tree 19ca27b2d85727299d5fa93f3a085bf3bef71b4a
parent 556a399406635566413f9c71b134d5d287b25b29
parent f932a8e4824b529e455b7e3eb3e5118beceb3e32
author Alexei Starovoitov <ast@kernel.org> 1737078670 -0800
committer Alexei Starovoitov <ast@kernel.org> 1737078671 -0800

Merge branch 'support-eliding-map-lookup-nullness'

Daniel Xu says:

====================
Support eliding map lookup nullness

This patch allows progs to elide a null check on statically known map
lookup keys. In other words, if the verifier can statically prove that
the lookup will be in-bounds, allow the prog to drop the null check.

This is useful for two reasons:

1. Large numbers of nullness checks (especially when they cannot fail)
   unnecessarily pushes prog towards BPF_COMPLEXITY_LIMIT_JMP_SEQ.
2. It forms a tighter contract between programmer and verifier.

For (1), bpftrace is starting to make heavier use of percpu scratch
maps. As a result, for user scripts with large number of unrolled loops,
we are starting to hit jump complexity verification errors.  These
percpu lookups cannot fail anyways, as we only use static key values.
Eliding nullness probably results in less work for verifier as well.

For (2), percpu scratch maps are often used as a larger stack, as the
currrent stack is limited to 512 bytes. In these situations, it is
desirable for the programmer to express: "this lookup should never fail,
and if it does, it means I messed up the code". By omitting the null
check, the programmer can "ask" the verifier to double check the logic.

=== Changelog ===

Changes in v7:
* Use more accurate frame number when marking precise
* Add test for non-stack key
* Test for marking stack slot precise

Changes in v6:
* Use is_spilled_scalar_reg() helper and remove unnecessary comment
* Add back deleted selftest with different helper to dirty dst buffer
* Check size of spill is exactly key_size and update selftests
* Read slot_type from correct offset into the spi
* Rewrite selftests in C where possible
* Mark constant map keys as precise

Changes in v5:
* Dropped all acks
* Use s64 instead of long for const_map_key
* Ensure stack slot contains spilled reg before accessing spilled_ptr
* Ensure spilled reg is a scalar before accessing tnum const value
* Fix verifier selftest for 32-bit write to write at 8 byte alignment
  to ensure spill is tracked
* Introduce more precise tracking of helper stack accesses
* Do constant map key extraction as part of helper argument processing
  and then remove duplicated stack checks
* Use ret_flag instead of regs[BPF_REG_0].type
* Handle STACK_ZERO
* Fix bug in bpf_load_hdr_opt() arg annotation

Changes in v4:
* Only allow for CAP_BPF
* Add test for stack growing upwards
* Improve comment about stack growing upwards

Changes in v3:
* Check if stack is (erroneously) growing upwards
* Mention in commit message why existing tests needed change

Changes in v2:
* Added a check for when R2 is not a ptr to stack
* Added a check for when stack is uninitialized (no stack slot yet)
* Updated existing tests to account for null elision
* Added test case for when R2 can be both const and non-const
====================

Link: https://patch.msgid.link/cover.1736886479.git.dxu@dxuuu.xyz
Signed-off-by: Alexei Starovoitov <ast@kernel.org>