Merge pull request #836 from shodoco/lruhash

bcc: add lru_hash and lru_percpu_hash map types

src/cc/compat/linux/bpf.h

@@ -85,6 +85,8 @@ enum bpf_map_type {
 	BPF_MAP_TYPE_PERCPU_ARRAY,
 	BPF_MAP_TYPE_STACK_TRACE,
 	BPF_MAP_TYPE_CGROUP_ARRAY,
+	BPF_MAP_TYPE_LRU_HASH,
+	BPF_MAP_TYPE_LRU_PERCPU_HASH,
 };
 
 enum bpf_prog_type {
@@ -106,6 +108,13 @@ enum bpf_prog_type {
 #define BPF_EXIST	2 /* update existing element */
 
 #define BPF_F_NO_PREALLOC	(1U << 0)
+/* Instead of having one common LRU list in the
+ * BPF_MAP_TYPE_LRU_[PERCPU_]HASH map, use a percpu LRU list
+ * which can scale and perform better.
+ * Note, the LRU nodes (including free nodes) cannot be moved
+ * across different LRU lists.
+ */
+#define BPF_F_NO_COMMON_LRU	(1U << 1)
 
 union bpf_attr {
 	struct { /* anonymous struct used by BPF_MAP_CREATE command */

src/cc/compat/linux/virtual_bpf.h

@@ -86,6 +86,8 @@ enum bpf_map_type {
 	BPF_MAP_TYPE_PERCPU_ARRAY,
 	BPF_MAP_TYPE_STACK_TRACE,
 	BPF_MAP_TYPE_CGROUP_ARRAY,
+	BPF_MAP_TYPE_LRU_HASH,
+	BPF_MAP_TYPE_LRU_PERCPU_HASH,
 };
 
 enum bpf_prog_type {
@@ -96,6 +98,7 @@ enum bpf_prog_type {
 	BPF_PROG_TYPE_SCHED_ACT,
 	BPF_PROG_TYPE_TRACEPOINT,
 	BPF_PROG_TYPE_XDP,
+	BPF_PROG_TYPE_PERF_EVENT,
 };
 
 #define BPF_PSEUDO_MAP_FD	1
@@ -106,6 +109,13 @@ enum bpf_prog_type {
 #define BPF_EXIST	2 /* update existing element */
 
 #define BPF_F_NO_PREALLOC	(1U << 0)
+/* Instead of having one common LRU list in the
+ * BPF_MAP_TYPE_LRU_[PERCPU_]HASH map, use a percpu LRU list
+ * which can scale and perform better.
+ * Note, the LRU nodes (including free nodes) cannot be moved
+ * across different LRU lists.
+ */
+#define BPF_F_NO_COMMON_LRU	(1U << 1)
 
 union bpf_attr {
 	struct { /* anonymous struct used by BPF_MAP_CREATE command */

src/cc/frontends/clang/b_frontend_action.cc

@@ -639,6 +639,10 @@ bool BTypeVisitor::VisitVarDecl(VarDecl *Decl) {
       map_type = BPF_MAP_TYPE_PERCPU_HASH;
     } else if (A->getName() == "maps/percpu_array") {
       map_type = BPF_MAP_TYPE_PERCPU_ARRAY;
+    } else if (A->getName() == "maps/lru_hash") {
+      map_type = BPF_MAP_TYPE_LRU_HASH;
+    } else if (A->getName() == "maps/lru_percpu_hash") {
+      map_type = BPF_MAP_TYPE_LRU_PERCPU_HASH;
     } else if (A->getName() == "maps/histogram") {
       if (table.key_desc == "\"int\"")
         map_type = BPF_MAP_TYPE_ARRAY;

src/lua/bcc/table.lua

@@ -26,6 +26,9 @@ BaseTable.static.BPF_MAP_TYPE_PERF_EVENT_ARRAY = 4
 BaseTable.static.BPF_MAP_TYPE_PERCPU_HASH = 5
 BaseTable.static.BPF_MAP_TYPE_PERCPU_ARRAY = 6
 BaseTable.static.BPF_MAP_TYPE_STACK_TRACE = 7
+BaseTable.static.BPF_MAP_TYPE_CGROUP_ARRAY = 8
+BaseTable.static.BPF_MAP_TYPE_LRU_HASH = 9
+BaseTable.static.BPF_MAP_TYPE_LRU_PERCPU_HASH = 10
 
 function BaseTable:initialize(t_type, bpf, map_id, map_fd, key_type, leaf_type)
   assert(t_type == libbcc.bpf_table_type_id(bpf.module, map_id))

src/lua/bpf/cdef.lua

@@ -138,9 +138,13 @@ else
 		S.c.BPF_MAP.PERCPU_ARRAY = 6
 		S.c.BPF_MAP.STACK_TRACE  = 7
 		S.c.BPF_MAP.CGROUP_ARRAY = 8
+		S.c.BPF_MAP.LRU_HASH     = 9
+		S.c.BPF_MAP.LRU_PERCPU_HASH = 10
 	end
 	if not S.c.BPF_PROG.TRACEPOINT then
 		S.c.BPF_PROG.TRACEPOINT  = 5
+		S.c.BPF_PROG.XDP         = 6
+		S.c.BPF_PROG.PERF_EVENT  = 7
 	end
 end
 

src/python/bcc/table.py

@@ -28,6 +28,9 @@ BPF_MAP_TYPE_PERF_EVENT_ARRAY = 4
 BPF_MAP_TYPE_PERCPU_HASH = 5
 BPF_MAP_TYPE_PERCPU_ARRAY = 6
 BPF_MAP_TYPE_STACK_TRACE = 7
+BPF_MAP_TYPE_CGROUP_ARRAY = 8
+BPF_MAP_TYPE_LRU_HASH = 9
+BPF_MAP_TYPE_LRU_PERCPU_HASH = 10
 
 stars_max = 40
 
@@ -97,6 +100,10 @@ def Table(bpf, map_id, map_fd, keytype, leaftype, **kwargs):
         t = PerCpuArray(bpf, map_id, map_fd, keytype, leaftype, **kwargs)
     elif ttype == BPF_MAP_TYPE_STACK_TRACE:
         t = StackTrace(bpf, map_id, map_fd, keytype, leaftype)
+    elif ttype == BPF_MAP_TYPE_LRU_HASH:
+        t = LruHash(bpf, map_id, map_fd, keytype, leaftype)
+    elif ttype == BPF_MAP_TYPE_LRU_PERCPU_HASH:
+        t = LruPerCpuHash(bpf, map_id, map_fd, keytype, leaftype)
     if t == None:
         raise Exception("Unknown table type %d" % ttype)
     return t
@@ -299,6 +306,9 @@ class HashTable(TableBase):
         if res < 0:
             raise KeyError
 
+class LruHash(HashTable):
+    def __init__(self, *args, **kwargs):
+        super(LruHash, self).__init__(*args, **kwargs)
 
 class ArrayBase(TableBase):
     def __init__(self, *args, **kwargs):
@@ -526,6 +536,10 @@ class PerCpuHash(HashTable):
         result.value/=self.total_cpu
         return result
 
+class LruPerCpuHash(PerCpuHash):
+    def __init__(self, *args, **kwargs):
+        super(LruPerCpuHash, self).__init__(*args, **kwargs)
+
 class PerCpuArray(ArrayBase):
     def __init__(self, *args, **kwargs):
         self.reducer = kwargs.pop("reducer", None)

tests/python/test_lru.py

+#!/usr/bin/env python
+# Copyright (c) PLUMgrid, Inc.
+# Licensed under the Apache License, Version 2.0 (the "License")
+
+import ctypes as ct
+import os
+import unittest
+from bcc import BPF
+import multiprocessing
+
+class TestLru(unittest.TestCase):
+    def test_lru_hash(self):
+        b = BPF(text="""BPF_TABLE("lru_hash", int, u64, lru, 1024);""")
+        t = b["lru"]
+        for i in range(1, 1032):
+            t[ct.c_int(i)] = ct.c_ulonglong(i)
+        for i, v in t.items():
+            self.assertEqual(v.value, i.value)
+        # BPF_MAP_TYPE_LRU_HASH eviction happens in batch and we expect less
+        # items than specified size.
+        self.assertLess(len(t), 1024);
+
+    def test_lru_percpu_hash(self):
+        test_prog1 = """
+        BPF_TABLE("lru_percpu_hash", u32, u32, stats, 1);
+        int hello_world(void *ctx) {
+            u32 key=0;
+            u32 value = 0, *val;
+            val = stats.lookup_or_init(&key, &value);
+            *val += 1;
+            return 0;
+        }
+        """
+        b = BPF(text=test_prog1)
+        stats_map = b.get_table("stats")
+        b.attach_kprobe(event="sys_clone", fn_name="hello_world")
+        ini = stats_map.Leaf()
+        for i in range(0, multiprocessing.cpu_count()):
+            ini[i] = 0
+        # First initialize with key 1
+        stats_map[ stats_map.Key(1) ] = ini
+        # Then initialize with key 0
+        stats_map[ stats_map.Key(0) ] = ini
+        # Key 1 should have been evicted
+        with self.assertRaises(KeyError):
+            val = stats_map[ stats_map.Key(1) ]
+        f = os.popen("hostname")
+        f.close()
+        self.assertEqual(len(stats_map),1)
+        val = stats_map[ stats_map.Key(0) ]
+        sum = stats_map.sum(stats_map.Key(0))
+        avg = stats_map.average(stats_map.Key(0))
+        max = stats_map.max(stats_map.Key(0))
+        self.assertGreater(sum.value, 0L)
+        self.assertGreater(max.value, 0L)
+        b.detach_kprobe("sys_clone")
+
+if __name__ == "__main__":
+    unittest.main()

tests/python/test_percpu.py

@@ -20,7 +20,6 @@ class TestPercpu(unittest.TestCase):
             return 0;
         }
         """
-        self.addCleanup(self.cleanup)
         bpf_code = BPF(text=test_prog1)
         stats_map = bpf_code.get_table("stats")
         bpf_code.attach_kprobe(event="sys_clone", fn_name="hello_world")
@@ -37,6 +36,7 @@ class TestPercpu(unittest.TestCase):
         max = stats_map.max(stats_map.Key(0))
         self.assertGreater(sum.value, 0L)
         self.assertGreater(max.value, 0L)
+        bpf_code.detach_kprobe("sys_clone")
 
     def test_u32(self):
         test_prog1 = """
@@ -49,7 +49,6 @@ class TestPercpu(unittest.TestCase):
             return 0;
         }
         """
-        self.addCleanup(self.cleanup)
         bpf_code = BPF(text=test_prog1)
         stats_map = bpf_code.get_table("stats")
         bpf_code.attach_kprobe(event="sys_clone", fn_name="hello_world")
@@ -66,6 +65,7 @@ class TestPercpu(unittest.TestCase):
         max = stats_map.max(stats_map.Key(0))
         self.assertGreater(sum.value, 0L)
         self.assertGreater(max.value, 0L)
+        bpf_code.detach_kprobe("sys_clone")
 
     def test_struct_custom_func(self):
         test_prog2 = """
@@ -83,7 +83,6 @@ class TestPercpu(unittest.TestCase):
             return 0;
         }
         """
-        self.addCleanup(self.cleanup)
         bpf_code = BPF(text=test_prog2)
         stats_map = bpf_code.get_table("stats",
                 reducer=lambda x,y: stats_map.sLeaf(x.c1+y.c1))
@@ -97,9 +96,7 @@ class TestPercpu(unittest.TestCase):
         self.assertEqual(len(stats_map),1)
         k = stats_map[ stats_map.Key(0) ]
         self.assertGreater(k.c1, 0L)
-
-    def cleanup(self):
-        BPF.detach_kprobe("sys_clone")
+        bpf_code.detach_kprobe("sys_clone")
 
 
 if __name__ == "__main__":