Skip to contentSkip to conversation

bpf: getsockopt and setsockopt hooks

PATCH Superseded Stanislav Fomichev 5 participants 15 messages Jun 10, 2019 – Jun 14, 2019

Subsystems (11)

v5 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v5 cover bpf: getsockopt and setsockopt hooks

This series implements two new per-cgroup hooks: getsockopt and setsockopt along with a new sockopt program type. The idea is pretty similar to recently introduced cgroup sysctl hooks, but implementation is simpler (no need to convert to/from strings).

What this can be applied to:
* move business logic of what tos/priority/etc can be set by containers (either pass or reject)
* handle existing options (or introduce new ones) differently by propagating some information in cgroup/socket local storage

Compared to a simple syscall/{g,s}etsockopt tracepoint, those hooks are context aware. Meaning, they can access underlying socket and use cgroup and socket local storage.

Stanislav Fomichev (8):
bpf: implement getsockopt and setsockopt hooks
bpf: sync bpf.h to tools/
libbpf: support sockopt hooks
selftests/bpf: test sockopt section name selftests/bpf: add sockopt test
selftests/bpf: add sockopt test that exercises sk helpers
bpf: add sockopt documentation
bpftool: support cgroup sockopt

 Documentation/bpf/index.rst                   |   1 +
 Documentation/bpf/prog_cgroup_sockopt.rst     |  39 +
 include/linux/bpf-cgroup.h                    |  29 +
 include/linux/bpf.h                           |  45 +
 include/linux/bpf_types.h                     |   1 +
 include/linux/filter.h                        |  13 +
 include/uapi/linux/bpf.h                      |  13 +
 kernel/bpf/cgroup.c                           | 260 ++++++
 kernel/bpf/core.c                             |   9 +
 kernel/bpf/syscall.c                          |  19 +
 kernel/bpf/verifier.c                         |  15 +
 net/core/filter.c                             |   2 +-
 net/socket.c                                  |  18 +
 .../bpftool/Documentation/bpftool-cgroup.rst  |   7 +-
 .../bpftool/Documentation/bpftool-prog.rst    |   2 +-
 tools/bpf/bpftool/bash-completion/bpftool     |   8 +-
 tools/bpf/bpftool/cgroup.c                    |   5 +-
 tools/bpf/bpftool/main.h                      |   1 +
 tools/bpf/bpftool/prog.c                      |   3 +-
 tools/include/uapi/linux/bpf.h                |  14 +
 tools/lib/bpf/libbpf.c                        |   5 +
 tools/lib/bpf/libbpf_probes.c                 |   1 +
 tools/testing/selftests/bpf/.gitignore        |   2 +
 tools/testing/selftests/bpf/Makefile          |   4 +-
 .../testing/selftests/bpf/progs/sockopt_sk.c  |  67 ++
 .../selftests/bpf/test_section_names.c        |  10 +
 tools/testing/selftests/bpf/test_sockopt.c    | 773 ++++++++++++++++++
 tools/testing/selftests/bpf/test_sockopt_sk.c | 156 ++++
 28 files changed, 1512 insertions(+), 10 deletions(-)
 create mode 100644 Documentation/bpf/prog_cgroup_sockopt.rst
 create mode 100644 tools/testing/selftests/bpf/progs/sockopt_sk.c
 create mode 100644 tools/testing/selftests/bpf/test_sockopt.c
 create mode 100644 tools/testing/selftests/bpf/test_sockopt_sk.c
Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
#1 lore raw

PATCH v5 1/8 bpf: implement getsockopt and setsockopt hooks

Implement new BPF_PROG_TYPE_CGROUP_SOCKOPT program type and BPF_CGROUP_{G,S}ETSOCKOPT cgroup hooks.

BPF_CGROUP_SETSOCKOPT get a read-only view of the setsockopt arguments. BPF_CGROUP_GETSOCKOPT can modify the supplied buffer.
Both of them reuse existing PTR_TO_PACKET{,_END} infrastructure.

The buffer memory is pre-allocated (because I don't think there is a precedent for working with __user memory from bpf). This might be slow to do for each {s,g}etsockopt call, that's why I've added __cgroup_bpf_prog_array_is_empty that exits early if there is nothing attached to a cgroup. Note, however, that there is a race between __cgroup_bpf_prog_array_is_empty and BPF_PROG_RUN_ARRAY where cgroup program layout might have changed; this should not be a problem because in general there is a race between multiple calls to {s,g}etsocktop and user adding/removing bpf progs from a cgroup.

The return code of the BPF program is handled as follows:
* 0: EPERM
* 1: success, execute kernel {s,g}etsockopt path after BPF prog exits
* 2: success, do _not_ execute kernel {s,g}etsockopt path after BPF prog exits

v5:
* skip copy_to_user() and put_user() when ret == 0 (Martin Lau)

v4:
* don't export bpf_sk_fullsock helper (Martin Lau)
* size != sizeof(__u64) for uapi pointers (Martin Lau)
* offsetof instead of bpf_ctx_range when checking ctx access (Martin Lau)

v3:
* typos in BPF_PROG_CGROUP_SOCKOPT_RUN_ARRAY comments (Andrii Nakryiko)
* reverse christmas tree in BPF_PROG_CGROUP_SOCKOPT_RUN_ARRAY (Andrii Nakryiko)
* use __bpf_md_ptr instead of __u32 for optval{,_end} (Martin Lau)
* use BPF_FIELD_SIZEOF() for consistency (Martin Lau)
* new CG_SOCKOPT_ACCESS macro to wrap repeated parts

v2:
* moved bpf_sockopt_kern fields around to remove a hole (Martin Lau)
* aligned bpf_sockopt_kern->buf to 8 bytes (Martin Lau)
* bpf_prog_array_is_empty instead of bpf_prog_array_length (Martin Lau)
* added [0,2] return code check to verifier (Martin Lau)
* dropped unused buf[64] from the stack (Martin Lau)
* use PTR_TO_SOCKET for bpf_sockopt->sk (Martin Lau)
* dropped bpf_target_off from ctx rewrites (Martin Lau)
* use return code for kernel bypass (Martin Lau & Andrii Nakryiko)

---
 include/linux/bpf-cgroup.h |  29 +++++
 include/linux/bpf.h        |  45 +++++++
 include/linux/bpf_types.h  |   1 +
 include/linux/filter.h     |  13 ++
 include/uapi/linux/bpf.h   |  13 ++
 kernel/bpf/cgroup.c        | 260 +++++++++++++++++++++++++++++++++++++
 kernel/bpf/core.c          |   9 ++
 kernel/bpf/syscall.c       |  19 +++
 kernel/bpf/verifier.c      |  15 +++
 net/core/filter.c          |   2 +-
 net/socket.c               |  18 +++
 11 files changed, 423 insertions(+), 1 deletion(-)
include/linux/bpf-cgroup.h +29 -0
index b631ee75762d..406f1ba82531 100644
@@ -124,6 +124,13 @@ int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
 				   loff_t *ppos, void **new_buf,
 				   enum bpf_attach_type type);
 
+int __cgroup_bpf_run_filter_setsockopt(struct sock *sock, int level,
+				       int optname, char __user *optval,
+				       unsigned int optlen);
+int __cgroup_bpf_run_filter_getsockopt(struct sock *sock, int level,
+				       int optname, char __user *optval,
+				       int __user *optlen);
+
 static inline enum bpf_cgroup_storage_type cgroup_storage_type(
 	struct bpf_map *map)
 {
@@ -280,6 +287,26 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key,
 	__ret;								       \
 })
 
+#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen)   \
+({									       \
+	int __ret = 0;							       \
+	if (cgroup_bpf_enabled)						       \
+		__ret = __cgroup_bpf_run_filter_setsockopt(sock, level,	       \
+							   optname, optval,    \
+							   optlen);	       \
+	__ret;								       \
+})
+
+#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, optlen)   \
+({									       \
+	int __ret = 0;							       \
+	if (cgroup_bpf_enabled)						       \
+		__ret = __cgroup_bpf_run_filter_getsockopt(sock, level,	       \
+							   optname, optval,    \
+							   optlen);	       \
+	__ret;								       \
+})
+
 int cgroup_bpf_prog_attach(const union bpf_attr *attr,
 			   enum bpf_prog_type ptype, struct bpf_prog *prog);
 int cgroup_bpf_prog_detach(const union bpf_attr *attr,
@@ -349,6 +376,8 @@ static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
 #define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) ({ 0; })
 #define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type,major,minor,access) ({ 0; })
 #define BPF_CGROUP_RUN_PROG_SYSCTL(head,table,write,buf,count,pos,nbuf) ({ 0; })
+#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, optlen) ({ 0; })
+#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen) ({ 0; })
 
 #define for_each_cgroup_storage_type(stype) for (; false; )
 
include/linux/bpf.h +45 -0

63 lines not on this page · raw message

include/linux/bpf_types.h +1 -0
index 5a9975678d6f..eec5aeeeaf92 100644
@@ -30,6 +30,7 @@ BPF_PROG_TYPE(BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE, raw_tracepoint_writable)
 #ifdef CONFIG_CGROUP_BPF
 BPF_PROG_TYPE(BPF_PROG_TYPE_CGROUP_DEVICE, cg_dev)
 BPF_PROG_TYPE(BPF_PROG_TYPE_CGROUP_SYSCTL, cg_sysctl)
+BPF_PROG_TYPE(BPF_PROG_TYPE_CGROUP_SOCKOPT, cg_sockopt)
 #endif
 #ifdef CONFIG_BPF_LIRC_MODE2
 BPF_PROG_TYPE(BPF_PROG_TYPE_LIRC_MODE2, lirc_mode2)
include/linux/filter.h +13 -0
index 43b45d6db36d..6e64d01e4e36 100644
@@ -1199,4 +1199,17 @@ struct bpf_sysctl_kern {
 	u64 tmp_reg;
 };
 
+struct bpf_sockopt_kern {
+	struct sock	*sk;
+	u8		*optval;
+	u8		*optval_end;
+	s32		level;
+	s32		optname;
+	u32		optlen;
+
+	/* Small on-stack optval buffer to avoid small allocations.
+	 */
+	u8 buf[64] __aligned(8);
+};
+
 #endif /* __LINUX_FILTER_H__ */
include/uapi/linux/bpf.h +13 -0
index 7c6aef253173..afaa7e28d1e4 100644
@@ -170,6 +170,7 @@ enum bpf_prog_type {
 	BPF_PROG_TYPE_FLOW_DISSECTOR,
 	BPF_PROG_TYPE_CGROUP_SYSCTL,
 	BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE,
+	BPF_PROG_TYPE_CGROUP_SOCKOPT,
 };
 
 enum bpf_attach_type {
@@ -192,6 +193,8 @@ enum bpf_attach_type {
 	BPF_LIRC_MODE2,
 	BPF_FLOW_DISSECTOR,
 	BPF_CGROUP_SYSCTL,
+	BPF_CGROUP_GETSOCKOPT,
+	BPF_CGROUP_SETSOCKOPT,
 	__MAX_BPF_ATTACH_TYPE
 };
 
@@ -3533,4 +3536,14 @@ struct bpf_sysctl {
 				 */
 };
 
+struct bpf_sockopt {
+	__bpf_md_ptr(struct bpf_sock *, sk);
+	__bpf_md_ptr(void *, optval);
+	__bpf_md_ptr(void *, optval_end);
+
+	__s32	level;
+	__s32	optname;
+	__u32	optlen;
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/cgroup.c +260 -0

275 lines not on this page · raw message

kernel/bpf/core.c +9 -0
index 33fb292f2e30..e9152ebd66bc 100644
@@ -1813,6 +1813,15 @@ int bpf_prog_array_length(struct bpf_prog_array *array)
 	return cnt;
 }
 
+bool bpf_prog_array_is_empty(struct bpf_prog_array *array)
+{
+	struct bpf_prog_array_item *item;
+
+	for (item = array->items; item->prog; item++)
+		if (item->prog != &dummy_bpf_prog.prog)
+			return false;
+	return true;
+}
 
 static bool bpf_prog_array_copy_core(struct bpf_prog_array *array,
 				     u32 *prog_ids,
kernel/bpf/syscall.c +19 -0
index 4c53cbd3329d..4ad2b5f1905f 100644
@@ -1596,6 +1596,14 @@ bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
 		default:
 			return -EINVAL;
 		}
+	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
+		switch (expected_attach_type) {
+		case BPF_CGROUP_SETSOCKOPT:
+		case BPF_CGROUP_GETSOCKOPT:
+			return 0;
+		default:
+			return -EINVAL;
+		}
 	default:
 		return 0;
 	}
@@ -1846,6 +1854,7 @@ static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
 	switch (prog->type) {
 	case BPF_PROG_TYPE_CGROUP_SOCK:
 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
+	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
 	case BPF_PROG_TYPE_CGROUP_SKB:
 		return prog->enforce_expected_attach_type &&
@@ -1916,6 +1925,10 @@ static int bpf_prog_attach(const union bpf_attr *attr)
 	case BPF_CGROUP_SYSCTL:
 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
 		break;
+	case BPF_CGROUP_GETSOCKOPT:
+	case BPF_CGROUP_SETSOCKOPT:
+		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
+		break;
 	default:
 		return -EINVAL;
 	}
@@ -1997,6 +2010,10 @@ static int bpf_prog_detach(const union bpf_attr *attr)
 	case BPF_CGROUP_SYSCTL:
 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
 		break;
+	case BPF_CGROUP_GETSOCKOPT:
+	case BPF_CGROUP_SETSOCKOPT:
+		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
+		break;
 	default:
 		return -EINVAL;
 	}
@@ -2031,6 +2048,8 @@ static int bpf_prog_query(const union bpf_attr *attr,
 	case BPF_CGROUP_SOCK_OPS:
 	case BPF_CGROUP_DEVICE:
 	case BPF_CGROUP_SYSCTL:
+	case BPF_CGROUP_GETSOCKOPT:
+	case BPF_CGROUP_SETSOCKOPT:
 		break;
 	case BPF_LIRC_MODE2:
 		return lirc_prog_query(attr, uattr);
kernel/bpf/verifier.c +15 -0
index 5c2cb5bd84ce..fffc668ef536 100644
@@ -1717,6 +1717,18 @@ static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
 
 		env->seen_direct_write = true;
 		return true;
+
+	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
+		if (t == BPF_WRITE) {
+			if (env->prog->expected_attach_type ==
+			    BPF_CGROUP_GETSOCKOPT) {
+				env->seen_direct_write = true;
+				return true;
+			}
+			return false;
+		}
+		return true;
+
 	default:
 		return false;
 	}
@@ -5524,6 +5536,9 @@ static int check_return_code(struct bpf_verifier_env *env)
 	case BPF_PROG_TYPE_CGROUP_DEVICE:
 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
 		break;
+	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
+		range = tnum_range(0, 2);
+		break;
 	default:
 		return 0;
 	}
net/core/filter.c +1 -1
index 55bfc941d17a..9ac0d0efe7ec 100644
@@ -5636,7 +5636,7 @@ BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
 	return (unsigned long)NULL;
 }
 
-static const struct bpf_func_proto bpf_tcp_sock_proto = {
+const struct bpf_func_proto bpf_tcp_sock_proto = {
 	.func		= bpf_tcp_sock,
 	.gpl_only	= false,
 	.ret_type	= RET_PTR_TO_TCP_SOCK_OR_NULL,
net/socket.c +18 -0
index 72372dc5dd70..e8654f1f70e6 100644
@@ -2069,6 +2069,15 @@ static int __sys_setsockopt(int fd, int level, int optname,
 		if (err)
 			goto out_put;
 
+		err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, level, optname,
+						     optval, optlen);
+		if (err < 0) {
+			goto out_put;
+		} else if (err > 0) {
+			err = 0;
+			goto out_put;
+		}
+
 		if (level == SOL_SOCKET)
 			err =
 			    sock_setsockopt(sock, level, optname, optval,
@@ -2106,6 +2115,15 @@ static int __sys_getsockopt(int fd, int level, int optname,
 		if (err)
 			goto out_put;
 
+		err = BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock->sk, level, optname,
+						     optval, optlen);
+		if (err < 0) {
+			goto out_put;
+		} else if (err > 0) {
+			err = 0;
+			goto out_put;
+		}
+
 		if (level == SOL_SOCKET)
 			err =
 			    sock_getsockopt(sock, level, optname, optval,
Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
#7 lore raw

PATCH v5 2/8 bpf: sync bpf.h to tools/

In reply to Stanislav Fomichev

Export new prog type and hook points to the libbpf.

---
 tools/include/uapi/linux/bpf.h | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
tools/include/uapi/linux/bpf.h +14 -0
index 7c6aef253173..174136aa6906 100644
@@ -170,6 +170,7 @@ enum bpf_prog_type {
 	BPF_PROG_TYPE_FLOW_DISSECTOR,
 	BPF_PROG_TYPE_CGROUP_SYSCTL,
 	BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE,
+	BPF_PROG_TYPE_CGROUP_SOCKOPT,
 };
 
 enum bpf_attach_type {
@@ -192,6 +193,8 @@ enum bpf_attach_type {
 	BPF_LIRC_MODE2,
 	BPF_FLOW_DISSECTOR,
 	BPF_CGROUP_SYSCTL,
+	BPF_CGROUP_GETSOCKOPT,
+	BPF_CGROUP_SETSOCKOPT,
 	__MAX_BPF_ATTACH_TYPE
 };
 
@@ -3533,4 +3536,15 @@ struct bpf_sysctl {
 				 */
 };
 
+struct bpf_sockopt {
+	__bpf_md_ptr(struct bpf_sock *, sk);
+	__bpf_md_ptr(void *, optval);
+	__bpf_md_ptr(void *, optval_end);
+
+	__s32	level;
+	__s32	optname;
+
+	__u32	optlen;
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */
Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
#8 lore raw

PATCH v5 3/8 libbpf: support sockopt hooks

In reply to Stanislav Fomichev

Make libbpf aware of new sockopt hooks so it can derive prog type and hook point from the section names.

---
 tools/lib/bpf/libbpf.c        | 5 +++++
 tools/lib/bpf/libbpf_probes.c | 1 +
 2 files changed, 6 insertions(+)
tools/lib/bpf/libbpf.c +5 -0
index ba89d9727137..cd3c692a8b5d 100644
@@ -2243,6 +2243,7 @@ static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
 	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
 	case BPF_PROG_TYPE_PERF_EVENT:
 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
+	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
 		return false;
 	case BPF_PROG_TYPE_KPROBE:
 	default:
@@ -3196,6 +3197,10 @@ static const struct {
 						BPF_CGROUP_UDP6_SENDMSG),
 	BPF_EAPROG_SEC("cgroup/sysctl",		BPF_PROG_TYPE_CGROUP_SYSCTL,
 						BPF_CGROUP_SYSCTL),
+	BPF_EAPROG_SEC("cgroup/getsockopt",	BPF_PROG_TYPE_CGROUP_SOCKOPT,
+						BPF_CGROUP_GETSOCKOPT),
+	BPF_EAPROG_SEC("cgroup/setsockopt",	BPF_PROG_TYPE_CGROUP_SOCKOPT,
+						BPF_CGROUP_SETSOCKOPT),
 };
 
 #undef BPF_PROG_SEC_IMPL
tools/lib/bpf/libbpf_probes.c +1 -0
index 5e2aa83f637a..7e21db11dde8 100644
@@ -101,6 +101,7 @@ probe_load(enum bpf_prog_type prog_type, const struct bpf_insn *insns,
 	case BPF_PROG_TYPE_SK_REUSEPORT:
 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
+	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
 	default:
 		break;
 	}
Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
#9 lore raw

PATCH v5 4/8 selftests/bpf: test sockopt section name

In reply to Stanislav Fomichev

Add tests that make sure libbpf section detection works.

---
 tools/testing/selftests/bpf/test_section_names.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
tools/testing/selftests/bpf/test_section_names.c +10 -0
index bebd4fbca1f4..5f84b3b8c90b 100644
@@ -124,6 +124,16 @@ static struct sec_name_test tests[] = {
 		{0, BPF_PROG_TYPE_CGROUP_SYSCTL, BPF_CGROUP_SYSCTL},
 		{0, BPF_CGROUP_SYSCTL},
 	},
+	{
+		"cgroup/getsockopt",
+		{0, BPF_PROG_TYPE_CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT},
+		{0, BPF_CGROUP_GETSOCKOPT},
+	},
+	{
+		"cgroup/setsockopt",
+		{0, BPF_PROG_TYPE_CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT},
+		{0, BPF_CGROUP_SETSOCKOPT},
+	},
 };
 
 static int test_prog_type_by_name(const struct sec_name_test *test)
Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
#10 lore raw

PATCH v5 5/8 selftests/bpf: add sockopt test

In reply to Stanislav Fomichev

Add sockopt selftests:
* require proper expected_attach_type
* enforce context field read/write access
* test bpf_sockopt_handled handler
* test EPERM
* test limiting optlen from getsockopt
* test out-of-bounds access

v3:
* use DW for optval{,_end} loads

v2:
* use return code 2 for kernel bypass

---
 tools/testing/selftests/bpf/.gitignore     |   1 +
 tools/testing/selftests/bpf/Makefile       |   3 +-
 tools/testing/selftests/bpf/test_sockopt.c | 773 +++++++++++++++++++++
 3 files changed, 776 insertions(+), 1 deletion(-)
 create mode 100644 tools/testing/selftests/bpf/test_sockopt.c
tools/testing/selftests/bpf/.gitignore +1 -0
index 7470327edcfe..3fe92601223d 100644
@@ -39,3 +39,4 @@ libbpf.so.*
 test_hashmap
 test_btf_dump
 xdping
+test_sockopt
tools/testing/selftests/bpf/Makefile +2 -1
index 2b426ae1cdc9..b982393b9181 100644
@@ -26,7 +26,7 @@ TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test
 	test_sock test_btf test_sockmap test_lirc_mode2_user get_cgroup_id_user \
 	test_socket_cookie test_cgroup_storage test_select_reuseport test_section_names \
 	test_netcnt test_tcpnotify_user test_sock_fields test_sysctl test_hashmap \
-	test_btf_dump test_cgroup_attach xdping
+	test_btf_dump test_cgroup_attach xdping test_sockopt
 
 BPF_OBJ_FILES = $(patsubst %.c,%.o, $(notdir $(wildcard progs/*.c)))
 TEST_GEN_FILES = $(BPF_OBJ_FILES)
@@ -101,6 +101,7 @@ $(OUTPUT)/test_netcnt: cgroup_helpers.c
 $(OUTPUT)/test_sock_fields: cgroup_helpers.c
 $(OUTPUT)/test_sysctl: cgroup_helpers.c
 $(OUTPUT)/test_cgroup_attach: cgroup_helpers.c
+$(OUTPUT)/test_sockopt: cgroup_helpers.c
 
 .PHONY: force
 
tools/testing/selftests/bpf/test_sockopt.c new file +773 -0

773 lines not on this page · raw message

Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
#11 lore raw

PATCH v5 6/8 selftests/bpf: add sockopt test that exercises sk helpers

In reply to Stanislav Fomichev

socktop test that introduces new SOL_CUSTOM sockopt level and stores whatever users sets in sk storage. Whenever getsockopt is called, the original value is retrieved.

v4:
* don't call bpf_sk_fullsock helper

v3:
* drop (__u8 *)(long) casts for optval{,_end}

v2:
* new test

---
 tools/testing/selftests/bpf/.gitignore        |   1 +
 tools/testing/selftests/bpf/Makefile          |   3 +-
 .../testing/selftests/bpf/progs/sockopt_sk.c  |  67 ++++++++
 tools/testing/selftests/bpf/test_sockopt_sk.c | 156 ++++++++++++++++++
 4 files changed, 226 insertions(+), 1 deletion(-)
 create mode 100644 tools/testing/selftests/bpf/progs/sockopt_sk.c
 create mode 100644 tools/testing/selftests/bpf/test_sockopt_sk.c
tools/testing/selftests/bpf/.gitignore +1 -0
index 3fe92601223d..8ac076c311d4 100644
@@ -40,3 +40,4 @@ test_hashmap
 test_btf_dump
 xdping
 test_sockopt
+test_sockopt_sk
tools/testing/selftests/bpf/Makefile +2 -1
index b982393b9181..9c167bd5ac18 100644
@@ -26,7 +26,7 @@ TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test
 	test_sock test_btf test_sockmap test_lirc_mode2_user get_cgroup_id_user \
 	test_socket_cookie test_cgroup_storage test_select_reuseport test_section_names \
 	test_netcnt test_tcpnotify_user test_sock_fields test_sysctl test_hashmap \
-	test_btf_dump test_cgroup_attach xdping test_sockopt
+	test_btf_dump test_cgroup_attach xdping test_sockopt test_sockopt_sk
 
 BPF_OBJ_FILES = $(patsubst %.c,%.o, $(notdir $(wildcard progs/*.c)))
 TEST_GEN_FILES = $(BPF_OBJ_FILES)
@@ -102,6 +102,7 @@ $(OUTPUT)/test_sock_fields: cgroup_helpers.c
 $(OUTPUT)/test_sysctl: cgroup_helpers.c
 $(OUTPUT)/test_cgroup_attach: cgroup_helpers.c
 $(OUTPUT)/test_sockopt: cgroup_helpers.c
+$(OUTPUT)/test_sockopt_sk: cgroup_helpers.c
 
 .PHONY: force
 
tools/testing/selftests/bpf/progs/sockopt_sk.c new file +67 -0

67 lines not on this page · raw message

tools/testing/selftests/bpf/test_sockopt_sk.c new file +156 -0

156 lines not on this page · raw message

Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
#12 lore raw

PATCH v5 7/8 bpf: add sockopt documentation

In reply to Stanislav Fomichev

Provide user documentation about sockopt prog type and cgroup hooks.

v2:
* use return code 2 for kernel bypass

---
 Documentation/bpf/index.rst               |  1 +
 Documentation/bpf/prog_cgroup_sockopt.rst | 39 +++++++++++++++++++++++
 2 files changed, 40 insertions(+)
 create mode 100644 Documentation/bpf/prog_cgroup_sockopt.rst
Documentation/bpf/index.rst +1 -0
index d3fe4cac0c90..801a6ed3f2e5 100644
@@ -42,6 +42,7 @@ Program types
 .. toctree::
    :maxdepth: 1
 
+   prog_cgroup_sockopt
    prog_cgroup_sysctl
    prog_flow_dissector
 
Documentation/bpf/prog_cgroup_sockopt.rst new file +39 -0
new file mode 100644
index 000000000000..b117451ab571
@@ -0,0 +1,39 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+=====================
+BPF_PROG_TYPE_SOCKOPT
+=====================
+
+``BPF_PROG_TYPE_SOCKOPT`` program type can be attached to two cgroup hooks:
+
+* ``BPF_CGROUP_GETSOCKOPT`` - called every time process executes ``getsockopt``
+  system call.
+* ``BPF_CGROUP_SETSOCKOPT`` - called every time process executes ``setsockopt``
+  system call.
+
+The context (``struct bpf_sockopt``) has associated socket (``sk``) and
+all input arguments: ``level``, ``optname``, ``optval`` and ``optlen``.
+
+BPF_CGROUP_SETSOCKOPT
+=====================
+
+``BPF_CGROUP_SETSOCKOPT`` has a read-only context and this hook has
+access to cgroup and socket local storage.
+
+BPF_CGROUP_GETSOCKOPT
+=====================
+
+``BPF_CGROUP_GETSOCKOPT`` has to fill in ``optval`` and adjust
+``optlen`` accordingly. Input ``optlen`` contains the maximum length
+of data that can be returned to the userspace. In other words, BPF
+program can't increase ``optlen``, it can only decrease it.
+
+Return Type
+===========
+
+* ``0`` - reject the syscall, ``EPERM`` will be returned to the userspace.
+* ``1`` - success: after returning from the BPF hook, kernel will also
+  handle this socket option.
+* ``2`` - success: after returning from the BPF hook, kernel will _not_
+  handle this socket option; control will be returned to the userspace
+  instead.
Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
#13 lore raw

PATCH v5 8/8 bpftool: support cgroup sockopt

In reply to Stanislav Fomichev

Support sockopt prog type and cgroup hooks in the bpftool.

---
 tools/bpf/bpftool/Documentation/bpftool-cgroup.rst | 7 +++++--
 tools/bpf/bpftool/Documentation/bpftool-prog.rst   | 2 +-
 tools/bpf/bpftool/bash-completion/bpftool          | 8 +++++---
 tools/bpf/bpftool/cgroup.c                         | 5 ++++-
 tools/bpf/bpftool/main.h                           | 1 +
 tools/bpf/bpftool/prog.c                           | 3 ++-
 6 files changed, 18 insertions(+), 8 deletions(-)
tools/bpf/bpftool/Documentation/bpftool-cgroup.rst +5 -2
index 36807735e2a5..cac088a320a6 100644
@@ -29,7 +29,8 @@ CGROUP COMMANDS
 |	*PROG* := { **id** *PROG_ID* | **pinned** *FILE* | **tag** *PROG_TAG* }
 |	*ATTACH_TYPE* := { **ingress** | **egress** | **sock_create** | **sock_ops** | **device** |
 |		**bind4** | **bind6** | **post_bind4** | **post_bind6** | **connect4** | **connect6** |
-|		**sendmsg4** | **sendmsg6** | **sysctl** }
+|		**sendmsg4** | **sendmsg6** | **sysctl** | **getsockopt** |
+|		**setsockopt** }
 |	*ATTACH_FLAGS* := { **multi** | **override** }
 
 DESCRIPTION
@@ -86,7 +87,9 @@ DESCRIPTION
 		  unconnected udp4 socket (since 4.18);
 		  **sendmsg6** call to sendto(2), sendmsg(2), sendmmsg(2) for an
 		  unconnected udp6 socket (since 4.18);
-		  **sysctl** sysctl access (since 5.2).
+		  **sysctl** sysctl access (since 5.2);
+		  **getsockopt** call to getsockopt (since 5.3);
+		  **setsockopt** call to setsockopt (since 5.3).
 
 	**bpftool cgroup detach** *CGROUP* *ATTACH_TYPE* *PROG*
 		  Detach *PROG* from the cgroup *CGROUP* and attach type
tools/bpf/bpftool/Documentation/bpftool-prog.rst +1 -1
index 228a5c863cc7..c6bade35032c 100644
@@ -40,7 +40,7 @@ PROG COMMANDS
 |		**lwt_seg6local** | **sockops** | **sk_skb** | **sk_msg** | **lirc_mode2** |
 |		**cgroup/bind4** | **cgroup/bind6** | **cgroup/post_bind4** | **cgroup/post_bind6** |
 |		**cgroup/connect4** | **cgroup/connect6** | **cgroup/sendmsg4** | **cgroup/sendmsg6** |
-|		**cgroup/sysctl**
+|		**cgroup/sysctl** | **cgroup/getsockopt** | **cgroup/setsockopt**
 |	}
 |       *ATTACH_TYPE* := {
 |		**msg_verdict** | **stream_verdict** | **stream_parser** | **flow_dissector**
tools/bpf/bpftool/bash-completion/bpftool +5 -3
index 2725e27dfa42..7afb8b6fbaaa 100644
@@ -378,7 +378,8 @@ _bpftool()
                                 cgroup/connect4 cgroup/connect6 \
                                 cgroup/sendmsg4 cgroup/sendmsg6 \
                                 cgroup/post_bind4 cgroup/post_bind6 \
-                                cgroup/sysctl" -- \
+                                cgroup/sysctl cgroup/getsockopt \
+                                cgroup/setsockopt" -- \
                                                    "$cur" ) )
                             return 0
                             ;;
@@ -688,7 +689,8 @@ _bpftool()
                 attach|detach)
                     local ATTACH_TYPES='ingress egress sock_create sock_ops \
                         device bind4 bind6 post_bind4 post_bind6 connect4 \
-                        connect6 sendmsg4 sendmsg6 sysctl'
+                        connect6 sendmsg4 sendmsg6 sysctl getsockopt \
+                        setsockopt'
                     local ATTACH_FLAGS='multi override'
                     local PROG_TYPE='id pinned tag'
                     case $prev in
@@ -698,7 +700,7 @@ _bpftool()
                             ;;
                         ingress|egress|sock_create|sock_ops|device|bind4|bind6|\
                         post_bind4|post_bind6|connect4|connect6|sendmsg4|\
-                        sendmsg6|sysctl)
+                        sendmsg6|sysctl|getsockopt|setsockopt)
                             COMPREPLY=( $( compgen -W "$PROG_TYPE" -- \
                                 "$cur" ) )
                             return 0
tools/bpf/bpftool/cgroup.c +4 -1
index 7e22f115c8c1..3083f2e4886e 100644
@@ -25,7 +25,8 @@
 	"       ATTACH_TYPE := { ingress | egress | sock_create |\n"	       \
 	"                        sock_ops | device | bind4 | bind6 |\n"	       \
 	"                        post_bind4 | post_bind6 | connect4 |\n"       \
-	"                        connect6 | sendmsg4 | sendmsg6 | sysctl }"
+	"                        connect6 | sendmsg4 | sendmsg6 | sysctl |\n"  \
+	"                        getsockopt | setsockopt }"
 
 static const char * const attach_type_strings[] = {
 	[BPF_CGROUP_INET_INGRESS] = "ingress",
@@ -42,6 +43,8 @@ static const char * const attach_type_strings[] = {
 	[BPF_CGROUP_UDP4_SENDMSG] = "sendmsg4",
 	[BPF_CGROUP_UDP6_SENDMSG] = "sendmsg6",
 	[BPF_CGROUP_SYSCTL] = "sysctl",
+	[BPF_CGROUP_GETSOCKOPT] = "getsockopt",
+	[BPF_CGROUP_SETSOCKOPT] = "setsockopt",
 	[__MAX_BPF_ATTACH_TYPE] = NULL,
 };
 
tools/bpf/bpftool/main.h +1 -0
index 28a2a5857e14..9c5d9c80f71e 100644
@@ -74,6 +74,7 @@ static const char * const prog_type_name[] = {
 	[BPF_PROG_TYPE_SK_REUSEPORT]		= "sk_reuseport",
 	[BPF_PROG_TYPE_FLOW_DISSECTOR]		= "flow_dissector",
 	[BPF_PROG_TYPE_CGROUP_SYSCTL]		= "cgroup_sysctl",
+	[BPF_PROG_TYPE_CGROUP_SOCKOPT]		= "cgroup_sockopt",
 };
 
 extern const char * const map_type_name[];
tools/bpf/bpftool/prog.c +2 -1
index 1f209c80d906..a201e1c83346 100644
@@ -1070,7 +1070,8 @@ static int do_help(int argc, char **argv)
 		"                 sk_reuseport | flow_dissector | cgroup/sysctl |\n"
 		"                 cgroup/bind4 | cgroup/bind6 | cgroup/post_bind4 |\n"
 		"                 cgroup/post_bind6 | cgroup/connect4 | cgroup/connect6 |\n"
-		"                 cgroup/sendmsg4 | cgroup/sendmsg6 }\n"
+		"                 cgroup/sendmsg4 | cgroup/sendmsg6 | cgroup/getsockopt |\n"
+		"                 cgroup/setsockopt }\n"
 		"       ATTACH_TYPE := { msg_verdict | stream_verdict | stream_parser |\n"
 		"                        flow_dissector }\n"
 		"       " HELP_SPEC_OPTIONS "\n"
Signature · 1 line
2.22.0.rc2.383.gf4fbbf30c2-goog
Acked-by on the v5 series
#14 lore raw
In reply to Stanislav Fomichev
Quoted 11 lines · Stanislav Fomichev

This series implements two new per-cgroup hooks: getsockopt and setsockopt along with a new sockopt program type. The idea is pretty similar to recently introduced cgroup sysctl hooks, but implementation is simpler (no need to convert to/from strings).

What this can be applied to:
* move business logic of what tos/priority/etc can be set by containers (either pass or reject)
* handle existing options (or introduce new ones) differently by propagating some information in cgroup/socket local storage

Quoted 3 lines

Compared to a simple syscall/{g,s}etsockopt tracepoint, those hooks are context aware. Meaning, they can access underlying socket and use cgroup and socket local storage.

#2 lore raw

Re: [PATCH bpf-next v5 1/8] bpf: implement getsockopt and setsockopt hooks

In reply to Stanislav Fomichev
Quoted 338 lines · Stanislav Fomichev

Implement new BPF_PROG_TYPE_CGROUP_SOCKOPT program type and BPF_CGROUP_{G,S}ETSOCKOPT cgroup hooks.

BPF_CGROUP_SETSOCKOPT get a read-only view of the setsockopt arguments. BPF_CGROUP_GETSOCKOPT can modify the supplied buffer.
Both of them reuse existing PTR_TO_PACKET{,_END} infrastructure.

The buffer memory is pre-allocated (because I don't think there is a precedent for working with __user memory from bpf). This might be slow to do for each {s,g}etsockopt call, that's why I've added __cgroup_bpf_prog_array_is_empty that exits early if there is nothing attached to a cgroup. Note, however, that there is a race between __cgroup_bpf_prog_array_is_empty and BPF_PROG_RUN_ARRAY where cgroup program layout might have changed; this should not be a problem because in general there is a race between multiple calls to {s,g}etsocktop and user adding/removing bpf progs from a cgroup.

The return code of the BPF program is handled as follows:
* 0: EPERM
* 1: success, execute kernel {s,g}etsockopt path after BPF prog exits
* 2: success, do _not_ execute kernel {s,g}etsockopt path after BPF prog exits

v5:
* skip copy_to_user() and put_user() when ret == 0 (Martin Lau)

v4:
* don't export bpf_sk_fullsock helper (Martin Lau)
* size != sizeof(__u64) for uapi pointers (Martin Lau)
* offsetof instead of bpf_ctx_range when checking ctx access (Martin Lau)

v3:
* typos in BPF_PROG_CGROUP_SOCKOPT_RUN_ARRAY comments (Andrii Nakryiko)
* reverse christmas tree in BPF_PROG_CGROUP_SOCKOPT_RUN_ARRAY (Andrii Nakryiko)
* use __bpf_md_ptr instead of __u32 for optval{,_end} (Martin Lau)
* use BPF_FIELD_SIZEOF() for consistency (Martin Lau)
* new CG_SOCKOPT_ACCESS macro to wrap repeated parts

v2:
* moved bpf_sockopt_kern fields around to remove a hole (Martin Lau)
* aligned bpf_sockopt_kern->buf to 8 bytes (Martin Lau)
* bpf_prog_array_is_empty instead of bpf_prog_array_length (Martin Lau)
* added [0,2] return code check to verifier (Martin Lau)
* dropped unused buf[64] from the stack (Martin Lau)
* use PTR_TO_SOCKET for bpf_sockopt->sk (Martin Lau)
* dropped bpf_target_off from ctx rewrites (Martin Lau)
* use return code for kernel bypass (Martin Lau & Andrii Nakryiko)

---
 include/linux/bpf-cgroup.h |  29 +++++
 include/linux/bpf.h        |  45 +++++++
 include/linux/bpf_types.h  |   1 +
 include/linux/filter.h     |  13 ++
 include/uapi/linux/bpf.h   |  13 ++
 kernel/bpf/cgroup.c        | 260 +++++++++++++++++++++++++++++++++++++
 kernel/bpf/core.c          |   9 ++
 kernel/bpf/syscall.c       |  19 +++
 kernel/bpf/verifier.c      |  15 +++
 net/core/filter.c          |   2 +-
 net/socket.c               |  18 +++
 11 files changed, 423 insertions(+), 1 deletion(-)
include/linux/bpf-cgroup.h +29 -0
index b631ee75762d..406f1ba82531 100644
@@ -124,6 +124,13 @@ int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
 				   loff_t *ppos, void **new_buf,
 				   enum bpf_attach_type type);
 
+int __cgroup_bpf_run_filter_setsockopt(struct sock *sock, int level,
+				       int optname, char __user *optval,
+				       unsigned int optlen);
+int __cgroup_bpf_run_filter_getsockopt(struct sock *sock, int level,
+				       int optname, char __user *optval,
+				       int __user *optlen);
+
 static inline enum bpf_cgroup_storage_type cgroup_storage_type(
 	struct bpf_map *map)
 {
@@ -280,6 +287,26 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key,
 	__ret;								       \
 })
 
+#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen)   \
+({									       \
+	int __ret = 0;							       \
+	if (cgroup_bpf_enabled)						       \
+		__ret = __cgroup_bpf_run_filter_setsockopt(sock, level,	       \
+							   optname, optval,    \
+							   optlen);	       \
+	__ret;								       \
+})
+
+#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, optlen)   \
+({									       \
+	int __ret = 0;							       \
+	if (cgroup_bpf_enabled)						       \
+		__ret = __cgroup_bpf_run_filter_getsockopt(sock, level,	       \
+							   optname, optval,    \
+							   optlen);	       \
+	__ret;								       \
+})
+
 int cgroup_bpf_prog_attach(const union bpf_attr *attr,
 			   enum bpf_prog_type ptype, struct bpf_prog *prog);
 int cgroup_bpf_prog_detach(const union bpf_attr *attr,
@@ -349,6 +376,8 @@ static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
 #define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) ({ 0; })
 #define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type,major,minor,access) ({ 0; })
 #define BPF_CGROUP_RUN_PROG_SYSCTL(head,table,write,buf,count,pos,nbuf) ({ 0; })
+#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, optlen) ({ 0; })
+#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen) ({ 0; })
 
 #define for_each_cgroup_storage_type(stype) for (; false; )
 
include/linux/bpf.h +45 -0
index e5a309e6a400..194a47ca622f 100644
@@ -520,6 +520,7 @@ struct bpf_prog_array {
 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags);
 void bpf_prog_array_free(struct bpf_prog_array *progs);
 int bpf_prog_array_length(struct bpf_prog_array *progs);
+bool bpf_prog_array_is_empty(struct bpf_prog_array *array);
 int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs,
 				__u32 __user *prog_ids, u32 cnt);
 
@@ -606,6 +607,49 @@ _out:							\
 		_ret;					\
 	})
 
+/* To be used by BPF_PROG_TYPE_CGROUP_SOCKOPT program type.
+ *
+ * Expected BPF program return values are:
+ *   0: return -EPERM to the userspace
+ *   1: sockopt was not handled by BPF, kernel should do it
+ *   2: sockopt was handled by BPF, kernel should _not_ do it and return
+ *      to the userspace instead
+ *
+ * Note, that return '0' takes precedence over everything else. In other
+ * words, if any single program in the prog array has returned 0,
+ * the userspace will get -EPERM (regardless of what other programs
+ * return).
+ *
+ * The macro itself returns:
+ *        0: sockopt was not handled by BPF, kernel should do it
+ *        1: sockopt was handled by BPF, kernel should _not_ do it
+ *   -EPERM: return error back to userspace
+ */
+#define BPF_PROG_CGROUP_SOCKOPT_RUN_ARRAY(array, ctx, func)		\
+	({								\
+		struct bpf_prog_array_item *_item;			\
+		struct bpf_prog_array *_array;				\
+		struct bpf_prog *_prog;					\
+		u32 _success = 1;					\
+		u32 _bypass = 0;					\
+		u32 ret;						\
+		preempt_disable();					\
+		rcu_read_lock();					\
+		_array = rcu_dereference(array);			\
+		_item = &_array->items[0];				\
+		while ((_prog = READ_ONCE(_item->prog))) {		\
+			bpf_cgroup_storage_set(_item->cgroup_storage);	\
+			ret = func(_prog, ctx);				\
+			_success &= (ret > 0);				\
+			_bypass |= (ret == 2);				\
+			_item++;					\
+		}							\
+		rcu_read_unlock();					\
+		preempt_enable();					\
+		ret = _success ? _bypass : -EPERM;			\
+		ret;							\
+	})
+
 #define BPF_PROG_RUN_ARRAY(array, ctx, func)		\
 	__BPF_PROG_RUN_ARRAY(array, ctx, func, false)
 
@@ -1054,6 +1098,7 @@ extern const struct bpf_func_proto bpf_spin_unlock_proto;
 extern const struct bpf_func_proto bpf_get_local_storage_proto;
 extern const struct bpf_func_proto bpf_strtol_proto;
 extern const struct bpf_func_proto bpf_strtoul_proto;
+extern const struct bpf_func_proto bpf_tcp_sock_proto;
 
 /* Shared helpers among cBPF and eBPF. */
 void bpf_user_rnd_init_once(void);
include/linux/bpf_types.h +1 -0
index 5a9975678d6f..eec5aeeeaf92 100644
@@ -30,6 +30,7 @@ BPF_PROG_TYPE(BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE, raw_tracepoint_writable)
 #ifdef CONFIG_CGROUP_BPF
 BPF_PROG_TYPE(BPF_PROG_TYPE_CGROUP_DEVICE, cg_dev)
 BPF_PROG_TYPE(BPF_PROG_TYPE_CGROUP_SYSCTL, cg_sysctl)
+BPF_PROG_TYPE(BPF_PROG_TYPE_CGROUP_SOCKOPT, cg_sockopt)
 #endif
 #ifdef CONFIG_BPF_LIRC_MODE2
 BPF_PROG_TYPE(BPF_PROG_TYPE_LIRC_MODE2, lirc_mode2)
include/linux/filter.h +13 -0
index 43b45d6db36d..6e64d01e4e36 100644
@@ -1199,4 +1199,17 @@ struct bpf_sysctl_kern {
 	u64 tmp_reg;
 };
 
+struct bpf_sockopt_kern {
+	struct sock	*sk;
+	u8		*optval;
+	u8		*optval_end;
+	s32		level;
+	s32		optname;
+	u32		optlen;
+
+	/* Small on-stack optval buffer to avoid small allocations.
+	 */
+	u8 buf[64] __aligned(8);
+};
+
 #endif /* __LINUX_FILTER_H__ */
include/uapi/linux/bpf.h +13 -0
index 7c6aef253173..afaa7e28d1e4 100644
@@ -170,6 +170,7 @@ enum bpf_prog_type {
 	BPF_PROG_TYPE_FLOW_DISSECTOR,
 	BPF_PROG_TYPE_CGROUP_SYSCTL,
 	BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE,
+	BPF_PROG_TYPE_CGROUP_SOCKOPT,
 };
 
 enum bpf_attach_type {
@@ -192,6 +193,8 @@ enum bpf_attach_type {
 	BPF_LIRC_MODE2,
 	BPF_FLOW_DISSECTOR,
 	BPF_CGROUP_SYSCTL,
+	BPF_CGROUP_GETSOCKOPT,
+	BPF_CGROUP_SETSOCKOPT,
 	__MAX_BPF_ATTACH_TYPE
 };
 
@@ -3533,4 +3536,14 @@ struct bpf_sysctl {
 				 */
 };
 
+struct bpf_sockopt {
+	__bpf_md_ptr(struct bpf_sock *, sk);
+	__bpf_md_ptr(void *, optval);
+	__bpf_md_ptr(void *, optval_end);
+
+	__s32	level;
+	__s32	optname;
+	__u32	optlen;
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/cgroup.c +63 -0
index 1b65ab0df457..9085a218a1a8 100644
@@ -18,6 +18,7 @@
 #include <linux/bpf.h>
 #include <linux/bpf-cgroup.h>
 #include <net/sock.h>
+#include <net/bpf_sk_storage.h>
 
 DEFINE_STATIC_KEY_FALSE(cgroup_bpf_enabled_key);
 EXPORT_SYMBOL(cgroup_bpf_enabled_key);
@@ -924,6 +925,140 @@ int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
 }
 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sysctl);
 
+static bool __cgroup_bpf_prog_array_is_empty(struct cgroup *cgrp,
+					     enum bpf_attach_type attach_type)
+{
+	struct bpf_prog_array *prog_array;
+	bool empty;
+
+	rcu_read_lock();
+	prog_array = rcu_dereference(cgrp->bpf.effective[attach_type]);
+	empty = bpf_prog_array_is_empty(prog_array);
+	rcu_read_unlock();
+
+	return empty;
+}
+
+static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen)
+{
+	if (unlikely(max_optlen > PAGE_SIZE))
+		return -EINVAL;
+
+	if (likely(max_optlen <= sizeof(ctx->buf))) {
+		ctx->optval = ctx->buf;
+	} else {
+		ctx->optval = kzalloc(max_optlen, GFP_USER);
+		if (!ctx->optval)
+			return -ENOMEM;
+	}
+
+	ctx->optval_end = ctx->optval + max_optlen;
+	ctx->optlen = max_optlen;
+
+	return 0;
+}
+
+static void sockopt_free_buf(struct bpf_sockopt_kern *ctx)
+{
+	if (unlikely(ctx->optval != ctx->buf))
+		kfree(ctx->optval);
+}
+
+int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int level,
+				       int optname, char __user *optval,
+				       unsigned int optlen)
+{
+	struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
+	struct bpf_sockopt_kern ctx = {
+		.sk = sk,
+		.level = level,
+		.optname = optname,
+	};
+	int ret;
+
+	/* Opportunistic check to see whether we have any BPF program
+	 * attached to the hook so we don't waste time allocating
+	 * memory and locking the socket.
+	 */
+	if (__cgroup_bpf_prog_array_is_empty(cgrp, BPF_CGROUP_SETSOCKOPT))
+		return 0;
+
+	ret = sockopt_alloc_buf(&ctx, optlen);
+	if (ret)
+		return ret;
+

kernel/bpf/cgroup.c int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head, · line 999

Quoted 10 lines
kernel/bpf/cgroup.c +10 -0
+	if (copy_from_user(ctx.optval, optval, optlen) != 0) {
+		sockopt_free_buf(&ctx);
+		return -EFAULT;
+	}
+
+	lock_sock(sk);
+	ret = BPF_PROG_CGROUP_SOCKOPT_RUN_ARRAY(
+		cgrp->bpf.effective[BPF_CGROUP_SETSOCKOPT],
+		&ctx, BPF_PROG_RUN);
+	release_sock(sk);

C based example doesn't use ret=1.
imo that's a sign that something is odd in the api.
In particular ret=1 doesn't prohibit bpf prog to modify the optval. Multiple progs can overwrite it and still return 1.
But that optval is not going to be processed by the kernel. Should we do copy_to_user(optval, ctx.optval, ctx.optlen) here and let kernel pick it up from there?
Should bpf prog be allowed to change optlen as well? ret=1 would mean that bpf prog did something and needs kernel to continue.

Now consider a sequence of bpf progs.
Some are doing ret=1. Some others are doing ret=2 ret=2 will supersede.
If first executed prog (child in cgroup) did ret=2 the parent has no way to tell kernel to handle it. Even if parent does ret=1, it's effectively ignored. Parent can enforce rejection with ret=0, but it's a weird discrepancy.
The rule for cgroup progs was 'all yes is yes, any no is no'.

So if ret=1 means 'kernel handles it'. Should it be almost as strong as 'reject it': any prog doing ret=1 means 'kernel does it' (unless some prog did ret=0. then reject it) ?
if ret=1 means 'bpf did some and needs kernel to continue' that's another story.
For ret=2 being 'bpf handled it completely', should parent overwrite it?

May be retval from child prog should be seen by parent prog?

In some sense kernel can be seen as another bpf prog in a sequence.

Whatever new behavior is with 3 values it needs to be documented in uapi/bpf.h
We were sloppy with such docs in the past, but that's not a reason to continue.

#3 lore raw

Re: [PATCH bpf-next v5 1/8] bpf: implement getsockopt and setsockopt hooks

Quoted 2 lines · Alexei Starovoitov

C based example doesn't use ret=1.
imo that's a sign that something is odd in the api.

I decided not to test ret=1 it because there are tests in the test_sockopt.c for ret=1 usecase. But I can certainly extend C based test to cover ret=1 as well. I agree that C based test can be used as an example, will extend that to cover ret=0/1/2.

Quoted 1 line

In particular ret=1 doesn't prohibit bpf prog to modify the optval.

That's a good point, Martin brought that up as well. We were trying to remedy it by doing copy_to_user only if any program returned 2 ("BPF handled that, bypass the kernel"). But I agree, the fact that the prog in the chain can modify optval and return 1 is suboptimal. Especially if the previous one filled in some valid data and returned 2.

Quoted 4 lines

Multiple progs can overwrite it and still return 1. But that optval is not going to be processed by the kernel. Should we do copy_to_user(optval, ctx.optval, ctx.optlen) here and let kernel pick it up from there?

I was thinking initially about that, that kernel can "transparently" modify user buffer and then kernel (or next BPF program in the chain) can run standard getsockopt on that.

But it sounds a bit complicated and I don't really have a good use case for that.

Quoted 4 lines

Should bpf prog be allowed to change optlen as well? ret=1 would mean that bpf prog did something and needs kernel to continue.

Quoted 9 lines

Now consider a sequence of bpf progs.
Some are doing ret=1. Some others are doing ret=2 ret=2 will supersede.
If first executed prog (child in cgroup) did ret=2 the parent has no way to tell kernel to handle it. Even if parent does ret=1, it's effectively ignored. Parent can enforce rejection with ret=0, but it's a weird discrepancy.
The rule for cgroup progs was 'all yes is yes, any no is no'.

My canonical example when reasoning about multiple progs was that each one of them would implement handling for a particular level+optname. So only a single one form the chain would return 2 or 0, the rest would return 1 without touching the buffer. I can't come up with a good use-case where two programs in the chain can both return 2 and fill out the buffer. The majority of the sockopts would still be handled by the kernel, we'd have only a handful of bpf progs that handle a tiny subset and delegate the rest to the kernel.

How about we stop further processing as soon as some program in the chain returned 2? I think that would address most of the concerns? Maybe, in this case, also stop further processing as soon as we get ret=0 (EPERM) for consistency?

Quoted 6 lines

So if ret=1 means 'kernel handles it'. Should it be almost as strong as 'reject it': any prog doing ret=1 means 'kernel does it' (unless some prog did ret=0. then reject it) ?
if ret=1 means 'bpf did some and needs kernel to continue' that's another story.
For ret=2 being 'bpf handled it completely', should parent overwrite it?

See above, I was thinking the opposite. Treat ret=1 from the BPF program as "I'm not interested in this level+optname, other BPF program or kernel should do the job". Essentially, as soon as bpf program returns 2, that means BPF had consumed the request and no further processing from neither BPF, nor kernel is requred; we can return to userspace.

There is a problem that some prog in the chain might do some "background" work and still return 1, but so far I don't see why that can be useful. The pattern should be: filter the option you want, handle it, otherwise return 1 to let the other progs/kernel run.

That BPF_F_ALLOW_MULTI use-case probably deserves another selftest :-/

Quoted 8 lines

May be retval from child prog should be seen by parent prog?

In some sense kernel can be seen as another bpf prog in a sequence.

Whatever new behavior is with 3 values it needs to be documented in uapi/bpf.h
We were sloppy with such docs in the past, but that's not a reason to continue.

Good point on documenting that, I was trying to document everything in Documentation/bpf/prog_cgroup_sockopt.rst, uapi/bpf.h seems too constrained (I didn't find a good place to put that ret 1 vs 2 info). Do you think having a file under Documentation/ with all the details is not enough? Where can I put this ret=0/1/2 handing info in the uapi/bpf.h?

#4 lore raw

Re: [PATCH bpf-next v5 1/8] bpf: implement getsockopt and setsockopt hooks

Quoted 49 lines · Stanislav Fomichev
Quoted 2 lines · Alexei Starovoitov

C based example doesn't use ret=1.
imo that's a sign that something is odd in the api.

I decided not to test ret=1 it because there are tests in the test_sockopt.c for ret=1 usecase. But I can certainly extend C based test to cover ret=1 as well. I agree that C based test can be used as an example, will extend that to cover ret=0/1/2.

Quoted 1 line

In particular ret=1 doesn't prohibit bpf prog to modify the optval.

That's a good point, Martin brought that up as well. We were trying to remedy it by doing copy_to_user only if any program returned 2 ("BPF handled that, bypass the kernel"). But I agree, the fact that the prog in the chain can modify optval and return 1 is suboptimal. Especially if the previous one filled in some valid data and returned 2.

Quoted 4 lines

Multiple progs can overwrite it and still return 1. But that optval is not going to be processed by the kernel. Should we do copy_to_user(optval, ctx.optval, ctx.optlen) here and let kernel pick it up from there?

I was thinking initially about that, that kernel can "transparently" modify user buffer and then kernel (or next BPF program in the chain) can run standard getsockopt on that.

But it sounds a bit complicated and I don't really have a good use case for that.

Quoted 4 lines

Should bpf prog be allowed to change optlen as well? ret=1 would mean that bpf prog did something and needs kernel to continue.

Quoted 9 lines

Now consider a sequence of bpf progs.
Some are doing ret=1. Some others are doing ret=2 ret=2 will supersede.
If first executed prog (child in cgroup) did ret=2 the parent has no way to tell kernel to handle it. Even if parent does ret=1, it's effectively ignored. Parent can enforce rejection with ret=0, but it's a weird discrepancy.
The rule for cgroup progs was 'all yes is yes, any no is no'.

My canonical example when reasoning about multiple progs was that each one of them would implement handling for a particular level+optname. So only a single one form the chain would return 2 or 0, the rest would return 1 without touching the buffer. I can't come up with a good use-case where two programs in the chain can both return 2 and fill out the buffer. The majority of the sockopts would still be handled by the kernel, we'd have only a handful of bpf progs that handle a tiny subset and delegate the rest to the kernel.

Quoted 2 lines

How about we stop further processing as soon as some program in the chain returned 2? I think that would address most of the concerns?

What about a case of passive "auditing" BPF programs, that are not modifying anything, but want to capture every single getsockopt/setsockopt call? This premature stop would render that whole approach broken.

Quoted 37 lines

Maybe, in this case, also stop further processing as soon as we get ret=0 (EPERM) for consistency?

Quoted 6 lines

So if ret=1 means 'kernel handles it'. Should it be almost as strong as 'reject it': any prog doing ret=1 means 'kernel does it' (unless some prog did ret=0. then reject it) ?
if ret=1 means 'bpf did some and needs kernel to continue' that's another story.
For ret=2 being 'bpf handled it completely', should parent overwrite it?

See above, I was thinking the opposite. Treat ret=1 from the BPF program as "I'm not interested in this level+optname, other BPF program or kernel should do the job". Essentially, as soon as bpf program returns 2, that means BPF had consumed the request and no further processing from neither BPF, nor kernel is requred; we can return to userspace.

There is a problem that some prog in the chain might do some "background" work and still return 1, but so far I don't see why that can be useful. The pattern should be: filter the option you want, handle it, otherwise return 1 to let the other progs/kernel run.

That BPF_F_ALLOW_MULTI use-case probably deserves another selftest :-/

Quoted 8 lines

May be retval from child prog should be seen by parent prog?

In some sense kernel can be seen as another bpf prog in a sequence.

Whatever new behavior is with 3 values it needs to be documented in uapi/bpf.h
We were sloppy with such docs in the past, but that's not a reason to continue.

Good point on documenting that, I was trying to document everything in Documentation/bpf/prog_cgroup_sockopt.rst, uapi/bpf.h seems too constrained (I didn't find a good place to put that ret 1 vs 2 info). Do you think having a file under Documentation/ with all the details is not enough? Where can I put this ret=0/1/2 handing info in the uapi/bpf.h?

#5 lore raw

Re: [PATCH bpf-next v5 1/8] bpf: implement getsockopt and setsockopt hooks

Quoted 50 lines · Andrii Nakryiko
Quoted 49 lines · Stanislav Fomichev
Quoted 2 lines · Alexei Starovoitov

C based example doesn't use ret=1.
imo that's a sign that something is odd in the api.

I decided not to test ret=1 it because there are tests in the test_sockopt.c for ret=1 usecase. But I can certainly extend C based test to cover ret=1 as well. I agree that C based test can be used as an example, will extend that to cover ret=0/1/2.

Quoted 1 line

In particular ret=1 doesn't prohibit bpf prog to modify the optval.

That's a good point, Martin brought that up as well. We were trying to remedy it by doing copy_to_user only if any program returned 2 ("BPF handled that, bypass the kernel"). But I agree, the fact that the prog in the chain can modify optval and return 1 is suboptimal. Especially if the previous one filled in some valid data and returned 2.

Quoted 4 lines

Multiple progs can overwrite it and still return 1. But that optval is not going to be processed by the kernel. Should we do copy_to_user(optval, ctx.optval, ctx.optlen) here and let kernel pick it up from there?

I was thinking initially about that, that kernel can "transparently" modify user buffer and then kernel (or next BPF program in the chain) can run standard getsockopt on that.

But it sounds a bit complicated and I don't really have a good use case for that.

Quoted 4 lines

Should bpf prog be allowed to change optlen as well? ret=1 would mean that bpf prog did something and needs kernel to continue.

Quoted 9 lines

Now consider a sequence of bpf progs.
Some are doing ret=1. Some others are doing ret=2 ret=2 will supersede.
If first executed prog (child in cgroup) did ret=2 the parent has no way to tell kernel to handle it. Even if parent does ret=1, it's effectively ignored. Parent can enforce rejection with ret=0, but it's a weird discrepancy.
The rule for cgroup progs was 'all yes is yes, any no is no'.

My canonical example when reasoning about multiple progs was that each one of them would implement handling for a particular level+optname. So only a single one form the chain would return 2 or 0, the rest would return 1 without touching the buffer. I can't come up with a good use-case where two programs in the chain can both return 2 and fill out the buffer. The majority of the sockopts would still be handled by the kernel, we'd have only a handful of bpf progs that handle a tiny subset and delegate the rest to the kernel.

Quoted 7 lines
Quoted 2 lines

How about we stop further processing as soon as some program in the chain returned 2? I think that would address most of the concerns?

What about a case of passive "auditing" BPF programs, that are not modifying anything, but want to capture every single getsockopt/setsockopt call? This premature stop would render that whole approach broken.

In that case you'd attach that program to the root of a cgroup (sub)tree what you want to audit (and it would be always executed and would return 1)? And you'd have to attach it first.

Quoted 37 lines
Quoted 37 lines

Maybe, in this case, also stop further processing as soon as we get ret=0 (EPERM) for consistency?

Quoted 6 lines

So if ret=1 means 'kernel handles it'. Should it be almost as strong as 'reject it': any prog doing ret=1 means 'kernel does it' (unless some prog did ret=0. then reject it) ?
if ret=1 means 'bpf did some and needs kernel to continue' that's another story.
For ret=2 being 'bpf handled it completely', should parent overwrite it?

See above, I was thinking the opposite. Treat ret=1 from the BPF program as "I'm not interested in this level+optname, other BPF program or kernel should do the job". Essentially, as soon as bpf program returns 2, that means BPF had consumed the request and no further processing from neither BPF, nor kernel is requred; we can return to userspace.

There is a problem that some prog in the chain might do some "background" work and still return 1, but so far I don't see why that can be useful. The pattern should be: filter the option you want, handle it, otherwise return 1 to let the other progs/kernel run.

That BPF_F_ALLOW_MULTI use-case probably deserves another selftest :-/

Quoted 8 lines

May be retval from child prog should be seen by parent prog?

In some sense kernel can be seen as another bpf prog in a sequence.

Whatever new behavior is with 3 values it needs to be documented in uapi/bpf.h
We were sloppy with such docs in the past, but that's not a reason to continue.

Good point on documenting that, I was trying to document everything in Documentation/bpf/prog_cgroup_sockopt.rst, uapi/bpf.h seems too constrained (I didn't find a good place to put that ret 1 vs 2 info). Do you think having a file under Documentation/ with all the details is not enough? Where can I put this ret=0/1/2 handing info in the uapi/bpf.h?

#6 lore raw

Re: [PATCH bpf-next v5 1/8] bpf: implement getsockopt and setsockopt hooks

Quoted 12 lines · Stanislav Fomichev
Quoted 12 lines
Quoted 11 lines

My canonical example when reasoning about multiple progs was that each one of them would implement handling for a particular level+optname. So only a single one form the chain would return 2 or 0, the rest would return 1 without touching the buffer. I can't come up with a good use-case where two programs in the chain can both return 2 and fill out the buffer. The majority of the sockopts would still be handled by the kernel, we'd have only a handful of bpf progs that handle a tiny subset and delegate the rest to the kernel.

How about we stop further processing as soon as some program in the chain returned 2? I think that would address most of the concerns?

Quoted 7 lines
Quoted 4 lines

What about a case of passive "auditing" BPF programs, that are not modifying anything, but want to capture every single getsockopt/setsockopt call? This premature stop would render that whole approach broken.

In that case you'd attach that program to the root of a cgroup (sub)tree what you want to audit (and it would be always executed and would return 1)? And you'd have to attach it first.

On a second thought, that's not true. BPF progs are executed from the bottom up, so attaching to the root cgroup wouldn't work for that auditing case :-/