[dpdk-dev] ebpf issues with secondary process.

[Stephen Hemminger]

Trying to use the existing BPF code in BPF for packet filter is possible.
But one road block is that capture wants to allow a secondary process to
insert a packet filter. The current code in bpf_load() won't work for that.
The BPF program needs to be in shared area to allow secondary process
to work.

Why does the code use mmap() as an allocator instead of rte_malloc?

Something like the diff below (untested). JIT would have the same problem
but you would have issues with the mprotect stuff with huge pages.

I also noticed that bpf code is not using rte_memcpy and it has unnecessary
casts to void *.


diff --git a/lib/librte_bpf/bpf.c b/lib/librte_bpf/bpf.c
index 7e1879ffa5b5..d6995bbf0ba9 100644
--- a/lib/librte_bpf/bpf.c
+++ b/lib/librte_bpf/bpf.c
@@ -22,7 +22,7 @@ rte_bpf_destroy(struct rte_bpf *bpf)
 	if (bpf != NULL) {
 		if (bpf->jit.func != NULL)
 			munmap(bpf->jit.func, bpf->jit.sz);
-		munmap(bpf, bpf->sz);
+		rte_free(bpf);
 	}
 }
 
diff --git a/lib/librte_bpf/bpf_load.c b/lib/librte_bpf/bpf_load.c
index 2a3b901d74c3..9a8e438a8963 100644
--- a/lib/librte_bpf/bpf_load.c
+++ b/lib/librte_bpf/bpf_load.c
@@ -32,9 +32,8 @@ bpf_load(const struct rte_bpf_prm *prm)
 	bsz = sizeof(bpf[0]);
 	sz = insz + xsz + bsz;
 
-	buf = mmap(NULL, sz, PROT_READ | PROT_WRITE,
-		MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
-	if (buf == MAP_FAILED)
+	buf = rte_malloc("bpf", sz, 0);
+	if (buf == NULL)
 		return NULL;
 
 	bpf = (void *)buf;

-- 
2.20.1