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  * How many times to try sending a request on a socket before waiting
  40  * for the socket buffer to clear.
  41  */
  42 #define XS_SENDMSG_RETRY        (10U)
  43
  44 #ifdef RPC_DEBUG
  45 # undef  RPC_DEBUG_DATA
  46 # define RPCDBG_FACILITY        RPCDBG_TRANS
  47 #endif
  48
  49 #ifdef RPC_DEBUG_DATA
  50 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
  51 {
  52         u8 *buf = (u8 *) packet;
  53         int j;
  54
  55         dprintk("RPC:      %s\n", msg);
  56         for (j = 0; j < count && j < 128; j += 4) {
  57                 if (!(j & 31)) {
  58                         if (j)
  59                                 dprintk("\n");
  60                         dprintk("0x%04x ", j);
  61                 }
  62                 dprintk("%02x%02x%02x%02x ",
  63                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
  64         }
  65         dprintk("\n");
  66 }
  67 #else
  68 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
  69 {
  70         /* NOP */
  71 }
  72 #endif
  73
  74 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
  75
  76 static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
  77 {
  78         struct kvec iov = {
  79                 .iov_base       = xdr->head[0].iov_base + base,
  80                 .iov_len        = len - base,
  81         };
  82         struct msghdr msg = {
  83                 .msg_name       = addr,
  84                 .msg_namelen    = addrlen,
  85                 .msg_flags      = XS_SENDMSG_FLAGS,
  86         };
  87
  88         if (xdr->len > len)
  89                 msg.msg_flags |= MSG_MORE;
  90
  91         if (likely(iov.iov_len))
  92                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
  93         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
  94 }
  95
  96 static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
  97 {
  98         struct kvec iov = {
  99                 .iov_base       = xdr->tail[0].iov_base + base,
 100                 .iov_len        = len - base,
 101         };
 102         struct msghdr msg = {
 103                 .msg_flags      = XS_SENDMSG_FLAGS,
 104         };
 105
 106         return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 107 }
 108
 109 /**
 110  * xs_sendpages - write pages directly to a socket
 111  * @sock: socket to send on
 112  * @addr: UDP only -- address of destination
 113  * @addrlen: UDP only -- length of destination address
 114  * @xdr: buffer containing this request
 115  * @base: starting position in the buffer
 116  *
 117  */
 118 static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 119 {
 120         struct page **ppage = xdr->pages;
 121         unsigned int len, pglen = xdr->page_len;
 122         int err, ret = 0;
 123         ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
 124
 125         if (unlikely(!sock))
 126                 return -ENOTCONN;
 127
 128         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 129
 130         len = xdr->head[0].iov_len;
 131         if (base < len || (addr != NULL && base == 0)) {
 132                 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
 133                 if (ret == 0)
 134                         ret = err;
 135                 else if (err > 0)
 136                         ret += err;
 137                 if (err != (len - base))
 138                         goto out;
 139                 base = 0;
 140         } else
 141                 base -= len;
 142
 143         if (unlikely(pglen == 0))
 144                 goto copy_tail;
 145         if (unlikely(base >= pglen)) {
 146                 base -= pglen;
 147                 goto copy_tail;
 148         }
 149         if (base || xdr->page_base) {
 150                 pglen -= base;
 151                 base += xdr->page_base;
 152                 ppage += base >> PAGE_CACHE_SHIFT;
 153                 base &= ~PAGE_CACHE_MASK;
 154         }
 155
 156         sendpage = sock->ops->sendpage ? : sock_no_sendpage;
 157         do {
 158                 int flags = XS_SENDMSG_FLAGS;
 159
 160                 len = PAGE_CACHE_SIZE;
 161                 if (base)
 162                         len -= base;
 163                 if (pglen < len)
 164                         len = pglen;
 165
 166                 if (pglen != len || xdr->tail[0].iov_len != 0)
 167                         flags |= MSG_MORE;
 168
 169                 /* Hmm... We might be dealing with highmem pages */
 170                 if (PageHighMem(*ppage))
 171                         sendpage = sock_no_sendpage;
 172                 err = sendpage(sock, *ppage, base, len, flags);
 173                 if (ret == 0)
 174                         ret = err;
 175                 else if (err > 0)
 176                         ret += err;
 177                 if (err != len)
 178                         goto out;
 179                 base = 0;
 180                 ppage++;
 181         } while ((pglen -= len) != 0);
 182 copy_tail:
 183         len = xdr->tail[0].iov_len;
 184         if (base < len) {
 185                 err = xs_send_tail(sock, xdr, base, len);
 186                 if (ret == 0)
 187                         ret = err;
 188                 else if (err > 0)
 189                         ret += err;
 190         }
 191 out:
 192         return ret;
 193 }
 194
 195 /**
 196  * xs_nospace - place task on wait queue if transmit was incomplete
 197  * @task: task to put to sleep
 198  *
 199  */
 200 static void xs_nospace(struct rpc_task *task)
 201 {
 202         struct rpc_rqst *req = task->tk_rqstp;
 203         struct rpc_xprt *xprt = req->rq_xprt;
 204
 205         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 206                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 207                         req->rq_slen);
 208
 209         if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
 210                 /* Protect against races with write_space */
 211                 spin_lock_bh(&xprt->transport_lock);
 212
 213                 /* Don't race with disconnect */
 214                 if (!xprt_connected(xprt))
 215                         task->tk_status = -ENOTCONN;
 216                 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
 217                         xprt_wait_for_buffer_space(task);
 218
 219                 spin_unlock_bh(&xprt->transport_lock);
 220         } else
 221                 /* Keep holding the socket if it is blocked */
 222                 rpc_delay(task, HZ>>4);
 223 }
 224
 225 /**
 226  * xs_udp_send_request - write an RPC request to a UDP socket
 227  * @task: address of RPC task that manages the state of an RPC request
 228  *
 229  * Return values:
 230  *        0:    The request has been sent
 231  *   EAGAIN:    The socket was blocked, please call again later to
 232  *              complete the request
 233  * ENOTCONN:    Caller needs to invoke connect logic then call again
 234  *    other:    Some other error occured, the request was not sent
 235  */
 236 static int xs_udp_send_request(struct rpc_task *task)
 237 {
 238         struct rpc_rqst *req = task->tk_rqstp;
 239         struct rpc_xprt *xprt = req->rq_xprt;
 240         struct xdr_buf *xdr = &req->rq_snd_buf;
 241         int status;
 242
 243         xs_pktdump("packet data:",
 244                                 req->rq_svec->iov_base,
 245                                 req->rq_svec->iov_len);
 246
 247         req->rq_xtime = jiffies;
 248         status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
 249                                 sizeof(xprt->addr), xdr, req->rq_bytes_sent);
 250
 251         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 252                         xdr->len - req->rq_bytes_sent, status);
 253
 254         if (likely(status >= (int) req->rq_slen))
 255                 return 0;
 256
 257         /* Still some bytes left; set up for a retry later. */
 258         if (status > 0)
 259                 status = -EAGAIN;
 260
 261         switch (status) {
 262         case -ENETUNREACH:
 263         case -EPIPE:
 264         case -ECONNREFUSED:
 265                 /* When the server has died, an ICMP port unreachable message
 266                  * prompts ECONNREFUSED. */
 267                 break;
 268         case -EAGAIN:
 269                 xs_nospace(task);
 270                 break;
 271         default:
 272                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 273                         -status);
 274                 break;
 275         }
 276
 277         return status;
 278 }
 279
 280 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 281 {
 282         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 283         rpc_fraghdr *base = buf->head[0].iov_base;
 284         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 285 }
 286
 287 /**
 288  * xs_tcp_send_request - write an RPC request to a TCP socket
 289  * @task: address of RPC task that manages the state of an RPC request
 290  *
 291  * Return values:
 292  *        0:    The request has been sent
 293  *   EAGAIN:    The socket was blocked, please call again later to
 294  *              complete the request
 295  * ENOTCONN:    Caller needs to invoke connect logic then call again
 296  *    other:    Some other error occured, the request was not sent
 297  *
 298  * XXX: In the case of soft timeouts, should we eventually give up
 299  *      if sendmsg is not able to make progress?
 300  */
 301 static int xs_tcp_send_request(struct rpc_task *task)
 302 {
 303         struct rpc_rqst *req = task->tk_rqstp;
 304         struct rpc_xprt *xprt = req->rq_xprt;
 305         struct xdr_buf *xdr = &req->rq_snd_buf;
 306         int status, retry = 0;
 307
 308         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 309
 310         xs_pktdump("packet data:",
 311                                 req->rq_svec->iov_base,
 312                                 req->rq_svec->iov_len);
 313
 314         /* Continue transmitting the packet/record. We must be careful
 315          * to cope with writespace callbacks arriving _after_ we have
 316          * called sendmsg(). */
 317         while (1) {
 318                 req->rq_xtime = jiffies;
 319                 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
 320                                                 req->rq_bytes_sent);
 321
 322                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 323                                 xdr->len - req->rq_bytes_sent, status);
 324
 325                 if (unlikely(status < 0))
 326                         break;
 327
 328                 /* If we've sent the entire packet, immediately
 329                  * reset the count of bytes sent. */
 330                 req->rq_bytes_sent += status;
 331                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 332                         req->rq_bytes_sent = 0;
 333                         return 0;
 334                 }
 335
 336                 status = -EAGAIN;
 337                 if (retry++ > XS_SENDMSG_RETRY)
 338                         break;
 339         }
 340
 341         switch (status) {
 342         case -EAGAIN:
 343                 xs_nospace(task);
 344                 break;
 345         case -ECONNREFUSED:
 346         case -ECONNRESET:
 347         case -ENOTCONN:
 348         case -EPIPE:
 349                 status = -ENOTCONN;
 350                 break;
 351         default:
 352                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 353                         -status);
 354                 xprt_disconnect(xprt);
 355                 break;
 356         }
 357
 358         return status;
 359 }
 360
 361 /**
 362  * xs_close - close a socket
 363  * @xprt: transport
 364  *
 365  */
 366 static void xs_close(struct rpc_xprt *xprt)
 367 {
 368         struct socket *sock = xprt->sock;
 369         struct sock *sk = xprt->inet;
 370
 371         if (!sk)
 372                 return;
 373
 374         dprintk("RPC:      xs_close xprt %p\n", xprt);
 375
 376         write_lock_bh(&sk->sk_callback_lock);
 377         xprt->inet = NULL;
 378         xprt->sock = NULL;
 379
 380         sk->sk_user_data = NULL;
 381         sk->sk_data_ready = xprt->old_data_ready;
 382         sk->sk_state_change = xprt->old_state_change;
 383         sk->sk_write_space = xprt->old_write_space;
 384         write_unlock_bh(&sk->sk_callback_lock);
 385
 386         sk->sk_no_check = 0;
 387
 388         sock_release(sock);
 389 }
 390
 391 /**
 392  * xs_destroy - prepare to shutdown a transport
 393  * @xprt: doomed transport
 394  *
 395  */
 396 static void xs_destroy(struct rpc_xprt *xprt)
 397 {
 398         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 399
 400         cancel_delayed_work(&xprt->connect_worker);
 401         flush_scheduled_work();
 402
 403         xprt_disconnect(xprt);
 404         xs_close(xprt);
 405         kfree(xprt->slot);
 406 }
 407
 408 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 409 {
 410         return (struct rpc_xprt *) sk->sk_user_data;
 411 }
 412
 413 /**
 414  * xs_udp_data_ready - "data ready" callback for UDP sockets
 415  * @sk: socket with data to read
 416  * @len: how much data to read
 417  *
 418  */
 419 static void xs_udp_data_ready(struct sock *sk, int len)
 420 {
 421         struct rpc_task *task;
 422         struct rpc_xprt *xprt;
 423         struct rpc_rqst *rovr;
 424         struct sk_buff *skb;
 425         int err, repsize, copied;
 426         u32 _xid, *xp;
 427
 428         read_lock(&sk->sk_callback_lock);
 429         dprintk("RPC:      xs_udp_data_ready...\n");
 430         if (!(xprt = xprt_from_sock(sk)))
 431                 goto out;
 432
 433         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 434                 goto out;
 435
 436         if (xprt->shutdown)
 437                 goto dropit;
 438
 439         repsize = skb->len - sizeof(struct udphdr);
 440         if (repsize < 4) {
 441                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 442                 goto dropit;
 443         }
 444
 445         /* Copy the XID from the skb... */
 446         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 447                                 sizeof(_xid), &_xid);
 448         if (xp == NULL)
 449                 goto dropit;
 450
 451         /* Look up and lock the request corresponding to the given XID */
 452         spin_lock(&xprt->transport_lock);
 453         rovr = xprt_lookup_rqst(xprt, *xp);
 454         if (!rovr)
 455                 goto out_unlock;
 456         task = rovr->rq_task;
 457
 458         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 459                 copied = repsize;
 460
 461         /* Suck it into the iovec, verify checksum if not done by hw. */
 462         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 463                 goto out_unlock;
 464
 465         /* Something worked... */
 466         dst_confirm(skb->dst);
 467
 468         xprt_adjust_cwnd(task, copied);
 469         xprt_update_rtt(task);
 470         xprt_complete_rqst(task, copied);
 471
 472  out_unlock:
 473         spin_unlock(&xprt->transport_lock);
 474  dropit:
 475         skb_free_datagram(sk, skb);
 476  out:
 477         read_unlock(&sk->sk_callback_lock);
 478 }
 479
 480 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 481 {
 482         if (len > desc->count)
 483                 len = desc->count;
 484         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 485                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 486                                 len, desc->count);
 487                 return 0;
 488         }
 489         desc->offset += len;
 490         desc->count -= len;
 491         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 492                         len, desc->count);
 493         return len;
 494 }
 495
 496 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 497 {
 498         size_t len, used;
 499         char *p;
 500
 501         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 502         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 503         used = xs_tcp_copy_data(desc, p, len);
 504         xprt->tcp_offset += used;
 505         if (used != len)
 506                 return;
 507
 508         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 509         if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 510                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 511         else
 512                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 513         xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 514
 515         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 516         xprt->tcp_offset = 0;
 517
 518         /* Sanity check of the record length */
 519         if (unlikely(xprt->tcp_reclen < 4)) {
 520                 dprintk("RPC:      invalid TCP record fragment length\n");
 521                 xprt_disconnect(xprt);
 522                 return;
 523         }
 524         dprintk("RPC:      reading TCP record fragment of length %d\n",
 525                         xprt->tcp_reclen);
 526 }
 527
 528 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 529 {
 530         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 531                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 532         if (xprt->tcp_offset == xprt->tcp_reclen) {
 533                 xprt->tcp_flags |= XPRT_COPY_RECM;
 534                 xprt->tcp_offset = 0;
 535                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 536                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 537                         xprt->tcp_flags |= XPRT_COPY_XID;
 538                         xprt->tcp_copied = 0;
 539                 }
 540         }
 541 }
 542
 543 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 544 {
 545         size_t len, used;
 546         char *p;
 547
 548         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 549         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 550         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 551         used = xs_tcp_copy_data(desc, p, len);
 552         xprt->tcp_offset += used;
 553         if (used != len)
 554                 return;
 555         xprt->tcp_flags &= ~XPRT_COPY_XID;
 556         xprt->tcp_flags |= XPRT_COPY_DATA;
 557         xprt->tcp_copied = 4;
 558         dprintk("RPC:      reading reply for XID %08x\n",
 559                                                 ntohl(xprt->tcp_xid));
 560         xs_tcp_check_recm(xprt);
 561 }
 562
 563 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 564 {
 565         struct rpc_rqst *req;
 566         struct xdr_buf *rcvbuf;
 567         size_t len;
 568         ssize_t r;
 569
 570         /* Find and lock the request corresponding to this xid */
 571         spin_lock(&xprt->transport_lock);
 572         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 573         if (!req) {
 574                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 575                 dprintk("RPC:      XID %08x request not found!\n",
 576                                 ntohl(xprt->tcp_xid));
 577                 spin_unlock(&xprt->transport_lock);
 578                 return;
 579         }
 580
 581         rcvbuf = &req->rq_private_buf;
 582         len = desc->count;
 583         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 584                 skb_reader_t my_desc;
 585
 586                 len = xprt->tcp_reclen - xprt->tcp_offset;
 587                 memcpy(&my_desc, desc, sizeof(my_desc));
 588                 my_desc.count = len;
 589                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 590                                           &my_desc, xs_tcp_copy_data);
 591                 desc->count -= r;
 592                 desc->offset += r;
 593         } else
 594                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 595                                           desc, xs_tcp_copy_data);
 596
 597         if (r > 0) {
 598                 xprt->tcp_copied += r;
 599                 xprt->tcp_offset += r;
 600         }
 601         if (r != len) {
 602                 /* Error when copying to the receive buffer,
 603                  * usually because we weren't able to allocate
 604                  * additional buffer pages. All we can do now
 605                  * is turn off XPRT_COPY_DATA, so the request
 606                  * will not receive any additional updates,
 607                  * and time out.
 608                  * Any remaining data from this record will
 609                  * be discarded.
 610                  */
 611                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 612                 dprintk("RPC:      XID %08x truncated request\n",
 613                                 ntohl(xprt->tcp_xid));
 614                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 615                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 616                 goto out;
 617         }
 618
 619         dprintk("RPC:      XID %08x read %Zd bytes\n",
 620                         ntohl(xprt->tcp_xid), r);
 621         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 622                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 623
 624         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 625                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 626         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 627                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 628                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 629         }
 630
 631 out:
 632         if (!(xprt->tcp_flags & XPRT_COPY_DATA))
 633                 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
 634         spin_unlock(&xprt->transport_lock);
 635         xs_tcp_check_recm(xprt);
 636 }
 637
 638 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 639 {
 640         size_t len;
 641
 642         len = xprt->tcp_reclen - xprt->tcp_offset;
 643         if (len > desc->count)
 644                 len = desc->count;
 645         desc->count -= len;
 646         desc->offset += len;
 647         xprt->tcp_offset += len;
 648         dprintk("RPC:      discarded %Zu bytes\n", len);
 649         xs_tcp_check_recm(xprt);
 650 }
 651
 652 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 653 {
 654         struct rpc_xprt *xprt = rd_desc->arg.data;
 655         skb_reader_t desc = {
 656                 .skb    = skb,
 657                 .offset = offset,
 658                 .count  = len,
 659                 .csum   = 0
 660         };
 661
 662         dprintk("RPC:      xs_tcp_data_recv started\n");
 663         do {
 664                 /* Read in a new fragment marker if necessary */
 665                 /* Can we ever really expect to get completely empty fragments? */
 666                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 667                         xs_tcp_read_fraghdr(xprt, &desc);
 668                         continue;
 669                 }
 670                 /* Read in the xid if necessary */
 671                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 672                         xs_tcp_read_xid(xprt, &desc);
 673                         continue;
 674                 }
 675                 /* Read in the request data */
 676                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 677                         xs_tcp_read_request(xprt, &desc);
 678                         continue;
 679                 }
 680                 /* Skip over any trailing bytes on short reads */
 681                 xs_tcp_read_discard(xprt, &desc);
 682         } while (desc.count);
 683         dprintk("RPC:      xs_tcp_data_recv done\n");
 684         return len - desc.count;
 685 }
 686
 687 /**
 688  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 689  * @sk: socket with data to read
 690  * @bytes: how much data to read
 691  *
 692  */
 693 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 694 {
 695         struct rpc_xprt *xprt;
 696         read_descriptor_t rd_desc;
 697
 698         read_lock(&sk->sk_callback_lock);
 699         dprintk("RPC:      xs_tcp_data_ready...\n");
 700         if (!(xprt = xprt_from_sock(sk)))
 701                 goto out;
 702         if (xprt->shutdown)
 703                 goto out;
 704
 705         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 706         rd_desc.arg.data = xprt;
 707         rd_desc.count = 65536;
 708         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 709 out:
 710         read_unlock(&sk->sk_callback_lock);
 711 }
 712
 713 /**
 714  * xs_tcp_state_change - callback to handle TCP socket state changes
 715  * @sk: socket whose state has changed
 716  *
 717  */
 718 static void xs_tcp_state_change(struct sock *sk)
 719 {
 720         struct rpc_xprt *xprt;
 721
 722         read_lock(&sk->sk_callback_lock);
 723         if (!(xprt = xprt_from_sock(sk)))
 724                 goto out;
 725         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 726         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 727                                 sk->sk_state, xprt_connected(xprt),
 728                                 sock_flag(sk, SOCK_DEAD),
 729                                 sock_flag(sk, SOCK_ZAPPED));
 730
 731         switch (sk->sk_state) {
 732         case TCP_ESTABLISHED:
 733                 spin_lock_bh(&xprt->transport_lock);
 734                 if (!xprt_test_and_set_connected(xprt)) {
 735                         /* Reset TCP record info */
 736                         xprt->tcp_offset = 0;
 737                         xprt->tcp_reclen = 0;
 738                         xprt->tcp_copied = 0;
 739                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 740                         xprt_wake_pending_tasks(xprt, 0);
 741                 }
 742                 spin_unlock_bh(&xprt->transport_lock);
 743                 break;
 744         case TCP_SYN_SENT:
 745         case TCP_SYN_RECV:
 746                 break;
 747         default:
 748                 xprt_disconnect(xprt);
 749                 break;
 750         }
 751  out:
 752         read_unlock(&sk->sk_callback_lock);
 753 }
 754
 755 /**
 756  * xs_udp_write_space - callback invoked when socket buffer space
 757  *                             becomes available
 758  * @sk: socket whose state has changed
 759  *
 760  * Called when more output buffer space is available for this socket.
 761  * We try not to wake our writers until they can make "significant"
 762  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 763  * with a bunch of small requests.
 764  */
 765 static void xs_udp_write_space(struct sock *sk)
 766 {
 767         read_lock(&sk->sk_callback_lock);
 768
 769         /* from net/core/sock.c:sock_def_write_space */
 770         if (sock_writeable(sk)) {
 771                 struct socket *sock;
 772                 struct rpc_xprt *xprt;
 773
 774                 if (unlikely(!(sock = sk->sk_socket)))
 775                         goto out;
 776                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 777                         goto out;
 778                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 779                         goto out;
 780
 781                 xprt_write_space(xprt);
 782         }
 783
 784  out:
 785         read_unlock(&sk->sk_callback_lock);
 786 }
 787
 788 /**
 789  * xs_tcp_write_space - callback invoked when socket buffer space
 790  *                             becomes available
 791  * @sk: socket whose state has changed
 792  *
 793  * Called when more output buffer space is available for this socket.
 794  * We try not to wake our writers until they can make "significant"
 795  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 796  * with a bunch of small requests.
 797  */
 798 static void xs_tcp_write_space(struct sock *sk)
 799 {
 800         read_lock(&sk->sk_callback_lock);
 801
 802         /* from net/core/stream.c:sk_stream_write_space */
 803         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 804                 struct socket *sock;
 805                 struct rpc_xprt *xprt;
 806
 807                 if (unlikely(!(sock = sk->sk_socket)))
 808                         goto out;
 809                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 810                         goto out;
 811                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 812                         goto out;
 813
 814                 xprt_write_space(xprt);
 815         }
 816
 817  out:
 818         read_unlock(&sk->sk_callback_lock);
 819 }
 820
 821 /**
 822  * xs_udp_set_buffer_size - set send and receive limits
 823  * @xprt: generic transport
 824  *
 825  * Set socket send and receive limits based on the
 826  * sndsize and rcvsize fields in the generic transport
 827  * structure.
 828  */
 829 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt)
 830 {
 831         struct sock *sk = xprt->inet;
 832
 833         if (xprt->rcvsize) {
 834                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 835                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 836         }
 837         if (xprt->sndsize) {
 838                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 839                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 840                 sk->sk_write_space(sk);
 841         }
 842 }
 843
 844 /**
 845  * xs_tcp_set_buffer_size - set send and receive limits
 846  * @xprt: generic transport
 847  *
 848  * Nothing to do for TCP.
 849  */
 850 static void xs_tcp_set_buffer_size(struct rpc_xprt *xprt)
 851 {
 852         return;
 853 }
 854
 855 /**
 856  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
 857  * @task: task that timed out
 858  *
 859  * Adjust the congestion window after a retransmit timeout has occurred.
 860  */
 861 static void xs_udp_timer(struct rpc_task *task)
 862 {
 863         xprt_adjust_cwnd(task, -ETIMEDOUT);
 864 }
 865
 866 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
 867 {
 868         struct sockaddr_in myaddr = {
 869                 .sin_family = AF_INET,
 870         };
 871         int err;
 872         unsigned short port = xprt->port;
 873
 874         do {
 875                 myaddr.sin_port = htons(port);
 876                 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
 877                                                 sizeof(myaddr));
 878                 if (err == 0) {
 879                         xprt->port = port;
 880                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
 881                                         port);
 882                         return 0;
 883                 }
 884                 if (port <= xprt_min_resvport)
 885                         port = xprt_max_resvport;
 886                 else
 887                         port--;
 888         } while (err == -EADDRINUSE && port != xprt->port);
 889
 890         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
 891         return err;
 892 }
 893
 894 /**
 895  * xs_udp_connect_worker - set up a UDP socket
 896  * @args: RPC transport to connect
 897  *
 898  * Invoked by a work queue tasklet.
 899  */
 900 static void xs_udp_connect_worker(void *args)
 901 {
 902         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
 903         struct socket *sock = xprt->sock;
 904         int err, status = -EIO;
 905
 906         if (xprt->shutdown || xprt->addr.sin_port == 0)
 907                 goto out;
 908
 909         dprintk("RPC:      xs_udp_connect_worker for xprt %p\n", xprt);
 910
 911         /* Start by resetting any existing state */
 912         xs_close(xprt);
 913
 914         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
 915                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
 916                 goto out;
 917         }
 918
 919         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
 920                 sock_release(sock);
 921                 goto out;
 922         }
 923
 924         if (!xprt->inet) {
 925                 struct sock *sk = sock->sk;
 926
 927                 write_lock_bh(&sk->sk_callback_lock);
 928
 929                 sk->sk_user_data = xprt;
 930                 xprt->old_data_ready = sk->sk_data_ready;
 931                 xprt->old_state_change = sk->sk_state_change;
 932                 xprt->old_write_space = sk->sk_write_space;
 933                 sk->sk_data_ready = xs_udp_data_ready;
 934                 sk->sk_write_space = xs_udp_write_space;
 935                 sk->sk_no_check = UDP_CSUM_NORCV;
 936
 937                 xprt_set_connected(xprt);
 938
 939                 /* Reset to new socket */
 940                 xprt->sock = sock;
 941                 xprt->inet = sk;
 942
 943                 write_unlock_bh(&sk->sk_callback_lock);
 944         }
 945         xs_udp_set_buffer_size(xprt);
 946         status = 0;
 947 out:
 948         xprt_wake_pending_tasks(xprt, status);
 949         xprt_clear_connecting(xprt);
 950 }
 951
 952 /**
 953  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
 954  * @args: RPC transport to connect
 955  *
 956  * Invoked by a work queue tasklet.
 957  */
 958 static void xs_tcp_connect_worker(void *args)
 959 {
 960         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
 961         struct socket *sock = xprt->sock;
 962         int err, status = -EIO;
 963
 964         if (xprt->shutdown || xprt->addr.sin_port == 0)
 965                 goto out;
 966
 967         dprintk("RPC:      xs_tcp_connect_worker for xprt %p\n", xprt);
 968
 969         /* Start by resetting any existing socket state */
 970         xs_close(xprt);
 971
 972         if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
 973                 dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
 974                 goto out;
 975         }
 976
 977         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
 978                 sock_release(sock);
 979                 goto out;
 980         }
 981
 982         if (!xprt->inet) {
 983                 struct sock *sk = sock->sk;
 984
 985                 write_lock_bh(&sk->sk_callback_lock);
 986
 987                 sk->sk_user_data = xprt;
 988                 xprt->old_data_ready = sk->sk_data_ready;
 989                 xprt->old_state_change = sk->sk_state_change;
 990                 xprt->old_write_space = sk->sk_write_space;
 991                 sk->sk_data_ready = xs_tcp_data_ready;
 992                 sk->sk_state_change = xs_tcp_state_change;
 993                 sk->sk_write_space = xs_tcp_write_space;
 994                 tcp_sk(sk)->nonagle = 1;
 995
 996                 xprt_clear_connected(xprt);
 997
 998                 /* Reset to new socket */
 999                 xprt->sock = sock;
1000                 xprt->inet = sk;
1001
1002                 write_unlock_bh(&sk->sk_callback_lock);
1003         }
1004
1005         /* Tell the socket layer to start connecting... */
1006         status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
1007                         sizeof(xprt->addr), O_NONBLOCK);
1008         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1009                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1010         if (status < 0) {
1011                 switch (status) {
1012                         case -EINPROGRESS:
1013                         case -EALREADY:
1014                                 goto out_clear;
1015                 }
1016         }
1017 out:
1018         xprt_wake_pending_tasks(xprt, status);
1019 out_clear:
1020         xprt_clear_connecting(xprt);
1021 }
1022
1023 /**
1024  * xs_connect - connect a socket to a remote endpoint
1025  * @task: address of RPC task that manages state of connect request
1026  *
1027  * TCP: If the remote end dropped the connection, delay reconnecting.
1028  */
1029 static void xs_connect(struct rpc_task *task)
1030 {
1031         struct rpc_xprt *xprt = task->tk_xprt;
1032
1033         if (xprt_test_and_set_connecting(xprt))
1034                 return;
1035
1036         if (xprt->sock != NULL) {
1037                 dprintk("RPC:      xs_connect delayed xprt %p\n", xprt);
1038                 schedule_delayed_work(&xprt->connect_worker,
1039                                         RPC_REESTABLISH_TIMEOUT);
1040         } else {
1041                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1042                 schedule_work(&xprt->connect_worker);
1043
1044                 /* flush_scheduled_work can sleep... */
1045                 if (!RPC_IS_ASYNC(task))
1046                         flush_scheduled_work();
1047         }
1048 }
1049
1050 static struct rpc_xprt_ops xs_udp_ops = {
1051         .set_buffer_size        = xs_udp_set_buffer_size,
1052         .reserve_xprt           = xprt_reserve_xprt_cong,
1053         .release_xprt           = xprt_release_xprt_cong,
1054         .connect                = xs_connect,
1055         .send_request           = xs_udp_send_request,
1056         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1057         .timer                  = xs_udp_timer,
1058         .release_request        = xprt_release_rqst_cong,
1059         .close                  = xs_close,
1060         .destroy                = xs_destroy,
1061 };
1062
1063 static struct rpc_xprt_ops xs_tcp_ops = {
1064         .set_buffer_size        = xs_tcp_set_buffer_size,
1065         .reserve_xprt           = xprt_reserve_xprt,
1066         .release_xprt           = xprt_release_xprt,
1067         .connect                = xs_connect,
1068         .send_request           = xs_tcp_send_request,
1069         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1070         .close                  = xs_close,
1071         .destroy                = xs_destroy,
1072 };
1073
1074 /**
1075  * xs_setup_udp - Set up transport to use a UDP socket
1076  * @xprt: transport to set up
1077  * @to:   timeout parameters
1078  *
1079  */
1080 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1081 {
1082         size_t slot_table_size;
1083
1084         dprintk("RPC:      setting up udp-ipv4 transport...\n");
1085
1086         xprt->max_reqs = xprt_udp_slot_table_entries;
1087         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1088         xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1089         if (xprt->slot == NULL)
1090                 return -ENOMEM;
1091         memset(xprt->slot, 0, slot_table_size);
1092
1093         xprt->prot = IPPROTO_UDP;
1094         xprt->port = xprt_max_resvport;
1095         xprt->tsh_size = 0;
1096         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1097         /* XXX: header size can vary due to auth type, IPv6, etc. */
1098         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1099
1100         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1101
1102         xprt->ops = &xs_udp_ops;
1103
1104         if (to)
1105                 xprt->timeout = *to;
1106         else
1107                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1108
1109         return 0;
1110 }
1111
1112 /**
1113  * xs_setup_tcp - Set up transport to use a TCP socket
1114  * @xprt: transport to set up
1115  * @to: timeout parameters
1116  *
1117  */
1118 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1119 {
1120         size_t slot_table_size;
1121
1122         dprintk("RPC:      setting up tcp-ipv4 transport...\n");
1123
1124         xprt->max_reqs = xprt_tcp_slot_table_entries;
1125         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1126         xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1127         if (xprt->slot == NULL)
1128                 return -ENOMEM;
1129         memset(xprt->slot, 0, slot_table_size);
1130
1131         xprt->prot = IPPROTO_TCP;
1132         xprt->port = xprt_max_resvport;
1133         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1134         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1135         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1136
1137         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1138
1139         xprt->ops = &xs_tcp_ops;
1140
1141         if (to)
1142                 xprt->timeout = *to;
1143         else
1144                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1145
1146         return 0;
1147 }