From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions When using eBPF to probe CS CALL instructions within a function, a crash can be triggered. The eBPF tool probes offset 257 of the __hrtimer_run_queues() function: <__hrtimer_run_queues+249>: nopl 0x0(%rax,%rax,1) <__hrtimer_run_queues+254>: mov %r14,%rdi <__hrtimer_run_queues+257>: cs call <__x86_indirect_thunk_r12> <__hrtimer_run_queues+263>: mov %eax,%r12d <__hrtimer_run_queues+266>: xchg %ax,%ax <__hrtimer_run_queues+268>: mov %r13,%rdi Which triggers this crash: BUG: unable to handle page fault for address: 00000000000f41c9 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P RIP: 0010:__hrtimer_run_queues+0x106/0x230 Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is at the 6th byte of the above CS CALL instruction. Since the CS CALL instruction occupies 6 bytes, the exception occurred in the middle of that call instruction. The root cause is that when using eBPF tools to probe in the middle of a function, a kprobe with INT3 is used as the underlying implementation. During single-step emulation of the original CALL instruction, int3_emulate_call() assumes that the probed CALL instruction is 5 bytes long. However, the actual CS-prefixed CALL instruction occupies 6 bytes, so it constructs an incorrect exception return address. When the CPU returns from the kprobe handler, the next instruction to be executed is at the address of the last byte of that CS CALL instruction. Coincidentally, starting from that address, the CPU fetches and decodes a completely different instruction, which ultimately triggers a kernel crash. Fix the issue by using the actual instruction length obtained from the instruction decoder when constructing the exception return address, rather than relying on the hardcoded CALL_INSN_SIZE macro. [ mingo: Refined the changelog ] Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step") Suggested-by: Masami Hiramatsu (Google) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- fs/tests/exec_kunit.c | 132 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 132 insertions(+) create mode 100644 fs/tests/exec_kunit.c (limited to 'fs/tests/exec_kunit.c') diff --git a/fs/tests/exec_kunit.c b/fs/tests/exec_kunit.c new file mode 100644 index 000000000..1c32cac09 --- /dev/null +++ b/fs/tests/exec_kunit.c @@ -0,0 +1,132 @@ +// SPDX-License-Identifier: GPL-2.0-only +#include + +struct bprm_stack_limits_result { + struct linux_binprm bprm; + int expected_rc; + unsigned long expected_argmin; +}; + +static const struct bprm_stack_limits_result bprm_stack_limits_results[] = { + /* Negative argc/envc counts produce -E2BIG */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ULONG_MAX, + .argc = INT_MIN, .envc = INT_MIN }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ULONG_MAX, + .argc = 5, .envc = -1 }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ULONG_MAX, + .argc = -1, .envc = 10 }, .expected_rc = -E2BIG }, + /* The max value of argc or envc is MAX_ARG_STRINGS. */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ULONG_MAX, + .argc = INT_MAX, .envc = INT_MAX }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ULONG_MAX, + .argc = MAX_ARG_STRINGS, .envc = MAX_ARG_STRINGS }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ULONG_MAX, + .argc = 0, .envc = MAX_ARG_STRINGS }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ULONG_MAX, + .argc = MAX_ARG_STRINGS, .envc = 0 }, .expected_rc = -E2BIG }, + /* + * On 32-bit system these argc and envc counts, while likely impossible + * to represent within the associated TASK_SIZE, could overflow the + * limit calculation, and bypass the ptr_size <= limit check. + */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ULONG_MAX, + .argc = 0x20000001, .envc = 0x20000001 }, .expected_rc = -E2BIG }, +#ifdef CONFIG_MMU + /* Make sure a pathological bprm->p doesn't cause an overflow. */ + { { .p = sizeof(void *), .rlim_stack.rlim_cur = ULONG_MAX, + .argc = 10, .envc = 10 }, .expected_rc = -E2BIG }, +#endif + /* + * 0 rlim_stack will get raised to ARG_MAX. With 1 string pointer, + * we should see p - ARG_MAX + sizeof(void *). + */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 0, + .argc = 1, .envc = 0 }, .expected_argmin = ULONG_MAX - ARG_MAX + sizeof(void *)}, + /* Validate that argc is always raised to a minimum of 1. */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 0, + .argc = 0, .envc = 0 }, .expected_argmin = ULONG_MAX - ARG_MAX + sizeof(void *)}, + /* + * 0 rlim_stack will get raised to ARG_MAX. With pointers filling ARG_MAX, + * we should see -E2BIG. (Note argc is always raised to at least 1.) + */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 0, + .argc = ARG_MAX / sizeof(void *), .envc = 0 }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 0, + .argc = 0, .envc = ARG_MAX / sizeof(void *) - 1 }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 0, + .argc = ARG_MAX / sizeof(void *) + 1, .envc = 0 }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 0, + .argc = 0, .envc = ARG_MAX / sizeof(void *) }, .expected_rc = -E2BIG }, + /* And with one less, we see space for exactly 1 pointer. */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 0, + .argc = (ARG_MAX / sizeof(void *)) - 1, .envc = 0 }, + .expected_argmin = ULONG_MAX - sizeof(void *) }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 0, + .argc = 0, .envc = (ARG_MAX / sizeof(void *)) - 2, }, + .expected_argmin = ULONG_MAX - sizeof(void *) }, + /* If we raise rlim_stack / 4 to exactly ARG_MAX, nothing changes. */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ARG_MAX * 4, + .argc = ARG_MAX / sizeof(void *), .envc = 0 }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ARG_MAX * 4, + .argc = 0, .envc = ARG_MAX / sizeof(void *) - 1 }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ARG_MAX * 4, + .argc = ARG_MAX / sizeof(void *) + 1, .envc = 0 }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ARG_MAX * 4, + .argc = 0, .envc = ARG_MAX / sizeof(void *) }, .expected_rc = -E2BIG }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ARG_MAX * 4, + .argc = (ARG_MAX / sizeof(void *)) - 1, .envc = 0 }, + .expected_argmin = ULONG_MAX - sizeof(void *) }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = ARG_MAX * 4, + .argc = 0, .envc = (ARG_MAX / sizeof(void *)) - 2, }, + .expected_argmin = ULONG_MAX - sizeof(void *) }, + /* But raising it another pointer * 4 will provide space for 1 more pointer. */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = (ARG_MAX + sizeof(void *)) * 4, + .argc = ARG_MAX / sizeof(void *), .envc = 0 }, + .expected_argmin = ULONG_MAX - sizeof(void *) }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = (ARG_MAX + sizeof(void *)) * 4, + .argc = 0, .envc = ARG_MAX / sizeof(void *) - 1 }, + .expected_argmin = ULONG_MAX - sizeof(void *) }, + /* Raising rlim_stack / 4 to _STK_LIM / 4 * 3 will see more space. */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 4 * (_STK_LIM / 4 * 3), + .argc = 0, .envc = 0 }, + .expected_argmin = ULONG_MAX - (_STK_LIM / 4 * 3) + sizeof(void *) }, + /* But raising it any further will see no increase. */ + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 4 * (_STK_LIM / 4 * 3 + sizeof(void *)), + .argc = 0, .envc = 0 }, + .expected_argmin = ULONG_MAX - (_STK_LIM / 4 * 3) + sizeof(void *) }, + { { .p = ULONG_MAX, .rlim_stack.rlim_cur = 4 * _STK_LIM, + .argc = 0, .envc = 0 }, + .expected_argmin = ULONG_MAX - (_STK_LIM / 4 * 3) + sizeof(void *) }, +}; + +static void exec_test_bprm_stack_limits(struct kunit *test) +{ + /* Double-check the constants. */ + KUNIT_EXPECT_EQ(test, _STK_LIM, SZ_8M); + KUNIT_EXPECT_EQ(test, ARG_MAX, 32 * SZ_4K); + KUNIT_EXPECT_EQ(test, MAX_ARG_STRINGS, 0x7FFFFFFF); + + for (int i = 0; i < ARRAY_SIZE(bprm_stack_limits_results); i++) { + const struct bprm_stack_limits_result *result = &bprm_stack_limits_results[i]; + struct linux_binprm bprm = result->bprm; + int rc; + + rc = bprm_stack_limits(&bprm); + KUNIT_EXPECT_EQ_MSG(test, rc, result->expected_rc, "on loop %d", i); +#ifdef CONFIG_MMU + KUNIT_EXPECT_EQ_MSG(test, bprm.argmin, result->expected_argmin, "on loop %d", i); +#endif + } +} + +static struct kunit_case exec_test_cases[] = { + KUNIT_CASE(exec_test_bprm_stack_limits), + {}, +}; + +static struct kunit_suite exec_test_suite = { + .name = "exec", + .test_cases = exec_test_cases, +}; + +kunit_test_suite(exec_test_suite); -- cgit v1.3.1