Re: [PATCH] bpf: emit audit messages upon successful prog load and unload

On Thu, Nov 21, 2019 at 7:23 PM Alexei Starovoitov <alexei.starovoitov(a)gmail.com&gt; wrote: > On Thu, Nov 21, 2019 at 06:41:31PM -0500, Paul Moore wrote: > > On Wed, Nov 20, 2019 at 4:49 PM Alexei Starovoitov > > <alexei.starovoitov(a)gmail.com&gt; wrote: > > > On Wed, Nov 20, 2019 at 1:46 PM Daniel Borkmann <daniel(a)iogearbox.net&gt; wrote: > > > > On 11/20/19 10:38 PM, Jiri Olsa wrote: > > > > > From: Daniel Borkmann <daniel(a)iogearbox.net&gt; > > > > > > > > > > Allow for audit messages to be emitted upon BPF program load and > > > > > unload for having a timeline of events. The load itself is in > > > > > syscall context, so additional info about the process initiating > > > > > the BPF prog creation can be logged and later directly correlated > > > > > to the unload event. > > > > > > > > > > The only info really needed from BPF side is the globally unique > > > > > prog ID where then audit user space tooling can query / dump all > > > > > info needed about the specific BPF program right upon load event > > > > > and enrich the record, thus these changes needed here can be kept > > > > > small and non-intrusive to the core. > > > > > > > > > > Raw example output: > > > > > > > > > > # auditctl -D > > > > > # auditctl -a always,exit -F arch=x86_64 -S bpf > > > > > # ausearch --start recent -m 1334 > > > > > [...] > > > > > ---- > > > > > time->Wed Nov 20 12:45:51 2019 > > > > > type=PROCTITLE msg=audit(1574271951.590:8974): proctitle="./test_verifier" > > > > > type=SYSCALL msg=audit(1574271951.590:8974): arch=c000003e syscall=321 success=yes exit=14 a0=5 a1=7ffe2d923e80 a2=78 a3=0 items=0 ppid=742 pid=949 auid=0 uid=0 gid=0 euid=0 suid=0 fsuid=0 egid=0 sgid=0 fsgid=0 tty=pts0 ses=2 comm="test_verifier" exe="/root/bpf-next/tools/testing/selftests/bpf/test_verifier" subj=unconfined_u:unconfined_r:unconfined_t:s0-s0:c0.c1023 key=(null) > > > > > type=UNKNOWN[1334] msg=audit(1574271951.590:8974): auid=0 uid=0 gid=0 ses=2 subj=unconfined_u:unconfined_r:unconfined_t:s0-s0:c0.c1023 pid=949 comm="test_verifier" exe="/root/bpf-next/tools/testing/selftests/bpf/test_verifier" prog-id=3260 event=LOAD > > > > > ---- > > > > > time->Wed Nov 20 12:45:51 2019 > > > > > type=UNKNOWN[1334] msg=audit(1574271951.590:8975): prog-id=3260 event=UNLOAD > > > > > ---- > > > > > [...] > > > > > > > > > > Signed-off-by: Daniel Borkmann <daniel(a)iogearbox.net&gt; > > > > > Signed-off-by: Jiri Olsa <jolsa(a)kernel.org&gt; > > > > > > > > LGTM, thanks for the rebase! > > > > > > Applied to bpf-next. Thanks! > > > > [NOTE: added linux-audit to the To/CC line] > > > > Wait a minute, why was the linux-audit list not CC'd on this? Why are > > you merging a patch into -next that adds to the uapi definition *and* > > creates a new audit record while we are at -rc8? > > > > Aside from that I'm concerned that you are relying on audit userspace > > changes that might not be okay; I see the PR below, but I don't see > > any comment on it from Steve (it is his audit userspace). I also > > don't see a corresponding test added to the audit-testsuite, which is > > a common requirement for new audit functionality (link below). I'm > > also fairly certain we don't want this new BPF record to look like how > > you've coded it up in bpf_audit_prog(); duplicating the fields with > > audit_log_task() is wrong, you've either already got them via an > > associated record (which you get from passing non-NULL as the first > > parameter to audit_log_start()), or you don't because there is no > > associated syscall/task (which you get from passing NULL as the first > > parameter). Please revert, un-merge, etc. this patch from bpf-next; > > it should not go into Linus' tree as written. > > Sorry I didn't realize there was a disagreement. > > Dave, could you please revert it in net-next? > > > Audit userspace PR: > > * https://github.com/linux-audit/audit-userspace/pull/104 > > This PR does not use this new audit. It's doing everything via existing > perf_event notification. My understanding of Jiri's email was that netlink > style is preferred vs perf_event. Did I get it wrong? Perhaps confusion on my part regarding the audit-userspace PR. The commit description mentioned the audit userspace (the daemon most likely) fetching additional info about the BPF program and this was the only outstanding audit-userspace PR that had any mention of BPF. However, getting back to netlink vs perf_event, if you want to generate an audit record, it should happen via the audit subsystem (and go up to the audit daemon via netlink).