**
 949  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
 950  * @args: RPC transport to connect
 951  *
 952  * Invoked by a work queue tasklet.
 953  */
 954 static void xs_tcp_connect_worker(void *args)
 955 {
 956         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
 957         struct socket *sock = xprt->sock;
 958         int err, status = -EIO;
 959
 960         if (xprt->shutdown || xprt->addr.sin_port == 0)
 961                 goto out;
 962
 963         dprintk("RPC:      xs_tcp_connect_worker for xprt %p\n", xprt);
 964
 965         /* Start by resetting any existing socket state */
 966         xs_close(xprt);
 967
 968         if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
 969                 dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
 970                 goto out;
 971         }
 972
 973         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
 974                 sock_release(sock);
 975                 goto out;
 976         }
 977
 978         if (!xprt->inet) {
 979                 struct sock *sk = sock->sk;
 980
 981                 write_lock_bh(&sk->sk_callback_lock);
 982
 983                 sk->sk_user_data = xprt;
 984                 xprt->old_data_ready = sk->sk_data_ready;
 985                 xprt->old_state_change = sk->sk_state_change;
 986                 xprt->old_write_space = sk->sk_write_space;
 987                 sk->sk_data_ready = xs_tcp_data_ready;
 988                 sk->sk_state_change = xs_tcp_state_change;
 989                 sk->sk_write_space = xs_tcp_write_space;
 990                 tcp_sk(sk)->nonagle = 1;
 991
 992                 xprt_clear_connected(xprt);
 993
 994                 /* Reset to new socket */
 995                 xprt->sock = sock;
 996                 xprt->inet = sk;
 997
 998                 write_unlock_bh(&sk->sk_callback_lock);
 999         }
1000
1001         /* Tell the socket layer to start connecting... */
1002         status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
1003                         sizeof(xprt->addr), O_NONBLOCK);
1004         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1005                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1006         if (status < 0) {
1007                 switch (status) {
1008                         case -EINPROGRESS:
1009                         case -EALREADY:
1010                                 goto out_clear;
1011                 }
1012         }
1013 out:
1014         xprt_wake_pending_tasks(xprt, status);
1015 out_clear:
1016         xprt_clear_connecting(xprt);
1017 }
1018
1019 /**
1020  * xs_connect - connect a socket to a remote endpoint
1021  * @task: address of RPC task that manages state of connect request
1022  *
1023  * TCP: If the remote end dropped the connection, delay reconnecting.
1024  */
1025 static void xs_connect(struct rpc_task *task)
1026 {
1027         struct rpc_xprt *xprt = task->tk_xprt;
1028
1029         if (xprt_test_and_set_connecting(xprt))
1030                 return;
1031
1032         if (xprt->sock != NULL) {
1033                 dprintk("RPC:      xs_connect delayed xprt %p\n", xprt);
1034                 schedule_delayed_work(&xprt->connect_worker,
1035                                         RPC_REESTABLISH_TIMEOUT);
1036         } else {
1037                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1038                 schedule_work(&xprt->connect_worker);
1039
1040                 /* flush_scheduled_work can sleep... */
1041                 if (!RPC_IS_ASYNC(task))
1042                         flush_scheduled_work();
1043         }
1044 }
1045
1046 static struct rpc_xprt_ops xs_udp_ops = {
1047         .set_buffer_size        = xs_udp_set_buffer_size,
1048         .connect                = xs_connect,
1049         .send_request           = xs_udp_send_request,
1050         .close                  = xs_close,
1051         .destroy                = xs_destroy,
1052 };
1053
1054 static struct rpc_xprt_ops xs_tcp_ops = {
1055         .set_buffer_size        = xs_tcp_set_buffer_size,
1056         .connect                = xs_connect,
1057         .send_request           = xs_tcp_send_request,
1058         .close                  = xs_close,
1059         .destroy                = xs_destroy,
1060 };
1061
1062 extern unsigned int xprt_udp_slot_table_entries;
1063 extern unsigned int xprt_tcp_slot_table_entries;
1064
1065 /**
1066  * xs_setup_udp - Set up transport to use a UDP socket
1067  * @xprt: transport to set up
1068  * @to:   timeout parameters
1069  *
1070  */
1071 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1072 {
1073         size_t slot_table_size;
1074
1075         dprintk("RPC:      setting up udp-ipv4 transport...\n");
1076
1077         xprt->max_reqs = xprt_udp_slot_table_entries;
1078         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1079         xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1080         if (xprt->slot == NULL)
1081                 return -ENOMEM;
1082         memset(xprt->slot, 0, slot_table_size);
1083
1084         xprt->prot = IPPROTO_UDP;
1085         xprt->port = XS_MAX_RESVPORT;
1086         xprt->tsh_size = 0;
1087         xprt->nocong = 0;
1088         xprt->cwnd = RPC_INITCWND;
1089         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1090         /* XXX: header size can vary due to auth type, IPv6, etc. */
1091         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1092
1093         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1094
1095         xprt->ops = &xs_udp_ops;
1096
1097         if (to)
1098                 xprt->timeout = *to;
1099         else
1100                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1101
1102         return 0;
1103 }
1104
1105 /**
1106  * xs_setup_tcp - Set up transport to use a TCP socket
1107  * @xprt: transport to set up
1108  * @to: timeout parameters
1109  *
1110  */
1111 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1112 {
1113         size_t slot_table_size;
1114
1115         dprintk("RPC:      setting up tcp-ipv4 transport...\n");
1116
1117         xprt->max_reqs = xprt_tcp_slot_table_entries;
1118         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1119         xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1120         if (xprt->slot == NULL)
1121                 return -ENOMEM;
1122         memset(xprt->slot, 0, slot_table_size);
1123
1124         xprt->prot = IPPROTO_TCP;
1125         xprt->port = XS_MAX_RESVPORT;
1126         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1127         xprt->nocong = 1;
1128         xprt->cwnd = RPC_MAXCWND(xprt);
1129         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1130         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1131
1132         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1133
1134         xprt->ops = &xs_tcp_ops;
1135
1136         if (to)
1137                 xprt->timeout = *to;
1138         else
1139                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1140
1141         return 0;
1142 }