Merge branch 'bpf: BTF support for ksyms'

Hao Luo says:

====================
v3 -> v4:
 - Rebasing
 - Cast bpf_[per|this]_cpu_ptr's parameter to void __percpu * before
   passing into per_cpu_ptr.

v2 -> v3:
 - Rename functions and variables in verifier for better readability.
 - Stick to logging message convention in libbpf.
 - Move bpf_per_cpu_ptr and bpf_this_cpu_ptr from trace-specific
   helper set to base helper set.
 - More specific test in ksyms_btf.
 - Fix return type cast in bpf_*_cpu_ptr.
 - Fix btf leak in ksyms_btf selftest.
 - Fix return error code for kallsyms_find().

v1 -> v2:
 - Move check_pseudo_btf_id from check_ld_imm() to
   replace_map_fd_with_map_ptr() and rename the latter.
 - Add bpf_this_cpu_ptr().
 - Use bpf_core_types_are_compat() in libbpf.c for checking type
   compatibility.
 - Rewrite typed ksym extern type in BTF with int to save space.
 - Minor revision of bpf_per_cpu_ptr()'s comments.
 - Avoid using long in tests that use skeleton.
 - Refactored test_ksyms.c by moving kallsyms_find() to trace_helpers.c
 - Fold the patches that sync include/linux/uapi and
   tools/include/linux/uapi.

rfc -> v1:
 - Encode VAR's btf_id for PSEUDO_BTF_ID.
 - More checks in verifier. Checking the btf_id passed as
   PSEUDO_BTF_ID is valid VAR, its name and type.
 - Checks in libbpf on type compatibility of ksyms.
 - Add bpf_per_cpu_ptr() to access kernel percpu vars. Introduced
   new ARG and RET types for this helper.

This patch series extends the previously added __ksym externs with
btf support.

Right now the __ksym externs are treated as pure 64-bit scalar value.
Libbpf replaces ld_imm64 insn of __ksym by its kernel address at load
time. This patch series extend those externs with their btf info. Note
that btf support for __ksym must come with the kernel btf that has
VARs encoded to work properly. The corresponding chagnes in pahole
is available at [1] (with a fix at [2] for gcc 4.9+).

The first 3 patches in this series add support for general kernel
global variables, which include verifier checking (01/06), libpf
support (02/06) and selftests for getting typed ksym extern's kernel
address (03/06).

The next 3 patches extends that capability further by introducing
helpers bpf_per_cpu_ptr() and bpf_this_cpu_ptr(), which allows accessing
kernel percpu variables correctly (04/06 and 05/06).

The tests of this feature were performed against pahole that is extended
with [1] and [2]. For kernel BTF that does not have VARs encoded, the
selftests will be skipped.

[1] https://git.kernel.org/pub/scm/devel/pahole/pahole.git/commit/?id=f3d9054ba8ff1df0fc44e507e3a01c0964cabd42
[2] https://www.spinics.net/lists/dwarves/msg00451.html
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>

3 files changed, 208 insertions, 39 deletions

diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index 4d0ee7839fdb..ed7d02e8bc93 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c
@@ -188,11 +188,6 @@
 	     i < btf_type_vlen(struct_type);				\
 	     i++, member++)
 
-#define for_each_vsi(i, struct_type, member)			\
-	for (i = 0, member = btf_type_var_secinfo(struct_type);	\
-	     i < btf_type_vlen(struct_type);			\
-	     i++, member++)
-
 #define for_each_vsi_from(i, from, struct_type, member)				\
 	for (i = from, member = btf_type_var_secinfo(struct_type) + from;	\
 	     i < btf_type_vlen(struct_type);					\
@@ -440,16 +435,6 @@ static bool btf_type_nosize_or_null(const struct btf_type *t)
 	return !t || btf_type_nosize(t);
 }
 
-/* union is only a special case of struct:
- * all its offsetof(member) == 0
- */
-static bool btf_type_is_struct(const struct btf_type *t)
-{
-	u8 kind = BTF_INFO_KIND(t->info);
-
-	return kind == BTF_KIND_STRUCT || kind == BTF_KIND_UNION;
-}
-
 static bool __btf_type_is_struct(const struct btf_type *t)
 {
 	return BTF_INFO_KIND(t->info) == BTF_KIND_STRUCT;
@@ -460,11 +445,6 @@ static bool btf_type_is_array(const struct btf_type *t)
 	return BTF_INFO_KIND(t->info) == BTF_KIND_ARRAY;
 }
 
-static bool btf_type_is_var(const struct btf_type *t)
-{
-	return BTF_INFO_KIND(t->info) == BTF_KIND_VAR;
-}
-
 static bool btf_type_is_datasec(const struct btf_type *t)
 {
 	return BTF_INFO_KIND(t->info) == BTF_KIND_DATASEC;
@@ -613,11 +593,6 @@ static const struct btf_var *btf_type_var(const struct btf_type *t)
 	return (const struct btf_var *)(t + 1);
 }
 
-static const struct btf_var_secinfo *btf_type_var_secinfo(const struct btf_type *t)
-{
-	return (const struct btf_var_secinfo *)(t + 1);
-}
-
 static const struct btf_kind_operations *btf_type_ops(const struct btf_type *t)
 {
 	return kind_ops[BTF_INFO_KIND(t->info)];
diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
index e825441781ab..25520f5eeaf6 100644
--- a/kernel/bpf/helpers.c
+++ b/kernel/bpf/helpers.c
@@ -623,6 +623,34 @@ const struct bpf_func_proto bpf_copy_from_user_proto = {
 	.arg3_type	= ARG_ANYTHING,
 };
 
+BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
+{
+	if (cpu >= nr_cpu_ids)
+		return (unsigned long)NULL;
+
+	return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
+}
+
+const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
+	.func		= bpf_per_cpu_ptr,
+	.gpl_only	= false,
+	.ret_type	= RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL,
+	.arg1_type	= ARG_PTR_TO_PERCPU_BTF_ID,
+	.arg2_type	= ARG_ANYTHING,
+};
+
+BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
+{
+	return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
+}
+
+const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
+	.func		= bpf_this_cpu_ptr,
+	.gpl_only	= false,
+	.ret_type	= RET_PTR_TO_MEM_OR_BTF_ID,
+	.arg1_type	= ARG_PTR_TO_PERCPU_BTF_ID,
+};
+
 const struct bpf_func_proto bpf_get_current_task_proto __weak;
 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
@@ -689,6 +717,10 @@ bpf_base_func_proto(enum bpf_func_id func_id)
 		return &bpf_snprintf_btf_proto;
 	case BPF_FUNC_jiffies64:
 		return &bpf_jiffies64_proto;
+	case BPF_FUNC_bpf_per_cpu_ptr:
+		return &bpf_per_cpu_ptr_proto;
+	case BPF_FUNC_bpf_this_cpu_ptr:
+		return &bpf_this_cpu_ptr_proto;
 	default:
 		break;
 	}
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 015a1c074b6b..d9dbf271ebab 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -238,6 +238,8 @@ struct bpf_call_arg_meta {
 	u64 msize_max_value;
 	int ref_obj_id;
 	int func_id;
+	u32 btf_id;
+	u32 ret_btf_id;
 };
 
 struct btf *btf_vmlinux;
@@ -517,6 +519,7 @@ static const char * const reg_type_str[] = {
 	[PTR_TO_XDP_SOCK]	= "xdp_sock",
 	[PTR_TO_BTF_ID]		= "ptr_",
 	[PTR_TO_BTF_ID_OR_NULL]	= "ptr_or_null_",
+	[PTR_TO_PERCPU_BTF_ID]	= "percpu_ptr_",
 	[PTR_TO_MEM]		= "mem",
 	[PTR_TO_MEM_OR_NULL]	= "mem_or_null",
 	[PTR_TO_RDONLY_BUF]	= "rdonly_buf",
@@ -583,7 +586,9 @@ static void print_verifier_state(struct bpf_verifier_env *env,
 			/* reg->off should be 0 for SCALAR_VALUE */
 			verbose(env, "%lld", reg->var_off.value + reg->off);
 		} else {
-			if (t == PTR_TO_BTF_ID || t == PTR_TO_BTF_ID_OR_NULL)
+			if (t == PTR_TO_BTF_ID ||
+			    t == PTR_TO_BTF_ID_OR_NULL ||
+			    t == PTR_TO_PERCPU_BTF_ID)
 				verbose(env, "%s", kernel_type_name(reg->btf_id));
 			verbose(env, "(id=%d", reg->id);
 			if (reg_type_may_be_refcounted_or_null(t))
@@ -2204,6 +2209,7 @@ static bool is_spillable_regtype(enum bpf_reg_type type)
 	case PTR_TO_RDONLY_BUF_OR_NULL:
 	case PTR_TO_RDWR_BUF:
 	case PTR_TO_RDWR_BUF_OR_NULL:
+	case PTR_TO_PERCPU_BTF_ID:
 		return true;
 	default:
 		return false;
@@ -4017,6 +4023,7 @@ static const struct bpf_reg_types alloc_mem_types = { .types = { PTR_TO_MEM } };
 static const struct bpf_reg_types const_map_ptr_types = { .types = { CONST_PTR_TO_MAP } };
 static const struct bpf_reg_types btf_ptr_types = { .types = { PTR_TO_BTF_ID } };
 static const struct bpf_reg_types spin_lock_types = { .types = { PTR_TO_MAP_VALUE } };
+static const struct bpf_reg_types percpu_btf_ptr_types = { .types = { PTR_TO_PERCPU_BTF_ID } };
 
 static const struct bpf_reg_types *compatible_reg_types[__BPF_ARG_TYPE_MAX] = {
 	[ARG_PTR_TO_MAP_KEY]		= &map_key_value_types,
@@ -4042,6 +4049,7 @@ static const struct bpf_reg_types *compatible_reg_types[__BPF_ARG_TYPE_MAX] = {
 	[ARG_PTR_TO_ALLOC_MEM_OR_NULL]	= &alloc_mem_types,
 	[ARG_PTR_TO_INT]		= &int_ptr_types,
 	[ARG_PTR_TO_LONG]		= &int_ptr_types,
+	[ARG_PTR_TO_PERCPU_BTF_ID]	= &percpu_btf_ptr_types,
 };
 
 static int check_reg_type(struct bpf_verifier_env *env, u32 regno,
@@ -4205,6 +4213,12 @@ skip_type_check:
 		err = check_helper_mem_access(env, regno,
 					      meta->map_ptr->value_size, false,
 					      meta);
+	} else if (arg_type == ARG_PTR_TO_PERCPU_BTF_ID) {
+		if (!reg->btf_id) {
+			verbose(env, "Helper has invalid btf_id in R%d\n", regno);
+			return -EACCES;
+		}
+		meta->ret_btf_id = reg->btf_id;
 	} else if (arg_type == ARG_PTR_TO_SPIN_LOCK) {
 		if (meta->func_id == BPF_FUNC_spin_lock) {
 			if (process_spin_lock(env, regno, true))
@@ -5114,6 +5128,35 @@ static int check_helper_call(struct bpf_verifier_env *env, int func_id, int insn
 		regs[BPF_REG_0].type = PTR_TO_MEM_OR_NULL;
 		regs[BPF_REG_0].id = ++env->id_gen;
 		regs[BPF_REG_0].mem_size = meta.mem_size;
+	} else if (fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL ||
+		   fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID) {
+		const struct btf_type *t;
+
+		mark_reg_known_zero(env, regs, BPF_REG_0);
+		t = btf_type_skip_modifiers(btf_vmlinux, meta.ret_btf_id, NULL);
+		if (!btf_type_is_struct(t)) {
+			u32 tsize;
+			const struct btf_type *ret;
+			const char *tname;
+
+			/* resolve the type size of ksym. */
+			ret = btf_resolve_size(btf_vmlinux, t, &tsize);
+			if (IS_ERR(ret)) {
+				tname = btf_name_by_offset(btf_vmlinux, t->name_off);
+				verbose(env, "unable to resolve the size of type '%s': %ld\n",
+					tname, PTR_ERR(ret));
+				return -EINVAL;
+			}
+			regs[BPF_REG_0].type =
+				fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID ?
+				PTR_TO_MEM : PTR_TO_MEM_OR_NULL;
+			regs[BPF_REG_0].mem_size = tsize;
+		} else {
+			regs[BPF_REG_0].type =
+				fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID ?
+				PTR_TO_BTF_ID : PTR_TO_BTF_ID_OR_NULL;
+			regs[BPF_REG_0].btf_id = meta.ret_btf_id;
+		}
 	} else if (fn->ret_type == RET_PTR_TO_BTF_ID_OR_NULL) {
 		int ret_btf_id;
 
@@ -7488,6 +7531,7 @@ static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
 {
 	struct bpf_insn_aux_data *aux = cur_aux(env);
 	struct bpf_reg_state *regs = cur_regs(env);
+	struct bpf_reg_state *dst_reg;
 	struct bpf_map *map;
 	int err;
 
@@ -7504,25 +7548,45 @@ static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
 	if (err)
 		return err;
 
+	dst_reg = &regs[insn->dst_reg];
 	if (insn->src_reg == 0) {
 		u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
 
-		regs[insn->dst_reg].type = SCALAR_VALUE;
+		dst_reg->type = SCALAR_VALUE;
 		__mark_reg_known(&regs[insn->dst_reg], imm);
 		return 0;
 	}
 
+	if (insn->src_reg == BPF_PSEUDO_BTF_ID) {
+		mark_reg_known_zero(env, regs, insn->dst_reg);
+
+		dst_reg->type = aux->btf_var.reg_type;
+		switch (dst_reg->type) {
+		case PTR_TO_MEM:
+			dst_reg->mem_size = aux->btf_var.mem_size;
+			break;
+		case PTR_TO_BTF_ID:
+		case PTR_TO_PERCPU_BTF_ID:
+			dst_reg->btf_id = aux->btf_var.btf_id;
+			break;
+		default:
+			verbose(env, "bpf verifier is misconfigured\n");
+			return -EFAULT;
+		}
+		return 0;
+	}
+
 	map = env->used_maps[aux->map_index];
 	mark_reg_known_zero(env, regs, insn->dst_reg);
-	regs[insn->dst_reg].map_ptr = map;
+	dst_reg->map_ptr = map;
 
 	if (insn->src_reg == BPF_PSEUDO_MAP_VALUE) {
-		regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
-		regs[insn->dst_reg].off = aux->map_off;
+		dst_reg->type = PTR_TO_MAP_VALUE;
+		dst_reg->off = aux->map_off;
 		if (map_value_has_spin_lock(map))
-			regs[insn->dst_reg].id = ++env->id_gen;
+			dst_reg->id = ++env->id_gen;
 	} else if (insn->src_reg == BPF_PSEUDO_MAP_FD) {
-		regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
+		dst_reg->type = CONST_PTR_TO_MAP;
 	} else {
 		verbose(env, "bpf verifier is misconfigured\n");
 		return -EINVAL;
@@ -9424,6 +9488,92 @@ process_bpf_exit:
 	return 0;
 }
 
+/* replace pseudo btf_id with kernel symbol address */
+static int check_pseudo_btf_id(struct bpf_verifier_env *env,
+			       struct bpf_insn *insn,
+			       struct bpf_insn_aux_data *aux)
+{
+	u32 datasec_id, type, id = insn->imm;
+	const struct btf_var_secinfo *vsi;
+	const struct btf_type *datasec;
+	const struct btf_type *t;
+	const char *sym_name;
+	bool percpu = false;
+	u64 addr;
+	int i;
+
+	if (!btf_vmlinux) {
+		verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n");
+		return -EINVAL;
+	}
+
+	if (insn[1].imm != 0) {
+		verbose(env, "reserved field (insn[1].imm) is used in pseudo_btf_id ldimm64 insn.\n");
+		return -EINVAL;
+	}
+
+	t = btf_type_by_id(btf_vmlinux, id);
+	if (!t) {
+		verbose(env, "ldimm64 insn specifies invalid btf_id %d.\n", id);
+		return -ENOENT;
+	}
+
+	if (!btf_type_is_var(t)) {
+		verbose(env, "pseudo btf_id %d in ldimm64 isn't KIND_VAR.\n",
+			id);
+		return -EINVAL;
+	}
+
+	sym_name = btf_name_by_offset(btf_vmlinux, t->name_off);
+	addr = kallsyms_lookup_name(sym_name);
+	if (!addr) {
+		verbose(env, "ldimm64 failed to find the address for kernel symbol '%s'.\n",
+			sym_name);
+		return -ENOENT;
+	}
+
+	datasec_id = btf_find_by_name_kind(btf_vmlinux, ".data..percpu",
+					   BTF_KIND_DATASEC);
+	if (datasec_id > 0) {
+		datasec = btf_type_by_id(btf_vmlinux, datasec_id);
+		for_each_vsi(i, datasec, vsi) {
+			if (vsi->type == id) {
+				percpu = true;
+				break;
+			}
+		}
+	}
+
+	insn[0].imm = (u32)addr;
+	insn[1].imm = addr >> 32;
+
+	type = t->type;
+	t = btf_type_skip_modifiers(btf_vmlinux, type, NULL);
+	if (percpu) {
+		aux->btf_var.reg_type = PTR_TO_PERCPU_BTF_ID;
+		aux->btf_var.btf_id = type;
+	} else if (!btf_type_is_struct(t)) {
+		const struct btf_type *ret;
+		const char *tname;
+		u32 tsize;
+
+		/* resolve the type size of ksym. */
+		ret = btf_resolve_size(btf_vmlinux, t, &tsize);
+		if (IS_ERR(ret)) {
+			tname = btf_name_by_offset(btf_vmlinux, t->name_off);
+			verbose(env, "ldimm64 unable to resolve the size of type '%s': %ld\n",
+				tname, PTR_ERR(ret));
+			return -EINVAL;
+		}
+		aux->btf_var.reg_type = PTR_TO_MEM;
+		aux->btf_var.mem_size = tsize;
+	} else {
+		aux->btf_var.reg_type = PTR_TO_BTF_ID;
+		aux->btf_var.btf_id = type;
+	}
+	return 0;
+}
+
 static int check_map_prealloc(struct bpf_map *map)
 {
 	return (map->map_type != BPF_MAP_TYPE_HASH &&
@@ -9534,10 +9684,14 @@ static bool bpf_map_is_cgroup_storage(struct bpf_map *map)
 		map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE);
 }
 
-/* look for pseudo eBPF instructions that access map FDs and
- * replace them with actual map pointers
+/* find and rewrite pseudo imm in ld_imm64 instructions:
+ *
+ * 1. if it accesses map FD, replace it with actual map pointer.
+ * 2. if it accesses btf_id of a VAR, replace it with pointer to the var.
+ *
+ * NOTE: btf_vmlinux is required for converting pseudo btf_id.
  */
-static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
+static int resolve_pseudo_ldimm64(struct bpf_verifier_env *env)
 {
 	struct bpf_insn *insn = env->prog->insnsi;
 	int insn_cnt = env->prog->len;
@@ -9578,6 +9732,14 @@ static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
 				/* valid generic load 64-bit imm */
 				goto next_insn;
 
+			if (insn[0].src_reg == BPF_PSEUDO_BTF_ID) {
+				aux = &env->insn_aux_data[i];
+				err = check_pseudo_btf_id(env, insn, aux);
+				if (err)
+					return err;
+				goto next_insn;
+			}
+
 			/* In final convert_pseudo_ld_imm64() step, this is
 			 * converted into regular 64-bit imm load insn.
 			 */
@@ -11633,10 +11795,6 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr,
 	if (is_priv)
 		env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ;
 
-	ret = replace_map_fd_with_map_ptr(env);
-	if (ret < 0)
-		goto skip_full_check;
-
 	if (bpf_prog_is_dev_bound(env->prog->aux)) {
 		ret = bpf_prog_offload_verifier_prep(env->prog);
 		if (ret)
@@ -11662,6 +11820,10 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr,
 	if (ret)
 		goto skip_full_check;
 
+	ret = resolve_pseudo_ldimm64(env);
+	if (ret < 0)
+		goto skip_full_check;
+
 	ret = check_cfg(env);
 	if (ret < 0)
 		goto skip_full_check;