@@ -277,22 +277,38 @@ void __test_init(unsigned int ntests, int family, unsigned int prefix,
/* /proc/sys/net/core/optmem_max artifically limits the amount of memory
* that can be allocated with sock_kmalloc() on each socket in the system.
- * It is not virtualized, so it has to written outside test namespaces.
- * To be nice a test will revert optmem back to the old value.
+ * It is not virtualized in v6.7, so it has to written outside test
+ * namespaces. To be nice a test will revert optmem back to the old value.
* Keeping it simple without any file lock, which means the tests that
* need to set/increase optmem value shouldn't run in parallel.
* Also, not re-entrant.
+ * Since commit f5769faeec36 ("net: Namespace-ify sysctl_optmem_max")
+ * it is per-namespace, keeping logic for non-virtualized optmem_max
+ * for v6.7, which supports TCP-AO.
*/
static const char *optmem_file = "/proc/sys/net/core/optmem_max";
static size_t saved_optmem;
+static int optmem_ns = -1;
+
+static bool is_optmem_namespaced(void)
+{
+ if (optmem_ns == -1) {
+ int old_ns = switch_save_ns(nsfd_child);
+
+ optmem_ns = !access(optmem_file, F_OK);
+ switch_ns(old_ns);
+ }
+ return !!optmem_ns;
+}
size_t test_get_optmem(void)
{
+ int old_ns = 0;
FILE *foptmem;
- int old_ns;
size_t ret;
- old_ns = switch_save_ns(nsfd_outside);
+ if (!is_optmem_namespaced())
+ old_ns = switch_save_ns(nsfd_outside);
foptmem = fopen(optmem_file, "r");
if (!foptmem)
test_error("failed to open %s", optmem_file);
@@ -300,19 +316,21 @@ size_t test_get_optmem(void)
if (fscanf(foptmem, "%zu", &ret) != 1)
test_error("can't read from %s", optmem_file);
fclose(foptmem);
- switch_ns(old_ns);
+ if (!is_optmem_namespaced())
+ switch_ns(old_ns);
return ret;
}
static void __test_set_optmem(size_t new, size_t *old)
{
+ int old_ns = 0;
FILE *foptmem;
- int old_ns;
if (old != NULL)
*old = test_get_optmem();
- old_ns = switch_save_ns(nsfd_outside);
+ if (!is_optmem_namespaced())
+ old_ns = switch_save_ns(nsfd_outside);
foptmem = fopen(optmem_file, "w");
if (!foptmem)
test_error("failed to open %s", optmem_file);
@@ -320,7 +338,8 @@ static void __test_set_optmem(size_t new, size_t *old)
if (fprintf(foptmem, "%zu", new) <= 0)
test_error("can't write %zu to %s", new, optmem_file);
fclose(foptmem);
- switch_ns(old_ns);
+ if (!is_optmem_namespaced())
+ switch_ns(old_ns);
}
static void test_revert_optmem(void)
Since commit f5769faeec36 ("net: Namespace-ify sysctl_optmem_max") optmem_max is per-netns, so need of switching to root namespace. It seems trivial to keep the old logic working, so going to keep it for a while (at least, until kernel with netns-optmem_max will be release). Currently, there is a test that checks that optmem_max limit applies to TCP-AO keys and a little benchmark that measures linked-list TCP-AO keys scaling, those are fixed by this. Cc: Eric Dumazet <edumazet@google.com> Signed-off-by: Dmitry Safonov <dima@arista.com> --- tools/testing/selftests/net/tcp_ao/lib/setup.c | 35 ++++++++++++++++++++------ 1 file changed, 27 insertions(+), 8 deletions(-)