On 15 November 2026 12:14:04 GMT, Bradley Morgan [off-list ref]
wrote:
On 15 November 2026 08:25:39 GMT, Renzo Davoli [off-list ref] wrote:
quoted
With is_error set, PTRACE_SET_SYSCALL_INFO passes info.exit.rval to
syscall_set_return_value() as an "int" error without checking it: a
positive value, or on 64-bit one outside the "int" range, turns the
error into a success or is truncated.
Reject with -ERANGE any error value outside [-MAX_ERRNO, -1].
Fixes: 26bb32768fe6 ("ptrace: introduce PTRACE_SET_SYSCALL_INFO request")
CC:stable?
Reviewed-by: Bradley Morgan <redacted>
quoted
Signed-off-by: Renzo Davoli <redacted>
---
v2: shorter commit message and comment, dropped Assisted-by (Bradley
Morgan)
quoted
v1:
https://lore.kernel.org/lkml/20261114184842.373304-1-renzo@cs.unibo.it/ (local)
quoted
Tested on x86_64: rval 0, 1, 3, -4096, INT_MIN and 0x100000000 with
is_error set are now rejected with -ERANGE; -ESRCH and -4095 still work.
I don't know why ai does these kinds of sections tbf. I think testing is
like a basic requirement, like so basic that it wouldn't need to be added
as a section
quoted
kernel/ptrace.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/kernel/ptrace.c b/kernel/ptrace.c
index 3f2fe68ccb14..1ffacfcef8d4 100644
--- a/kernel/ptrace.c
+++ b/kernel/ptrace.c
@@ -1111,6 +1111,10 @@ ptrace_set_syscall_info_exit(struct task_struct
*child, struct pt_regs *regs,
quoted
if (rval != info->exit.rval)
return -ERANGE;
+ /* An error must be a negative errno value. */
+ if (info->exit.is_error && (rval < -MAX_ERRNO || rval >= 0))
+ return -ERANGE;
+
please consider Renzos comment and add a test case.
oops Dmitry.. sorry, consider dmitrys comment
quoted hunk ↗ jump to hunk
quoted
if (info->exit.is_error)
syscall_set_return_value(child, regs, rval, 0);
else
base-commit: 7e1392a42ea288d70a266f6553fe06f57f8647ea