[PATCH bpf-next 0/3] bpf: Introduce minimal support for sleepable progs

[PATCH bpf-next 0/3] bpf: Introduce minimal support for sleepable progs

[Alexei Starovoitov]

From: Alexei Starovoitov <ast@kernel.org>

This patch set introduces the minimal viable support for sleepable bpf programs.
In this patch only fentry/fexit/fmod_ret and lsm progs can be sleepable.
Only array and pre-allocated hash and lru maps allowed.

Here is 'perf report' difference of sleepable vs non-sleepable:
   3.86%  bench     [k] __srcu_read_unlock
   3.22%  bench     [k] __srcu_read_lock
   0.92%  bench     [k] bpf_prog_740d4210cdcd99a3_bench_trigger_fentry_sleep
   0.50%  bench     [k] bpf_trampoline_10297
   0.26%  bench     [k] __bpf_prog_exit_sleepable
   0.21%  bench     [k] __bpf_prog_enter_sleepable
vs
   0.88%  bench     [k] bpf_prog_740d4210cdcd99a3_bench_trigger_fentry
   0.84%  bench     [k] bpf_trampoline_10297
   0.13%  bench     [k] __bpf_prog_enter
   0.12%  bench     [k] __bpf_prog_exit

Clearly sleepable vs non-sleepable program invocation overhead is high, but it
could be acceptable in certain cases until rcu_trace is available.

bpf_copy_from_user(), bpf_msleep() will be coming in the future patches.
Support for perf_event_array and bpf_perf_event_output() will follow as well.

Alexei Starovoitov (3):
  bpf: Introduce sleepable BPF programs
  libbpf: support sleepable progs
  selftests/bpf: basic sleepable tests

 arch/x86/net/bpf_jit_comp.c                   | 36 ++++++++----
 include/linux/bpf.h                           |  4 ++
 include/uapi/linux/bpf.h                      |  8 +++
 kernel/bpf/arraymap.c                         |  5 ++
 kernel/bpf/hashtab.c                          | 19 +++++--
 kernel/bpf/syscall.c                          | 12 +++-
 kernel/bpf/trampoline.c                       | 33 ++++++++++-
 kernel/bpf/verifier.c                         | 56 ++++++++++++++-----
 tools/include/uapi/linux/bpf.h                |  8 +++
 tools/lib/bpf/libbpf.c                        | 25 ++++++++-
 tools/testing/selftests/bpf/bench.c           |  2 +
 .../selftests/bpf/benchs/bench_trigger.c      | 17 ++++++
 tools/testing/selftests/bpf/progs/lsm.c       |  4 +-
 .../selftests/bpf/progs/trigger_bench.c       |  7 +++
 14 files changed, 199 insertions(+), 37 deletions(-)

-- 
2.23.0

[PATCH bpf-next 1/3] bpf: Introduce sleepable BPF programs

[Alexei Starovoitov]

From: Alexei Starovoitov <ast@kernel.org>

Introduce sleepable BPF programs that can request such property for themselves
via BPF_F_SLEEPABLE flag at program load time. In such case they will be able
to use helpers like bpf_copy_from_user() that might sleep. At present only
fentry/fexit/fmod_ret and lsm programs can request to be sleepable and only
when they are attached to kernel functions that are known to allow sleeping.

The non-sleepable programs are relying on implicit rcu_read_lock() and
migrate_disable() to protect life time of programs, maps that they use and
per-cpu kernel structures used to pass info between bpf programs and the
kernel. The sleepable programs cannot be enclosed into rcu_read_lock().
migrate_disable() maps to preempt_disable() in non-RT kernels, so the progs
should not be enclosed in migrate_disable() as well. Therefore bpf_srcu is used
to protect the life time of sleepable progs.

There are many networking and tracing program types. In many cases the
'struct bpf_prog *' pointer itself is rcu protected within some other kernel
data structure and the kernel code is using rcu_dereference() to load that
program pointer and call BPF_PROG_RUN() on it. All these cases are not touched.
Instead sleepable bpf programs are allowed with bpf trampoline only. The
program pointers are hard-coded into generated assembly of bpf trampoline and
synchronize_srcu(&bpf_srcu) is used to protect the life time of the program.
The same trampoline can hold both sleepable and non-sleepable progs.

When bpf_srcu lock is held it means that some sleepable bpf program is running
from bpf trampoline. Those programs can use bpf arrays and preallocated hash/lru
maps. These map types are waiting on programs to complete via
synchronize_srcu(&bpf_srcu);

Updates to trampoline now has to do synchronize_srcu + synchronize_rcu_tasks
to wait for sleepable progs to finish and for trampoline assembly to finish.

In the future srcu will be replaced with upcoming rcu_trace.
That will complete the first step of introducing sleepable progs.

After that dynamically allocated hash maps can be allowed. All map elements
would have to be srcu protected instead of normal rcu.
per-cpu maps will be allowed. Either via the following pattern:
void *elem = bpf_map_lookup_elem(map, key);
if (elem) {
   // access elem
   bpf_map_release_elem(map, elem);
}
where modified lookup() helper will do migrate_disable() and
new bpf_map_release_elem() will do corresponding migrate_enable().
Or explicit bpf_migrate_disable/enable() helpers will be introduced.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
---
 arch/x86/net/bpf_jit_comp.c    | 36 +++++++++++++++-------
 include/linux/bpf.h            |  4 +++
 include/uapi/linux/bpf.h       |  8 +++++
 kernel/bpf/arraymap.c          |  5 +++
 kernel/bpf/hashtab.c           | 19 ++++++++----
 kernel/bpf/syscall.c           | 12 ++++++--
 kernel/bpf/trampoline.c        | 33 +++++++++++++++++++-
 kernel/bpf/verifier.c          | 56 ++++++++++++++++++++++++++--------
 tools/include/uapi/linux/bpf.h |  8 +++++
 9 files changed, 147 insertions(+), 34 deletions(-)

diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index 42b6709e6dc7..3fdb62c89d6f 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -1379,10 +1379,17 @@ static int invoke_bpf_prog(const struct btf_func_model *m, u8 **pprog,
 	u8 *prog = *pprog;
 	int cnt = 0;
 
-	if (emit_call(&prog, __bpf_prog_enter, prog))
-		return -EINVAL;
-	/* remember prog start time returned by __bpf_prog_enter */
-	emit_mov_reg(&prog, true, BPF_REG_6, BPF_REG_0);
+	if (p->aux->sleepable) {
+		if (emit_call(&prog, __bpf_prog_enter_sleepable, prog))
+			return -EINVAL;
+		/* remember srcu idx returned by __bpf_prog_enter_sleepable */
+		emit_mov_reg(&prog, true, BPF_REG_6, BPF_REG_0);
+	} else {
+		if (emit_call(&prog, __bpf_prog_enter, prog))
+			return -EINVAL;
+		/* remember prog start time returned by __bpf_prog_enter */
+		emit_mov_reg(&prog, true, BPF_REG_6, BPF_REG_0);
+	}
 
 	/* arg1: lea rdi, [rbp - stack_size] */
 	EMIT4(0x48, 0x8D, 0x7D, -stack_size);
@@ -1402,13 +1409,20 @@ static int invoke_bpf_prog(const struct btf_func_model *m, u8 **pprog,
 	if (mod_ret)
 		emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -8);
 
-	/* arg1: mov rdi, progs[i] */
-	emit_mov_imm64(&prog, BPF_REG_1, (long) p >> 32,
-		       (u32) (long) p);
-	/* arg2: mov rsi, rbx <- start time in nsec */
-	emit_mov_reg(&prog, true, BPF_REG_2, BPF_REG_6);
-	if (emit_call(&prog, __bpf_prog_exit, prog))
-		return -EINVAL;
+	if (p->aux->sleepable) {
+		/* arg1: mov rdi, rbx <- srcu idx */
+		emit_mov_reg(&prog, true, BPF_REG_1, BPF_REG_6);
+		if (emit_call(&prog, __bpf_prog_exit_sleepable, prog))
+			return -EINVAL;
+	} else {
+		/* arg1: mov rdi, progs[i] */
+		emit_mov_imm64(&prog, BPF_REG_1, (long) p >> 32,
+			       (u32) (long) p);
+		/* arg2: mov rsi, rbx <- start time in nsec */
+		emit_mov_reg(&prog, true, BPF_REG_2, BPF_REG_6);
+		if (emit_call(&prog, __bpf_prog_exit, prog))
+			return -EINVAL;
+	}
 
 	*pprog = prog;
 	return 0;
diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index efe8836b5c48..7b8291a96838 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -36,6 +36,7 @@ struct seq_operations;
 
 extern struct idr btf_idr;
 extern spinlock_t btf_idr_lock;
+extern struct srcu_struct bpf_srcu;
 
 /* map is generic key/value storage optionally accesible by eBPF programs */
 struct bpf_map_ops {
@@ -476,6 +477,8 @@ int arch_prepare_bpf_trampoline(void *image, void *image_end,
 /* these two functions are called from generated trampoline */
 u64 notrace __bpf_prog_enter(void);
 void notrace __bpf_prog_exit(struct bpf_prog *prog, u64 start);
+u64 notrace __bpf_prog_enter_sleepable(void);
+void notrace __bpf_prog_exit_sleepable(u64 idx);
 
 struct bpf_ksym {
 	unsigned long		 start;
@@ -668,6 +671,7 @@ struct bpf_prog_aux {
 	bool offload_requested;
 	bool attach_btf_trace; /* true if attaching to BTF-enabled raw tp */
 	bool func_proto_unreliable;
+	bool sleepable;
 	enum bpf_tramp_prog_type trampoline_prog_type;
 	struct bpf_trampoline *trampoline;
 	struct hlist_node tramp_hlist;
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index 54b93f8b49b8..cc08a2064d4e 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -329,6 +329,14 @@ enum bpf_link_type {
 /* The verifier internal test flag. Behavior is undefined */
 #define BPF_F_TEST_STATE_FREQ	(1U << 3)
 
+/* If BPF_F_SLEEPABLE is used in BPF_PROG_LOAD command, the verifier will
+ * restrict map and helper usage for such programs. Sleepable BPF programs can
+ * only be attached to hooks where kernel execution context allows sleeping.
+ * Such programs are allowed to use helpers that may sleep like
+ * bpf_copy_from_user().
+ */
+#define BPF_F_SLEEPABLE		(1U << 4)
+
 /* When BPF ldimm64's insn[0].src_reg != 0 then this can have
  * two extensions:
  *
diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
index 11584618e861..26b18b6a3dbc 100644
--- a/kernel/bpf/arraymap.c
+++ b/kernel/bpf/arraymap.c
@@ -393,6 +393,11 @@ static void array_map_free(struct bpf_map *map)
 	 */
 	synchronize_rcu();
 
+	/* arrays could have been used by both sleepable and non-sleepable bpf
+	 * progs. Make sure to wait for both prog types to finish executing.
+	 */
+	synchronize_srcu(&bpf_srcu);
+
 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 		bpf_array_free_percpu(array);
 
diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
index b4b288a3c3c9..b001957fdcbf 100644
--- a/kernel/bpf/hashtab.c
+++ b/kernel/bpf/hashtab.c
@@ -577,8 +577,8 @@ static void *__htab_map_lookup_elem(struct bpf_map *map, void *key)
 	struct htab_elem *l;
 	u32 hash, key_size;
 
-	/* Must be called with rcu_read_lock. */
-	WARN_ON_ONCE(!rcu_read_lock_held());
+	/* Must be called with s?rcu_read_lock. */
+	WARN_ON_ONCE(!rcu_read_lock_held() && !srcu_read_lock_held(&bpf_srcu));
 
 	key_size = map->key_size;
 
@@ -935,7 +935,7 @@ static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
 		/* unknown flags */
 		return -EINVAL;
 
-	WARN_ON_ONCE(!rcu_read_lock_held());
+	WARN_ON_ONCE(!rcu_read_lock_held() && !srcu_read_lock_held(&bpf_srcu));
 
 	key_size = map->key_size;
 
@@ -1026,7 +1026,7 @@ static int htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value,
 		/* unknown flags */
 		return -EINVAL;
 
-	WARN_ON_ONCE(!rcu_read_lock_held());
+	WARN_ON_ONCE(!rcu_read_lock_held() && !srcu_read_lock_held(&bpf_srcu));
 
 	key_size = map->key_size;
 
@@ -1214,7 +1214,7 @@ static int htab_map_delete_elem(struct bpf_map *map, void *key)
 	u32 hash, key_size;
 	int ret = -ENOENT;
 
-	WARN_ON_ONCE(!rcu_read_lock_held());
+	WARN_ON_ONCE(!rcu_read_lock_held() && !srcu_read_lock_held(&bpf_srcu));
 
 	key_size = map->key_size;
 
@@ -1246,7 +1246,7 @@ static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
 	u32 hash, key_size;
 	int ret = -ENOENT;
 
-	WARN_ON_ONCE(!rcu_read_lock_held());
+	WARN_ON_ONCE(!rcu_read_lock_held() && !srcu_read_lock_held(&bpf_srcu));
 
 	key_size = map->key_size;
 
@@ -1297,6 +1297,13 @@ static void htab_map_free(struct bpf_map *map)
 	 */
 	synchronize_rcu();
 
+	/* preallocated hash map could have been used by both sleepable and
+	 * non-sleepable bpf progs. Make sure to wait for both prog types
+	 * to finish executing.
+	 */
+	if (htab_is_prealloc(htab))
+		synchronize_srcu(&bpf_srcu);
+
 	/* some of free_htab_elem() callbacks for elements of this map may
 	 * not have executed. Wait for them.
 	 */
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index d13b804ff045..0820115315d1 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -1724,10 +1724,14 @@ static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
 	btf_put(prog->aux->btf);
 	bpf_prog_free_linfo(prog);
 
-	if (deferred)
-		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
-	else
+	if (deferred) {
+		if (prog->aux->sleepable)
+			call_srcu(&bpf_srcu, &prog->aux->rcu, __bpf_prog_put_rcu);
+		else
+			call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
+	} else {
 		__bpf_prog_put_rcu(&prog->aux->rcu);
+	}
 }
 
 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
@@ -2092,6 +2096,7 @@ static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
 				 BPF_F_ANY_ALIGNMENT |
 				 BPF_F_TEST_STATE_FREQ |
+				 BPF_F_SLEEPABLE |
 				 BPF_F_TEST_RND_HI32))
 		return -EINVAL;
 
@@ -2147,6 +2152,7 @@ static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
 	}
 
 	prog->aux->offload_requested = !!attr->prog_ifindex;
+	prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
 
 	err = security_bpf_prog_alloc(prog->aux);
 	if (err)
diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
index 9be85aa4ec5f..d35d26e9693d 100644
--- a/kernel/bpf/trampoline.c
+++ b/kernel/bpf/trampoline.c
@@ -205,13 +205,21 @@ static int bpf_trampoline_update(struct bpf_trampoline *tr)
 	    tprogs[BPF_TRAMP_MODIFY_RETURN].nr_progs)
 		flags = BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_SKIP_FRAME;
 
+	/* the same trampoline can hold both sleepable and non-sleepable progs.
+	 * synchronize_srcu() is cheap when sleepable progs are not running, so
+	 * just call it here to make sure all sleepable progs finish executing.
+	 * Then call synchronize_rcu_tasks() to make sure that the rest of
+	 * generated tramopline assembly finishes too before updating
+	 * trampoline.
+	 */
+	synchronize_srcu(&bpf_srcu);
+
 	/* Though the second half of trampoline page is unused a task could be
 	 * preempted in the middle of the first half of trampoline and two
 	 * updates to trampoline would change the code from underneath the
 	 * preempted task. Hence wait for tasks to voluntarily schedule or go
 	 * to userspace.
 	 */
-
 	synchronize_rcu_tasks();
 
 	err = arch_prepare_bpf_trampoline(new_image, new_image + PAGE_SIZE / 2,
@@ -344,6 +352,11 @@ void bpf_trampoline_put(struct bpf_trampoline *tr)
 	if (WARN_ON_ONCE(!hlist_empty(&tr->progs_hlist[BPF_TRAMP_FEXIT])))
 		goto out;
 	bpf_image_ksym_del(&tr->ksym);
+	/* This code will be executed when all bpf progs (both sleepable and
+	 * non-sleepable) went through
+	 * bpf_prog_put()->call_s?rcu()->bpf_prog_free_deferred().
+	 * Hence no need for another synchronize_srcu(&bpf_srcu) here.
+	 */
 	/* wait for tasks to get out of trampoline before freeing it */
 	synchronize_rcu_tasks();
 	bpf_jit_free_exec(tr->image);
@@ -394,6 +407,23 @@ void notrace __bpf_prog_exit(struct bpf_prog *prog, u64 start)
 	rcu_read_unlock();
 }
 
+/* when bpf_srcu lock is held it means that some sleepable bpf program is
+ * running. Those programs can use bpf arrays and preallocated hash maps. These
+ * map types are waiting on programs to complete via
+ * synchronize_srcu(&bpf_srcu);
+ */
+struct srcu_struct bpf_srcu;
+
+u64 notrace __bpf_prog_enter_sleepable(void)
+{
+	return srcu_read_lock(&bpf_srcu);
+}
+
+void notrace __bpf_prog_exit_sleepable(u64 idx)
+{
+	srcu_read_unlock(&bpf_srcu, idx);
+}
+
 int __weak
 arch_prepare_bpf_trampoline(void *image, void *image_end,
 			    const struct btf_func_model *m, u32 flags,
@@ -409,6 +439,7 @@ static int __init init_trampolines(void)
 
 	for (i = 0; i < TRAMPOLINE_TABLE_SIZE; i++)
 		INIT_HLIST_HEAD(&trampoline_table[i]);
+	init_srcu_struct(&bpf_srcu);
 	return 0;
 }
 late_initcall(init_trampolines);
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index d2e27dba4ac6..01c7d3634151 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -8922,6 +8922,23 @@ static int check_map_prog_compatibility(struct bpf_verifier_env *env,
 		return -EINVAL;
 	}
 
+	if (prog->aux->sleepable)
+		switch (map->map_type) {
+		case BPF_MAP_TYPE_HASH:
+		case BPF_MAP_TYPE_LRU_HASH:
+		case BPF_MAP_TYPE_ARRAY:
+			if (!is_preallocated_map(map)) {
+				verbose(env,
+					"Sleepable programs can only use preallocated hash maps\n");
+				return -EINVAL;
+			}
+			break;
+		default:
+			verbose(env,
+				"Sleepable programs can only use array and hash maps\n");
+			return -EINVAL;
+		}
+
 	return 0;
 }
 
@@ -10522,22 +10539,13 @@ static int check_struct_ops_btf_id(struct bpf_verifier_env *env)
 }
 #define SECURITY_PREFIX "security_"
 
-static int check_attach_modify_return(struct bpf_verifier_env *env)
+static int check_attach_modify_return(struct bpf_prog *prog, unsigned long addr)
 {
-	struct bpf_prog *prog = env->prog;
-	unsigned long addr = (unsigned long) prog->aux->trampoline->func.addr;
-
-	/* This is expected to be cleaned up in the future with the KRSI effort
-	 * introducing the LSM_HOOK macro for cleaning up lsm_hooks.h.
-	 */
 	if (within_error_injection_list(addr) ||
 	    !strncmp(SECURITY_PREFIX, prog->aux->attach_func_name,
 		     sizeof(SECURITY_PREFIX) - 1))
 		return 0;
 
-	verbose(env, "fmod_ret attach_btf_id %u (%s) is not modifiable\n",
-		prog->aux->attach_btf_id, prog->aux->attach_func_name);
-
 	return -EINVAL;
 }
 
@@ -10558,6 +10566,12 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
 	long addr;
 	u64 key;
 
+	if (prog->aux->sleepable && prog->type != BPF_PROG_TYPE_TRACING &&
+	    prog->type != BPF_PROG_TYPE_LSM) {
+		verbose(env, "Only fentry/fexit/fmod_ret and lsm programs can be sleepable\n");
+		return -EINVAL;
+	}
+
 	if (prog->type == BPF_PROG_TYPE_STRUCT_OPS)
 		return check_struct_ops_btf_id(env);
 
@@ -10765,11 +10779,27 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
 				goto out;
 			}
 		}
+
+		if (prog->expected_attach_type == BPF_MODIFY_RETURN) {
+			ret = check_attach_modify_return(prog, addr);
+			if (ret)
+				verbose(env, "%s() is not modifiable\n",
+					prog->aux->attach_func_name);
+		} else if (prog->aux->sleepable && prog->type == BPF_PROG_TYPE_TRACING) {
+			/* fentry/fexit progs can be sleepable only if they are
+			 * attached to ALLOW_ERROR_INJECTION or security_*() funcs.
+			 * LSM progs check that they are attached to bpf_lsm_*() funcs
+			 * which are sleepable too.
+			 */
+			ret = check_attach_modify_return(prog, addr);
+			if (ret)
+				verbose(env, "%s is not sleepable\n",
+					prog->aux->attach_func_name);
+		}
+		if (ret)
+			goto out;
 		tr->func.addr = (void *)addr;
 		prog->aux->trampoline = tr;
-
-		if (prog->expected_attach_type == BPF_MODIFY_RETURN)
-			ret = check_attach_modify_return(env);
 out:
 		mutex_unlock(&tr->mutex);
 		if (ret)
diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index 54b93f8b49b8..cc08a2064d4e 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h
@@ -329,6 +329,14 @@ enum bpf_link_type {
 /* The verifier internal test flag. Behavior is undefined */
 #define BPF_F_TEST_STATE_FREQ	(1U << 3)
 
+/* If BPF_F_SLEEPABLE is used in BPF_PROG_LOAD command, the verifier will
+ * restrict map and helper usage for such programs. Sleepable BPF programs can
+ * only be attached to hooks where kernel execution context allows sleeping.
+ * Such programs are allowed to use helpers that may sleep like
+ * bpf_copy_from_user().
+ */
+#define BPF_F_SLEEPABLE		(1U << 4)
+
 /* When BPF ldimm64's insn[0].src_reg != 0 then this can have
  * two extensions:
  *
-- 
2.23.0

Re: [PATCH bpf-next 1/3] bpf: Introduce sleepable BPF programs

[KP Singh]

On 27-May 22:33, Alexei Starovoitov wrote:
> From: Alexei Starovoitov <ast@kernel.org>
> 
> Introduce sleepable BPF programs that can request such property for themselves
> via BPF_F_SLEEPABLE flag at program load time. In such case they will be able
> to use helpers like bpf_copy_from_user() that might sleep. At present only
> fentry/fexit/fmod_ret and lsm programs can request to be sleepable and only
> when they are attached to kernel functions that are known to allow sleeping.
> 
> The non-sleepable programs are relying on implicit rcu_read_lock() and
> migrate_disable() to protect life time of programs, maps that they use and
> per-cpu kernel structures used to pass info between bpf programs and the
> kernel. The sleepable programs cannot be enclosed into rcu_read_lock().
> migrate_disable() maps to preempt_disable() in non-RT kernels, so the progs
> should not be enclosed in migrate_disable() as well. Therefore bpf_srcu is used
> to protect the life time of sleepable progs.
> 
> There are many networking and tracing program types. In many cases the
> 'struct bpf_prog *' pointer itself is rcu protected within some other kernel
> data structure and the kernel code is using rcu_dereference() to load that
> program pointer and call BPF_PROG_RUN() on it. All these cases are not touched.
> Instead sleepable bpf programs are allowed with bpf trampoline only. The
> program pointers are hard-coded into generated assembly of bpf trampoline and
> synchronize_srcu(&bpf_srcu) is used to protect the life time of the program.
> The same trampoline can hold both sleepable and non-sleepable progs.
> 
> When bpf_srcu lock is held it means that some sleepable bpf program is running
> from bpf trampoline. Those programs can use bpf arrays and preallocated hash/lru
> maps. These map types are waiting on programs to complete via
> synchronize_srcu(&bpf_srcu);
> 
> Updates to trampoline now has to do synchronize_srcu + synchronize_rcu_tasks
> to wait for sleepable progs to finish and for trampoline assembly to finish.
> 
> In the future srcu will be replaced with upcoming rcu_trace.
> That will complete the first step of introducing sleepable progs.
> 
> After that dynamically allocated hash maps can be allowed. All map elements
> would have to be srcu protected instead of normal rcu.
> per-cpu maps will be allowed. Either via the following pattern:
> void *elem = bpf_map_lookup_elem(map, key);
> if (elem) {
>    // access elem
>    bpf_map_release_elem(map, elem);
> }
> where modified lookup() helper will do migrate_disable() and
> new bpf_map_release_elem() will do corresponding migrate_enable().
> Or explicit bpf_migrate_disable/enable() helpers will be introduced.
> 
> Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Thanks! This will be really helpful for LSM programs.

Acked-by: KP Singh <kpsingh@google.com>

> ---
>  arch/x86/net/bpf_jit_comp.c    | 36 +++++++++++++++-------
>  include/linux/bpf.h            |  4 +++
>  include/uapi/linux/bpf.h       |  8 +++++
>  kernel/bpf/arraymap.c          |  5 +++
>  kernel/bpf/hashtab.c           | 19 ++++++++----
>  kernel/bpf/syscall.c           | 12 ++++++--
>  kernel/bpf/trampoline.c        | 33 +++++++++++++++++++-
>  kernel/bpf/verifier.c          | 56 ++++++++++++++++++++++++++--------
>  tools/include/uapi/linux/bpf.h |  8 +++++
>  9 files changed, 147 insertions(+), 34 deletions(-)

[...]

> +			if (ret)
> +				verbose(env, "%s() is not modifiable\n",
> +					prog->aux->attach_func_name);
> +		} else if (prog->aux->sleepable && prog->type == BPF_PROG_TYPE_TRACING) {
> +			/* fentry/fexit progs can be sleepable only if they are
> +			 * attached to ALLOW_ERROR_INJECTION or security_*() funcs.
> +			 * LSM progs check that they are attached to bpf_lsm_*() funcs
> +			 * which are sleepable too.

I know of one LSM hook which is not sleepable and is executed in an
RCU callback i.e. task_free. I don't think t's a problem to run under
SRCU for that (I tried it and it does not cause any issues).

We can add a blacklisting mechanism later for the sleepable flags or
just the sleeping helpers (based on some of the work going on to
whitelist functions for helper usage).

- KP

> +			 */
> +			ret = check_attach_modify_return(prog, addr);
> +			if (ret)
> +				verbose(env, "%s is not sleepable\n",

[...]

>   * two extensions:
>   *
> -- 
> 2.23.0
> 

Re: [PATCH bpf-next 1/3] bpf: Introduce sleepable BPF programs

[Alexei Starovoitov]

On Fri, May 29, 2020 at 12:12:27AM +0200, KP Singh wrote:
> > +			if (ret)
> > +				verbose(env, "%s() is not modifiable\n",
> > +					prog->aux->attach_func_name);
> > +		} else if (prog->aux->sleepable && prog->type == BPF_PROG_TYPE_TRACING) {
> > +			/* fentry/fexit progs can be sleepable only if they are
> > +			 * attached to ALLOW_ERROR_INJECTION or security_*() funcs.
> > +			 * LSM progs check that they are attached to bpf_lsm_*() funcs
> > +			 * which are sleepable too.
> 
> I know of one LSM hook which is not sleepable and is executed in an
> RCU callback i.e. task_free. I don't think t's a problem to run under
> SRCU for that (I tried it and it does not cause any issues).
> 
> We can add a blacklisting mechanism later for the sleepable flags or
> just the sleeping helpers (based on some of the work going on to
> whitelist functions for helper usage).

Good catch. *_task_free() are not sleepable. I'll introduce a simple
blacklist for now. Since I'm not adding actual sleeping helpers in this
patch set nothing is broken, but it will give us the base to build stuff on top.

[PATCH bpf-next 2/3] libbpf: support sleepable progs

[Alexei Starovoitov]

From: Alexei Starovoitov <ast@kernel.org>

Pass request to load program as sleepable via ".s" suffix in the section name.
If it happens in the future that all map types and helpers are allowed with
BPF_F_SLEEPABLE flag "fmod_ret/" and "lsm/" can be aliased to "fmod_ret.s/" and
"lsm.s/" to make all lsm and fmod_ret programs sleepable by default. The fentry
and fexit programs would always need to have sleepable vs non-sleepable
distinction, since not all fentry/fexit progs will be attached to sleepable
kernel functions.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
---
 tools/lib/bpf/libbpf.c | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)

diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
index 5d60de6fd818..d5dc9a1abea3 100644
--- a/tools/lib/bpf/libbpf.c
+++ b/tools/lib/bpf/libbpf.c
@@ -209,6 +209,7 @@ struct bpf_sec_def {
 	bool is_exp_attach_type_optional;
 	bool is_attachable;
 	bool is_attach_btf;
+	bool is_sleepable;
 	attach_fn_t attach_fn;
 };
 
@@ -5451,6 +5452,8 @@ __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
 			/* couldn't guess, but user might manually specify */
 			continue;
 
+		if (prog->sec_def->is_sleepable)
+			prog->prog_flags |= BPF_F_SLEEPABLE;
 		bpf_program__set_type(prog, prog->sec_def->prog_type);
 		bpf_program__set_expected_attach_type(prog,
 				prog->sec_def->expected_attach_type);
@@ -6646,6 +6649,21 @@ static const struct bpf_sec_def section_defs[] = {
 		.expected_attach_type = BPF_TRACE_FEXIT,
 		.is_attach_btf = true,
 		.attach_fn = attach_trace),
+	SEC_DEF("fentry.s/", TRACING,
+		.expected_attach_type = BPF_TRACE_FENTRY,
+		.is_attach_btf = true,
+		.is_sleepable = true,
+		.attach_fn = attach_trace),
+	SEC_DEF("fmod_ret.s/", TRACING,
+		.expected_attach_type = BPF_MODIFY_RETURN,
+		.is_attach_btf = true,
+		.is_sleepable = true,
+		.attach_fn = attach_trace),
+	SEC_DEF("fexit.s/", TRACING,
+		.expected_attach_type = BPF_TRACE_FEXIT,
+		.is_attach_btf = true,
+		.is_sleepable = true,
+		.attach_fn = attach_trace),
 	SEC_DEF("freplace/", EXT,
 		.is_attach_btf = true,
 		.attach_fn = attach_trace),
@@ -6653,6 +6671,11 @@ static const struct bpf_sec_def section_defs[] = {
 		.is_attach_btf = true,
 		.expected_attach_type = BPF_LSM_MAC,
 		.attach_fn = attach_lsm),
+	SEC_DEF("lsm.s/", LSM,
+		.is_attach_btf = true,
+		.is_sleepable = true,
+		.expected_attach_type = BPF_LSM_MAC,
+		.attach_fn = attach_lsm),
 	SEC_DEF("iter/", TRACING,
 		.expected_attach_type = BPF_TRACE_ITER,
 		.is_attach_btf = true,
@@ -7292,7 +7315,7 @@ int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
 
 		prog->prog_ifindex = attr->ifindex;
 		prog->log_level = attr->log_level;
-		prog->prog_flags = attr->prog_flags;
+		prog->prog_flags |= attr->prog_flags;
 		if (!first_prog)
 			first_prog = prog;
 	}
-- 
2.23.0

Re: [PATCH bpf-next 2/3] libbpf: support sleepable progs

[KP Singh]

On 27-May 22:33, Alexei Starovoitov wrote:
> From: Alexei Starovoitov <ast@kernel.org>
> 
> Pass request to load program as sleepable via ".s" suffix in the section name.
> If it happens in the future that all map types and helpers are allowed with
> BPF_F_SLEEPABLE flag "fmod_ret/" and "lsm/" can be aliased to "fmod_ret.s/" and
> "lsm.s/" to make all lsm and fmod_ret programs sleepable by default. The fentry
> and fexit programs would always need to have sleepable vs non-sleepable
> distinction, since not all fentry/fexit progs will be attached to sleepable
> kernel functions.
> 
> Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Acked-by: KP Singh <kpsingh@google.com>

> ---
>  tools/lib/bpf/libbpf.c | 25 ++++++++++++++++++++++++-
>  1 file changed, 24 insertions(+), 1 deletion(-)

[...]

> -		prog->prog_flags = attr->prog_flags;
> +		prog->prog_flags |= attr->prog_flags;
>  		if (!first_prog)
>  			first_prog = prog;
>  	}
> -- 
> 2.23.0
> 

[PATCH bpf-next 3/3] selftests/bpf: basic sleepable tests

[Alexei Starovoitov]

From: Alexei Starovoitov <ast@kernel.org>

Modify few tests to sanity test sleepable bpf functionality.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
---
 tools/testing/selftests/bpf/bench.c             |  2 ++
 .../selftests/bpf/benchs/bench_trigger.c        | 17 +++++++++++++++++
 tools/testing/selftests/bpf/progs/lsm.c         |  4 ++--
 .../testing/selftests/bpf/progs/trigger_bench.c |  7 +++++++
 4 files changed, 28 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/bpf/bench.c b/tools/testing/selftests/bpf/bench.c
index 14390689ef90..f6a75cd47f01 100644
--- a/tools/testing/selftests/bpf/bench.c
+++ b/tools/testing/selftests/bpf/bench.c
@@ -309,6 +309,7 @@ extern const struct bench bench_trig_tp;
 extern const struct bench bench_trig_rawtp;
 extern const struct bench bench_trig_kprobe;
 extern const struct bench bench_trig_fentry;
+extern const struct bench bench_trig_fentry_sleep;
 extern const struct bench bench_trig_fmodret;
 
 static const struct bench *benchs[] = {
@@ -326,6 +327,7 @@ static const struct bench *benchs[] = {
 	&bench_trig_rawtp,
 	&bench_trig_kprobe,
 	&bench_trig_fentry,
+	&bench_trig_fentry_sleep,
 	&bench_trig_fmodret,
 };
 
diff --git a/tools/testing/selftests/bpf/benchs/bench_trigger.c b/tools/testing/selftests/bpf/benchs/bench_trigger.c
index 49c22832f216..2a0b6c9885a4 100644
--- a/tools/testing/selftests/bpf/benchs/bench_trigger.c
+++ b/tools/testing/selftests/bpf/benchs/bench_trigger.c
@@ -90,6 +90,12 @@ static void trigger_fentry_setup()
 	attach_bpf(ctx.skel->progs.bench_trigger_fentry);
 }
 
+static void trigger_fentry_sleep_setup()
+{
+	setup_ctx();
+	attach_bpf(ctx.skel->progs.bench_trigger_fentry_sleep);
+}
+
 static void trigger_fmodret_setup()
 {
 	setup_ctx();
@@ -155,6 +161,17 @@ const struct bench bench_trig_fentry = {
 	.report_final = hits_drops_report_final,
 };
 
+const struct bench bench_trig_fentry_sleep = {
+	.name = "trig-fentry-sleep",
+	.validate = trigger_validate,
+	.setup = trigger_fentry_sleep_setup,
+	.producer_thread = trigger_producer,
+	.consumer_thread = trigger_consumer,
+	.measure = trigger_measure,
+	.report_progress = hits_drops_report_progress,
+	.report_final = hits_drops_report_final,
+};
+
 const struct bench bench_trig_fmodret = {
 	.name = "trig-fmodret",
 	.validate = trigger_validate,
diff --git a/tools/testing/selftests/bpf/progs/lsm.c b/tools/testing/selftests/bpf/progs/lsm.c
index b4598d4bc4f7..55815d0cc5fb 100644
--- a/tools/testing/selftests/bpf/progs/lsm.c
+++ b/tools/testing/selftests/bpf/progs/lsm.c
@@ -15,7 +15,7 @@ int monitored_pid = 0;
 int mprotect_count = 0;
 int bprm_count = 0;
 
-SEC("lsm/file_mprotect")
+SEC("lsm.s/file_mprotect")
 int BPF_PROG(test_int_hook, struct vm_area_struct *vma,
 	     unsigned long reqprot, unsigned long prot, int ret)
 {
@@ -36,7 +36,7 @@ int BPF_PROG(test_int_hook, struct vm_area_struct *vma,
 	return ret;
 }
 
-SEC("lsm/bprm_committed_creds")
+SEC("lsm.s/bprm_committed_creds")
 int BPF_PROG(test_void_hook, struct linux_binprm *bprm)
 {
 	__u32 pid = bpf_get_current_pid_tgid() >> 32;
diff --git a/tools/testing/selftests/bpf/progs/trigger_bench.c b/tools/testing/selftests/bpf/progs/trigger_bench.c
index 8b36b6640e7e..9a4d09590b3d 100644
--- a/tools/testing/selftests/bpf/progs/trigger_bench.c
+++ b/tools/testing/selftests/bpf/progs/trigger_bench.c
@@ -39,6 +39,13 @@ int bench_trigger_fentry(void *ctx)
 	return 0;
 }
 
+SEC("fentry.s/__x64_sys_getpgid")
+int bench_trigger_fentry_sleep(void *ctx)
+{
+	__sync_add_and_fetch(&hits, 1);
+	return 0;
+}
+
 SEC("fmod_ret/__x64_sys_getpgid")
 int bench_trigger_fmodret(void *ctx)
 {
-- 
2.23.0

Re: [PATCH bpf-next 3/3] selftests/bpf: basic sleepable tests

[KP Singh]

On 27-May 22:33, Alexei Starovoitov wrote:
> From: Alexei Starovoitov <ast@kernel.org>
> 
> Modify few tests to sanity test sleepable bpf functionality.
> 
> Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Acked-by: KP Singh <kpsingh@google.com>

> ---
>  tools/testing/selftests/bpf/bench.c             |  2 ++
>  .../selftests/bpf/benchs/bench_trigger.c        | 17 +++++++++++++++++
>  tools/testing/selftests/bpf/progs/lsm.c         |  4 ++--
>  .../testing/selftests/bpf/progs/trigger_bench.c |  7 +++++++
>  4 files changed, 28 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/testing/selftests/bpf/bench.c b/tools/testing/selftests/bpf/bench.c
>  

[...]

> -SEC("lsm/bprm_committed_creds")
> +SEC("lsm.s/bprm_committed_creds")
>  int BPF_PROG(test_void_hook, struct linux_binprm *bprm)
>  {
>  	__u32 pid = bpf_get_current_pid_tgid() >> 32;
> diff --git a/tools/testing/selftests/bpf/progs/trigger_bench.c b/tools/testing/selftests/bpf/progs/trigger_bench.c
> index 8b36b6640e7e..9a4d09590b3d 100644
> --- a/tools/testing/selftests/bpf/progs/trigger_bench.c
> +++ b/tools/testing/selftests/bpf/progs/trigger_bench.c
> @@ -39,6 +39,13 @@ int bench_trigger_fentry(void *ctx)
>  	return 0;
>  }
>  
> +SEC("fentry.s/__x64_sys_getpgid")
> +int bench_trigger_fentry_sleep(void *ctx)
> +{
> +	__sync_add_and_fetch(&hits, 1);
> +	return 0;
> +}
> +
>  SEC("fmod_ret/__x64_sys_getpgid")
>  int bench_trigger_fmodret(void *ctx)
>  {
> -- 
> 2.23.0
>