service: don't walk out of bounds when checking services
authorAaron Conole <aconole@redhat.com>
Tue, 3 Dec 2019 21:15:44 +0000 (16:15 -0500)
committerDavid Marchand <david.marchand@redhat.com>
Fri, 20 Dec 2019 14:09:35 +0000 (15:09 +0100)
The service_valid call is used without properly bounds checking the
input parameter.  Almost all instances of the service_valid call are
inside a for() loop that prevents excessive walks, but some of the
public APIs don't bounds check and will pass invalid arguments.

Prevent this by using SERVICE_GET_OR_ERR_RET where it makes sense,
and adding a bounds check to one service_valid() use.

Fixes: 8d39d3e237c2 ("service: fix race in service on app lcore function")
Fixes: e9139a32f6e8 ("service: add function to run on app lcore")
Fixes: e30dd31847d2 ("service: add mechanism for quiescing")
Cc: stable@dpdk.org
Signed-off-by: Aaron Conole <aconole@redhat.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
lib/librte_eal/common/rte_service.c

index 79235c0..7e537b8 100644 (file)
@@ -137,6 +137,12 @@ service_valid(uint32_t id)
        return !!(rte_services[id].internal_flags & SERVICE_F_REGISTERED);
 }
 
+static struct rte_service_spec_impl *
+service_get(uint32_t id)
+{
+       return &rte_services[id];
+}
+
 /* validate ID and retrieve service pointer, or return error value */
 #define SERVICE_VALID_GET_OR_ERR_RET(id, service, retval) do {          \
        if (id >= RTE_SERVICE_NUM_MAX || !service_valid(id))            \
@@ -344,12 +350,14 @@ rte_service_runner_do_callback(struct rte_service_spec_impl *s,
 }
 
 
-static inline int32_t
-service_run(uint32_t i, struct core_state *cs, uint64_t service_mask)
+/* Expects the service 's' is valid. */
+static int32_t
+service_run(uint32_t i, struct core_state *cs, uint64_t service_mask,
+           struct rte_service_spec_impl *s)
 {
-       if (!service_valid(i))
+       if (!s)
                return -EINVAL;
-       struct rte_service_spec_impl *s = &rte_services[i];
+
        if (s->comp_runstate != RUNSTATE_RUNNING ||
                        s->app_runstate != RUNSTATE_RUNNING ||
                        !(service_mask & (UINT64_C(1) << i))) {
@@ -383,7 +391,7 @@ rte_service_may_be_active(uint32_t id)
        int32_t lcore_count = rte_service_lcore_list(ids, RTE_MAX_LCORE);
        int i;
 
-       if (!service_valid(id))
+       if (id >= RTE_SERVICE_NUM_MAX || !service_valid(id))
                return -EINVAL;
 
        for (i = 0; i < lcore_count; i++) {
@@ -397,12 +405,10 @@ rte_service_may_be_active(uint32_t id)
 int32_t
 rte_service_run_iter_on_app_lcore(uint32_t id, uint32_t serialize_mt_unsafe)
 {
-       /* run service on calling core, using all-ones as the service mask */
-       if (!service_valid(id))
-               return -EINVAL;
-
        struct core_state *cs = &lcore_states[rte_lcore_id()];
-       struct rte_service_spec_impl *s = &rte_services[id];
+       struct rte_service_spec_impl *s;
+
+       SERVICE_VALID_GET_OR_ERR_RET(id, s, -EINVAL);
 
        /* Atomically add this core to the mapped cores first, then examine if
         * we can run the service. This avoids a race condition between
@@ -418,7 +424,7 @@ rte_service_run_iter_on_app_lcore(uint32_t id, uint32_t serialize_mt_unsafe)
                return -EBUSY;
        }
 
-       int ret = service_run(id, cs, UINT64_MAX);
+       int ret = service_run(id, cs, UINT64_MAX, s);
 
        if (serialize_mt_unsafe)
                rte_atomic32_dec(&s->num_mapped_cores);
@@ -438,8 +444,10 @@ rte_service_runner_func(void *arg)
                const uint64_t service_mask = cs->service_mask;
 
                for (i = 0; i < RTE_SERVICE_NUM_MAX; i++) {
+                       if (!service_valid(i))
+                               continue;
                        /* return value ignored as no change to code flow */
-                       service_run(i, cs, service_mask);
+                       service_run(i, cs, service_mask, service_get(i));
                }
 
                cs->loops++;