1. 08 Sep, 2006 1 commit
  2. 28 Apr, 2006 1 commit
    • Rémi Denis-Courmont's avatar
      Hmm, damn stupid fd_set implementation by Winsock does of course not · 81ced014
      Rémi Denis-Courmont authored
      work like any other one: it's actually thirty two times as big than the
      POSIX ones for any given FD_SETSIZE, plus FD_SET and FD_ISSET are O(n)
      instead of O(1). But at least, the mingw implementations of FD_SET has
      built-in overflow checks.
      
      Also, Winsock insists on returning big socket numbers, so we can't drop
      those above FD_SETSIZE or we have no TCP/IP support at all, anyway.
      
      
      81ced014
  3. 27 Apr, 2006 1 commit
    • Rémi Denis-Courmont's avatar
      Whenever using select() companion macro FD_SET(), one must ensure that · 0ee3f461
      Rémi Denis-Courmont authored
      the specified file descriptor does not exceeds the fd_set limit (FD_SETSIZE).
      Registering a file descriptor above this limit is not defined (read: this is
      a buffer overflow).
      
      This is not normally exploitable on Linux because FD_SETSIZE equals the
      default limit for file descriptors per processes, but should work on BSDs.
      Cygwin is also very prone to this problem (but right, we don't use it). As
      for Windows, it will overflow if you try to select() more than 64 sockets at
      a time (regardless of the total numbers of socket), but for the sake of
      simplicity and portability, this patch actually prevent opening more than
      64 sockets at all.
      
      There are still quite a bunch of plugins that needs fixing.
      0ee3f461
  4. 12 Apr, 2006 1 commit
  5. 01 Apr, 2006 1 commit
  6. 12 Jan, 2006 1 commit
  7. 05 Dec, 2005 2 commits
  8. 04 Dec, 2005 1 commit
  9. 23 Oct, 2005 1 commit
  10. 22 Sep, 2005 1 commit
  11. 17 Sep, 2005 2 commits
  12. 22 Aug, 2005 1 commit
  13. 21 Aug, 2005 1 commit
  14. 19 Aug, 2005 1 commit
  15. 18 Aug, 2005 2 commits
  16. 14 Aug, 2005 2 commits
  17. 02 Aug, 2005 1 commit
    • Rémi Denis-Courmont's avatar
      - When both IPv4 and IPv6 plugins return a socket, use the IPv6 one if · 331d097f
      Rémi Denis-Courmont authored
        IPV6_ONLY can be turned off. It is supposed to be off by default, and
        it actually is. Unfortunately, OpenBSD and NetBSD people think
        network software programmer like us are too stupid to use this
        feature properly, so they just removed it. Lets say the most stupid
        are not always the ones one think. Also, Windows doesn't support this
        yet. On all these OSes, keep the old behavior; that is prefer IPv4.
        Closes #166.
      
      - Windows protection level compile fix
      331d097f
  18. 29 Jul, 2005 1 commit
  19. 28 Jul, 2005 1 commit
  20. 27 Jul, 2005 2 commits
  21. 17 Jul, 2005 1 commit
  22. 15 Jul, 2005 1 commit
  23. 13 Jul, 2005 1 commit
  24. 12 Jul, 2005 1 commit
  25. 10 Jul, 2005 1 commit
  26. 09 Jul, 2005 1 commit
  27. 08 Jul, 2005 1 commit
  28. 06 Jul, 2005 1 commit
  29. 05 Jul, 2005 2 commits
  30. 04 Jul, 2005 1 commit
  31. 01 Jul, 2005 2 commits
  32. 30 Jun, 2005 1 commit
  33. 26 Jun, 2005 1 commit