Skip to content
transfer.c 80.8 KiB
Newer Older
            data->info.header_size += (long)k->hbuflen;
            k->headerbytecount += (long)k->hbuflen;
            /* reset hbufp pointer && hbuflen */
            k->hbufp = data->state.headerbuff;
            k->hbuflen = 0;
          }
          while (!stop_reading && *k->str); /* header line within buffer */
          if(stop_reading)
            /* We've stopped dealing with input, get out of the do-while loop */
            break;
          /* We might have reached the end of the header part here, but
             there might be a non-header part left in the end of the read
             buffer. */
Daniel Stenberg's avatar
Daniel Stenberg committed
        }                       /* end if header mode */
        /* This is not an 'else if' since it may be a rest from the header
           parsing, where the beginning of the buffer is headers and the end
           is non-headers. */
        if (k->str && !k->header && (nread > 0 || is_empty_data)) {
          if(0 == k->bodywrites && !is_empty_data) {
            /* These checks are only made the first time we are about to
               write a piece of the body */
            if(conn->protocol&PROT_HTTP) {
              /* HTTP-only checks */
Daniel Stenberg's avatar
Daniel Stenberg committed
                if(conn->bits.close) {
                  /* Abort after the headers if "follow Location" is set
                     and we're set to close anyway. */
                  k->keepon &= ~KEEP_READ;
                  *done = TRUE;
                  return CURLE_OK;
                }
                /* We have a new url to load, but since we want to be able
                   to re-use this connection properly, we read the full
                   response in "ignore more" */
                k->ignorebody = TRUE;
                infof(data, "Ignoring the response-body\n");
              if (data->reqdata.resume_from && !k->content_range &&
                  (data->set.httpreq==HTTPREQ_GET) &&
                  !k->ignorebody) {
                /* we wanted to resume a download, although the server doesn't
                 * seem to support this and we did this with a GET (if it
                 * wasn't a GET we did a POST or PUT resume) */
                failf(data, "HTTP server doesn't seem to support "
                      "byte ranges. Cannot resume.");
              if(data->set.timecondition && !data->reqdata.range) {
                /* A time condition has been set AND no ranges have been
                   requested. This seems to be what chapter 13.3.4 of
                   RFC 2616 defines to be the correct action for a
                   HTTP/1.1 client */
                if((k->timeofdoc > 0) && (data->set.timevalue > 0)) {
                  switch(data->set.timecondition) {
                  case CURL_TIMECOND_IFMODSINCE:
                  default:
                    if(k->timeofdoc < data->set.timevalue) {
                      infof(data,
                            "The requested document is not new enough\n");
                      *done = TRUE;
                      return CURLE_OK;
                    }
                    break;
                  case CURL_TIMECOND_IFUNMODSINCE:
                    if(k->timeofdoc > data->set.timevalue) {
                      infof(data,
                            "The requested document is not old enough\n");
                      *done = TRUE;
                      return CURLE_OK;
                    }
                    break;
                  } /* switch */
                } /* two valid time strings */
              } /* we have a time condition */

            } /* this is HTTP */
          } /* this is the first time we write a body part */
          k->bodywrites++;

          /* pass data to the debug function before it gets "dechunked" */
          if(data->set.verbose) {
            if(k->badheader) {
              Curl_debug(data, CURLINFO_DATA_IN, data->state.headerbuff,
              if(k->badheader == HEADER_PARTHEADER)
                Curl_debug(data, CURLINFO_DATA_IN, k->str, nread, conn);
              Curl_debug(data, CURLINFO_DATA_IN, k->str, nread, conn);
          if(conn->bits.chunk) {
            /*
             * Bless me father for I have sinned. Here comes a chunked
             * transfer flying and we need to decode this properly.  While
             * the name says read, this function both reads and writes away
             * the data. The returned 'nread' holds the number of actual
             * data it wrote to the client.  */
            CHUNKcode res =
              Curl_httpchunk_read(conn, k->str, nread, &nread);

            if(CHUNKE_OK < res) {
              if(CHUNKE_WRITE_ERROR == res) {
                failf(data, "Failed writing data");
                return CURLE_WRITE_ERROR;
              }
              failf(data, "Received problem %d in the chunky parser", res);
            else if(CHUNKE_STOP == res) {
              /* we're done reading chunks! */
              k->keepon &= ~KEEP_READ; /* read no more */
              /* There are now possibly N number of bytes at the end of the
                 str buffer that weren't written to the client, but we don't
                 care about them right now. */
            }
            /* If it returned OK, we just keep going */
          if((-1 != k->maxdownload) &&
             (k->bytecount + nread >= k->maxdownload)) {
            size_t excess = k->bytecount + nread - k->maxdownload;

            if (excess > 0) {
                infof(data, "Rewinding stream by : %d bytes\n", excess);
                Curl_read_rewind(conn, excess);
                conn->bits.stream_was_rewound = TRUE;
            }

            nread = (ssize_t) (k->maxdownload - k->bytecount);
            if(nread < 0 ) /* this should be unusual */
              nread = 0;
            k->keepon &= ~KEEP_READ; /* we're done reading */
          }
          Curl_pgrsSetDownloadCounter(data, k->bytecount);
          if(!conn->bits.chunk && (nread || k->badheader || is_empty_data)) {
            /* If this is chunky transfer, it was already written */

Daniel Stenberg's avatar
Daniel Stenberg committed
            if(k->badheader && !k->ignorebody) {
              /* we parsed a piece of data wrongly assuming it was a header
                 and now we output it as body instead */
              result = Curl_client_write(conn, CLIENTWRITE_BODY,
            }
            if(k->badheader < HEADER_ALLBAD) {
              /* This switch handles various content encodings. If there's an
                 error here, be sure to check over the almost identical code
                 Make sure that ALL_CONTENT_ENCODINGS contains all the
                 encodings handled here. */
#ifdef HAVE_LIBZ
              switch (k->content_encoding) {
              case IDENTITY:
                /* This is the default when the server sends no
                   Content-Encoding header. See Curl_readwrite_init; the
                   memset() call initializes k->content_encoding to zero. */
Daniel Stenberg's avatar
Daniel Stenberg committed
                if(!k->ignorebody)
                  result = Curl_client_write(conn, CLIENTWRITE_BODY, k->str,
Daniel Stenberg's avatar
Daniel Stenberg committed
                                             nread);
#ifdef HAVE_LIBZ
                /* Assume CLIENTWRITE_BODY; headers are not encoded. */
                  result = Curl_unencode_deflate_write(conn, k, nread);
              case GZIP:
                /* Assume CLIENTWRITE_BODY; headers are not encoded. */
                  result = Curl_unencode_gzip_write(conn, k, nread);
              default:
                failf (data, "Unrecognized content encoding type. "
                       "libcurl understands `identity', `deflate' and `gzip' "
                       "content encodings.");
                result = CURLE_BAD_CONTENT_ENCODING;
                break;
              }
            }
            k->badheader = HEADER_NORMAL; /* taken care of now */
        } /* if (! header and data to read ) */
        if (is_empty_data) {
          /* if we received nothing, the server closed the connection and we
             are done */
          k->keepon &= ~KEEP_READ;
        }

    } /* if( read from socket ) */
    /* If we still have writing to do, we check if we have a writable
       socket. */
    if((k->keepon & KEEP_WRITE) && (select_res & CSELECT_OUT)) {
Daniel Stenberg's avatar
Daniel Stenberg committed
      bool writedone=TRUE;
      if ((k->bytecount == 0) && (k->writebytecount == 0))
        Curl_pgrsTime(data, TIMER_STARTTRANSFER);
      didwhat |= KEEP_WRITE;
      /*
       * We loop here to do the READ and SEND loop until we run out of
       * data to send or until we get EWOULDBLOCK back
       */
      do {
        /* only read more data if there's no upload data already
           present in the upload buffer */
        if(0 == data->reqdata.upload_present) {
          /* init the "upload from here" pointer */
          data->reqdata.upload_fromhere = k->uploadbuf;
          if(!k->upload_done) {
            /* HTTP pollution, this should be written nicer to become more
               protocol agnostic. */
               (data->reqdata.proto.http->sending == HTTPSEND_BODY)) {
              /* If this call is to send body data, we must take some action:
                 We have sent off the full HTTP 1.1 request, and we shall now
                 go into the Expect: 100 state and await such a header */
              k->wait100_after_headers = FALSE; /* headers sent */
              k->write_after_100_header = TRUE; /* wait for the header */
              k->keepon &= ~KEEP_WRITE;         /* disable writing */
              k->start100 = Curl_tvnow();       /* timeout count starts now */
              didwhat &= ~KEEP_WRITE;  /* we didn't write anything actually */
              break;
            }

            result = Curl_fillreadbuffer(conn, BUFSIZE, &fillcount);
          else
            nread = 0; /* we're done uploading/reading */

          /* the signed int typecase of nread of for systems that has
             unsigned size_t */
          if (nread<=0) {
            /* done */
            k->keepon &= ~KEEP_WRITE; /* we're done writing */
            writedone = TRUE;

            if(conn->bits.rewindaftersend) {
              result = Curl_readrewind(conn);
              if(result)
                return result;
            }
          /* store number of bytes available for upload */
          data->reqdata.upload_present = nread;
#ifdef CURL_DO_LINEEND_CONV
          /* always convert if we're FTPing in ASCII mode */
          if ((data->set.crlf) || (data->set.prefer_ascii)) {
#endif /* CURL_DO_LINEEND_CONV */
              if(data->state.scratch == NULL)
                data->state.scratch = malloc(2*BUFSIZE);
              if(data->state.scratch == NULL) {
                failf (data, "Failed to alloc scratch buffer!");
                return CURLE_OUT_OF_MEMORY;
              }
              /*
               * ASCII/EBCDIC Note: This is presumably a text (not binary)
               * transfer so the data should already be in ASCII.
               * That means the hex values for ASCII CR (0x0d) & LF (0x0a)
               * must be used instead of the escape sequences \r & \n.
               */
            for(i = 0, si = 0; i < nread; i++, si++) {
              if (data->reqdata.upload_fromhere[i] == 0x0a) {
                data->state.scratch[si++] = 0x0d;
                data->state.scratch[si] = 0x0a;
                if (!data->set.crlf) {
                  /* we're here only because FTP is in ASCII mode...
                     bump infilesize for the LF we just added */
                  data->set.infilesize++;
                }
                data->state.scratch[si] = data->reqdata.upload_fromhere[i];
            if(si != nread) {
              /* only perform the special operation if we really did replace
                 anything */
              nread = si;
              /* upload from the new (replaced) buffer instead */
              data->reqdata.upload_fromhere = data->state.scratch;
              data->reqdata.upload_present = nread;
        else {
          /* We have a partial buffer left from a previous "round". Use
             that instead of reading more data */
        }

        /* write to socket (send away data) */
                            conn->writesockfd,     /* socket to send to */
                            data->reqdata.upload_fromhere, /* buffer pointer */
                            data->reqdata.upload_present,  /* buffer size */
                            &bytes_written);       /* actually send away */

        if(data->set.verbose)
          /* show the data before we change the pointer upload_fromhere */
          Curl_debug(data, CURLINFO_DATA_OUT, data->reqdata.upload_fromhere,
        if(data->reqdata.upload_present != bytes_written) {
          /* we only wrote a part of the buffer (if anything), deal with it! */

          /* store the amount of bytes left in the buffer to write */
          data->reqdata.upload_present -= bytes_written;

          /* advance the pointer where to find the buffer when the next send
             is to happen */
          data->reqdata.upload_fromhere += bytes_written;

          writedone = TRUE; /* we are done, stop the loop */
        }
        else {
          /* we've uploaded that buffer now */
          data->reqdata.upload_fromhere = k->uploadbuf;
          data->reqdata.upload_present = 0; /* no more bytes left */

          if(k->upload_done) {
            /* switch off writing, we're done! */
            k->keepon &= ~KEEP_WRITE; /* we're done writing */
            writedone = TRUE;
          }
        Curl_pgrsSetUploadCounter(data, k->writebytecount);
      } while(!writedone); /* loop until we're done writing! */
  } while(0); /* just to break out from! */
  if(didwhat) {
    /* Update read/write counters */
    if(k->bytecountp)
      *k->bytecountp = k->bytecount; /* read count */
    if(k->writebytecountp)
      *k->writebytecountp = k->writebytecount; /* write count */
  }
  else {
    /* no read no write, this is a timeout? */
    if (k->write_after_100_header) {
      /* This should allow some time for the header to arrive, but only a
         very short time as otherwise it'll be too much wasted times too
         often. */

      /* Quoting RFC2616, section "8.2.3 Use of the 100 (Continue) Status":
      Therefore, when a client sends this header field to an origin server
      (possibly via a proxy) from which it has never seen a 100 (Continue)
      status, the client SHOULD NOT wait for an indefinite period before
      sending the request body.

      */

      long ms = Curl_tvdiff(k->now, k->start100);
      if(ms > CURL_TIMEOUT_EXPECT_100) {
        /* we've waited long enough, continue anyway */
        k->write_after_100_header = FALSE;
        k->keepon |= KEEP_WRITE;
      }
  if(Curl_pgrsUpdate(conn))
    result = CURLE_ABORTED_BY_CALLBACK;
  else
    result = Curl_speedcheck(data, k->now);
  if (result)
    return result;
  if (data->set.timeout &&
      ((Curl_tvdiff(k->now, k->start)/1000) >= data->set.timeout)) {
      failf(data, "Operation timed out after %d seconds with %"
            FORMAT_OFF_T " out of %" FORMAT_OFF_T " bytes received",
            data->set.timeout, k->bytecount, k->size);
    } else {
      failf(data, "Operation timed out after %d seconds with %"
            FORMAT_OFF_T " bytes received",
            data->set.timeout, k->bytecount);
    }
    return CURLE_OPERATION_TIMEOUTED;
  }

  if(!k->keepon) {
    /*
     * The transfer has been performed. Just make some general checks before
     * returning.
     */

    if(!(conn->bits.no_body) && (k->size != -1) &&
       (k->bytecount != k->size) &&
#ifdef CURL_DO_LINEEND_CONV
       /* Most FTP servers don't adjust their file SIZE response for CRLFs,
          so we'll check to see if the discrepancy can be explained
          by the number of CRLFs we've changed to LFs.
        */
       (k->bytecount != (k->size + data->state.crlf_conversions)) &&
#endif /* CURL_DO_LINEEND_CONV */
      failf(data, "transfer closed with %" FORMAT_OFF_T
            " bytes remaining to read",
      return CURLE_PARTIAL_FILE;
    }
    else if(!(conn->bits.no_body) &&
            conn->bits.chunk &&
            (data->reqdata.proto.http->chunk.state != CHUNK_STOP)) {
      /*
       * In chunked mode, return an error if the connection is closed prior to
       * the empty (terminiating) chunk is read.
       *
       * The condition above used to check for
       * conn->proto.http->chunk.datasize != 0 which is true after reading
       * *any* chunk, not just the empty chunk.
       *
       */
      failf(data, "transfer closed with outstanding read data remaining");
      return CURLE_PARTIAL_FILE;
    if(Curl_pgrsUpdate(conn))
      return CURLE_ABORTED_BY_CALLBACK;
  /* Now update the "done" boolean we return */
Yang Tse's avatar
Yang Tse committed
  *done = (bool)(0 == k->keepon);
Daniel Stenberg's avatar
Daniel Stenberg committed
/*
 * Curl_readwrite_init() inits the readwrite session. This is inited each time for a
 * transfer, sometimes multiple times on the same SessionHandle
CURLcode Curl_readwrite_init(struct connectdata *conn)
{
  struct SessionHandle *data = conn->data;
  struct Curl_transfer_keeper *k = &data->reqdata.keep;
  /* NB: the content encoding software depends on this initialization of
  memset(k, 0, sizeof(struct Curl_transfer_keeper));

  k->start = Curl_tvnow(); /* start time */
  k->now = k->start;   /* current time is now */
  k->header = TRUE; /* assume header */
  k->httpversion = -1; /* unknown at this point */

  k->size = data->reqdata.size;
  k->maxdownload = data->reqdata.maxdownload;
  k->bytecountp = data->reqdata.bytecountp;
  k->writebytecountp = data->reqdata.writebytecountp;

  k->bytecount = 0;
  k->headerbytecount = 0;

  k->buf = data->state.buffer;
  k->uploadbuf = data->state.uploadbuffer;
  k->maxfd = (conn->sockfd>conn->writesockfd?
              conn->sockfd:conn->writesockfd)+1;
  k->hbufp = data->state.headerbuff;
Daniel Stenberg's avatar
Daniel Stenberg committed
  k->ignorebody=FALSE;

  Curl_pgrsTime(data, TIMER_PRETRANSFER);
  Curl_speedinit(data);

  Curl_pgrsSetUploadCounter(data, 0);
  Curl_pgrsSetDownloadCounter(data, 0);

    if(k->size > 0)
      Curl_pgrsSetDownloadSize(data, k->size);
  /* we want header and/or body, if neither then don't do this! */
  if(conn->bits.getheader || !conn->bits.no_body) {
    if(conn->sockfd != CURL_SOCKET_BAD) {
    if(conn->writesockfd != CURL_SOCKET_BAD) {
      /* HTTP 1.1 magic:

         Even if we require a 100-return code before uploading data, we might
         need to write data before that since the REQUEST may not have been
         finished sent off just yet.

         Thus, we must check if the request has been sent before we set the
         state info where we wait for the 100-return code
      */
          (data->reqdata.proto.http->sending == HTTPSEND_BODY)) {
        /* wait with write until we either got 100-continue or a timeout */
        k->write_after_100_header = TRUE;
          /* when we've sent off the rest of the headers, we must await a
             100-continue */
          k->wait100_after_headers = TRUE;
Daniel Stenberg's avatar
Daniel Stenberg committed
/*
 * Curl_single_getsock() gets called by the multi interface code when the app
 * has requested to get the sockets for the current connection. This function
Daniel Stenberg's avatar
Daniel Stenberg committed
 * will then be called once for every connection that the multi interface
 * keeps track of. This function will only be called for connections that are
 * in the proper state to have this information available.
 */
int Curl_single_getsock(struct connectdata *conn,
                        curl_socket_t *sock, /* points to numsocks number
                                                of sockets */
                        int numsocks)
  struct SessionHandle *data = conn->data;
  int bitmap = GETSOCK_BLANK;
  int index = 0;

  if(numsocks < 2)
    /* simple check but we might need two slots */
    return GETSOCK_BLANK;

  if(data->reqdata.keep.keepon & KEEP_READ) {
    bitmap |= GETSOCK_READSOCK(index);
    sock[index] = conn->sockfd;

  if(data->reqdata.keep.keepon & KEEP_WRITE) {
    if((conn->sockfd != conn->writesockfd) ||
       !(data->reqdata.keep.keepon & KEEP_READ)) {
      /* only if they are not the same socket or we didn't have a readable
         one, we increase index */
      if(data->reqdata.keep.keepon & KEEP_READ)
        index++; /* increase index if we need two entries */
      sock[index] = conn->writesockfd;
    }

    bitmap |= GETSOCK_WRITESOCK(index);
}


/*
 * Transfer()
 *
 * This function is what performs the actual transfer. It is capable of
 * doing both ways simultaneously.
 * The transfer must already have been setup by a call to Curl_setup_transfer().
 *
 * Note that headers are created in a preallocated buffer of a default size.
 * That buffer can be enlarged on demand, but it is never shrunken again.
 *
 * Parts of this function was once written by the friendly Mark Butler
 * <butlerm@xmission.com>.
 */

static CURLcode
Transfer(struct connectdata *conn)
{
  CURLcode result;
  struct SessionHandle *data = conn->data;
  struct Curl_transfer_keeper *k = &data->reqdata.keep;
  if(!(conn->protocol & PROT_FILE))
    /* Only do this if we are not transferring FILE:, since the file: treatment
       is different*/
    Curl_readwrite_init(conn);
Daniel Stenberg's avatar
Daniel Stenberg committed
  if((conn->sockfd == CURL_SOCKET_BAD) &&
     (conn->writesockfd == CURL_SOCKET_BAD))
    /* nothing to read, nothing to write, we're already OK! */
    return CURLE_OK;

  /* we want header and/or body, if neither then don't do this! */
  if(!conn->bits.getheader && conn->bits.no_body)
    return CURLE_OK;

  while (!done) {
    curl_socket_t fd_read;
    curl_socket_t fd_write;
    int interval_ms;

    interval_ms = 1 * 1000;

    /* limit-rate logic: if speed exceeds threshold, then do not include fd in
       select set */
    if ( (conn->data->set.max_send_speed > 0) &&
         (conn->data->progress.ulspeed > conn->data->set.max_send_speed) )  {
      fd_write = CURL_SOCKET_BAD;
      Curl_pgrsUpdate(conn);
    }
    else {
      if(k->keepon & KEEP_WRITE)
        fd_write = conn->writesockfd;
      else
        fd_write = CURL_SOCKET_BAD;
    }

    if ( (conn->data->set.max_recv_speed > 0) &&
         (conn->data->progress.dlspeed > conn->data->set.max_recv_speed) ) {
      fd_read = CURL_SOCKET_BAD;
      Curl_pgrsUpdate(conn);
    }
    else {
      if(k->keepon & KEEP_READ)
        fd_read = conn->sockfd;
      else
        fd_read = CURL_SOCKET_BAD;
    }
    switch (Curl_select(fd_read, fd_write, interval_ms)) {
    case -1: /* select() error, stop reading */
#ifdef EINTR
      /* The EINTR is not serious, and it seems you might get this more
         ofen when using the lib in a multi-threaded environment! */
      if(errno == EINTR)
        ;
      else
        done = TRUE; /* no more read or write */
      continue;
    case 0:  /* timeout */
    default: /* readable descriptors */
      result = Curl_readwrite(conn, &done);
      break;
    }
    if(result)
      return result;
    /* "done" signals to us if the transfer(s) are ready */
  }
Daniel Stenberg's avatar
Daniel Stenberg committed
/*
 * Curl_pretransfer() is called immediately before a transfer starts.
 */
CURLcode Curl_pretransfer(struct SessionHandle *data)
{
  if(!data->change.url) {
    /* we can't do anything wihout URL */
    failf(data, "No URL set!\n");
  /* Init the SSL session ID cache here. We do it here since we want to do it
     after the *_setopt() calls (that could change the size of the cache) but
     before any transfer takes place. */
  res = Curl_ssl_initsessions(data, data->set.ssl.numsessions);
  if(res)
    return res;
  data->set.followlocation=0; /* reset the location-follow counter */
  data->state.this_is_a_follow = FALSE; /* reset this */
  data->state.errorbuf = FALSE; /* no error has occurred */
  data->state.authhost.want = data->set.httpauth;
  data->state.authproxy.want = data->set.proxyauth;
  /* If there is a list of cookie files to read, do it now! */
  if(data->change.cookielist) {
 /* Allow data->set.use_port to set which port to use. This needs to be
  * disabled for example when we follow Location: headers to URLs using
  * different ports! */
  data->state.allow_port = TRUE;

#if defined(HAVE_SIGNAL) && defined(SIGPIPE) && !defined(HAVE_MSG_NOSIGNAL)
  /*************************************************************
   * Tell signal handler to ignore SIGPIPE
   *************************************************************/
  if(!data->set.no_signal)
    data->state.prev_signal = signal(SIGPIPE, SIG_IGN);
  Curl_initinfo(data); /* reset session-specific information "variables" */
Daniel Stenberg's avatar
Daniel Stenberg committed
/*
 * Curl_posttransfer() is called immediately after a transfer ends
 */
CURLcode Curl_posttransfer(struct SessionHandle *data)
{
#if defined(HAVE_SIGNAL) && defined(SIGPIPE) && !defined(HAVE_MSG_NOSIGNAL)
  /* restore the signal handler for SIGPIPE before we get back */
  if(!data->set.no_signal)
    signal(SIGPIPE, data->state.prev_signal);
Daniel Stenberg's avatar
Daniel Stenberg committed
#else
  (void)data; /* unused parameter */
  if(!(data->progress.flags & PGRS_HIDE) &&
     !data->progress.callback)
    /* only output if we don't use a progress callback and we're not hidden */
    fprintf(data->set.err, "\n");

Daniel Stenberg's avatar
Daniel Stenberg committed
/*
 * strlen_url() returns the length of the given URL if the spaces within the
 * URL were properly URL encoded.
 */
static int strlen_url(char *url)
{
  char *ptr;
  int newlen=0;
  bool left=TRUE; /* left side of the ? */

  for(ptr=url; *ptr; ptr++) {
    switch(*ptr) {
    case '?':
      left=FALSE;
    default:
      newlen++;
      break;
    case ' ':
      if(left)
        newlen+=3;
      else
        newlen++;
      break;
    }
  }
  return newlen;
}

Daniel Stenberg's avatar
Daniel Stenberg committed
/* strcpy_url() copies a url to a output buffer and URL-encodes the spaces in
 * the source URL accordingly.
 */
static void strcpy_url(char *output, char *url)
{
  /* we must add this with whitespace-replacing */
  bool left=TRUE;
  char *iptr;
  char *optr = output;
  for(iptr = url;    /* read from here */
      *iptr;         /* until zero byte */
      iptr++) {
    switch(*iptr) {
    case '?':
      left=FALSE;
    default:
      *optr++=*iptr;
      break;
    case ' ':
      if(left) {
        *optr++='%'; /* add a '%' */
        *optr++='2'; /* add a '2' */
        *optr++='0'; /* add a '0' */
      }
      else
        *optr++='+'; /* add a '+' here */
      break;
    }
  }
  *optr=0; /* zero terminate output buffer */

}

Daniel Stenberg's avatar
Daniel Stenberg committed
/*
 * Curl_follow() handles the URL redirect magic. Pass in the 'newurl' string
 * as given by the remote server and set up the new URL to request.
 */
CURLcode Curl_follow(struct SessionHandle *data,
                     char *newurl, /* this 'newurl' is the Location: string,
                                      and it must be malloc()ed before passed
                                      here */
                     bool retry) /* set TRUE if this is a request retry as
                                    opposed to a real redirect following */
{
  /* Location: redirect */
  char prot[16]; /* URL protocol string storage */
  char letter;   /* used for a silly sscanf */
  if(!retry) {
    if ((data->set.maxredirs != -1) &&
        (data->set.followlocation >= data->set.maxredirs)) {
      failf(data,"Maximum (%d) redirects followed", data->set.maxredirs);
      return CURLE_TOO_MANY_REDIRECTS;
    }
    /* mark the next request as a followed location: */
    data->state.this_is_a_follow = TRUE;
    data->set.followlocation++; /* count location-followers */
  }

  if(data->set.http_auto_referer) {
    /* We are asked to automatically set the previous URL as the
       referer when we get the next URL. We pick the ->url field,
       which may or may not be 100% correct */

    if(data->change.referer_alloc)
      /* If we already have an allocated referer, free this first */
      free(data->change.referer);

    data->change.referer = strdup(data->change.url);
    data->change.referer_alloc = TRUE; /* yes, free this later */
  }

  if(2 != sscanf(newurl, "%15[^?&/:]://%c", prot, &letter)) {
    /***
     *DANG* this is an RFC 2068 violation. The URL is supposed
     to be absolute and this doesn't seem to be that!
     ***
     Instead, we have to TRY to append this new path to the old URL
     to the right of the host part. Oh crap, this is doomed to cause
     problems in the future...
    */
    char *protsep;
    char *pathsep;

    char *useurl = newurl;

    /* we must make our own copy of the URL to play with, as it may
       point to read-only data */
    char *url_clone=strdup(data->change.url);

    if(!url_clone)
      return CURLE_OUT_OF_MEMORY; /* skip out of this NOW */

    /* protsep points to the start of the host name */
    protsep=strstr(url_clone, "//");
    if(!protsep)
      protsep=url_clone;
    else
      protsep+=2; /* pass the slashes */

    if('/' != newurl[0]) {
      int level=0;

      /* First we need to find out if there's a ?-letter in the URL,
         and cut it and the right-side of that off */
      /* we have a relative path to append to the last slash if
         there's one available */
      pathsep = strrchr(protsep, '/');
      if(pathsep)
        *pathsep=0;

      /* Check if there's any slash after the host name, and if so,
         remember that position instead */
      pathsep = strchr(protsep, '/');
      if(pathsep)
        protsep = pathsep+1;
      else
        protsep = NULL;

      /* now deal with one "./" or any amount of "../" in the newurl
         and act accordingly */

      if((useurl[0] == '.') && (useurl[1] == '/'))
        useurl+=2; /* just skip the "./" */
      while((useurl[0] == '.') &&
            (useurl[1] == '.') &&
            (useurl[2] == '/')) {
        useurl+=3; /* pass the "../" */
      }

      if(protsep) {
        while(level--) {
          /* cut off one more level from the right of the original URL */
          pathsep = strrchr(protsep, '/');
          if(pathsep)
            *pathsep=0;
          else {
            *protsep=0;
            break;
          }
        }
      }
    }
    else {
      /* We got a new absolute path for this server, cut off from the
         first slash */
      pathsep = strchr(protsep, '/');
      if(pathsep) {
        /* When people use badly formatted URLs, such as
           "http://www.url.com?dir=/home/daniel" we must not use the first
           slash, if there's a ?-letter before it! */
        char *sep = strchr(protsep, '?');
        if(sep && (sep < pathsep))
          pathsep = sep;
      else {
        /* There was no slash. Now, since we might be operating on a badly
           formatted URL, such as "http://www.url.com?id=2380" which doesn't
           use a slash separator as it is supposed to, we need to check for a
           ?-letter as well! */
        pathsep = strchr(protsep, '?');
        if(pathsep)
          *pathsep=0;
      }
    /* If the new part contains a space, this is a mighty stupid redirect
       but we still make an effort to do "right". To the left of a '?'
       letter we replace each space with %20 while it is replaced with '+'
       on the right side of the '?' letter.
    */
    newlen = strlen_url(useurl);

    urllen = strlen(url_clone);
    newest=(char *)malloc( urllen + 1 + /* possible slash */
                           newlen + 1 /* zero byte */);