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
  16 #include <linux/types.h>
  17 #include <linux/slab.h>
  18 #include <linux/capability.h>
  19 #include <linux/sched.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/errno.h>
  22 #include <linux/socket.h>
  23 #include <linux/in.h>
  24 #include <linux/net.h>
  25 #include <linux/mm.h>
  26 #include <linux/udp.h>
  27 #include <linux/tcp.h>
  28 #include <linux/sunrpc/clnt.h>
  29 #include <linux/file.h>
  30
  31 #include <net/sock.h>
  32 #include <net/checksum.h>
  33 #include <net/udp.h>
  34 #include <net/tcp.h>
  35
  36 /*
  37  * Maximum port number to use when requesting a reserved port.
  38  */
  39 #define XS_MAX_RESVPORT         (800U)
  40
  41 #ifdef RPC_DEBUG
  42 # undef  RPC_DEBUG_DATA
  43 # define RPCDBG_FACILITY        RPCDBG_TRANS
  44 #endif
  45
  46 #ifdef RPC_DEBUG_DATA
  47 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
  48 {
  49         u8 *buf = (u8 *) packet;
  50         int j;
  51
  52         dprintk("RPC:      %s\n", msg);
  53         for (j = 0; j < count && j < 128; j += 4) {
  54                 if (!(j & 31)) {
  55                         if (j)
  56                                 dprintk("\n");
  57                         dprintk("0x%04x ", j);
  58                 }
  59                 dprintk("%02x%02x%02x%02x ",
  60                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
  61         }
  62         dprintk("\n");
  63 }
  64 #else
  65 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
  66 {
  67         /* NOP */
  68 }
  69 #endif
  70
  71 /**
  72  * xs_sendpages - write pages directly to a socket
  73  * @sock: socket to send on
  74  * @addr: UDP only -- address of destination
  75  * @addrlen: UDP only -- length of destination address
  76  * @xdr: buffer containing this request
  77  * @base: starting position in the buffer
  78  *
  79  */
  80 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, int msgflags)
  81 {
  82         struct page **ppage = xdr->pages;
  83         unsigned int len, pglen = xdr->page_len;
  84         int err, ret = 0;
  85         ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
  86
  87         len = xdr->head[0].iov_len;
  88         if (base < len || (addr != NULL && base == 0)) {
  89                 struct kvec iov = {
  90                         .iov_base = xdr->head[0].iov_base + base,
  91                         .iov_len  = len - base,
  92                 };
  93                 struct msghdr msg = {
  94                         .msg_name    = addr,
  95                         .msg_namelen = addrlen,
  96                         .msg_flags   = msgflags,
  97                 };
  98                 if (xdr->len > len)
  99                         msg.msg_flags |= MSG_MORE;
 100
 101                 if (iov.iov_len != 0)
 102                         err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 103                 else
 104                         err = kernel_sendmsg(sock, &msg, NULL, 0, 0);
 105                 if (ret == 0)
 106                         ret = err;
 107                 else if (err > 0)
 108                         ret += err;
 109                 if (err != iov.iov_len)
 110                         goto out;
 111                 base = 0;
 112         } else
 113                 base -= len;
 114
 115         if (pglen == 0)
 116                 goto copy_tail;
 117         if (base >= pglen) {
 118                 base -= pglen;
 119                 goto copy_tail;
 120         }
 121         if (base || xdr->page_base) {
 122                 pglen -= base;
 123                 base += xdr->page_base;
 124                 ppage += base >> PAGE_CACHE_SHIFT;
 125                 base &= ~PAGE_CACHE_MASK;
 126         }
 127
 128         sendpage = sock->ops->sendpage ? : sock_no_sendpage;
 129         do {
 130                 int flags = msgflags;
 131
 132                 len = PAGE_CACHE_SIZE;
 133                 if (base)
 134                         len -= base;
 135                 if (pglen < len)
 136                         len = pglen;
 137
 138                 if (pglen != len || xdr->tail[0].iov_len != 0)
 139                         flags |= MSG_MORE;
 140
 141                 /* Hmm... We might be dealing with highmem pages */
 142                 if (PageHighMem(*ppage))
 143                         sendpage = sock_no_sendpage;
 144                 err = sendpage(sock, *ppage, base, len, flags);
 145                 if (ret == 0)
 146                         ret = err;
 147                 else if (err > 0)
 148                         ret += err;
 149                 if (err != len)
 150                         goto out;
 151                 base = 0;
 152                 ppage++;
 153         } while ((pglen -= len) != 0);
 154 copy_tail:
 155         len = xdr->tail[0].iov_len;
 156         if (base < len) {
 157                 struct kvec iov = {
 158                         .iov_base = xdr->tail[0].iov_base + base,
 159                         .iov_len  = len - base,
 160                 };
 161                 struct msghdr msg = {
 162                         .msg_flags   = msgflags,
 163                 };
 164                 err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 165                 if (ret == 0)
 166                         ret = err;
 167                 else if (err > 0)
 168                         ret += err;
 169         }
 170 out:
 171         return ret;
 172 }
 173
 174 /**
 175  * xs_sendmsg - write an RPC request to a socket
 176  * @xprt: generic transport
 177  * @req: the RPC request to write
 178  *
 179  */
 180 static int xs_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req)
 181 {
 182         struct socket *sock = xprt->sock;
 183         struct xdr_buf *xdr = &req->rq_snd_buf;
 184         struct sockaddr *addr = NULL;
 185         int addrlen = 0;
 186         unsigned int skip;
 187         int result;
 188
 189         if (!sock)
 190                 return -ENOTCONN;
 191
 192         xs_pktdump("packet data:",
 193                                 req->rq_svec->iov_base,
 194                                 req->rq_svec->iov_len);
 195
 196         /* For UDP, we need to provide an address */
 197         if (!xprt->stream) {
 198                 addr = (struct sockaddr *) &xprt->addr;
 199                 addrlen = sizeof(xprt->addr);
 200         }
 201         /* Don't repeat bytes */
 202         skip = req->rq_bytes_sent;
 203
 204         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 205         result = xs_sendpages(sock, addr, addrlen, xdr, skip, MSG_DONTWAIT);
 206
 207         dprintk("RPC:      xs_sendmsg(%d) = %d\n", xdr->len - skip, result);
 208
 209         if (result >= 0)
 210                 return result;
 211
 212         switch (result) {
 213         case -ECONNREFUSED:
 214                 /* When the server has died, an ICMP port unreachable message
 215                  * prompts ECONNREFUSED. */
 216         case -EAGAIN:
 217                 break;
 218         case -ECONNRESET:
 219         case -ENOTCONN:
 220         case -EPIPE:
 221                 /* connection broken */
 222                 if (xprt->stream)
 223                         result = -ENOTCONN;
 224                 break;
 225         default:
 226                 break;
 227         }
 228         return result;
 229 }
 230
 231 /**
 232  * xs_send_request - write an RPC request to a socket
 233  * @task: address of RPC task that manages the state of an RPC request
 234  *
 235  * Return values:
 236  *      0:  The request has been sent
 237  * EAGAIN:  The socket was blocked, please call again later to
 238  *          complete the request
 239  *  other:  Some other error occured, the request was not sent
 240  *
 241  * XXX: In the case of soft timeouts, should we eventually give up
 242  *      if the socket is not able to make progress?
 243  */
 244 static int xs_send_request(struct rpc_task *task)
 245 {
 246         struct rpc_rqst *req = task->tk_rqstp;
 247         struct rpc_xprt *xprt = req->rq_xprt;
 248         int status, retry = 0;
 249
 250         /* set up everything as needed. */
 251         /* Write the record marker */
 252         if (xprt->stream) {
 253                 u32 *marker = req->rq_svec[0].iov_base;
 254
 255                 *marker = htonl(0x80000000|(req->rq_slen-sizeof(*marker)));
 256         }
 257
 258         /* Continue transmitting the packet/record. We must be careful
 259          * to cope with writespace callbacks arriving _after_ we have
 260          * called sendmsg().
 261          */
 262         while (1) {
 263                 req->rq_xtime = jiffies;
 264                 status = xs_sendmsg(xprt, req);
 265
 266                 if (status < 0)
 267                         break;
 268
 269                 if (xprt->stream) {
 270                         req->rq_bytes_sent += status;
 271
 272                         /* If we've sent the entire packet, immediately
 273                          * reset the count of bytes sent. */
 274                         if (req->rq_bytes_sent >= req->rq_slen) {
 275                                 req->rq_bytes_sent = 0;
 276                                 return 0;
 277                         }
 278                 } else {
 279                         if (status >= req->rq_slen)
 280                                 return 0;
 281                         status = -EAGAIN;
 282                         break;
 283                 }
 284
 285                 dprintk("RPC: %4d xmit incomplete (%d left of %d)\n",
 286                                 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 287                                 req->rq_slen);
 288
 289                 status = -EAGAIN;
 290                 if (retry++ > 50)
 291                         break;
 292         }
 293
 294         if (status == -EAGAIN) {
 295                 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
 296                         /* Protect against races with xs_write_space */
 297                         spin_lock_bh(&xprt->sock_lock);
 298                         /* Don't race with disconnect */
 299                         if (!xprt_connected(xprt))
 300                                 task->tk_status = -ENOTCONN;
 301                         else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags)) {
 302                                 task->tk_timeout = req->rq_timeout;
 303                                 rpc_sleep_on(&xprt->pending, task, NULL, NULL);
 304                         }
 305                         spin_unlock_bh(&xprt->sock_lock);
 306                         return status;
 307                 }
 308                 /* Keep holding the socket if it is blocked */
 309                 rpc_delay(task, HZ>>4);
 310         }
 311         return status;
 312 }
 313
 314 /**
 315  * xs_close - close a socket
 316  * @xprt: transport
 317  *
 318  */
 319 static void xs_close(struct rpc_xprt *xprt)
 320 {
 321         struct socket *sock = xprt->sock;
 322         struct sock *sk = xprt->inet;
 323
 324         if (!sk)
 325                 return;
 326
 327         dprintk("RPC:      xs_close xprt %p\n", xprt);
 328
 329         write_lock_bh(&sk->sk_callback_lock);
 330         xprt->inet = NULL;
 331         xprt->sock = NULL;
 332
 333         sk->sk_user_data = NULL;
 334         sk->sk_data_ready = xprt->old_data_ready;
 335         sk->sk_state_change = xprt->old_state_change;
 336         sk->sk_write_space = xprt->old_write_space;
 337         write_unlock_bh(&sk->sk_callback_lock);
 338
 339         sk->sk_no_check = 0;
 340
 341         sock_release(sock);
 342 }
 343
 344 /**
 345  * xs_destroy - prepare to shutdown a transport
 346  * @xprt: doomed transport
 347  *
 348  */
 349 static void xs_destroy(struct rpc_xprt *xprt)
 350 {
 351         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 352
 353         cancel_delayed_work(&xprt->sock_connect);
 354         flush_scheduled_work();
 355
 356         xprt_disconnect(xprt);
 357         xs_close(xprt);
 358         kfree(xprt->slot);
 359 }
 360
 361 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 362 {
 363         return (struct rpc_xprt *) sk->sk_user_data;
 364 }
 365
 366 /**
 367  * xs_udp_data_ready - "data ready" callback for UDP sockets
 368  * @sk: socket with data to read
 369  * @len: how much data to read
 370  *
 371  */
 372 static void xs_udp_data_ready(struct sock *sk, int len)
 373 {
 374         struct rpc_task *task;
 375         struct rpc_xprt *xprt;
 376         struct rpc_rqst *rovr;
 377         struct sk_buff *skb;
 378         int err, repsize, copied;
 379         u32 _xid, *xp;
 380
 381         read_lock(&sk->sk_callback_lock);
 382         dprintk("RPC:      xs_udp_data_ready...\n");
 383         if (!(xprt = xprt_from_sock(sk)))
 384                 goto out;
 385
 386         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 387                 goto out;
 388
 389         if (xprt->shutdown)
 390                 goto dropit;
 391
 392         repsize = skb->len - sizeof(struct udphdr);
 393         if (repsize < 4) {
 394                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 395                 goto dropit;
 396         }
 397
 398         /* Copy the XID from the skb... */
 399         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 400                                 sizeof(_xid), &_xid);
 401         if (xp == NULL)
 402                 goto dropit;
 403
 404         /* Look up and lock the request corresponding to the given XID */
 405         spin_lock(&xprt->sock_lock);
 406         rovr = xprt_lookup_rqst(xprt, *xp);
 407         if (!rovr)
 408                 goto out_unlock;
 409         task = rovr->rq_task;
 410
 411         dprintk("RPC: %4d received reply\n", task->tk_pid);
 412
 413         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 414                 copied = repsize;
 415
 416         /* Suck it into the iovec, verify checksum if not done by hw. */
 417         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 418                 goto out_unlock;
 419
 420         /* Something worked... */
 421         dst_confirm(skb->dst);
 422
 423         xprt_complete_rqst(xprt, rovr, copied);
 424
 425  out_unlock:
 426         spin_unlock(&xprt->sock_lock);
 427  dropit:
 428         skb_free_datagram(sk, skb);
 429  out:
 430         read_unlock(&sk->sk_callback_lock);
 431 }
 432
 433 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 434 {
 435         if (len > desc->count)
 436                 len = desc->count;
 437         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 438                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 439                                 len, desc->count);
 440                 return 0;
 441         }
 442         desc->offset += len;
 443         desc->count -= len;
 444         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 445                         len, desc->count);
 446         return len;
 447 }
 448
 449 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 450 {
 451         size_t len, used;
 452         char *p;
 453
 454         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 455         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 456         used = xs_tcp_copy_data(desc, p, len);
 457         xprt->tcp_offset += used;
 458         if (used != len)
 459                 return;
 460         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 461         if (xprt->tcp_reclen & 0x80000000)
 462                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 463         else
 464                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 465         xprt->tcp_reclen &= 0x7fffffff;
 466         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 467         xprt->tcp_offset = 0;
 468         /* Sanity check of the record length */
 469         if (xprt->tcp_reclen < 4) {
 470                 dprintk("RPC:      invalid TCP record fragment length\n");
 471                 xprt_disconnect(xprt);
 472                 return;
 473         }
 474         dprintk("RPC:      reading TCP record fragment of length %d\n",
 475                         xprt->tcp_reclen);
 476 }
 477
 478 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 479 {
 480         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 481                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 482         if (xprt->tcp_offset == xprt->tcp_reclen) {
 483                 xprt->tcp_flags |= XPRT_COPY_RECM;
 484                 xprt->tcp_offset = 0;
 485                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 486                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 487                         xprt->tcp_flags |= XPRT_COPY_XID;
 488                         xprt->tcp_copied = 0;
 489                 }
 490         }
 491 }
 492
 493 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 494 {
 495         size_t len, used;
 496         char *p;
 497
 498         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 499         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 500         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 501         used = xs_tcp_copy_data(desc, p, len);
 502         xprt->tcp_offset += used;
 503         if (used != len)
 504                 return;
 505         xprt->tcp_flags &= ~XPRT_COPY_XID;
 506         xprt->tcp_flags |= XPRT_COPY_DATA;
 507         xprt->tcp_copied = 4;
 508         dprintk("RPC:      reading reply for XID %08x\n",
 509                                                 ntohl(xprt->tcp_xid));
 510         xs_tcp_check_recm(xprt);
 511 }
 512
 513 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 514 {
 515         struct rpc_rqst *req;
 516         struct xdr_buf *rcvbuf;
 517         size_t len;
 518         ssize_t r;
 519
 520         /* Find and lock the request corresponding to this xid */
 521         spin_lock(&xprt->sock_lock);
 522         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 523         if (!req) {
 524                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 525                 dprintk("RPC:      XID %08x request not found!\n",
 526                                 ntohl(xprt->tcp_xid));
 527                 spin_unlock(&xprt->sock_lock);
 528                 return;
 529         }
 530
 531         rcvbuf = &req->rq_private_buf;
 532         len = desc->count;
 533         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 534                 skb_reader_t my_desc;
 535
 536                 len = xprt->tcp_reclen - xprt->tcp_offset;
 537                 memcpy(&my_desc, desc, sizeof(my_desc));
 538                 my_desc.count = len;
 539                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 540                                           &my_desc, xs_tcp_copy_data);
 541                 desc->count -= r;
 542                 desc->offset += r;
 543         } else
 544                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 545                                           desc, xs_tcp_copy_data);
 546
 547         if (r > 0) {
 548                 xprt->tcp_copied += r;
 549                 xprt->tcp_offset += r;
 550         }
 551         if (r != len) {
 552                 /* Error when copying to the receive buffer,
 553                  * usually because we weren't able to allocate
 554                  * additional buffer pages. All we can do now
 555                  * is turn off XPRT_COPY_DATA, so the request
 556                  * will not receive any additional updates,
 557                  * and time out.
 558                  * Any remaining data from this record will
 559                  * be discarded.
 560                  */
 561                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 562                 dprintk("RPC:      XID %08x truncated request\n",
 563                                 ntohl(xprt->tcp_xid));
 564                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 565                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 566                 goto out;
 567         }
 568
 569         dprintk("RPC:      XID %08x read %Zd bytes\n",
 570                         ntohl(xprt->tcp_xid), r);
 571         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 572                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 573
 574         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 575                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 576         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 577                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 578                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 579         }
 580
 581 out:
 582         if (!(xprt->tcp_flags & XPRT_COPY_DATA)) {
 583                 dprintk("RPC: %4d received reply complete\n",
 584                                 req->rq_task->tk_pid);
 585                 xprt_complete_rqst(xprt, req, xprt->tcp_copied);
 586         }
 587         spin_unlock(&xprt->sock_lock);
 588         xs_tcp_check_recm(xprt);
 589 }
 590
 591 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 592 {
 593         size_t len;
 594
 595         len = xprt->tcp_reclen - xprt->tcp_offset;
 596         if (len > desc->count)
 597                 len = desc->count;
 598         desc->count -= len;
 599         desc->offset += len;
 600         xprt->tcp_offset += len;
 601         dprintk("RPC:      discarded %Zu bytes\n", len);
 602         xs_tcp_check_recm(xprt);
 603 }
 604
 605 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 606 {
 607         struct rpc_xprt *xprt = rd_desc->arg.data;
 608         skb_reader_t desc = {
 609                 .skb    = skb,
 610                 .offset = offset,
 611                 .count  = len,
 612                 .csum   = 0
 613         };
 614
 615         dprintk("RPC:      xs_tcp_data_recv started\n");
 616         do {
 617                 /* Read in a new fragment marker if necessary */
 618                 /* Can we ever really expect to get completely empty fragments? */
 619                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 620                         xs_tcp_read_fraghdr(xprt, &desc);
 621                         continue;
 622                 }
 623                 /* Read in the xid if necessary */
 624                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 625                         xs_tcp_read_xid(xprt, &desc);
 626                         continue;
 627                 }
 628                 /* Read in the request data */
 629                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 630                         xs_tcp_read_request(xprt, &desc);
 631                         continue;
 632                 }
 633                 /* Skip over any trailing bytes on short reads */
 634                 xs_tcp_read_discard(xprt, &desc);
 635         } while (desc.count);
 636         dprintk("RPC:      xs_tcp_data_recv done\n");
 637         return len - desc.count;
 638 }
 639
 640 /**
 641  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 642  * @sk: socket with data to read
 643  * @bytes: how much data to read
 644  *
 645  */
 646 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 647 {
 648         struct rpc_xprt *xprt;
 649         read_descriptor_t rd_desc;
 650
 651         read_lock(&sk->sk_callback_lock);
 652         dprintk("RPC:      xs_tcp_data_ready...\n");
 653         if (!(xprt = xprt_from_sock(sk)))
 654                 goto out;
 655         if (xprt->shutdown)
 656                 goto out;
 657
 658         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 659         rd_desc.arg.data = xprt;
 660         rd_desc.count = 65536;
 661         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 662 out:
 663         read_unlock(&sk->sk_callback_lock);
 664 }
 665
 666 /**
 667  * xs_tcp_state_change - callback to handle TCP socket state changes
 668  * @sk: socket whose state has changed
 669  *
 670  */
 671 static void xs_tcp_state_change(struct sock *sk)
 672 {
 673         struct rpc_xprt *xprt;
 674
 675         read_lock(&sk->sk_callback_lock);
 676         if (!(xprt = xprt_from_sock(sk)))
 677                 goto out;
 678         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 679         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 680                                 sk->sk_state, xprt_connected(xprt),
 681                                 sock_flag(sk, SOCK_DEAD),
 682                                 sock_flag(sk, SOCK_ZAPPED));
 683
 684         switch (sk->sk_state) {
 685         case TCP_ESTABLISHED:
 686                 spin_lock_bh(&xprt->sock_lock);
 687                 if (!xprt_test_and_set_connected(xprt)) {
 688                         /* Reset TCP record info */
 689                         xprt->tcp_offset = 0;
 690                         xprt->tcp_reclen = 0;
 691                         xprt->tcp_copied = 0;
 692                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 693                         rpc_wake_up(&xprt->pending);
 694                 }
 695                 spin_unlock_bh(&xprt->sock_lock);
 696                 break;
 697         case TCP_SYN_SENT:
 698         case TCP_SYN_RECV:
 699                 break;
 700         default:
 701                 xprt_disconnect(xprt);
 702                 break;
 703         }
 704  out:
 705         read_unlock(&sk->sk_callback_lock);
 706 }
 707
 708 /**
 709  * xs_write_space - callback invoked when socket buffer space becomes
 710  *                         available
 711  * @sk: socket whose state has changed
 712  *
 713  * Called when more output buffer space is available for this socket.
 714  * We try not to wake our writers until they can make "significant"
 715  * progress, otherwise we'll waste resources thrashing sock_sendmsg
 716  * with a bunch of small requests.
 717  */
 718 static void xs_write_space(struct sock *sk)
 719 {
 720         struct rpc_xprt *xprt;
 721         struct socket *sock;
 722
 723         read_lock(&sk->sk_callback_lock);
 724         if (!(xprt = xprt_from_sock(sk)) || !(sock = sk->sk_socket))
 725                 goto out;
 726         if (xprt->shutdown)
 727                 goto out;
 728
 729         /* Wait until we have enough socket memory */
 730         if (xprt->stream) {
 731                 /* from net/core/stream.c:sk_stream_write_space */
 732                 if (sk_stream_wspace(sk) < sk_stream_min_wspace(sk))
 733                         goto out;
 734         } else {
 735                 /* from net/core/sock.c:sock_def_write_space */
 736                 if (!sock_writeable(sk))
 737                         goto out;
 738         }
 739
 740         if (!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))
 741                 goto out;
 742
 743         spin_lock_bh(&xprt->sock_lock);
 744         if (xprt->snd_task)
 745                 rpc_wake_up_task(xprt->snd_task);
 746         spin_unlock_bh(&xprt->sock_lock);
 747 out:
 748         read_unlock(&sk->sk_callback_lock);
 749 }
 750
 751 /**
 752  * xs_set_buffer_size - set send and receive limits
 753  * @xprt: generic transport
 754  *
 755  * Set socket send and receive limits based on the
 756  * sndsize and rcvsize fields in the generic transport
 757  * structure. This applies only to UDP sockets.
 758  */
 759 static void xs_set_buffer_size(struct rpc_xprt *xprt)
 760 {
 761         struct sock *sk = xprt->inet;
 762
 763         if (xprt->stream)
 764                 return;
 765         if (xprt->rcvsize) {
 766                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 767                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 768         }
 769         if (xprt->sndsize) {
 770                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 771                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 772                 sk->sk_write_space(sk);
 773         }
 774 }
 775
 776 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
 777 {
 778         struct sockaddr_in myaddr = {
 779                 .sin_family = AF_INET,
 780         };
 781         int err, port;
 782
 783         /* Were we already bound to a given port? Try to reuse it */
 784         port = xprt->port;
 785         do {
 786                 myaddr.sin_port = htons(port);
 787                 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
 788                                                 sizeof(myaddr));
 789                 if (err == 0) {
 790                         xprt->port = port;
 791                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
 792                                         port);
 793                         return 0;
 794                 }
 795                 if (--port == 0)
 796                         port = XS_MAX_RESVPORT;
 797         } while (err == -EADDRINUSE && port != xprt->port);
 798
 799         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
 800         return err;
 801 }
 802
 803 static struct socket *xs_create(struct rpc_xprt *xprt, int proto, int resvport)
 804 {
 805         struct socket *sock;
 806         int type, err;
 807
 808         dprintk("RPC:      xs_create(%s %d)\n",
 809                            (proto == IPPROTO_UDP)? "udp" : "tcp", proto);
 810
 811         type = (proto == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
 812
 813         if ((err = sock_create_kern(PF_INET, type, proto, &sock)) < 0) {
 814                 dprintk("RPC:      can't create socket (%d).\n", -err);
 815                 return NULL;
 816         }
 817
 818         /* If the caller has the capability, bind to a reserved port */
 819         if (resvport && xs_bindresvport(xprt, sock) < 0)
 820                 goto failed;
 821
 822         return sock;
 823
 824 failed:
 825         sock_release(sock);
 826         return NULL;
 827 }
 828
 829 static void xs_bind(struct rpc_xprt *xprt, struct socket *sock)
 830 {
 831         struct sock *sk = sock->sk;
 832
 833         if (xprt->inet)
 834                 return;
 835
 836         write_lock_bh(&sk->sk_callback_lock);
 837         sk->sk_user_data = xprt;
 838         xprt->old_data_ready = sk->sk_data_ready;
 839         xprt->old_state_change = sk->sk_state_change;
 840         xprt->old_write_space = sk->sk_write_space;
 841         if (xprt->prot == IPPROTO_UDP) {
 842                 sk->sk_data_ready = xs_udp_data_ready;
 843                 sk->sk_no_check = UDP_CSUM_NORCV;
 844                 xprt_set_connected(xprt);
 845         } else {
 846                 tcp_sk(sk)->nonagle = 1;        /* disable Nagle's algorithm */
 847                 sk->sk_data_ready = xs_tcp_data_ready;
 848                 sk->sk_state_change = xs_tcp_state_change;
 849                 xprt_clear_connected(xprt);
 850         }
 851         sk->sk_write_space = xs_write_space;
 852
 853         /* Reset to new socket */
 854         xprt->sock = sock;
 855         xprt->inet = sk;
 856         write_unlock_bh(&sk->sk_callback_lock);
 857
 858         return;
 859 }
 860
 861 /**
 862  * xs_connect_worker - try to connect a socket to a remote endpoint
 863  * @args: RPC transport to connect
 864  *
 865  * Invoked by a work queue tasklet.
 866  */
 867 static void xs_connect_worker(void *args)
 868 {
 869         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
 870         struct socket *sock = xprt->sock;
 871         int status = -EIO;
 872
 873         if (xprt->shutdown || xprt->addr.sin_port == 0)
 874                 goto out;
 875
 876         dprintk("RPC:      xs_connect_worker xprt %p\n", xprt);
 877
 878         /*
 879          * Start by resetting any existing state
 880          */
 881         xs_close(xprt);
 882         sock = xs_create(xprt, xprt->prot, xprt->resvport);
 883         if (sock == NULL) {
 884                 /* couldn't create socket or bind to reserved port;
 885                  * this is likely a permanent error, so cause an abort */
 886                 goto out;
 887         }
 888         xs_bind(xprt, sock);
 889         xs_set_buffer_size(xprt);
 890
 891         status = 0;
 892         if (!xprt->stream)
 893                 goto out;
 894
 895         /*
 896          * Tell the socket layer to start connecting...
 897          */
 898         status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
 899                         sizeof(xprt->addr), O_NONBLOCK);
 900         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
 901                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
 902         if (status < 0) {
 903                 switch (status) {
 904                         case -EINPROGRESS:
 905                         case -EALREADY:
 906                                 goto out_clear;
 907                 }
 908         }
 909 out:
 910         if (status < 0)
 911                 rpc_wake_up_status(&xprt->pending, status);
 912         else
 913                 rpc_wake_up(&xprt->pending);
 914 out_clear:
 915         smp_mb__before_clear_bit();
 916         clear_bit(XPRT_CONNECTING, &xprt->sockstate);
 917         smp_mb__after_clear_bit();
 918 }
 919
 920 /**
 921  * xs_connect - connect a socket to a remote endpoint
 922  * @task: address of RPC task that manages state of connect request
 923  *
 924  * TCP: If the remote end dropped the connection, delay reconnecting.
 925  */
 926 static void xs_connect(struct rpc_task *task)
 927 {
 928         struct rpc_xprt *xprt = task->tk_xprt;
 929
 930         if (!test_and_set_bit(XPRT_CONNECTING, &xprt->sockstate)) {
 931                 if (xprt->sock != NULL) {
 932                         dprintk("RPC:      xs_connect delayed xprt %p\n", xprt);
 933                         schedule_delayed_work(&xprt->sock_connect,
 934                                         RPC_REESTABLISH_TIMEOUT);
 935                 } else {
 936                         dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
 937                         schedule_work(&xprt->sock_connect);
 938                         /* flush_scheduled_work can sleep... */
 939                         if (!RPC_IS_ASYNC(task))
 940                                 flush_scheduled_work();
 941                 }
 942         }
 943 }
 944
 945 static struct rpc_xprt_ops xs_ops = {
 946         .set_buffer_size        = xs_set_buffer_size,
 947         .connect                = xs_connect,
 948         .send_request           = xs_send_request,
 949         .close                  = xs_close,
 950         .destroy                = xs_destroy,
 951 };
 952
 953 extern unsigned int xprt_udp_slot_table_entries;
 954 extern unsigned int xprt_tcp_slot_table_entries;
 955
 956 /**
 957  * xs_setup_udp - Set up transport to use a UDP socket
 958  * @xprt: transport to set up
 959  * @to:   timeout parameters
 960  *
 961  */
 962 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
 963 {
 964         size_t slot_table_size;
 965
 966         dprintk("RPC:      setting up udp-ipv4 transport...\n");
 967
 968         xprt->max_reqs = xprt_udp_slot_table_entries;
 969         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
 970         xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
 971         if (xprt->slot == NULL)
 972                 return -ENOMEM;
 973         memset(xprt->slot, 0, slot_table_size);
 974
 975         xprt->prot = IPPROTO_UDP;
 976         xprt->port = XS_MAX_RESVPORT;
 977         xprt->stream = 0;
 978         xprt->nocong = 0;
 979         xprt->cwnd = RPC_INITCWND;
 980         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
 981         /* XXX: header size can vary due to auth type, IPv6, etc. */
 982         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
 983
 984         INIT_WORK(&xprt->sock_connect, xs_connect_worker, xprt);
 985
 986         xprt->ops = &xs_ops;
 987
 988         if (to)
 989                 xprt->timeout = *to;
 990         else
 991                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
 992
 993         return 0;
 994 }
 995
 996 /**
 997  * xs_setup_tcp - Set up transport to use a TCP socket
 998  * @xprt: transport to set up
 999  * @to: timeout parameters
1000  *
1001  */
1002 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1003 {
1004         size_t slot_table_size;
1005
1006         dprintk("RPC:      setting up tcp-ipv4 transport...\n");
1007
1008         xprt->max_reqs = xprt_tcp_slot_table_entries;
1009         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1010         xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1011         if (xprt->slot == NULL)
1012                 return -ENOMEM;
1013         memset(xprt->slot, 0, slot_table_size);
1014
1015         xprt->prot = IPPROTO_TCP;
1016         xprt->port = XS_MAX_RESVPORT;
1017         xprt->stream = 1;
1018         xprt->nocong = 1;
1019         xprt->cwnd = RPC_MAXCWND(xprt);
1020         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1021         xprt->max_payload = (1U << 31) - 1;
1022
1023         INIT_WORK(&xprt->sock_connect, xs_connect_worker, xprt);
1024
1025         xprt->ops = &xs_ops;
1026
1027         if (to)
1028                 xprt->timeout = *to;
1029         else
1030                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1031
1032         return 0;
1033 }