file watch question

Reading the man page for auditctl, looking at file watch rules I see this:

-w path
Insert a watch for the file system object at path. You
cannot insert a watch to the top level directory. This is prohibited by the kernel.
Wildcards are not supported either and will generate a
warning. The way that watches work is by tracking the inode internally. If you place
a watch on a file, its the same as using the -F path
option on a syscall rule. If you place a watch on a directory, its the same as using
the -F dir option on a syscall rule. The -w form of
writing watches is for backwards compatibility and the syscall based form is more
expressive. Unlike most syscall auditing rules, watches do
not impact performance based on the number of rules sent to the kernel. The only
valid options when using a watch are the -p and -k. If
you need to anything fancy like audit a specific user accessing a file, then use the
syscall auditing form with the path or dir fields. See the
EXAMPLES section for an example of converting one form to another.

I assume if the "-w form" is just backwards-compatible, it is preferred to use the syscall method.
Question -
The line saying, "Unlike most syscall auditing rules, watches do not impact performance based on the number of rules sent to the kernel" - does this mean that BOTH the "-w" and "syscall" rules have no performance impact?

Thx,
LCB

LC (Lenny) Bruzenak
[email protected]

Nevermind; I see that the rule in force (with auditctl -l) is the same regardless of the way it is specified in the auditctl line.

Thx,
LCB

LC (Lenny) Bruzenak
[email protected]

The -w form is for backward compatibility with the original patch. It also turns out to be more concise to type. But auditctl translates it to a syscall rule with the -F path or -F dir depending on what it detects the object to be at rule loading time. The Syscall form does allow more fine grained control.

The reason it doesn't have performance impact like normal syscall rules is because it gets moved to a list that is not evaluated every syscall. A normal syscall rule will get evaluated for every syscall because it has to see if the syscall number is of interest and then it checks the next rule. All of my attempts to measure performance impact of watches showed that its in the cache noise.

Last time I measured was about 7 years ago but I don't think anything has changed that would affect the old results.

http://people.redhat.com/sgrubb/files/watch-perf.tar.gz

-Steve