On Fri, 2023-06-02 at 12:27 -0700, Alexei Starovoitov wrote: > On Fri, Jun 2, 2023 at 12:13 PM Eduard Zingerman <eddyz87@xxxxxxxxx> wrote: > > > > On Fri, 2023-06-02 at 11:50 -0700, Andrii Nakryiko wrote: > > [...] > > > > > The thread is long. Could you please describe it again in pseudo code? > > > > > > > > - Add a function mark_precise_scalar_ids(struct bpf_verifier_env *env, > > > > struct bpf_verifier_state *st) > > > > such that it: > > > > - collect PRECISE_IDS: a set of IDs of all registers marked in env->bt > > > > - visit all registers with ids from PRECISE_IDS and make sure > > > > that these registers are marked in env->bt > > > > - Call mark_precise_scalar_ids() from __mark_chain_precision() > > > > for each state 'st' visited by states chain processing loop, > > > > so that: > > > > - mark_precise_scalar_ids() is called for current state when > > > > __mark_chain_precision() is entered, reusing id assignments in > > > > current state; > > > > - mark_precise_scalar_ids() is called for each parent state, reusing > > > > id assignments valid at 'last_idx' instruction of that state. > > > > > > > > The idea is that in situations like below: > > > > > > > > 4: if (r6 > r7) goto +1 > > > > 5: r7 = r6 > > > > --- checkpoint #1 --- > > > > 6: <something> > > > > 7: if (r7 > X) goto ... > > > > 8: r7 = 0 > > > > 9: r9 += r6 > > > > > > > > The mark_precise_scalar_ids() would be called at: > > > > - (9) and current id assignments would be used. > > > > - (6) and id assignments saved in checkpoint #1 would be used. > > > > > > > > If <something> is the code that modifies r6/r7 the link would be > > > > broken and we would overestimate the set of precise registers. > > > > > > > > > > To avoid this we need to recalculate these IDs on each new parent > > > state, based on requested precision marks. If we keep a simple and > > > small array of IDs and do a quick linear search over them for each > > > SCALAR register, I suspect it should be very fast. I don't think in > > > practice we'll have more than 1-2 IDs in that array, right? > > > > I'm not sure I understand, could you please describe how it should > > work for e.g.?: > > > > 3: r6 &= 0xf // assume safe bound > > 4: if (r6 > r7) goto +1 > > 5: r7 = r6 > > --- checkpoint #1 --- > > 6: r7 = 0 > > 7: if (r7 > 10) goto exit; > > 8: r7 = 0 > > 9: r9 += r6 > > > > __mark_chain_precision() would get to checkpoint #1 with only r6 as > > precise, what should happen next? > > > > As a side note: I added several optimizations: > > - avoid allocation of scalar ids for constants; > > +1 > > > - remove sole scalar ids from cached states; > > - do a check as follows: > > if (rold->precise && rold->id && !check_ids(idmap, rold, rcur)) > > Ignoring rcur->id > 0 ? Is it safe? Well, I thought about it a bit and arrived to the following reasoning: - suppose checkpoint C exists, is proven safe and has registers r6=Pscalar(range1),id=0 and r7=Pscalar(range2),id=0 - this means that C is proven safe for any value of r6 in range1 and any value of r7 in range2 - having same id on r6 and r7 means that r6 and r7 share same value - so this is just a special case of what's already proven. But having written this down, it looks like I also need to verify that range1 and range2 overlap :( > > > return false; > > > > And I'm seeing almost zero performance overhead now. > > So, maybe what we figured so far is good enough. > > Need to add more tests, though.
