[Qemu-devel] [PATCH v3] linux-user: do setrlimit selectively

[Max Filippov]

setrlimit guest calls that affect memory resources
(RLIMIT_{AS,DATA,STACK}) may interfere with QEMU internal memory
management. They may result in QEMU lockup because mprotect call in
page_unprotect would fail with ENOMEM error code, causing infinite loop
of SIGSEGV. E.g. it happens when running libstdc++ testsuite for xtensa
target on x86_64 host.

Don't call host setrlimit for memory-related resources.

Reviewed-by: Peter Maydell <address@hidden>
Signed-off-by: Max Filippov <address@hidden>
---
Changes v2->v3:
- add comment sugessted by Peter

Changes v1->v2:
- don't limit change to 32-bit guest on 64-bit host case

 linux-user/syscall.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index bb42a225eb58..f06307ebd945 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -9275,7 +9275,21 @@ abi_long do_syscall(void *cpu_env, int num, abi_long 
arg1,
             rlim.rlim_cur = target_to_host_rlim(target_rlim->rlim_cur);
             rlim.rlim_max = target_to_host_rlim(target_rlim->rlim_max);
             unlock_user_struct(target_rlim, arg2, 0);
-            ret = get_errno(setrlimit(resource, &rlim));
+            /*
+             * If we just passed through resource limit settings for memory 
then
+             * they would also apply to QEMU's own allocations, and QEMU will
+             * crash or hang or die if its allocations fail. Ideally we would
+             * track the guest allocations in QEMU and apply the limits 
ourselves.
+             * For now, just tell the guest the call succeeded but don't 
actually
+             * limit anything.
+             */
+            if (resource != RLIMIT_AS &&
+                resource != RLIMIT_DATA &&
+                resource != RLIMIT_STACK) {
+                return get_errno(setrlimit(resource, &rlim));
+            } else {
+                return 0;
+            }
         }
         break;
 #endif
-- 
2.11.0