> Hi. Does anyone remember the old version for sendmail that had the > wizard bug? A lot of the books that have come out recently make passing > references to it, but none ever describe what it did. So, did anyone > ever take advantage of this in years gone by, and if so, how? > Thanks. > --Dave > > Note: To anyone who thinks that this is just a request for a hack to make it > easier to break into something, keep in mind that this bug was > removed from sendmail years ago (so please, no "Find out yourself, > LAMER!" replies). Thanks. Indeed. It was fixed well before the Internet Worm hit. Anyway -- the intended behavior of wizard mode was that if you supplied the right password, some other non-standard SMTP commands were enabled, notably one to give you a shell. The hashed password -- one-way encrypted exactly as per /etc/passwd -- was stored in the sendmail configuration file. But there was this bug; to explain it, I need to discuss some arcana relating to sendmail and the C compiler. In order to save the expense of reading and parsing the configuration file each time, sendmail has what's known as a ``frozen configuration file''. The concept is fine; the implementation isn't. To freeze the configuration file, sendmail just wrote out to disk the entire dynamic memory area (used by malloc) and the `bss' area -- the area that took up no space in the executable file, but was initialized to all zeros by the UNIX kernel when the program was executed. The bss area held all variables that were not given explicit initial values by the C source. Naturally, when delivering mail, sendmail just read these whole chunks back in, in two giant reads. It was therefore necessary to store all configuration file information in the bss or malloc areas, which demanded a fair amount of care in coding. The wizard mode password was stored in malloc'ed memory, so it was frozen properly. But the pointer to it was explicitly set to NULL in the source: char *wiz = NULL; That meant that it was in the initialized data area, *not* the bss. And it was therefore *not* saved with the frozen configuration. So -- when the configuration file is parsed and frozen, the password is read, and written out. The next time sendmail is run, though, the pointer will be reset to NULL. (The password is present, of course, but there's no way to find it.) And the code stupidly believed in the concept of no password for the back door. One more point is worth noting -- during testing, sendmail did the right thing with wizard mode. That is, it did check the password -- because if you didn't happen to do the wizard mode test with a frozen configuration file -- and most testing would not be done that way, since you have to refreeze after each compilation -- the pointer would be correct. --Steve Bellovin