Deal with CopySpecs that include kernel file system directories better about sos HOT 4 CLOSED

Can we not use the addForbiddenPath method to avoid these paths?

from sos.

Yes - that's what I've been doing for e.g. the deprecated sysctls, see commit 8db01b2 on rhel-6 branch. The problem with this approach is it gets hairy when you have a large number of such paths to manage. For now it may be the best way forward but there are a number of deprecated paths to avoid in proc and a much larger set of paths that we shouldn't attempt to copy in sysfs - sys is ten years old now and we really should up our game in terms of the information we collect from there. Right now we have bigger problems (our recursive copy function isn't safe for trees containing looping symlinks for e.g.) but this is something that I think we need to improve.

from sos.

This is now about 90% fixed. We now handle symlinks that point to directories (but we don't copy the target by default - that way lies madness) and block and character device nodes and other special node types are now handled properly. This means it's now safe to do e.g.:

        self.add_copy_spec("/sys")
        self.add_copy_spec("/dev")
        self.add_copy_spec("/proc")

There's still some work required to make this robust; in particular scooping up the whole of /sys and /proc will require the global blacklists (Issue #316) to avoid tripping over various pipes and comms channels that hang on read() IO.

from sos.

The fix for Issue #266 means that we handle unreadable files correctly. I'll do some more testing on this today and close the issue out if there's nothing more to do.

from sos.