Hrm. A good many years ago, I worked for an "outsourcing" company out in West London. Mainly we dealt with hosting physical systems and doing maintenance. One of our clients was a manufacturer of domestic appliances, and their stock control system ran on a rather ancient piece of IBM hardware. Ancient enough that it had a hard drive around the size of a small fridge, and an 8" floppy drive, but I digress. It was also ancient enough to have been created in earlier, less paranoid, days, where everyone had access to everything on the system, and security rights were down to "you have physical access to a keyboard". A major part of the joy of this particular system was that *everything* was editable, including the script used to start up the machine itself. Not the startup scripts that ran after the OS was loaded, but the actual scripts that loaded the OS. And if the first script were, by "chance", to end up calling itself recursively, all hell broke loose.
Well, not quite.
What actually happened was we got on the phone to IBM at the South Bank, and they would taxi over this greying old man with a beard, who would power off the machine, disconnect the hard drive, connect the floppy drive, hold the 3-phase breakers closed and power the machine up again, plug in the hard drive whilst the machine was powered on, then reprogram the boot script using a hex keypad. All the time using language that would make a sailor blush. He would then deposit a large invoice, have a coffee, get back in a taxi, and leave. The entire process took around 3 hours.
Oddly enough, this happened fairly regularly. Usually at around the time the entry clerks came back from a leaving "do" on a Friday afternoon. At one point there was a round of compulsory redundancies at client HQ and it happened every day for a week.
This was the same place where (for another client, a food & wine distributor) I was called out to fix a breaking batch order processing job at 3AM. As usual for an outsourcing house, the crashing binary bore no resemblance to the source that was supposed to have generated it, or any other version of the source we had. The crash was being caused by an order for 100,000,000 items overflowing a number field. Remove line, rerun job, all looks OK, go back home to bed and deal with it in the morning. After some investigation later in the day, it turned out some tit had modified the ordering parameters such that cocktail sticks were now being ordered in units, rather than boxes of 500,000. <sigh>
Ah, and back in the TSO days, sending a fake crashdump to someone else's terminal.
 Motto - "Nothing sucks like Electrolux"
 No, it wasn't Electrolux.