Also, we would need to have a separate rule for open and openat family with something like

-S open,openat -F dir=/etc -F perm=wa -k ETC_WATCH.

So 2 syscall rules instead of 1 watch rule but we replace -S all with specific syscalls.

Or we could combine all these open,openat,write family syscalls into a single syscall rule but with the permissions flag.

On Wed, Mar 15, 2023 at 3:29 PM Amjad Gabbar <amjadgabbar11@gmail.com> wrote:
Yup....I was able to find the dummy check you are referring to and the audit_reset_context() that is called next(which immediately exits in case of dummy).
Thanks for the help folks....have a much better understanding of how the audit context is allocated on enabling syscall auditing and the whole flow post that.

Had just 1 question wrt watches. IIUC, for watch rules we evaluate all syscalls (Snippet from audit-userspace: https://github.com/linux-audit/audit-userspace/blob/1482cec74f2d9472f81dd4f0533484bd0c26decd/lib/libaudit.c#L805).
But based on the permissions in the rule, we evaluate if the syscall belongs to a specific Audit Class using audit_match_class() and only log if the syscalls match/ are part of the class. This also explains why I see audit_filter_rules() called for watches even if the syscall being performed is not at all related to file system auditing.

  1. I was wondering why do we not automatically identify if the syscall is of interest or not in audit_in_mask() itself based on the rule permissions of the watch? In this way we would avoid the additional overhead of each syscall going into audit_filter_rules() and then evaluating on the AUDIT_PERM case as well.

Currently a watch rule for "wa" permissions for /etc is similar to :
-a always,exit -F arch=b64 -S all -F dir=/etc -F perm=wa -k ETC_WATCH

We only log if the syscall is part of the WRITE and ATTR permissions set. Instead what I was suggesting was something like this:

-a always,exit -F arch=b64 -S <all syscalls part of the write and attr classes> -F dir=/etc  -k ETC_WATCH

Please correct me if my understanding in any of the above is incorrect.

On Fri, Mar 10, 2023 at 3:54 PM Richard Guy Briggs <rgb@redhat.com> wrote:
On 2023-02-17 16:50, Steve Grubb wrote:
> Hello,
>
> On Tuesday, February 14, 2023 3:55:58 PM EST Amjad Gabbar wrote:
> > Thanks for the reply.
> > I was trying to evaluate the same via Flamegraphs and what I noticed was
> > that :
> >
> > 1. Despite deleting all rules (auditctl -D), there were still calls to
> > audit_filter_syscall() on each syscall. I assume this is because syscall
> > auditing is enabled and despite no rules, there still will be some
> > performance impact and calls to syscall filtering functions on each
> > syscall.
>
> Yes.
>
> > 2. For a single watch rule as well without any syscall rules, I could see
> > calls to audit_filter_syscall() followed by audit_filter_rules() for
> > unrelated syscalls such as futex() and recvmsg() - not present in
> > include/asm-generic/audit_*.h
> > Why would these functions be called for a single watch rule for syscalls
> > unrelated to the permissions?
>
> If auditing is enabled, it will go into the syscall filter for *any* syscall.
> It will go into __audit_syscall_exit for every syscall. If there is an audit
> context, it will go into audit_filter_syscall. The documentation in the
> comments above these functions is informative.
>
> My guess is that this code path might benefit from adding a list_empty check.
> A long time ago, I think we kept a variable that denoted if there were any
> rules and short-circuited if none.

There is essentially an empty list check in __audit_syscall_exit() with
the dummy check, based on the number of syscall (or io_uring) rules in
place tracked in audit_n_rules.  Unfortunately, we can't bail from
__audit_syscall_entry() right after setting dummy because other
hardwired records can cancel the dummy flag.

> -Steve
>
> > On Tue, Feb 14, 2023 at 8:29 AM Steve Grubb <sgrubb@redhat.com> wrote:
> > > Hello,
> > >
> > > On Monday, February 13, 2023 4:24:02 PM EST Amjad Gabbar wrote:
> > > > I wanted some help in better understanding the workflow of file system
> > > > auditing(watch rules) vs Syscall Auditing(syscall rules). I know in
> > >
> > > general
> > >
> > > > file system auditing does not have the same performance impact as
> > > > syscall
> > > > auditing, even though both make use of syscall exits for their
> > >
> > > evaluation.
> > >
> > > > From the manpage - "Unlike most syscall auditing rules, watches do not
> > > > impact performance based on the number of rules sent to the kernel."
> > > >
> > > > From a previous thread, I found this excerpt regarding file watch rules
> > >
> > > vs
> > >
> > > > sycall rules -
> > > >
> > > > "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."
> > > >
> > > > Based on this I had a couple of questions:
> > > >
> > > > For normal syscall rules, the evaluation happens as
> > > > __audit_syscall_exit
> > > > <https://elixir.bootlin.com/linux/v6.1.10/C/ident/__audit_syscall_exit>
> > > > calls audit_filter_syscall
> > > > (https://elixir.bootlin.com/linux/v6.1.10/source/kernel/auditsc.c#L841)
> > > >
> > > > Here, we check if the syscall is of interest or not in the
> > > > audit_in_mask
> > > > <https://elixir.bootlin.com/linux/v6.1.10/C/ident/audit_in_mask>
> > >
> > > function.
> > >
> > > > Only if the syscall is of interest do we proceed with examining the
> > > > task
> > > > and return on the first rule match.
> > > >
> > > > 1. What is the process or code path for watch rules?
> > > > audit_filter_syscall
> > > > <https://elixir.bootlin.com/linux/v6.1.10/C/ident/audit_filter_syscall>
> > >
> > > is
> > >
> > > > called for watch rules as well. Then how is it that these are not
> > > > called
> > > > for every syscall? Could you point me to the code where the evaluation
> > > > happens only once?
> > >
> > > There is a file, kernel/audit_watch.c, that implements the interface
> > > between
> > > audit and fsnotify. You would want to learn how fsnotify works to
> > > understand
> > > how it avoids the syscall filter.
> > >
> > > > 2. Also, do file watches only involve the open system call family
> > > > (open,
> > > > openat etc). The man page implies the same, so just wanted to confirm.
> > > >
> > > > I assume -w /etc -p wa is the same as -a always,exit -S open -S openat
> > > > -F
> > > > dir=/etc?
> > >
> > > It depends on the flag passed for perm as to what syscall it wants. See:
> > >
> > > include/asm-generic/audit_*.h
> > >
> > > -Steve
>
>
>
>
> --
> Linux-audit mailing list
> Linux-audit@redhat.com
> https://listman.redhat.com/mailman/listinfo/linux-audit
>

- RGB

--
Richard Guy Briggs <rgb@redhat.com>
Sr. S/W Engineer, Kernel Security, Base Operating Systems
Remote, Ottawa, Red Hat Canada
IRC: rgb, SunRaycer
Voice: +1.647.777.2635, Internal: (81) 32635