[CWS] add kernel bpf filter for raw packet (#30288)

LICENSE-3rdparty.csv

@@ -690,6 +690,7 @@ core,github.com/cilium/ebpf/perf,MIT,"Copyright (c) 2017 Nathan Sweet | Copyrigh
 core,github.com/cilium/ebpf/ringbuf,MIT,"Copyright (c) 2017 Nathan Sweet | Copyright (c) 2018, 2019 Cloudflare | Copyright (c) 2019 Authors of Cilium"
 core,github.com/cilium/ebpf/rlimit,MIT,"Copyright (c) 2017 Nathan Sweet | Copyright (c) 2018, 2019 Cloudflare | Copyright (c) 2019 Authors of Cilium"
 core,github.com/clbanning/mxj,MIT,Copyright (c) 2012-2016 Charles Banning <[email protected]>.  All rights reserved | Copyright 2009 The Go Authors. All rights reserved
+core,github.com/cloudflare/cbpfc,BSD-3-Clause,"Copyright (c) 2019, Cloudflare. All rights reserved"
 core,github.com/cloudflare/circl/dh/x25519,BSD-3-Clause,Copyright (c) 2009 The Go Authors. All rights reserved | Copyright (c) 2019 Cloudflare. All rights reserved
 core,github.com/cloudflare/circl/dh/x448,BSD-3-Clause,Copyright (c) 2009 The Go Authors. All rights reserved | Copyright (c) 2019 Cloudflare. All rights reserved
 core,github.com/cloudflare/circl/ecc/goldilocks,BSD-3-Clause,Copyright (c) 2009 The Go Authors. All rights reserved | Copyright (c) 2019 Cloudflare. All rights reserved

go.mod

@@ -510,7 +510,7 @@ require (
 	github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
 	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/rs/cors v1.11.1 // indirect
-	github.com/safchain/baloum v0.0.0-20221229104256-b1fc8f70a86b
+	github.com/safchain/baloum v0.0.0-20241120122234-f22c9bd19f3b
 	github.com/saracen/walker v0.1.3 // indirect
 	github.com/sassoftware/go-rpmutils v0.3.0 // indirect
 	github.com/secure-systems-lab/go-securesystemslib v0.8.0 // indirect
@@ -604,6 +604,7 @@ require (
 	github.com/DataDog/datadog-agent/pkg/util/defaultpaths v0.0.0-00010101000000-000000000000
 	github.com/DataDog/datadog-agent/pkg/util/utilizationtracker v0.0.0
 	github.com/NVIDIA/go-nvml v0.12.4-0
+	github.com/cloudflare/cbpfc v0.0.0-20240920015331-ff978e94500b
 	github.com/containerd/containerd/api v1.8.0
 	github.com/containerd/errdefs v1.0.0
 	github.com/distribution/reference v0.6.0

pkg/security/ebpf/c/include/constants/custom.h

@@ -61,12 +61,14 @@ enum DENTRY_ERPC_RESOLUTION_CODE
 
 enum TC_TAIL_CALL_KEYS
 {
-    UNKNOWN,
-    DNS_REQUEST,
+    DNS_REQUEST = 1,
     DNS_REQUEST_PARSER,
     IMDS_REQUEST,
-    RAW_PACKET,
+};
+
+enum TC_RAWPACKET_KEYS {
     RAW_PACKET_FILTER,
+    // reserved keys for raw packet filter tail calls
 };
 
 #define DNS_MAX_LENGTH 256

pkg/security/ebpf/c/include/helpers/network.h

@@ -82,10 +82,6 @@ __attribute__((always_inline)) void fill_network_context(struct network_context_
     fill_network_device_context(&net_ctx->device, skb, pkt);
 }
 
-__attribute__((always_inline)) void tail_call_to_classifier(struct __sk_buff *skb, int classifier_id) {
-    bpf_tail_call_compat(skb, &classifier_router, classifier_id);
-}
-
 __attribute__((always_inline)) void parse_tuple(struct nf_conntrack_tuple *tuple, struct flow_t *flow) {
     flow->sport = tuple->src.u.all;
     flow->dport = tuple->dst.u.all;
@@ -94,7 +90,6 @@ __attribute__((always_inline)) void parse_tuple(struct nf_conntrack_tuple *tuple
     bpf_probe_read(&flow->daddr, sizeof(flow->daddr), &tuple->dst.u3.all);
 }
 
-
 __attribute__((always_inline)) struct packet_t * parse_packet(struct __sk_buff *skb, int direction) {
     struct cursor c = {};
     tc_cursor_init(&c, skb);

pkg/security/ebpf/c/include/hooks/network/dns.h

@@ -80,7 +80,7 @@ int classifier_dns_request(struct __sk_buff *skb) {
     evt->id = htons(header.id);
 
     // tail call to the dns request parser
-    tail_call_to_classifier(skb, DNS_REQUEST_PARSER);
+    bpf_tail_call_compat(skb, &classifier_router, DNS_REQUEST_PARSER);
 
     // tail call failed, ignore packet
     return ACT_OK;
@@ -116,7 +116,7 @@ int classifier_dns_request_parser(struct __sk_buff *skb) {
     send_event_with_size_ptr(skb, EVENT_DNS, evt, offsetof(struct dns_event_t, name) + qname_length);
 
     if (!is_dns_request_parsing_done(skb, pkt)) {
-        tail_call_to_classifier(skb, DNS_REQUEST_PARSER);
+        bpf_tail_call_compat(skb, &classifier_router, DNS_REQUEST_PARSER);
     }
 
     return ACT_OK;

pkg/security/ebpf/c/include/hooks/network/raw.h

@@ -4,61 +4,44 @@
 #include "helpers/network.h"
 #include "perf_ring.h"
 
-__attribute__((always_inline)) struct raw_packet_t *get_raw_packet_event() {
+__attribute__((always_inline)) struct raw_packet_event_t *get_raw_packet_event() {
     u32 key = 0;
-    return bpf_map_lookup_elem(&raw_packets, &key);
+    return bpf_map_lookup_elem(&raw_packet_event, &key);
 }
 
-SEC("classifier/raw_packet")
-int classifier_raw_packet(struct __sk_buff *skb) {
+SEC("classifier/raw_packet_sender")
+int classifier_raw_packet_sender(struct __sk_buff *skb) {
     struct packet_t *pkt = get_packet();
     if (pkt == NULL) {
         // should never happen
         return ACT_OK;
     }
 
-    struct raw_packet_t *evt = get_raw_packet_event();
-    if ((evt == NULL) || (skb == NULL)) {
+    struct raw_packet_event_t *evt = get_raw_packet_event();
+    if (evt == NULL || skb == NULL || evt->len == 0) {
         // should never happen
         return ACT_OK;
     }
 
-    bpf_skb_pull_data(skb, 0);
+    // process context
+    fill_network_process_context(&evt->process, pkt);
 
-    u32 len = *(u32 *)(skb + offsetof(struct __sk_buff, len));
-    if (len > sizeof(evt->data)) {
-        len = sizeof(evt->data);
+    struct proc_cache_t *entry = get_proc_cache(evt->process.pid);
+    if (entry == NULL) {
+        evt->container.container_id[0] = 0;
+    } else {
+        copy_container_id_no_tracing(entry->container.container_id, &evt->container.container_id);
     }
 
-    // NOTE(safchain) inline asm because clang isn't generating the proper instructions for :
-    // if (len == 0) return ACT_OK;
-    /*asm ("r4 = %[len]\n"
-         "if r4 > 0 goto + 2\n"
-         "r0 = 0\n"
-         "exit\n" :: [len]"r"((u64)len));*/
-
-    if (len > 1) {
-        if (bpf_skb_load_bytes(skb, 0, evt->data, len) < 0) {
-            return ACT_OK;
-        }
-        evt->len = skb->len;
-
-        // process context
-        fill_network_process_context(&evt->process, pkt);
-
-        struct proc_cache_t *entry = get_proc_cache(evt->process.pid);
-        if (entry == NULL) {
-            evt->container.container_id[0] = 0;
-        } else {
-            copy_container_id_no_tracing(entry->container.container_id, &evt->container.container_id);
-        }
+    fill_network_device_context(&evt->device, skb, pkt);
 
-        fill_network_device_context(&evt->device, skb, pkt);
-
-        u32 size = offsetof(struct raw_packet_t, data) + len;
-        send_event_with_size_ptr(skb, EVENT_RAW_PACKET, evt, size);
+    u32 len = evt->len;
+    if (len > sizeof(evt->data)) {
+        len = sizeof(evt->data);
     }
 
+    send_event_with_size_ptr(skb, EVENT_RAW_PACKET, evt, offsetof(struct raw_packet_event_t, data) + len);
+
     return ACT_OK;
 }
 

pkg/security/ebpf/c/include/hooks/network/router.h

@@ -9,14 +9,14 @@ __attribute__((always_inline)) int route_pkt(struct __sk_buff *skb, struct packe
     // route DNS requests
     if (is_event_enabled(EVENT_DNS)) {
         if (pkt->l4_protocol == IPPROTO_UDP && pkt->translated_ns_flow.flow.dport == htons(53)) {
-            tail_call_to_classifier(skb, DNS_REQUEST);
+            bpf_tail_call_compat(skb, &classifier_router, DNS_REQUEST);
         }
     }
 
     // route IMDS requests
     if (is_event_enabled(EVENT_IMDS)) {
         if (pkt->l4_protocol == IPPROTO_TCP && ((pkt->ns_flow.flow.saddr[0] & 0xFFFFFFFF) == get_imds_ip() || (pkt->ns_flow.flow.daddr[0] & 0xFFFFFFFF) == get_imds_ip())) {
-            tail_call_to_classifier(skb, IMDS_REQUEST);
+            bpf_tail_call_compat(skb, &classifier_router, IMDS_REQUEST);
         }
     }
 

pkg/security/ebpf/c/include/hooks/network/tc.h

@@ -26,14 +26,45 @@ int classifier_egress(struct __sk_buff *skb) {
     return route_pkt(skb, pkt, EGRESS);
 };
 
+__attribute__((always_inline)) int prepare_raw_packet_event(struct __sk_buff *skb) {
+    struct raw_packet_event_t *evt = get_raw_packet_event();
+    if (evt == NULL) {
+        // should never happen
+        return ACT_OK;
+    }
+
+    bpf_skb_pull_data(skb, 0);
+
+    u32 len = *(u32 *)(skb + offsetof(struct __sk_buff, len));
+    if (len > sizeof(evt->data)) {
+        len = sizeof(evt->data);
+    }
+
+    if (len > 1) {
+        if (bpf_skb_load_bytes(skb, 0, evt->data, len) < 0) {
+            return ACT_OK;
+        }
+        evt->len = skb->len;
+    } else {
+        evt->len = 0;
+    }
+
+    return ACT_OK;
+}
+
+
 SEC("classifier/ingress")
 int classifier_raw_packet_ingress(struct __sk_buff *skb) {
     struct packet_t *pkt = parse_packet(skb, INGRESS);
     if (!pkt) {
         return ACT_OK;
     }
 
-    tail_call_to_classifier(skb, RAW_PACKET_FILTER);
+    if (prepare_raw_packet_event(skb) != ACT_OK) {
+        return ACT_OK;
+    }
+
+    bpf_tail_call_compat(skb, &raw_packet_classifier_router, RAW_PACKET_FILTER);
 
     return ACT_OK;
 };
@@ -45,7 +76,11 @@ int classifier_raw_packet_egress(struct __sk_buff *skb) {
         return ACT_OK;
     }
 
-    tail_call_to_classifier(skb, RAW_PACKET_FILTER);
+    if (prepare_raw_packet_event(skb) != ACT_OK) {
+        return ACT_OK;
+    }
+
+    bpf_tail_call_compat(skb, &raw_packet_classifier_router, RAW_PACKET_FILTER);
 
     return ACT_OK;
 };

pkg/security/ebpf/c/include/maps.h

@@ -88,14 +88,15 @@ BPF_PERCPU_ARRAY_MAP(packets, struct packet_t, 1)
 BPF_PERCPU_ARRAY_MAP(selinux_write_buffer, struct selinux_write_buffer_t, 1)
 BPF_PERCPU_ARRAY_MAP(is_new_kthread, u32, 1)
 BPF_PERCPU_ARRAY_MAP(syscalls_stats, struct syscalls_stats_t, EVENT_MAX)
-BPF_PERCPU_ARRAY_MAP(raw_packets, struct raw_packet_t, 1)
+BPF_PERCPU_ARRAY_MAP(raw_packet_event, struct raw_packet_event_t, 1)
 
 BPF_PROG_ARRAY(args_envs_progs, 3)
 BPF_PROG_ARRAY(dentry_resolver_kprobe_or_fentry_callbacks, EVENT_MAX)
 BPF_PROG_ARRAY(dentry_resolver_tracepoint_callbacks, EVENT_MAX)
 BPF_PROG_ARRAY(dentry_resolver_kprobe_or_fentry_progs, 6)
 BPF_PROG_ARRAY(dentry_resolver_tracepoint_progs, 3)
-BPF_PROG_ARRAY(classifier_router, 100)
+BPF_PROG_ARRAY(classifier_router, 10)
 BPF_PROG_ARRAY(sys_exit_progs, 64)
+BPF_PROG_ARRAY(raw_packet_classifier_router, 32)
 
 #endif

pkg/security/ebpf/c/include/structs/network.h

@@ -83,7 +83,7 @@ struct network_context_t {
     u16 l4_protocol;
 };
 
-struct raw_packet_t {
+struct raw_packet_event_t {
     struct kevent_t event;
     struct process_context_t process;
     struct span_context_t span;

pkg/security/ebpf/c/include/tests/activity_dump_ratelimiter_test.h

@@ -39,7 +39,7 @@ int test_ad_ratelimiter_basic() {
         assert_zero(activity_dump_rate_limiter_allow(&config, cookie, now, 0),
             "event allowed which should not be");
     }
-    return 0;
+    return 1;
 }
 
 SEC("test/ad_ratelimiter_basic_half")
@@ -73,7 +73,7 @@ int test_ad_ratelimiter_basic_half() {
         assert_zero(activity_dump_rate_limiter_allow(&config, cookie, now, 0),
             "event allowed which should not be");
     }
-    return 0;
+    return 1;
 }
 
 __attribute__((always_inline)) int test_ad_ratelimiter_variable_droprate(int algo) {
@@ -106,7 +106,7 @@ __attribute__((always_inline)) int test_ad_ratelimiter_variable_droprate(int alg
         assert_greater_than(total_allowed, AD_RL_TEST_RATE * 3 / 4, "nope");
         assert_lesser_than(total_allowed, AD_RL_TEST_RATE / 10, "nope");
     }
-    return 0;
+    return 1;
 }
 
 SEC("test/ad_ratelimiter_decreasing_droprate")

pkg/security/ebpf/c/include/tests/baloum.h

@@ -15,6 +15,7 @@ static int (*baloum_call)(struct baloum_ctx *ctx, const char *section) = (void *
 static int (*baloum_strcmp)(const char *s1, const char *s2) = (void *)0xfffd;
 static int (*baloum_memcmp)(const void *b1, const void *b2, __u32 size) = (void *)0xfffc;
 static int (*baloum_sleep)(__u64 ns) = (void *)0xfffb;
+static int (*baloum_memcpy)(const void *b1, const void *b2, __u32 size) = (void *)0xfffa;
 
 #define assert_memcmp(b1, b2, s, msg)                                \
     if (baloum_memcmp(b1, b2, s) != 0) {                             \

pkg/security/ebpf/c/include/tests/discarders_test.h

@@ -53,7 +53,7 @@ int test_discarders_event_mask() {
     ret = _is_discarded_by_inode(EVENT_CHMOD, mount_id, inode);
     assert_not_zero(ret, "inode should be discarded");
 
-    return 0;
+    return 1;
 }
 
 SEC("test/discarders_retention")
@@ -93,7 +93,7 @@ int test_discarders_retention() {
     ret = _is_discarded_by_inode(EVENT_OPEN, mount_id, inode);
     assert_not_zero(ret, "inode should be discarded");
 
-    return 0;
+    return 1;
 }
 
 SEC("test/discarders_revision")
@@ -142,7 +142,7 @@ int test_discarders_revision() {
     ret = _is_discarded_by_inode(EVENT_OPEN, mount_id1, inode1);
     assert_not_zero(ret, "inode should be discarded");
 
-    return 0;
+    return 1;
 }
 
 SEC("test/discarders_mount_revision")
@@ -183,7 +183,7 @@ int test_discarders_mount_revision() {
     ret = _is_discarded_by_inode(EVENT_OPEN, mount_id1, inode1);
     assert_not_zero(ret, "inode should be discarded");
 
-    return 0;
+    return 1;
 }
 
 #endif

pkg/security/ebpf/c/include/tests/raw_packet_test.h

@@ -0,0 +1,30 @@
+#ifndef _RAW_PACKET_TEST_H
+#define _RAW_PACKET_TEST_H
+
+#include "helpers/network.h"
+#include "baloum.h"
+
+SEC("test/raw_packet_tail_calls")
+int raw_packet_tail_calls(struct __sk_buff *skb) {
+    struct raw_packet_event_t *evt = get_raw_packet_event();
+    assert_not_null(evt, "unable to get raw packet event")
+
+    // tcp dst port 5555 and tcp[tcpflags] == tcp-syn
+    unsigned char data[] = {
+        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+        0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x45, 0x10,
+        0x00, 0x30, 0xf4, 0xa2, 0x40, 0x00, 0x40, 0x06,
+        0x48, 0x13, 0x7f, 0x00, 0x00, 0x01, 0x7f, 0x00,
+        0x00, 0x01, 0xa2, 0x36, 0x15, 0xb3, 0x1c, 0x5b,
+        0x89, 0x33, 0x00, 0x00, 0x00, 0x00, 0x70, 0x02,
+        0xff, 0xd7, 0xfe, 0x24, 0x00, 0x00, 0x02, 0x04,
+        0xff, 0xd7, 0x01, 0x03, 0x03, 0x07
+    };
+    baloum_memcpy(evt->data, data, sizeof(data));
+
+    bpf_tail_call_compat(skb, &raw_packet_classifier_router, RAW_PACKET_FILTER);
+
+    return 1;
+}
+
+#endif

pkg/security/ebpf/c/include/tests/tests.h

@@ -3,5 +3,6 @@
 
 #include "discarders_test.h"
 #include "activity_dump_ratelimiter_test.h"
+#include "raw_packet_test.h"
 
 #endif