OSDN Git Service

samples/bpf: fix broken map lookup probe
authorDaniel T. Lee <danieltimlee@gmail.com>
Fri, 18 Aug 2023 09:01:17 +0000 (18:01 +0900)
committerAlexei Starovoitov <ast@kernel.org>
Mon, 21 Aug 2023 22:39:09 +0000 (15:39 -0700)
In the commit 7c4cd051add3 ("bpf: Fix syscall's stackmap lookup
potential deadlock"), a potential deadlock issue was addressed, which
resulted in *_map_lookup_elem not triggering BPF programs.
(prior to lookup, bpf_disable_instrumentation() is used)

To resolve the broken map lookup probe using "htab_map_lookup_elem",
this commit introduces an alternative approach. Instead, it utilize
"bpf_map_copy_value" and apply a filter specifically for the hash table
with map_type.

Signed-off-by: Daniel T. Lee <danieltimlee@gmail.com>
Fixes: 7c4cd051add3 ("bpf: Fix syscall's stackmap lookup potential deadlock")
Link: https://lore.kernel.org/r/20230818090119.477441-8-danieltimlee@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
samples/bpf/tracex6.bpf.c

index 6ad82e6..9b23b47 100644 (file)
@@ -1,6 +1,8 @@
 #include "vmlinux.h"
 #include <linux/version.h>
 #include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+#include <bpf/bpf_core_read.h>
 
 struct {
        __uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
@@ -44,13 +46,24 @@ int bpf_prog1(struct pt_regs *ctx)
        return 0;
 }
 
-SEC("kprobe/htab_map_lookup_elem")
-int bpf_prog2(struct pt_regs *ctx)
+/*
+ * Since *_map_lookup_elem can't be expected to trigger bpf programs
+ * due to potential deadlocks (bpf_disable_instrumentation), this bpf
+ * program will be attached to bpf_map_copy_value (which is called
+ * from map_lookup_elem) and will only filter the hashtable type.
+ */
+SEC("kprobe/bpf_map_copy_value")
+int BPF_KPROBE(bpf_prog2, struct bpf_map *map)
 {
        u32 key = bpf_get_smp_processor_id();
        struct bpf_perf_event_value *val, buf;
+       enum bpf_map_type type;
        int error;
 
+       type = BPF_CORE_READ(map, map_type);
+       if (type != BPF_MAP_TYPE_HASH)
+               return 0;
+
        error = bpf_perf_event_read_value(&counters, key, &buf, sizeof(buf));
        if (error)
                return 0;