tcpdump mailing list archives

Re: [clearview-discuss] libdlpi with libpcap


From: sagun shakya <Sagun.Shakya () sun com>
Date: Sun, 27 Jan 2008 20:53:11 -0500

Due to other work priority I wasn't able to work on fixing and addressing these comments. I have finally addressed the comments from Meem and made a few other changes. My comments are inline and please find an update webrev at:
http://cr.opensolaris.org/~sagun/libpcap/

If you'd like to see the changes made from the last round of review you can find a webrev at:

http://cr.opensolaris.org/~sagun/codereview_fix/


-Thanks,

Sagun


Peter Memishian wrote:
I took a quick look at this.  Some offhand observations:

        * There's too much duplicated code betwen pcap-libdlpi.c and
          pcap-dlpi.c.  For instance, both strioctl() and the pcap_stats()
          functions are identical, as is the bufmod-related logic
          (starting with "Loop through packets") in pcap_read(), and
          the bufmod configuration in pcap_open_live().  I'd move the
          relevant bits of logic to a new pcap-streams.c file and call
          those routines from pcap-*dlpi.c.
Created a new file pcap-streams.c and moved common functions into this file.
        * The linklist structure should be renamed; it's too close to
"linked list".
Changed to linkname_list.
        * The link chaining logic in list_interfaces() doesn't seem like
          it would work for more than two links.  In particular, seems
          like you need:

                if (lwp->lw_list == NULL) {
                        lwp->lw_list = entry;
                } else {
                        entry->ll_next = lwp->lw_list;
                        lwp->lw_list = entry;
                }

          (Also, there's no need to initialize `entry' or cast `lwp'.)
Fixed.
        * In pcap_read_libdlpi(), I'm not sure what the special error
          handling after dlpi_recv() is about.  It looks to be some
          holdover from the old pcap-dlpi.c code, but I can't see how it
          applies to dlpi_recv() since dlpi_recv() only calls getmsg()
          when poll() has indicated there's data to read, and thus
          it should never get EINTR or EAGAIN -- and if it did, that is
          an issue that should be handled inside dlpi_recv() itself.
The special error handling was left so that when the read operation was terminated a warning/error message wouldn't be printed. But a case fo EAGAIN isn't needed so I've changed it to handle only EINTR.
        * Seems error handling would be cleaned up quite a bit by
          introducing a utility function like:

          static void
          pcap_libdlpi_err(dlpi_handle_t *dh, const char *func, int err,
              char *errbuf);

Added as well as a similar function in pcap-streams.c
        * Please check for spelling errors -- e.g. "promsic" on lines 330
          and 341.

        * In pcap_open_live(), `flags' should be removed and DLPI_RAW |
          DLPI_PASSIVE directly passed into dlpi_open().  Also, the
          comment above dlpi_open() should be updated to say something
          more appropriate like "Enable Solaris raw and passive DLPI
          extensions; see the Solaris dlpi(7P) manpage for details".

        * Surely line 317 is supposed to set `retv' so that line 320 will
          print something meaningful?

        * Would be good to be consistent with either `ret' or `retv' as
          a variable to house the return value.

All of the above have been fixed.

-
This is the tcpdump-workers list.
Visit https://cod.sandelman.ca/ to unsubscribe.


Current thread: