What's the difference between /tmp and /run?
The directories /tmp and /usr/tmp (later /var/tmp) used to be the dumping ground for everything and everybody. The only protection mechanism for files in these directories is the sticky bit which restricts deletion or renaming of files there to their owners. As marcelm pointed out in a comment, there's in principle nothing that prevents someone to create files with names that are used by services (such as nginx.pid or sshd.pid). (In practice, the startup scripts could remove such bogus files first, though.)
/run was established for non-persistent runtime data of long lived services such as locks, sockets, pid files and the like. Since it is not writable for the public, it shields service runtime data from the mess in /tmp and jobs that clean up there. Indeed: Two distributions that I run (no pun intended) have permissions 755 on /run, while /tmp and /var/tmp (and /dev/shm for that matter) have permissions 1777.
/tmp is the location for creation of temporary files and directories. It's not usable for storing "well-known names" (i.e. names another process could be aware of without you having to convey the name to it somehow) because nobody has ownership over the namespace; anyone can create files there. As such you generally use it when you have a utility that needs a file (i.e. not a pipe or such) as input or output, where any (randomly generated) name will work as long as you pass the name in.
Historically, some things (like X) violated this principle and put well-known names (like .X11-unix) in /tmp. This is of course buggy and allows any user to DoS the service needing to do so simply by racing to create a file by the desired name first. Such things belong under /run (or equivalently /var/run if you don't subscribe to Freedesktop.org revisionism). Of course even better would be to fix them not to use well-known names in a global namespace but instead pass around a pathname.
No reason to have both /run and /tmp
I think you're right. /tmp is essentially deprecated now we have /run. If your program is in a position to do so (which requires that it was installed as a privileged operation), then nowadays you would use a sub-directory of /run. This is for security reasons.
E.g. the CUPS printing daemon does not run as root, but is generally installed from an OS package. The package installs /usr/lib/tmpfiles.d/cups.conf, and systemd-tmpfiles creates a directory it can access. Since the directory is under /run, the name cannot have been maliciously claimed by an unprivileged user, unlike /tmp which is world-writable.
"Unprivileged programs" which can't use /run directly
The real distinction is if your program is being run by an arbitrary unprivileged user, under their own user ID. But you still generally don't want to use /tmp, because it can be accessed by other unprivileged users. You would prefer to use $XDG_RUNTIME_DIR. Typically this is implemented as /run/user/$(id -u) - so it happens to be a subdirectory of /run as well. The location isn't guaranteed though; programs should always use the environment variable.
/tmp would only be useful for ad-hoc co-operation between different unprivileged users on the system. Such ad-hoc systems are vulnerable to a malicious user refusing to co-operate and spoiling things for everyone :). One example would be unprivileged users deciding to run a version of the talk daemon, using a unix socket.
Original information from Lennart Poettering
Note, Poettering's checklist below claimed that /tmp would be useful for "small files", whereas /run should only be used for "communication primitives". I don't think this distinction is true either. The poster-boy for /run is udev, and I'm pretty sure /run/udev includes internal databases . Once you have a /run directory, I don't think anyone wants to follow the claimed distinction and create another directory, to clutter /tmp. So in practice we just use /run nowadays.
Usage of world-writable shared namespaces [like /tmp] for communication purposes has always been problematic, since to establish communication you need stable names, but stable names open the doors for DoS attacks. This can be corrected partially, by establishing protected per-app directories for certain services during early boot (like we do for X11), but this only fixes the problem partially, since this only works correctly if every package installation is followed by a reboot.
...
Another Fedora feature (for Fedora 17) changed the semantics of /tmp for many system services to make them more secure, by isolating the /tmp namespaces of the various services
...
Because /tmp is no longer necessarily a shared namespace it is generally unsuitable as a location for communication primitives.
...
[/run] is guaranteed to be a tmpfs and is hence automatically flushed at boots. No automatic clean-up is done beyond that.
...
Here's a rough guide how we suggest you (a Linux application developer) pick the right directory to use:
- You need a place to put your socket (or other communication primitive) and your code runs privileged: use a subdirectory beneath /run. (Or beneath /var/run for extra compatibility.)
- You need a place to put your socket (or other communication primitive) and your code runs unprivileged: use a subdirectory beneath $XDG_RUNTIME_DIR.
- You need a place to put your larger downloads and downloads in progress and run unprivileged: use $XDG_DOWNLOAD_DIR.
- You need a place to put cache files which should be persistent and run unprivileged: use $XDG_CACHE_HOME.
- Nothing of the above applies and you need to place a small file that needs no persistency: use $TMPDIR with a fallback on /tmp. And use mkstemp(), and mkdtemp() and nothing homegrown.
- Otherwise use $TMPDIR with a fallback on /var/tmp. Also use mkstemp()/mkdtemp().
Note that these rules above are only suggested by us. These rules take into account everything we know about this topic and avoid problems with current and future distributions, as far as we can see them. Please consider updating your projects to follow these rules, and keep them in mind if you write new code.
One thing we'd like to stress is that /tmp and /var/tmp more often than not are actually not the right choice for your usecase. There are valid uses of these directories, but quite often another directory might actually be the better place. So, be careful, consider the other options, but if you do go for /tmp or /var/tmp then at least make sure to use mkstemp()/mkdtemp().
We kind of get away with the legacy I misread /tmp socket used by the X window system, as described above.tmpfiles.d/x11.conf. Looks more like it relies on co-operation :). I assume the code's been audited, such that denial of service is the worst that can happen.