{"id":"CVE-2026-98058","published":"2026-09-25T11:17:34.767","lastModified":"2026-09-25T11:17:34.767","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nbpf: Mark syscall helpers as sleepable\n\nbpf_sys_bpf() executes the bpf(2) syscall body, which can take mutexes,\nallocate with GFP_KERNEL, and wait for an RCU grace period.\nbpf_sys_close() reaches close_fd() and filp_close(), which can sleep as\nwell.\n\nBoth helpers are limited to BPF_PROG_TYPE_SYSCALL, whose main program is\nsleepable. That does not make every callback sleepable: a syscall program\ncan register a bpf_timer callback, and the verifier checks that callback\nin a non-sleepable context while retaining the syscall helper set.\n\nWithout .might_sleep on the prototypes, such a callback can invoke\nbpf_sys_bpf() from hrtimer softirq context and trigger a\nscheduling-while-atomic failure. bpf_sys_close() is exposed through the\nsame missing context check.\n\nSet .might_sleep on both prototypes so the existing helper-context check\nrejects them from timer callbacks and other atomic regions. Calls from the\nsleepable main body remain valid.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/3408e98f06ce06ea6a58ca6c39da9c23f0029c06","tags":[]},{"url":"https://git.kernel.org/stable/c/d05524794240b52fdc3b6c1220dd05505715824d","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}