{"id":"CVE-2026-98084","published":"2026-09-25T11:17:37.980","lastModified":"2026-09-25T11:17:37.980","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nbpf: backtracking shouldn't clear outer frame R1-R5 for callbacks\n\nWhen processing calls to bpf_loop() verifier marks R1 (and R4) as\nprecise. R1 tracks loop iterations number and because of the\n'callback_depth < R1' mechanics in check_helper_call() must be marked\nprecise. However, precision propagation for R1 was broken,\nwhen bpf_loop() call was verified on a second iteration.\n\nConsider the following verification trace:\n- main: bpf_loop(nr_loops, callback ...)\n- callback: BPF_EXIT\n- main: bpf_loop(nr_loops, callback ...)\n- ...\n\nWhile the first visit of the call to bpf_loop() propagated R1\nprecision as expected, the second call to mark_chain_precision() in\nthe check_helper_call() set R1, but it was immediately reset when\nbacktrack_insn() processed preceding BPF_EXIT in the loop deleted in\nthis patch.\n\nBecause of that, the second visit of the call to bpf_loop() injected\ncheckpoint with R1 not marked as precise. Which could trick the\nverifier into accepting unsafe programs. See the next patch for an\nexample of such program.\n\nCommit is structured in a way to minimize conflicts when\n'bpf' would be eventually merged with 'bpf-next'.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/e3e4f66cc4b72333d0886ae2673c360248987889","tags":[]},{"url":"https://git.kernel.org/stable/c/f83890dfed9efc5e19df1dbc1ad707a142936923","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}