On Sun, Nov 9, 2025 at 6:25 AM <[email protected]> wrote: > i execute > some_long_running_command | mail billy_bob@localhost > where is the output of the command stored while it is running before it's > mailed With | the shell handles opening pipe between the two processes, in your example some_long_running_command and mail billy_bob@localhost The kernel then handles that - there's a certain amount of buffer size the kernel allocates to the pipe, e.g. 64KiB. That's in RAM (or at least virtual memory). That size may also possibly be adjusted. See also, e.g.: pipe(2), pipe2(2), pipesz(1), etc. The mail program in such mode, taking stdin for body of mail to be sent, collects and stores the stdin data it receives, generally to a file somewhere under /var (per FHS), though precise location may depend upon the particular mail program one may be using. And, at my particular Debian (12.12 amd64) where I peeked at that, and what I have installed there, for mail program I have bsd-mailx. Having a look with strace(1), that mail program first stored the data (message body) in /tmp/mail.RsXXXX7I9ipp location of which may possibly be altered by configuration for the program and/or TMPDIR in the environment. Once it had processed all the input data and successfully written it there, it then opened and wrote another file: /tmp/mail.RsXXXXUjcCyg and wrote email headers and message body to that file, It then manipulates file descriptor(s) so that becomes fd 0 (stdin) and then it execve(2)ed /usr/sbin/sendmail For brevity, I omitted many details, e.g. options, arguments, various unlinking and closing of files, etc. And, your mileage may/will vary, depending exactly which programs and versions you have installed for these various bits of functionality (e.g. mail, MTA, etc.), and configurations and/or environmental settings thereof.
Anyway, in general, pipe buffers a bit, mail program will buffer to a file, then typically hand off to MTA. And there may be additional steps between (like adding mail headers - not sure why it didn't do that first, as it didn't include time, length, or other variable headers - though perhaps that might've been different depending upon what content it was handed). And (not covered), MTA will typically spool to file, and then make delivery attempt(s).
