[PATCH v5 0/9] tracing: Improbe BTF support on probe events

[PATCH v5 0/9] tracing: Improbe BTF support on probe events

[Masami Hiramatsu (Google)]

Hi,

Here is the 5th version of series to improve the BTF support on probe events.
The previous series is here:

https://lore.kernel.org/all/169078860386.173706.3091034523220945605.stgit@devnote2/

This version introduces kernel/trace/trace_btf.c to separate the btf generic
functions. These functions will be moved to btf.c next merge window.
This fixes the member-search function to return the bit-offset of the
parent anonymous union/structure. Thus the caller can calculate the real
bit-offset from the root data structure.
This also fixes the ftrace selftest issue which fails if the kernel
supports only BTF args but not support field access.

This series can be applied on top of "probes/core" branch of
https://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace.git/

You can also get this series from:

git://git.kernel.org/pub/scm/linux/kernel/git/mhiramat/linux.git topic/fprobe-event-ext

Thank you,

---

Masami Hiramatsu (Google) (9):
      tracing/probes: Support BTF argument on module functions
      tracing/probes: Move finding func-proto API and getting func-param API to trace_btf
      tracing/probes: Add a function to search a member of a struct/union
      tracing/probes: Support BTF based data structure field access
      tracing/probes: Support BTF field access from $retval
      tracing/probes: Add string type check with BTF
      tracing/fprobe-event: Assume fprobe is a return event by $retval
      selftests/ftrace: Add BTF fields access testcases
      Documentation: tracing: Update fprobe event example with BTF field


 Documentation/trace/fprobetrace.rst                |   64 ++-
 include/linux/btf.h                                |    1 
 kernel/bpf/btf.c                                   |    2 
 kernel/trace/Makefile                              |    1 
 kernel/trace/trace.c                               |    3 
 kernel/trace/trace_btf.c                           |  109 ++++
 kernel/trace/trace_btf.h                           |   11 
 kernel/trace/trace_eprobe.c                        |    4 
 kernel/trace/trace_fprobe.c                        |   59 ++
 kernel/trace/trace_kprobe.c                        |    1 
 kernel/trace/trace_probe.c                         |  499 +++++++++++++++-----
 kernel/trace/trace_probe.h                         |   27 +
 kernel/trace/trace_uprobe.c                        |    1 
 .../ftrace/test.d/dynevent/add_remove_btfarg.tc    |   20 +
 .../ftrace/test.d/dynevent/fprobe_syntax_errors.tc |   10 
 15 files changed, 637 insertions(+), 175 deletions(-)
 create mode 100644 kernel/trace/trace_btf.c
 create mode 100644 kernel/trace/trace_btf.h

--
Masami Hiramatsu (Google) <mhiramat@kernel.org>

[PATCH v5 1/9] tracing/probes: Support BTF argument on module functions

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Since the btf returned from bpf_get_btf_vmlinux() only covers functions in
the vmlinux, BTF argument is not available on the functions in the modules.
Use bpf_find_btf_id() instead of bpf_get_btf_vmlinux()+btf_find_name_kind()
so that BTF argument can find the correct struct btf and btf_type in it.
With this fix, fprobe events can use `$arg*` on module functions as below

 # grep nf_log_ip_packet /proc/kallsyms
ffffffffa0005c00 t nf_log_ip_packet	[nf_log_syslog]
ffffffffa0005bf0 t __pfx_nf_log_ip_packet	[nf_log_syslog]
 # echo 'f nf_log_ip_packet $arg*' > dynamic_events
 # cat dynamic_events
f:fprobes/nf_log_ip_packet__entry nf_log_ip_packet net=net pf=pf hooknum=hooknum skb=skb in=in out=out loginfo=loginfo prefix=prefix

To support the module's btf which is removable, the struct btf needs to be
ref-counted. So this also records the btf in the traceprobe_parse_context
and returns the refcount when the parse has done.

Suggested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
---
 Changes in v3:
  - Newly added.
---
 include/linux/btf.h         |    1 
 kernel/bpf/btf.c            |    2 -
 kernel/trace/trace_eprobe.c |    4 --
 kernel/trace/trace_fprobe.c |    1 
 kernel/trace/trace_kprobe.c |    1 
 kernel/trace/trace_probe.c  |  100 +++++++++++++++++++++++++------------------
 kernel/trace/trace_probe.h  |   14 +++++-
 kernel/trace/trace_uprobe.c |    1 
 8 files changed, 75 insertions(+), 49 deletions(-)

diff --git a/include/linux/btf.h b/include/linux/btf.h
index cac9f304e27a..dbfe41a09c4b 100644
--- a/include/linux/btf.h
+++ b/include/linux/btf.h
@@ -211,6 +211,7 @@ struct btf_record *btf_parse_fields(const struct btf *btf, const struct btf_type
 int btf_check_and_fixup_fields(const struct btf *btf, struct btf_record *rec);
 bool btf_type_is_void(const struct btf_type *t);
 s32 btf_find_by_name_kind(const struct btf *btf, const char *name, u8 kind);
+s32 bpf_find_btf_id(const char *name, u32 kind, struct btf **btf_p);
 const struct btf_type *btf_type_skip_modifiers(const struct btf *btf,
 					       u32 id, u32 *res_id);
 const struct btf_type *btf_type_resolve_ptr(const struct btf *btf,
diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index 817204d53372..b9b0eb1189bb 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c
@@ -552,7 +552,7 @@ s32 btf_find_by_name_kind(const struct btf *btf, const char *name, u8 kind)
 	return -ENOENT;
 }
 
-static s32 bpf_find_btf_id(const char *name, u32 kind, struct btf **btf_p)
+s32 bpf_find_btf_id(const char *name, u32 kind, struct btf **btf_p)
 {
 	struct btf *btf;
 	s32 ret;
diff --git a/kernel/trace/trace_eprobe.c b/kernel/trace/trace_eprobe.c
index a0a704ba27db..09bac836d72b 100644
--- a/kernel/trace/trace_eprobe.c
+++ b/kernel/trace/trace_eprobe.c
@@ -807,13 +807,11 @@ static int trace_eprobe_tp_update_arg(struct trace_eprobe *ep, const char *argv[
 	int ret;
 
 	ret = traceprobe_parse_probe_arg(&ep->tp, i, argv[i], &ctx);
-	if (ret)
-		return ret;
-
 	/* Handle symbols "@" */
 	if (!ret)
 		ret = traceprobe_update_arg(&ep->tp.args[i]);
 
+	traceprobe_finish_parse(&ctx);
 	return ret;
 }
 
diff --git a/kernel/trace/trace_fprobe.c b/kernel/trace/trace_fprobe.c
index dfe2e546acdc..8f43f1f65b1b 100644
--- a/kernel/trace/trace_fprobe.c
+++ b/kernel/trace/trace_fprobe.c
@@ -1096,6 +1096,7 @@ static int __trace_fprobe_create(int argc, const char *argv[])
 	}
 
 out:
+	traceprobe_finish_parse(&ctx);
 	trace_probe_log_clear();
 	kfree(new_argv);
 	kfree(symbol);
diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
index 23dba01831f7..cc822f69bfe8 100644
--- a/kernel/trace/trace_kprobe.c
+++ b/kernel/trace/trace_kprobe.c
@@ -907,6 +907,7 @@ static int __trace_kprobe_create(int argc, const char *argv[])
 	}
 
 out:
+	traceprobe_finish_parse(&ctx);
 	trace_probe_log_clear();
 	kfree(new_argv);
 	kfree(symbol);
diff --git a/kernel/trace/trace_probe.c b/kernel/trace/trace_probe.c
index c68a72707852..ecbe28f8d676 100644
--- a/kernel/trace/trace_probe.c
+++ b/kernel/trace/trace_probe.c
@@ -304,16 +304,6 @@ static int parse_trace_event_arg(char *arg, struct fetch_insn *code,
 
 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
 
-static struct btf *traceprobe_get_btf(void)
-{
-	struct btf *btf = bpf_get_btf_vmlinux();
-
-	if (IS_ERR_OR_NULL(btf))
-		return NULL;
-
-	return btf;
-}
-
 static u32 btf_type_int(const struct btf_type *t)
 {
 	return *(u32 *)(t + 1);
@@ -371,42 +361,49 @@ static const char *type_from_btf_id(struct btf *btf, s32 id)
 	return NULL;
 }
 
-static const struct btf_type *find_btf_func_proto(const char *funcname)
+static const struct btf_type *find_btf_func_proto(const char *funcname,
+						  struct btf **btf_p)
 {
-	struct btf *btf = traceprobe_get_btf();
 	const struct btf_type *t;
+	struct btf *btf = NULL;
 	s32 id;
 
-	if (!btf || !funcname)
+	if (!funcname)
 		return ERR_PTR(-EINVAL);
 
-	id = btf_find_by_name_kind(btf, funcname, BTF_KIND_FUNC);
+	id = bpf_find_btf_id(funcname, BTF_KIND_FUNC, &btf);
 	if (id <= 0)
 		return ERR_PTR(-ENOENT);
 
 	/* Get BTF_KIND_FUNC type */
 	t = btf_type_by_id(btf, id);
 	if (!t || !btf_type_is_func(t))
-		return ERR_PTR(-ENOENT);
+		goto err;
 
 	/* The type of BTF_KIND_FUNC is BTF_KIND_FUNC_PROTO */
 	t = btf_type_by_id(btf, t->type);
 	if (!t || !btf_type_is_func_proto(t))
-		return ERR_PTR(-ENOENT);
+		goto err;
 
+	*btf_p = btf;
 	return t;
+
+err:
+	btf_put(btf);
+	return ERR_PTR(-ENOENT);
 }
 
 static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr,
-						   bool tracepoint)
+						   struct btf **btf_p, bool tracepoint)
 {
 	const struct btf_param *param;
 	const struct btf_type *t;
+	struct btf *btf;
 
 	if (!funcname || !nr)
 		return ERR_PTR(-EINVAL);
 
-	t = find_btf_func_proto(funcname);
+	t = find_btf_func_proto(funcname, &btf);
 	if (IS_ERR(t))
 		return (const struct btf_param *)t;
 
@@ -419,29 +416,37 @@ static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr
 		param++;
 	}
 
-	if (*nr > 0)
+	if (*nr > 0) {
+		*btf_p = btf;
 		return param;
-	else
-		return NULL;
+	}
+
+	btf_put(btf);
+	return NULL;
+}
+
+static void clear_btf_context(struct traceprobe_parse_context *ctx)
+{
+	if (ctx->btf) {
+		btf_put(ctx->btf);
+		ctx->btf = NULL;
+		ctx->params = NULL;
+		ctx->nr_params = 0;
+	}
 }
 
 static int parse_btf_arg(const char *varname, struct fetch_insn *code,
 			 struct traceprobe_parse_context *ctx)
 {
-	struct btf *btf = traceprobe_get_btf();
 	const struct btf_param *params;
 	int i;
 
-	if (!btf) {
-		trace_probe_log_err(ctx->offset, NOSUP_BTFARG);
-		return -EOPNOTSUPP;
-	}
-
 	if (WARN_ON_ONCE(!ctx->funcname))
 		return -EINVAL;
 
 	if (!ctx->params) {
-		params = find_btf_func_param(ctx->funcname, &ctx->nr_params,
+		params = find_btf_func_param(ctx->funcname,
+					     &ctx->nr_params, &ctx->btf,
 					     ctx->flags & TPARG_FL_TPOINT);
 		if (IS_ERR_OR_NULL(params)) {
 			trace_probe_log_err(ctx->offset, NO_BTF_ENTRY);
@@ -452,7 +457,7 @@ static int parse_btf_arg(const char *varname, struct fetch_insn *code,
 		params = ctx->params;
 
 	for (i = 0; i < ctx->nr_params; i++) {
-		const char *name = btf_name_by_offset(btf, params[i].name_off);
+		const char *name = btf_name_by_offset(ctx->btf, params[i].name_off);
 
 		if (name && !strcmp(name, varname)) {
 			code->op = FETCH_OP_ARG;
@@ -470,7 +475,7 @@ static int parse_btf_arg(const char *varname, struct fetch_insn *code,
 static const struct fetch_type *parse_btf_arg_type(int arg_idx,
 					struct traceprobe_parse_context *ctx)
 {
-	struct btf *btf = traceprobe_get_btf();
+	struct btf *btf = ctx->btf;
 	const char *typestr = NULL;
 
 	if (btf && ctx->params) {
@@ -485,14 +490,17 @@ static const struct fetch_type *parse_btf_arg_type(int arg_idx,
 static const struct fetch_type *parse_btf_retval_type(
 					struct traceprobe_parse_context *ctx)
 {
-	struct btf *btf = traceprobe_get_btf();
 	const char *typestr = NULL;
 	const struct btf_type *t;
+	struct btf *btf;
 
-	if (btf && ctx->funcname) {
-		t = find_btf_func_proto(ctx->funcname);
-		if (!IS_ERR(t))
+	if (ctx->funcname) {
+		/* Do not use ctx->btf, because it must be used with ctx->param */
+		t = find_btf_func_proto(ctx->funcname, &btf);
+		if (!IS_ERR(t)) {
 			typestr = type_from_btf_id(btf, t->type);
+			btf_put(btf);
+		}
 	}
 
 	return find_fetch_type(typestr, ctx->flags);
@@ -501,21 +509,25 @@ static const struct fetch_type *parse_btf_retval_type(
 static bool is_btf_retval_void(const char *funcname)
 {
 	const struct btf_type *t;
+	struct btf *btf;
+	bool ret;
 
-	t = find_btf_func_proto(funcname);
+	t = find_btf_func_proto(funcname, &btf);
 	if (IS_ERR(t))
 		return false;
 
-	return t->type == 0;
+	ret = (t->type == 0);
+	btf_put(btf);
+	return ret;
 }
 #else
-static struct btf *traceprobe_get_btf(void)
+static void clear_btf_context(struct traceprobe_parse_context *ctx)
 {
-	return NULL;
+	ctx->btf = NULL;
 }
 
 static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr,
-						   bool tracepoint)
+						   struct btf **btf_p, bool tracepoint)
 {
 	return ERR_PTR(-EOPNOTSUPP);
 }
@@ -1231,7 +1243,6 @@ static int sprint_nth_btf_arg(int idx, const char *type,
 			      char *buf, int bufsize,
 			      struct traceprobe_parse_context *ctx)
 {
-	struct btf *btf = traceprobe_get_btf();
 	const char *name;
 	int ret;
 
@@ -1239,7 +1250,7 @@ static int sprint_nth_btf_arg(int idx, const char *type,
 		trace_probe_log_err(0, NO_BTFARG);
 		return -ENOENT;
 	}
-	name = btf_name_by_offset(btf, ctx->params[idx].name_off);
+	name = btf_name_by_offset(ctx->btf, ctx->params[idx].name_off);
 	if (!name) {
 		trace_probe_log_err(0, NO_BTF_ENTRY);
 		return -ENOENT;
@@ -1271,7 +1282,7 @@ const char **traceprobe_expand_meta_args(int argc, const char *argv[],
 		return NULL;
 	}
 
-	params = find_btf_func_param(ctx->funcname, &nr_params,
+	params = find_btf_func_param(ctx->funcname, &nr_params, &ctx->btf,
 				     ctx->flags & TPARG_FL_TPOINT);
 	if (IS_ERR_OR_NULL(params)) {
 		if (args_idx != -1) {
@@ -1337,6 +1348,11 @@ const char **traceprobe_expand_meta_args(int argc, const char *argv[],
 	return ERR_PTR(ret);
 }
 
+void traceprobe_finish_parse(struct traceprobe_parse_context *ctx)
+{
+	clear_btf_context(ctx);
+}
+
 int traceprobe_update_arg(struct probe_arg *arg)
 {
 	struct fetch_insn *code = arg->code;
diff --git a/kernel/trace/trace_probe.h b/kernel/trace/trace_probe.h
index 01ea148723de..4dc91460a75d 100644
--- a/kernel/trace/trace_probe.h
+++ b/kernel/trace/trace_probe.h
@@ -383,9 +383,11 @@ static inline bool tparg_is_function_entry(unsigned int flags)
 
 struct traceprobe_parse_context {
 	struct trace_event_call *event;
-	const struct btf_param *params;
-	s32 nr_params;
-	const char *funcname;
+	/* BTF related parameters */
+	const char *funcname;		/* Function name in BTF */
+	const struct btf_param *params;	/* Parameter of the function */
+	s32 nr_params;			/* The number of the parameters */
+	struct btf *btf;		/* The BTF to be used */
 	unsigned int flags;
 	int offset;
 };
@@ -400,6 +402,12 @@ const char **traceprobe_expand_meta_args(int argc, const char *argv[],
 extern int traceprobe_update_arg(struct probe_arg *arg);
 extern void traceprobe_free_probe_arg(struct probe_arg *arg);
 
+/*
+ * If either traceprobe_parse_probe_arg() or traceprobe_expand_meta_args() is called,
+ * this MUST be called for clean up the context and return a resource.
+ */
+void traceprobe_finish_parse(struct traceprobe_parse_context *ctx);
+
 extern int traceprobe_split_symbol_offset(char *symbol, long *offset);
 int traceprobe_parse_event_name(const char **pevent, const char **pgroup,
 				char *buf, int offset);
diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
index 688bf579f2f1..9790f8f0a32d 100644
--- a/kernel/trace/trace_uprobe.c
+++ b/kernel/trace/trace_uprobe.c
@@ -693,6 +693,7 @@ static int __trace_uprobe_create(int argc, const char **argv)
 
 		trace_probe_log_set_index(i + 2);
 		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
+		traceprobe_finish_parse(&ctx);
 		if (ret)
 			goto error;
 	}

[PATCH v5 2/9] tracing/probes: Move finding func-proto API and getting func-param API to trace_btf

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Move generic function-proto find API and getting function parameter API
to BTF library code from trace_probe.c. This will avoid redundant efforts
on different feature.

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
---
 Changes in v3:
  - Remove perameter check.
  - Fix a typo.
  - Add a type check for btf_get_func_param() and add comment for that.
  - Use bpf_find_btf_id() and add bpf_put().
  - Move the code before btf_show() related code.
 Changes in v5:
  - Move the generic functions into a new file, trace_btf.c, for avoiding
    tree-merge conflict.
---
 kernel/trace/Makefile      |    1 +
 kernel/trace/trace_btf.c   |   52 ++++++++++++++++++++++++++++++++++++++++++++
 kernel/trace/trace_btf.h   |    7 ++++++
 kernel/trace/trace_probe.c |   51 +++++++++----------------------------------
 4 files changed, 71 insertions(+), 40 deletions(-)
 create mode 100644 kernel/trace/trace_btf.c
 create mode 100644 kernel/trace/trace_btf.h

diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index 64b61f67a403..057cd975d014 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -99,6 +99,7 @@ obj-$(CONFIG_KGDB_KDB) += trace_kdb.o
 endif
 obj-$(CONFIG_DYNAMIC_EVENTS) += trace_dynevent.o
 obj-$(CONFIG_PROBE_EVENTS) += trace_probe.o
+obj-$(CONFIG_PROBE_EVENTS_BTF_ARGS) += trace_btf.o
 obj-$(CONFIG_UPROBE_EVENTS) += trace_uprobe.o
 obj-$(CONFIG_BOOTTIME_TRACING) += trace_boot.o
 obj-$(CONFIG_FTRACE_RECORD_RECURSION) += trace_recursion_record.o
diff --git a/kernel/trace/trace_btf.c b/kernel/trace/trace_btf.c
new file mode 100644
index 000000000000..82cbbeeda071
--- /dev/null
+++ b/kernel/trace/trace_btf.c
@@ -0,0 +1,52 @@
+#include <linux/btf.h>
+#include <linux/kernel.h>
+
+#include "trace_btf.h"
+
+/*
+ * Find a function proto type by name, and return the btf_type with its btf
+ * in *@btf_p. Return NULL if not found.
+ * Note that caller has to call btf_put(*@btf_p) after using the btf_type.
+ */
+const struct btf_type *btf_find_func_proto(const char *func_name, struct btf **btf_p)
+{
+	const struct btf_type *t;
+	s32 id;
+
+	id = bpf_find_btf_id(func_name, BTF_KIND_FUNC, btf_p);
+	if (id < 0)
+		return NULL;
+
+	/* Get BTF_KIND_FUNC type */
+	t = btf_type_by_id(*btf_p, id);
+	if (!t || !btf_type_is_func(t))
+		goto err;
+
+	/* The type of BTF_KIND_FUNC is BTF_KIND_FUNC_PROTO */
+	t = btf_type_by_id(*btf_p, t->type);
+	if (!t || !btf_type_is_func_proto(t))
+		goto err;
+
+	return t;
+err:
+	btf_put(*btf_p);
+	return NULL;
+}
+
+/*
+ * Get function parameter with the number of parameters.
+ * This can return NULL if the function has no parameters.
+ * It can return -EINVAL if the @func_proto is not a function proto type.
+ */
+const struct btf_param *btf_get_func_param(const struct btf_type *func_proto, s32 *nr)
+{
+	if (!btf_type_is_func_proto(func_proto))
+		return ERR_PTR(-EINVAL);
+
+	*nr = btf_type_vlen(func_proto);
+	if (*nr > 0)
+		return (const struct btf_param *)(func_proto + 1);
+	else
+		return NULL;
+}
+
diff --git a/kernel/trace/trace_btf.h b/kernel/trace/trace_btf.h
new file mode 100644
index 000000000000..2ef1b2367e1e
--- /dev/null
+++ b/kernel/trace/trace_btf.h
@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/btf.h>
+
+const struct btf_type *btf_find_func_proto(const char *func_name,
+					   struct btf **btf_p);
+const struct btf_param *btf_get_func_param(const struct btf_type *func_proto,
+					   s32 *nr);
diff --git a/kernel/trace/trace_probe.c b/kernel/trace/trace_probe.c
index ecbe28f8d676..c3ac5698e80b 100644
--- a/kernel/trace/trace_probe.c
+++ b/kernel/trace/trace_probe.c
@@ -12,6 +12,7 @@
 #define pr_fmt(fmt)	"trace_probe: " fmt
 
 #include <linux/bpf.h>
+#include "trace_btf.h"
 
 #include "trace_probe.h"
 
@@ -361,38 +362,6 @@ static const char *type_from_btf_id(struct btf *btf, s32 id)
 	return NULL;
 }
 
-static const struct btf_type *find_btf_func_proto(const char *funcname,
-						  struct btf **btf_p)
-{
-	const struct btf_type *t;
-	struct btf *btf = NULL;
-	s32 id;
-
-	if (!funcname)
-		return ERR_PTR(-EINVAL);
-
-	id = bpf_find_btf_id(funcname, BTF_KIND_FUNC, &btf);
-	if (id <= 0)
-		return ERR_PTR(-ENOENT);
-
-	/* Get BTF_KIND_FUNC type */
-	t = btf_type_by_id(btf, id);
-	if (!t || !btf_type_is_func(t))
-		goto err;
-
-	/* The type of BTF_KIND_FUNC is BTF_KIND_FUNC_PROTO */
-	t = btf_type_by_id(btf, t->type);
-	if (!t || !btf_type_is_func_proto(t))
-		goto err;
-
-	*btf_p = btf;
-	return t;
-
-err:
-	btf_put(btf);
-	return ERR_PTR(-ENOENT);
-}
-
 static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr,
 						   struct btf **btf_p, bool tracepoint)
 {
@@ -403,12 +372,13 @@ static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr
 	if (!funcname || !nr)
 		return ERR_PTR(-EINVAL);
 
-	t = find_btf_func_proto(funcname, &btf);
-	if (IS_ERR(t))
+	t = btf_find_func_proto(funcname, &btf);
+	if (!t)
 		return (const struct btf_param *)t;
 
-	*nr = btf_type_vlen(t);
-	param = (const struct btf_param *)(t + 1);
+	param = btf_get_func_param(t, nr);
+	if (IS_ERR_OR_NULL(param))
+		goto err;
 
 	/* Hide the first 'data' argument of tracepoint */
 	if (tracepoint) {
@@ -421,6 +391,7 @@ static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr
 		return param;
 	}
 
+err:
 	btf_put(btf);
 	return NULL;
 }
@@ -496,8 +467,8 @@ static const struct fetch_type *parse_btf_retval_type(
 
 	if (ctx->funcname) {
 		/* Do not use ctx->btf, because it must be used with ctx->param */
-		t = find_btf_func_proto(ctx->funcname, &btf);
-		if (!IS_ERR(t)) {
+		t = btf_find_func_proto(ctx->funcname, &btf);
+		if (t) {
 			typestr = type_from_btf_id(btf, t->type);
 			btf_put(btf);
 		}
@@ -512,8 +483,8 @@ static bool is_btf_retval_void(const char *funcname)
 	struct btf *btf;
 	bool ret;
 
-	t = find_btf_func_proto(funcname, &btf);
-	if (IS_ERR(t))
+	t = btf_find_func_proto(funcname, &btf);
+	if (!t)
 		return false;
 
 	ret = (t->type == 0);

[PATCH v5 3/9] tracing/probes: Add a function to search a member of a struct/union

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Add btf_find_struct_member() API to search a member of a given data structure
or union from the member's name.

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
---
 Changes in v3:
  - Remove simple input check.
  - Fix unneeded IS_ERR_OR_NULL() check for btf_type_by_id().
  - Move the code next to btf_get_func_param().
  - Use for_each_member() macro instead of for-loop.
  - Use btf_type_skip_modifiers() instead of btf_type_by_id().
 Changes in v4:
  - Use a stack for searching in anonymous members instead of nested call.
 Changes in v5:
  - Move the generic functions into a new file, trace_btf.c, for avoiding
    tree-merge conflict.
  - Return anon_offset for calculating correct offset from root struct.
---
 kernel/trace/trace_btf.c |   57 ++++++++++++++++++++++++++++++++++++++++++++++
 kernel/trace/trace_btf.h |    4 +++
 2 files changed, 61 insertions(+)

diff --git a/kernel/trace/trace_btf.c b/kernel/trace/trace_btf.c
index 82cbbeeda071..109a238437cf 100644
--- a/kernel/trace/trace_btf.c
+++ b/kernel/trace/trace_btf.c
@@ -50,3 +50,60 @@ const struct btf_param *btf_get_func_param(const struct btf_type *func_proto, s3
 		return NULL;
 }
 
+#define BTF_ANON_STACK_MAX	16
+
+/*
+ * Find a member of data structure/union by name and return it.
+ * Return NULL if not found, or -EINVAL if parameter is invalid.
+ * If the member is an member of anonymous union/structure, the offset
+ * of that anonymous union/structure is stored into @anon_offset. Caller
+ * can calculate the correct offset from the root data structure by
+ * adding anon_offset to the member's offset.
+ */
+const struct btf_member *btf_find_struct_member(struct btf *btf,
+						const struct btf_type *type,
+						const char *member_name,
+						u32 *anon_offset)
+{
+	struct {
+		u32 tid;
+		u32 offset;
+	} anon_stack[BTF_ANON_STACK_MAX];
+	const struct btf_member *member;
+	u32 tid, cur_offset = 0;
+	const char *name;
+	int i, top = 0;
+
+retry:
+	if (!btf_type_is_struct(type))
+		return ERR_PTR(-EINVAL);
+
+	for_each_member(i, type, member) {
+		if (!member->name_off) {
+			/* Anonymous union/struct: push it for later use */
+			type = btf_type_skip_modifiers(btf, member->type, &tid);
+			if (type && top < BTF_ANON_STACK_MAX) {
+				anon_stack[top].tid = tid;
+				anon_stack[top++].offset =
+					cur_offset + member->offset;
+			}
+		} else {
+			name = btf_name_by_offset(btf, member->name_off);
+			if (name && !strcmp(member_name, name)) {
+				if (anon_offset)
+					*anon_offset = cur_offset;
+				return member;
+			}
+		}
+	}
+	if (top > 0) {
+		/* Pop from the anonymous stack and retry */
+		tid = anon_stack[--top].tid;
+		cur_offset = anon_stack[top].offset;
+		type = btf_type_by_id(btf, tid);
+		goto retry;
+	}
+
+	return NULL;
+}
+
diff --git a/kernel/trace/trace_btf.h b/kernel/trace/trace_btf.h
index 2ef1b2367e1e..fb2919a74225 100644
--- a/kernel/trace/trace_btf.h
+++ b/kernel/trace/trace_btf.h
@@ -5,3 +5,7 @@ const struct btf_type *btf_find_func_proto(const char *func_name,
 					   struct btf **btf_p);
 const struct btf_param *btf_get_func_param(const struct btf_type *func_proto,
 					   s32 *nr);
+const struct btf_member *btf_find_struct_member(struct btf *btf,
+						const struct btf_type *type,
+						const char *member_name,
+						u32 *anon_offset);

Re: [PATCH v5 3/9] tracing/probes: Add a function to search a member of a struct/union

[Steven Rostedt]

On Mon,  7 Aug 2023 11:54:58 +0900
"Masami Hiramatsu (Google)" <mhiramat@kernel.org> wrote:

> --- a/kernel/trace/trace_btf.c
> +++ b/kernel/trace/trace_btf.c
> @@ -50,3 +50,60 @@ const struct btf_param *btf_get_func_param(const struct btf_type *func_proto, s3
>  		return NULL;
>  }
>  
> +#define BTF_ANON_STACK_MAX	16
> +
> +/*
> + * Find a member of data structure/union by name and return it.
> + * Return NULL if not found, or -EINVAL if parameter is invalid.
> + * If the member is an member of anonymous union/structure, the offset
> + * of that anonymous union/structure is stored into @anon_offset. Caller
> + * can calculate the correct offset from the root data structure by
> + * adding anon_offset to the member's offset.
> + */
> +const struct btf_member *btf_find_struct_member(struct btf *btf,
> +						const struct btf_type *type,
> +						const char *member_name,
> +						u32 *anon_offset)
> +{
> +	struct {
> +		u32 tid;
> +		u32 offset;
> +	} anon_stack[BTF_ANON_STACK_MAX];

Where is this called as the above is 128 bytes, which is a bit large
for the stack. It may not be bad if it's not that generic of a
function. But if the stack is getting tight, this could still be an
issue.

-- Steve


> +	const struct btf_member *member;
> +	u32 tid, cur_offset = 0;
> +	const char *name;
> +	int i, top = 0;
> +
> +retry:
> +	if (!btf_type_is_struct(type))
> +		return ERR_PTR(-EINVAL);
> +
> +	for_each_member(i, type, member) {
> +		if (!member->name_off) {
> +			/* Anonymous union/struct: push it for later use */
> +			type = btf_type_skip_modifiers(btf, member->type, &tid);
> +			if (type && top < BTF_ANON_STACK_MAX) {
> +				anon_stack[top].tid = tid;
> +				anon_stack[top++].offset =
> +					cur_offset + member->offset;
> +			}
> +		} else {
> +			name = btf_name_by_offset(btf, member->name_off);
> +			if (name && !strcmp(member_name, name)) {
> +				if (anon_offset)
> +					*anon_offset = cur_offset;
> +				return member;
> +			}
> +		}
> +	}
> +	if (top > 0) {
> +		/* Pop from the anonymous stack and retry */
> +		tid = anon_stack[--top].tid;
> +		cur_offset = anon_stack[top].offset;
> +		type = btf_type_by_id(btf, tid);
> +		goto retry;
> +	}
> +
> +	return NULL;
> +}
> +

Re: [PATCH v5 3/9] tracing/probes: Add a function to search a member of a struct/union

[Masami Hiramatsu]

On Tue, 22 Aug 2023 09:37:20 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:

> On Mon,  7 Aug 2023 11:54:58 +0900
> "Masami Hiramatsu (Google)" <mhiramat@kernel.org> wrote:
> 
> > --- a/kernel/trace/trace_btf.c
> > +++ b/kernel/trace/trace_btf.c
> > @@ -50,3 +50,60 @@ const struct btf_param *btf_get_func_param(const struct btf_type *func_proto, s3
> >  		return NULL;
> >  }
> >  
> > +#define BTF_ANON_STACK_MAX	16
> > +
> > +/*
> > + * Find a member of data structure/union by name and return it.
> > + * Return NULL if not found, or -EINVAL if parameter is invalid.
> > + * If the member is an member of anonymous union/structure, the offset
> > + * of that anonymous union/structure is stored into @anon_offset. Caller
> > + * can calculate the correct offset from the root data structure by
> > + * adding anon_offset to the member's offset.
> > + */
> > +const struct btf_member *btf_find_struct_member(struct btf *btf,
> > +						const struct btf_type *type,
> > +						const char *member_name,
> > +						u32 *anon_offset)
> > +{
> > +	struct {
> > +		u32 tid;
> > +		u32 offset;
> > +	} anon_stack[BTF_ANON_STACK_MAX];
> 
> Where is this called as the above is 128 bytes, which is a bit large
> for the stack. It may not be bad if it's not that generic of a
> function. But if the stack is getting tight, this could still be an
> issue.

OK, let me allocate an array then.

Thank you,

> 
> -- Steve
> 
> 
> > +	const struct btf_member *member;
> > +	u32 tid, cur_offset = 0;
> > +	const char *name;
> > +	int i, top = 0;
> > +
> > +retry:
> > +	if (!btf_type_is_struct(type))
> > +		return ERR_PTR(-EINVAL);
> > +
> > +	for_each_member(i, type, member) {
> > +		if (!member->name_off) {
> > +			/* Anonymous union/struct: push it for later use */
> > +			type = btf_type_skip_modifiers(btf, member->type, &tid);
> > +			if (type && top < BTF_ANON_STACK_MAX) {
> > +				anon_stack[top].tid = tid;
> > +				anon_stack[top++].offset =
> > +					cur_offset + member->offset;
> > +			}
> > +		} else {
> > +			name = btf_name_by_offset(btf, member->name_off);
> > +			if (name && !strcmp(member_name, name)) {
> > +				if (anon_offset)
> > +					*anon_offset = cur_offset;
> > +				return member;
> > +			}
> > +		}
> > +	}
> > +	if (top > 0) {
> > +		/* Pop from the anonymous stack and retry */
> > +		tid = anon_stack[--top].tid;
> > +		cur_offset = anon_stack[top].offset;
> > +		type = btf_type_by_id(btf, tid);
> > +		goto retry;
> > +	}
> > +
> > +	return NULL;
> > +}
> > +


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>

[PATCH v5 4/9] tracing/probes: Support BTF based data structure field access

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Using BTF to access the fields of a data structure. You can use this
for accessing the field with '->' or '.' operation with BTF argument.

 # echo 't sched_switch next=next->pid vruntime=next->se.vruntime' \
   > dynamic_events
 # echo 1 > events/tracepoints/sched_switch/enable
 # head -n 40 trace | tail
          <idle>-0       [000] d..3.   272.565382: sched_switch: (__probestub_sched_switch+0x4/0x10) next=26 vruntime=956533179
      kcompactd0-26      [000] d..3.   272.565406: sched_switch: (__probestub_sched_switch+0x4/0x10) next=0 vruntime=0
          <idle>-0       [000] d..3.   273.069441: sched_switch: (__probestub_sched_switch+0x4/0x10) next=9 vruntime=956533179
     kworker/0:1-9       [000] d..3.   273.069464: sched_switch: (__probestub_sched_switch+0x4/0x10) next=26 vruntime=956579181
      kcompactd0-26      [000] d..3.   273.069480: sched_switch: (__probestub_sched_switch+0x4/0x10) next=0 vruntime=0
          <idle>-0       [000] d..3.   273.141434: sched_switch: (__probestub_sched_switch+0x4/0x10) next=22 vruntime=956533179
    kworker/u2:1-22      [000] d..3.   273.141461: sched_switch: (__probestub_sched_switch+0x4/0x10) next=0 vruntime=0
          <idle>-0       [000] d..3.   273.480872: sched_switch: (__probestub_sched_switch+0x4/0x10) next=22 vruntime=956585857
    kworker/u2:1-22      [000] d..3.   273.480905: sched_switch: (__probestub_sched_switch+0x4/0x10) next=70 vruntime=959533179
              sh-70      [000] d..3.   273.481102: sched_switch: (__probestub_sched_switch+0x4/0x10) next=0 vruntime=0

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
---
 Changes in v2:
  - Use new BTF API for finding the member.
 Changes in v3:
  - Update according to previous changes in the series.
 Changes in v5:
  - Fix offset for anonymous union/structure members.
  - Add dot and allow operators in README.
---
 kernel/trace/trace.c       |    3 -
 kernel/trace/trace_probe.c |  230 +++++++++++++++++++++++++++++++++++++++-----
 kernel/trace/trace_probe.h |   11 ++
 3 files changed, 216 insertions(+), 28 deletions(-)

diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index be847d45d81c..5ed90855a3e3 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -5716,7 +5716,8 @@ static const char readme_msg[] =
 	"\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
-	"\t           $stack<index>, $stack, $retval, $comm, $arg<N>, <argname>\n"
+	"\t           $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
+	"\t           <argname>[->field[->field|.field...]],\n"
 #else
 	"\t           $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
 #endif
diff --git a/kernel/trace/trace_probe.c b/kernel/trace/trace_probe.c
index c3ac5698e80b..821f43e5c52b 100644
--- a/kernel/trace/trace_probe.c
+++ b/kernel/trace/trace_probe.c
@@ -310,16 +310,14 @@ static u32 btf_type_int(const struct btf_type *t)
 	return *(u32 *)(t + 1);
 }
 
-static const char *type_from_btf_id(struct btf *btf, s32 id)
+static const char *fetch_type_from_btf_type(struct btf *btf,
+					const struct btf_type *type,
+					struct traceprobe_parse_context *ctx)
 {
-	const struct btf_type *t;
 	u32 intdata;
-	s32 tid;
 
 	/* TODO: const char * could be converted as a string */
-	t = btf_type_skip_modifiers(btf, id, &tid);
-
-	switch (BTF_INFO_KIND(t->info)) {
+	switch (BTF_INFO_KIND(type->info)) {
 	case BTF_KIND_ENUM:
 		/* enum is "int", so convert to "s32" */
 		return "s32";
@@ -332,7 +330,7 @@ static const char *type_from_btf_id(struct btf *btf, s32 id)
 		else
 			return "x32";
 	case BTF_KIND_INT:
-		intdata = btf_type_int(t);
+		intdata = btf_type_int(type);
 		if (BTF_INT_ENCODING(intdata) & BTF_INT_SIGNED) {
 			switch (BTF_INT_BITS(intdata)) {
 			case 8:
@@ -355,6 +353,10 @@ static const char *type_from_btf_id(struct btf *btf, s32 id)
 			case 64:
 				return "u64";
 			}
+			/* bitfield, size is encoded in the type */
+			ctx->last_bitsize = BTF_INT_BITS(intdata);
+			ctx->last_bitoffs += BTF_INT_OFFSET(intdata);
+			return "u64";
 		}
 	}
 	/* TODO: support other types */
@@ -406,15 +408,136 @@ static void clear_btf_context(struct traceprobe_parse_context *ctx)
 	}
 }
 
-static int parse_btf_arg(const char *varname, struct fetch_insn *code,
+/* Return 1 if the field separater is arrow operator ('->') */
+static int split_next_field(char *varname, char **next_field,
+			    struct traceprobe_parse_context *ctx)
+{
+	char *field;
+	int ret = 0;
+
+	field = strpbrk(varname, ".-");
+	if (field) {
+		if (field[0] == '-' && field[1] == '>') {
+			field[0] = '\0';
+			field += 2;
+			ret = 1;
+		} else if (field[0] == '.') {
+			field[0] = '\0';
+			field += 1;
+		} else {
+			trace_probe_log_err(ctx->offset + field - varname, BAD_HYPHEN);
+			return -EINVAL;
+		}
+		*next_field = field;
+	}
+
+	return ret;
+}
+
+/*
+ * Parse the field of data structure. The @type must be a pointer type
+ * pointing the target data structure type.
+ */
+static int parse_btf_field(char *fieldname, const struct btf_type *type,
+			   struct fetch_insn **pcode, struct fetch_insn *end,
+			   struct traceprobe_parse_context *ctx)
+{
+	struct fetch_insn *code = *pcode;
+	const struct btf_member *field;
+	u32 bitoffs, anon_offs;
+	char *next;
+	int is_ptr;
+	s32 tid;
+
+	do {
+		/* Outer loop for solving arrow operator ('->') */
+		if (BTF_INFO_KIND(type->info) != BTF_KIND_PTR) {
+			trace_probe_log_err(ctx->offset, NO_PTR_STRCT);
+			return -EINVAL;
+		}
+		/* Convert a struct pointer type to a struct type */
+		type = btf_type_skip_modifiers(ctx->btf, type->type, &tid);
+		if (!type) {
+			trace_probe_log_err(ctx->offset, BAD_BTF_TID);
+			return -EINVAL;
+		}
+
+		bitoffs = 0;
+		do {
+			/* Inner loop for solving dot operator ('.') */
+			next = NULL;
+			is_ptr = split_next_field(fieldname, &next, ctx);
+			if (is_ptr < 0)
+				return is_ptr;
+
+			anon_offs = 0;
+			field = btf_find_struct_member(ctx->btf, type, fieldname,
+						       &anon_offs);
+			if (!field) {
+				trace_probe_log_err(ctx->offset, NO_BTF_FIELD);
+				return -ENOENT;
+			}
+			/* Add anonymous structure/union offset */
+			bitoffs += anon_offs;
+
+			/* Accumulate the bit-offsets of the dot-connected fields */
+			if (btf_type_kflag(type)) {
+				bitoffs += BTF_MEMBER_BIT_OFFSET(field->offset);
+				ctx->last_bitsize = BTF_MEMBER_BITFIELD_SIZE(field->offset);
+			} else {
+				bitoffs += field->offset;
+				ctx->last_bitsize = 0;
+			}
+
+			type = btf_type_skip_modifiers(ctx->btf, field->type, &tid);
+			if (!type) {
+				trace_probe_log_err(ctx->offset, BAD_BTF_TID);
+				return -EINVAL;
+			}
+
+			ctx->offset += next - fieldname;
+			fieldname = next;
+		} while (!is_ptr && fieldname);
+
+		if (++code == end) {
+			trace_probe_log_err(ctx->offset, TOO_MANY_OPS);
+			return -EINVAL;
+		}
+		code->op = FETCH_OP_DEREF;	/* TODO: user deref support */
+		code->offset = bitoffs / 8;
+		*pcode = code;
+
+		ctx->last_bitoffs = bitoffs % 8;
+		ctx->last_type = type;
+	} while (fieldname);
+
+	return 0;
+}
+
+static int parse_btf_arg(char *varname,
+			 struct fetch_insn **pcode, struct fetch_insn *end,
 			 struct traceprobe_parse_context *ctx)
 {
+	struct fetch_insn *code = *pcode;
 	const struct btf_param *params;
-	int i;
+	const struct btf_type *type;
+	char *field = NULL;
+	int i, is_ptr;
+	u32 tid;
 
 	if (WARN_ON_ONCE(!ctx->funcname))
 		return -EINVAL;
 
+	is_ptr = split_next_field(varname, &field, ctx);
+	if (is_ptr < 0)
+		return is_ptr;
+	if (!is_ptr && field) {
+		/* dot-connected field on an argument is not supported. */
+		trace_probe_log_err(ctx->offset + field - varname,
+				    NOSUP_DAT_ARG);
+		return -EOPNOTSUPP;
+	}
+
 	if (!ctx->params) {
 		params = find_btf_func_param(ctx->funcname,
 					     &ctx->nr_params, &ctx->btf,
@@ -436,24 +559,39 @@ static int parse_btf_arg(const char *varname, struct fetch_insn *code,
 				code->param = i + 1;
 			else
 				code->param = i;
-			return 0;
+
+			tid = params[i].type;
+			goto found;
 		}
 	}
 	trace_probe_log_err(ctx->offset, NO_BTFARG);
 	return -ENOENT;
+
+found:
+	type = btf_type_skip_modifiers(ctx->btf, tid, &tid);
+	if (!type) {
+		trace_probe_log_err(ctx->offset, BAD_BTF_TID);
+		return -EINVAL;
+	}
+	/* Initialize the last type information */
+	ctx->last_type = type;
+	ctx->last_bitoffs = 0;
+	ctx->last_bitsize = 0;
+	if (field) {
+		ctx->offset += field - varname;
+		return parse_btf_field(field, type, pcode, end, ctx);
+	}
+	return 0;
 }
 
-static const struct fetch_type *parse_btf_arg_type(int arg_idx,
+static const struct fetch_type *parse_btf_arg_type(
 					struct traceprobe_parse_context *ctx)
 {
 	struct btf *btf = ctx->btf;
 	const char *typestr = NULL;
 
-	if (btf && ctx->params) {
-		if (ctx->flags & TPARG_FL_TPOINT)
-			arg_idx--;
-		typestr = type_from_btf_id(btf, ctx->params[arg_idx].type);
-	}
+	if (btf && ctx->last_type)
+		typestr = fetch_type_from_btf_type(btf, ctx->last_type, ctx);
 
 	return find_fetch_type(typestr, ctx->flags);
 }
@@ -462,14 +600,16 @@ static const struct fetch_type *parse_btf_retval_type(
 					struct traceprobe_parse_context *ctx)
 {
 	const char *typestr = NULL;
-	const struct btf_type *t;
+	const struct btf_type *type;
 	struct btf *btf;
 
 	if (ctx->funcname) {
 		/* Do not use ctx->btf, because it must be used with ctx->param */
-		t = btf_find_func_proto(ctx->funcname, &btf);
-		if (t) {
-			typestr = type_from_btf_id(btf, t->type);
+		type = btf_find_func_proto(ctx->funcname, &btf);
+		if (type) {
+			type = btf_type_skip_modifiers(btf, type->type, NULL);
+			if (!IS_ERR_OR_NULL(type))
+				typestr = fetch_type_from_btf_type(btf, type, ctx);
 			btf_put(btf);
 		}
 	}
@@ -477,6 +617,28 @@ static const struct fetch_type *parse_btf_retval_type(
 	return find_fetch_type(typestr, ctx->flags);
 }
 
+static int parse_btf_bitfield(struct fetch_insn **pcode,
+			      struct traceprobe_parse_context *ctx)
+{
+	struct fetch_insn *code = *pcode;
+
+	if ((ctx->last_bitsize % 8 == 0) && ctx->last_bitoffs == 0)
+		return 0;
+
+	code++;
+	if (code->op != FETCH_OP_NOP) {
+		trace_probe_log_err(ctx->offset, TOO_MANY_OPS);
+		return -EINVAL;
+	}
+	*pcode = code;
+
+	code->op = FETCH_OP_MOD_BF;
+	code->lshift = 64 - (ctx->last_bitsize + ctx->last_bitoffs);
+	code->rshift = 64 - ctx->last_bitsize;
+	code->basesize = 64 / 8;
+	return 0;
+}
+
 static bool is_btf_retval_void(const char *funcname)
 {
 	const struct btf_type *t;
@@ -503,14 +665,22 @@ static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr
 	return ERR_PTR(-EOPNOTSUPP);
 }
 
-static int parse_btf_arg(const char *varname, struct fetch_insn *code,
+static int parse_btf_arg(char *varname,
+			 struct fetch_insn **pcode, struct fetch_insn *end,
 			 struct traceprobe_parse_context *ctx)
 {
 	trace_probe_log_err(ctx->offset, NOSUP_BTFARG);
 	return -EOPNOTSUPP;
 }
 
-#define parse_btf_arg_type(idx, ctx)		\
+static int parse_btf_bitfield(struct fetch_insn **pcode,
+			      struct traceprobe_parse_context *ctx)
+{
+	trace_probe_log_err(ctx->offset, NOSUP_BTFARG);
+	return -EOPNOTSUPP;
+}
+
+#define parse_btf_arg_type(ctx)		\
 	find_fetch_type(NULL, ctx->flags)
 
 #define parse_btf_retval_type(ctx)		\
@@ -778,6 +948,8 @@ parse_probe_arg(char *arg, const struct fetch_type *type,
 
 			code->op = deref;
 			code->offset = offset;
+			/* Reset the last type if used */
+			ctx->last_type = NULL;
 		}
 		break;
 	case '\\':	/* Immediate value */
@@ -801,7 +973,7 @@ parse_probe_arg(char *arg, const struct fetch_type *type,
 				trace_probe_log_err(ctx->offset, NOSUP_BTFARG);
 				return -EINVAL;
 			}
-			ret = parse_btf_arg(arg, code, ctx);
+			ret = parse_btf_arg(arg, pcode, end, ctx);
 			break;
 		}
 	}
@@ -947,6 +1119,7 @@ static int traceprobe_parse_probe_arg_body(const char *argv, ssize_t *size,
 		goto out;
 	code[FETCH_INSN_MAX - 1].op = FETCH_OP_END;
 
+	ctx->last_type = NULL;
 	ret = parse_probe_arg(arg, parg->type, &code, &code[FETCH_INSN_MAX - 1],
 			      ctx);
 	if (ret)
@@ -954,9 +1127,9 @@ static int traceprobe_parse_probe_arg_body(const char *argv, ssize_t *size,
 
 	/* Update storing type if BTF is available */
 	if (IS_ENABLED(CONFIG_PROBE_EVENTS_BTF_ARGS) && !t) {
-		if (code->op == FETCH_OP_ARG)
-			parg->type = parse_btf_arg_type(code->param, ctx);
-		else if (code->op == FETCH_OP_RETVAL)
+		if (ctx->last_type)
+			parg->type = parse_btf_arg_type(ctx);
+		else if (ctx->flags & TPARG_FL_RETURN)
 			parg->type = parse_btf_retval_type(ctx);
 	}
 
@@ -1031,6 +1204,11 @@ static int traceprobe_parse_probe_arg_body(const char *argv, ssize_t *size,
 			trace_probe_log_err(ctx->offset + t - arg, BAD_BITFIELD);
 			goto fail;
 		}
+	} else if (IS_ENABLED(CONFIG_PROBE_EVENTS_BTF_ARGS) &&
+		   ctx->last_type) {
+		ret = parse_btf_bitfield(&code, ctx);
+		if (ret)
+			goto fail;
 	}
 	ret = -EINVAL;
 	/* Loop(Array) operation */
diff --git a/kernel/trace/trace_probe.h b/kernel/trace/trace_probe.h
index 4dc91460a75d..6111f1ffca6c 100644
--- a/kernel/trace/trace_probe.h
+++ b/kernel/trace/trace_probe.h
@@ -388,6 +388,9 @@ struct traceprobe_parse_context {
 	const struct btf_param *params;	/* Parameter of the function */
 	s32 nr_params;			/* The number of the parameters */
 	struct btf *btf;		/* The BTF to be used */
+	const struct btf_type *last_type;	/* Saved type */
+	u32 last_bitoffs;		/* Saved bitoffs */
+	u32 last_bitsize;		/* Saved bitsize */
 	unsigned int flags;
 	int offset;
 };
@@ -503,7 +506,13 @@ extern int traceprobe_define_arg_fields(struct trace_event_call *event_call,
 	C(BAD_VAR_ARGS,		"$arg* must be an independent parameter without name etc."),\
 	C(NOFENTRY_ARGS,	"$arg* can be used only on function entry"),	\
 	C(DOUBLE_ARGS,		"$arg* can be used only once in the parameters"),	\
-	C(ARGS_2LONG,		"$arg* failed because the argument list is too long"),
+	C(ARGS_2LONG,		"$arg* failed because the argument list is too long"),	\
+	C(ARGIDX_2BIG,		"$argN index is too big"),		\
+	C(NO_PTR_STRCT,		"This is not a pointer to union/structure."),	\
+	C(NOSUP_DAT_ARG,	"Non pointer structure/union argument is not supported."),\
+	C(BAD_HYPHEN,		"Failed to parse single hyphen. Forgot '>'?"),	\
+	C(NO_BTF_FIELD,		"This field is not found."),	\
+	C(BAD_BTF_TID,		"Failed to get BTF type info."),
 
 #undef C
 #define C(a, b)		TP_ERR_##a

[PATCH v5 5/9] tracing/probes: Support BTF field access from $retval

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Support BTF argument on '$retval' for function return events including
kretprobe and fprobe for accessing the return value.
This also allows user to access its fields if the return value is a
pointer of a data structure.

E.g.
 # echo 'f getname_flags%return +0($retval->name):string' \
   > dynamic_events
 # echo 1 > events/fprobes/getname_flags__exit/enable
 # ls > /dev/null
 # head -n 40 trace | tail
              ls-87      [000] ...1.  8067.616101: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./function_profile_enabled"
              ls-87      [000] ...1.  8067.616108: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./trace_stat"
              ls-87      [000] ...1.  8067.616115: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./set_graph_notrace"
              ls-87      [000] ...1.  8067.616122: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./set_graph_function"
              ls-87      [000] ...1.  8067.616129: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./set_ftrace_notrace"
              ls-87      [000] ...1.  8067.616135: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./set_ftrace_filter"
              ls-87      [000] ...1.  8067.616143: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./touched_functions"
              ls-87      [000] ...1.  8067.616237: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./enabled_functions"
              ls-87      [000] ...1.  8067.616245: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./available_filter_functions"
              ls-87      [000] ...1.  8067.616253: getname_flags__exit: (vfs_fstatat+0x3c/0x70 <- getname_flags) arg1="./set_ftrace_notrace_pid"


Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
---
 Changes in v2:
  - Use '$retval' instead of 'retval' because it is confusing.
 Changes in v3:
  - Introduce query_btf_context() to cache the btf related data (function
    prototype) for using common field analyzing code with function
    parameters.
 Changes in v3.1:
  - Return int error code from query_btf_context() if !CONFIG_PROBE_EVENTS_BTF_ARGS
 Changes in v4:
  - Fix wrong BTF access if query_btf_context() is failed in parse_btf_arg().
  - Return error if $retval accesses a field but BTF is not found.
---
 kernel/trace/trace_probe.c |  187 ++++++++++++++++++++------------------------
 kernel/trace/trace_probe.h |    1 
 2 files changed, 86 insertions(+), 102 deletions(-)

diff --git a/kernel/trace/trace_probe.c b/kernel/trace/trace_probe.c
index 821f43e5c52b..7345e1af4db2 100644
--- a/kernel/trace/trace_probe.c
+++ b/kernel/trace/trace_probe.c
@@ -364,38 +364,46 @@ static const char *fetch_type_from_btf_type(struct btf *btf,
 	return NULL;
 }
 
-static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr,
-						   struct btf **btf_p, bool tracepoint)
+static int query_btf_context(struct traceprobe_parse_context *ctx)
 {
 	const struct btf_param *param;
-	const struct btf_type *t;
+	const struct btf_type *type;
 	struct btf *btf;
+	s32 nr;
 
-	if (!funcname || !nr)
-		return ERR_PTR(-EINVAL);
+	if (ctx->btf)
+		return 0;
 
-	t = btf_find_func_proto(funcname, &btf);
-	if (!t)
-		return (const struct btf_param *)t;
+	if (!ctx->funcname)
+		return -EINVAL;
 
-	param = btf_get_func_param(t, nr);
-	if (IS_ERR_OR_NULL(param))
-		goto err;
+	type = btf_find_func_proto(ctx->funcname, &btf);
+	if (!type)
+		return -ENOENT;
 
-	/* Hide the first 'data' argument of tracepoint */
-	if (tracepoint) {
-		(*nr)--;
-		param++;
+	ctx->btf = btf;
+	ctx->proto = type;
+
+	/* ctx->params is optional, since func(void) will not have params. */
+	nr = 0;
+	param = btf_get_func_param(type, &nr);
+	if (!IS_ERR_OR_NULL(param)) {
+		/* Hide the first 'data' argument of tracepoint */
+		if (ctx->flags & TPARG_FL_TPOINT) {
+			nr--;
+			param++;
+		}
 	}
 
-	if (*nr > 0) {
-		*btf_p = btf;
-		return param;
+	if (nr > 0) {
+		ctx->nr_params = nr;
+		ctx->params = param;
+	} else {
+		ctx->nr_params = 0;
+		ctx->params = NULL;
 	}
 
-err:
-	btf_put(btf);
-	return NULL;
+	return 0;
 }
 
 static void clear_btf_context(struct traceprobe_parse_context *ctx)
@@ -403,6 +411,7 @@ static void clear_btf_context(struct traceprobe_parse_context *ctx)
 	if (ctx->btf) {
 		btf_put(ctx->btf);
 		ctx->btf = NULL;
+		ctx->proto = NULL;
 		ctx->params = NULL;
 		ctx->nr_params = 0;
 	}
@@ -522,7 +531,7 @@ static int parse_btf_arg(char *varname,
 	const struct btf_param *params;
 	const struct btf_type *type;
 	char *field = NULL;
-	int i, is_ptr;
+	int i, is_ptr, ret;
 	u32 tid;
 
 	if (WARN_ON_ONCE(!ctx->funcname))
@@ -538,17 +547,37 @@ static int parse_btf_arg(char *varname,
 		return -EOPNOTSUPP;
 	}
 
-	if (!ctx->params) {
-		params = find_btf_func_param(ctx->funcname,
-					     &ctx->nr_params, &ctx->btf,
-					     ctx->flags & TPARG_FL_TPOINT);
-		if (IS_ERR_OR_NULL(params)) {
+	if (ctx->flags & TPARG_FL_RETURN) {
+		if (strcmp(varname, "$retval") != 0) {
+			trace_probe_log_err(ctx->offset, NO_BTFARG);
+			return -ENOENT;
+		}
+		code->op = FETCH_OP_RETVAL;
+		/* Check whether the function return type is not void */
+		if (query_btf_context(ctx) == 0) {
+			if (ctx->proto->type == 0) {
+				trace_probe_log_err(ctx->offset, NO_RETVAL);
+				return -ENOENT;
+			}
+			tid = ctx->proto->type;
+			goto found;
+		}
+		if (field) {
+			trace_probe_log_err(ctx->offset + field - varname,
+					    NO_BTF_ENTRY);
+			return -ENOENT;
+		}
+		return 0;
+	}
+
+	if (!ctx->btf) {
+		ret = query_btf_context(ctx);
+		if (ret < 0 || ctx->nr_params == 0) {
 			trace_probe_log_err(ctx->offset, NO_BTF_ENTRY);
 			return PTR_ERR(params);
 		}
-		ctx->params = params;
-	} else
-		params = ctx->params;
+	}
+	params = ctx->params;
 
 	for (i = 0; i < ctx->nr_params; i++) {
 		const char *name = btf_name_by_offset(ctx->btf, params[i].name_off);
@@ -559,7 +588,6 @@ static int parse_btf_arg(char *varname,
 				code->param = i + 1;
 			else
 				code->param = i;
-
 			tid = params[i].type;
 			goto found;
 		}
@@ -584,7 +612,7 @@ static int parse_btf_arg(char *varname,
 	return 0;
 }
 
-static const struct fetch_type *parse_btf_arg_type(
+static const struct fetch_type *find_fetch_type_from_btf_type(
 					struct traceprobe_parse_context *ctx)
 {
 	struct btf *btf = ctx->btf;
@@ -596,27 +624,6 @@ static const struct fetch_type *parse_btf_arg_type(
 	return find_fetch_type(typestr, ctx->flags);
 }
 
-static const struct fetch_type *parse_btf_retval_type(
-					struct traceprobe_parse_context *ctx)
-{
-	const char *typestr = NULL;
-	const struct btf_type *type;
-	struct btf *btf;
-
-	if (ctx->funcname) {
-		/* Do not use ctx->btf, because it must be used with ctx->param */
-		type = btf_find_func_proto(ctx->funcname, &btf);
-		if (type) {
-			type = btf_type_skip_modifiers(btf, type->type, NULL);
-			if (!IS_ERR_OR_NULL(type))
-				typestr = fetch_type_from_btf_type(btf, type, ctx);
-			btf_put(btf);
-		}
-	}
-
-	return find_fetch_type(typestr, ctx->flags);
-}
-
 static int parse_btf_bitfield(struct fetch_insn **pcode,
 			      struct traceprobe_parse_context *ctx)
 {
@@ -639,30 +646,15 @@ static int parse_btf_bitfield(struct fetch_insn **pcode,
 	return 0;
 }
 
-static bool is_btf_retval_void(const char *funcname)
-{
-	const struct btf_type *t;
-	struct btf *btf;
-	bool ret;
-
-	t = btf_find_func_proto(funcname, &btf);
-	if (!t)
-		return false;
-
-	ret = (t->type == 0);
-	btf_put(btf);
-	return ret;
-}
 #else
 static void clear_btf_context(struct traceprobe_parse_context *ctx)
 {
 	ctx->btf = NULL;
 }
 
-static const struct btf_param *find_btf_func_param(const char *funcname, s32 *nr,
-						   struct btf **btf_p, bool tracepoint)
+static int query_btf_context(struct traceprobe_parse_context *ctx)
 {
-	return ERR_PTR(-EOPNOTSUPP);
+	return -EOPNOTSUPP;
 }
 
 static int parse_btf_arg(char *varname,
@@ -680,24 +672,23 @@ static int parse_btf_bitfield(struct fetch_insn **pcode,
 	return -EOPNOTSUPP;
 }
 
-#define parse_btf_arg_type(ctx)		\
-	find_fetch_type(NULL, ctx->flags)
-
-#define parse_btf_retval_type(ctx)		\
+#define find_fetch_type_from_btf_type(ctx)		\
 	find_fetch_type(NULL, ctx->flags)
 
-#define is_btf_retval_void(funcname)	(false)
-
 #endif
 
 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 
-static int parse_probe_vars(char *arg, const struct fetch_type *t,
-			    struct fetch_insn *code,
+/* Parse $vars. @orig_arg points '$', which syncs to @ctx->offset */
+static int parse_probe_vars(char *orig_arg, const struct fetch_type *t,
+			    struct fetch_insn **pcode,
+			    struct fetch_insn *end,
 			    struct traceprobe_parse_context *ctx)
 {
-	unsigned long param;
+	struct fetch_insn *code = *pcode;
 	int err = TP_ERR_BAD_VAR;
+	char *arg = orig_arg + 1;
+	unsigned long param;
 	int ret = 0;
 	int len;
 
@@ -716,18 +707,17 @@ static int parse_probe_vars(char *arg, const struct fetch_type *t,
 		goto inval;
 	}
 
-	if (strcmp(arg, "retval") == 0) {
-		if (ctx->flags & TPARG_FL_RETURN) {
-			if ((ctx->flags & TPARG_FL_KERNEL) &&
-			    is_btf_retval_void(ctx->funcname)) {
-				err = TP_ERR_NO_RETVAL;
-				goto inval;
-			}
+	if (str_has_prefix(arg, "retval")) {
+		if (!(ctx->flags & TPARG_FL_RETURN)) {
+			err = TP_ERR_RETVAL_ON_PROBE;
+			goto inval;
+		}
+		if (!(ctx->flags & TPARG_FL_KERNEL) ||
+		    !IS_ENABLED(CONFIG_PROBE_EVENTS_BTF_ARGS)) {
 			code->op = FETCH_OP_RETVAL;
 			return 0;
 		}
-		err = TP_ERR_RETVAL_ON_PROBE;
-		goto inval;
+		return parse_btf_arg(orig_arg, pcode, end, ctx);
 	}
 
 	len = str_has_prefix(arg, "stack");
@@ -829,7 +819,7 @@ parse_probe_arg(char *arg, const struct fetch_type *type,
 
 	switch (arg[0]) {
 	case '$':
-		ret = parse_probe_vars(arg + 1, type, code, ctx);
+		ret = parse_probe_vars(arg, type, pcode, end, ctx);
 		break;
 
 	case '%':	/* named register */
@@ -1126,12 +1116,9 @@ static int traceprobe_parse_probe_arg_body(const char *argv, ssize_t *size,
 		goto fail;
 
 	/* Update storing type if BTF is available */
-	if (IS_ENABLED(CONFIG_PROBE_EVENTS_BTF_ARGS) && !t) {
-		if (ctx->last_type)
-			parg->type = parse_btf_arg_type(ctx);
-		else if (ctx->flags & TPARG_FL_RETURN)
-			parg->type = parse_btf_retval_type(ctx);
-	}
+	if (IS_ENABLED(CONFIG_PROBE_EVENTS_BTF_ARGS) &&
+	    !t && ctx->last_type)
+		parg->type = find_fetch_type_from_btf_type(ctx);
 
 	ret = -EINVAL;
 	/* Store operation */
@@ -1420,7 +1407,6 @@ const char **traceprobe_expand_meta_args(int argc, const char *argv[],
 	const struct btf_param *params = NULL;
 	int i, j, n, used, ret, args_idx = -1;
 	const char **new_argv = NULL;
-	int nr_params;
 
 	ret = argv_has_var_arg(argc, argv, &args_idx, ctx);
 	if (ret < 0)
@@ -1431,9 +1417,8 @@ const char **traceprobe_expand_meta_args(int argc, const char *argv[],
 		return NULL;
 	}
 
-	params = find_btf_func_param(ctx->funcname, &nr_params, &ctx->btf,
-				     ctx->flags & TPARG_FL_TPOINT);
-	if (IS_ERR_OR_NULL(params)) {
+	ret = query_btf_context(ctx);
+	if (ret < 0 || ctx->nr_params == 0) {
 		if (args_idx != -1) {
 			/* $arg* requires BTF info */
 			trace_probe_log_err(0, NOSUP_BTFARG);
@@ -1442,8 +1427,6 @@ const char **traceprobe_expand_meta_args(int argc, const char *argv[],
 		*new_argc = argc;
 		return NULL;
 	}
-	ctx->params = params;
-	ctx->nr_params = nr_params;
 
 	if (args_idx >= 0)
 		*new_argc = argc + ctx->nr_params - 1;
@@ -1458,7 +1441,7 @@ const char **traceprobe_expand_meta_args(int argc, const char *argv[],
 	for (i = 0, j = 0; i < argc; i++) {
 		trace_probe_log_set_index(i + 2);
 		if (i == args_idx) {
-			for (n = 0; n < nr_params; n++) {
+			for (n = 0; n < ctx->nr_params; n++) {
 				ret = sprint_nth_btf_arg(n, "", buf + used,
 							 bufsize - used, ctx);
 				if (ret < 0)
diff --git a/kernel/trace/trace_probe.h b/kernel/trace/trace_probe.h
index 6111f1ffca6c..9184c84833f8 100644
--- a/kernel/trace/trace_probe.h
+++ b/kernel/trace/trace_probe.h
@@ -385,6 +385,7 @@ struct traceprobe_parse_context {
 	struct trace_event_call *event;
 	/* BTF related parameters */
 	const char *funcname;		/* Function name in BTF */
+	const struct btf_type  *proto;	/* Prototype of the function */
 	const struct btf_param *params;	/* Parameter of the function */
 	s32 nr_params;			/* The number of the parameters */
 	struct btf *btf;		/* The BTF to be used */

[PATCH v5 6/9] tracing/probes: Add string type check with BTF

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Add a string type checking with BTF information if possible.
This will check whether the given BTF argument (and field) is
signed char array or pointer to signed char. If not, it reject
the 'string' type. If it is pointer to signed char, it adds
a dereference opration so that it can correctly fetch the
string data from memory.

 # echo 'f getname_flags%return retval->name:string' >> dynamic_events
 # echo 't sched_switch next->comm:string' >> dynamic_events

The above cases, 'struct filename::name' is 'char *' and
'struct task_struct::comm' is 'char []'. But in both case,
user can specify ':string' to fetch the string data.

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
---
 Changes in v3:
  - Use ctx->btf instead of traceprobe_get_btf().
---
 kernel/trace/trace_probe.c |   89 +++++++++++++++++++++++++++++++++++++++++++-
 kernel/trace/trace_probe.h |    3 +
 2 files changed, 89 insertions(+), 3 deletions(-)

diff --git a/kernel/trace/trace_probe.c b/kernel/trace/trace_probe.c
index 7345e1af4db2..4dc74d73fc1d 100644
--- a/kernel/trace/trace_probe.c
+++ b/kernel/trace/trace_probe.c
@@ -310,6 +310,77 @@ static u32 btf_type_int(const struct btf_type *t)
 	return *(u32 *)(t + 1);
 }
 
+static bool btf_type_is_char_ptr(struct btf *btf, const struct btf_type *type)
+{
+	const struct btf_type *real_type;
+	u32 intdata;
+	s32 tid;
+
+	real_type = btf_type_skip_modifiers(btf, type->type, &tid);
+	if (!real_type)
+		return false;
+
+	if (BTF_INFO_KIND(real_type->info) != BTF_KIND_INT)
+		return false;
+
+	intdata = btf_type_int(real_type);
+	return !(BTF_INT_ENCODING(intdata) & BTF_INT_SIGNED)
+		&& BTF_INT_BITS(intdata) == 8;
+}
+
+static bool btf_type_is_char_array(struct btf *btf, const struct btf_type *type)
+{
+	const struct btf_type *real_type;
+	const struct btf_array *array;
+	u32 intdata;
+	s32 tid;
+
+	if (BTF_INFO_KIND(type->info) != BTF_KIND_ARRAY)
+		return false;
+
+	array = (const struct btf_array *)(type + 1);
+
+	real_type = btf_type_skip_modifiers(btf, array->type, &tid);
+
+	intdata = btf_type_int(real_type);
+	return !(BTF_INT_ENCODING(intdata) & BTF_INT_SIGNED)
+		&& BTF_INT_BITS(intdata) == 8;
+}
+
+static int check_prepare_btf_string_fetch(char *typename,
+				struct fetch_insn **pcode,
+				struct traceprobe_parse_context *ctx)
+{
+	struct btf *btf = ctx->btf;
+
+	if (!btf || !ctx->last_type)
+		return 0;
+
+	/* char [] does not need any change. */
+	if (btf_type_is_char_array(btf, ctx->last_type))
+		return 0;
+
+	/* char * requires dereference the pointer. */
+	if (btf_type_is_char_ptr(btf, ctx->last_type)) {
+		struct fetch_insn *code = *pcode + 1;
+
+		if (code->op == FETCH_OP_END) {
+			trace_probe_log_err(ctx->offset, TOO_MANY_OPS);
+			return -E2BIG;
+		}
+		if (typename[0] == 'u')
+			code->op = FETCH_OP_UDEREF;
+		else
+			code->op = FETCH_OP_DEREF;
+		code->offset = 0;
+		*pcode = code;
+		return 0;
+	}
+	/* Other types are not available for string */
+	trace_probe_log_err(ctx->offset, BAD_TYPE4STR);
+	return -EINVAL;
+}
+
 static const char *fetch_type_from_btf_type(struct btf *btf,
 					const struct btf_type *type,
 					struct traceprobe_parse_context *ctx)
@@ -675,6 +746,13 @@ static int parse_btf_bitfield(struct fetch_insn **pcode,
 #define find_fetch_type_from_btf_type(ctx)		\
 	find_fetch_type(NULL, ctx->flags)
 
+static int check_prepare_btf_string_fetch(char *typename,
+				struct fetch_insn **pcode,
+				struct traceprobe_parse_context *ctx)
+{
+	return 0;
+}
+
 #endif
 
 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
@@ -1117,8 +1195,15 @@ static int traceprobe_parse_probe_arg_body(const char *argv, ssize_t *size,
 
 	/* Update storing type if BTF is available */
 	if (IS_ENABLED(CONFIG_PROBE_EVENTS_BTF_ARGS) &&
-	    !t && ctx->last_type)
-		parg->type = find_fetch_type_from_btf_type(ctx);
+	    ctx->last_type) {
+		if (!t) {
+			parg->type = find_fetch_type_from_btf_type(ctx);
+		} else if (strstr(t, "string")) {
+			ret = check_prepare_btf_string_fetch(t, &code, ctx);
+			if (ret)
+				goto fail;
+		}
+	}
 
 	ret = -EINVAL;
 	/* Store operation */
diff --git a/kernel/trace/trace_probe.h b/kernel/trace/trace_probe.h
index 9184c84833f8..7f929482e8d4 100644
--- a/kernel/trace/trace_probe.h
+++ b/kernel/trace/trace_probe.h
@@ -513,7 +513,8 @@ extern int traceprobe_define_arg_fields(struct trace_event_call *event_call,
 	C(NOSUP_DAT_ARG,	"Non pointer structure/union argument is not supported."),\
 	C(BAD_HYPHEN,		"Failed to parse single hyphen. Forgot '>'?"),	\
 	C(NO_BTF_FIELD,		"This field is not found."),	\
-	C(BAD_BTF_TID,		"Failed to get BTF type info."),
+	C(BAD_BTF_TID,		"Failed to get BTF type info."),\
+	C(BAD_TYPE4STR,		"This type does not fit for string."),
 
 #undef C
 #define C(a, b)		TP_ERR_##a

[PATCH v5 7/9] tracing/fprobe-event: Assume fprobe is a return event by $retval

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Assume the fprobe event is a return event if there is $retval is
used in the probe's argument without %return. e.g.

echo 'f:myevent vfs_read $retval' >> dynamic_events

then 'myevent' is a return probe event.

Suggested-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
---
 kernel/trace/trace_fprobe.c                        |   58 +++++++++++++++-----
 .../ftrace/test.d/dynevent/fprobe_syntax_errors.tc |    2 -
 2 files changed, 45 insertions(+), 15 deletions(-)

diff --git a/kernel/trace/trace_fprobe.c b/kernel/trace/trace_fprobe.c
index 8f43f1f65b1b..8bfe23af9c73 100644
--- a/kernel/trace/trace_fprobe.c
+++ b/kernel/trace/trace_fprobe.c
@@ -898,6 +898,46 @@ static struct tracepoint *find_tracepoint(const char *tp_name)
 	return data.tpoint;
 }
 
+static int parse_symbol_and_return(int argc, const char *argv[],
+				   char **symbol, bool *is_return,
+				   bool is_tracepoint)
+{
+	char *tmp = strchr(argv[1], '%');
+	int i;
+
+	if (tmp) {
+		int len = tmp - argv[1];
+
+		if (!is_tracepoint && !strcmp(tmp, "%return")) {
+			*is_return = true;
+		} else {
+			trace_probe_log_err(len, BAD_ADDR_SUFFIX);
+			return -EINVAL;
+		}
+		*symbol = kmemdup_nul(argv[1], len, GFP_KERNEL);
+	} else
+		*symbol = kstrdup(argv[1], GFP_KERNEL);
+	if (!*symbol)
+		return -ENOMEM;
+
+	if (*is_return)
+		return 0;
+
+	/* If there is $retval, this should be a return fprobe. */
+	for (i = 2; i < argc; i++) {
+		tmp = strstr(argv[i], "$retval");
+		if (tmp && !isalnum(tmp[7]) && tmp[7] != '_') {
+			*is_return = true;
+			/*
+			 * NOTE: Don't check is_tracepoint here, because it will
+			 * be checked when the argument is parsed.
+			 */
+			break;
+		}
+	}
+	return 0;
+}
+
 static int __trace_fprobe_create(int argc, const char *argv[])
 {
 	/*
@@ -927,7 +967,7 @@ static int __trace_fprobe_create(int argc, const char *argv[])
 	struct trace_fprobe *tf = NULL;
 	int i, len, new_argc = 0, ret = 0;
 	bool is_return = false;
-	char *symbol = NULL, *tmp = NULL;
+	char *symbol = NULL;
 	const char *event = NULL, *group = FPROBE_EVENT_SYSTEM;
 	const char **new_argv = NULL;
 	int maxactive = 0;
@@ -983,20 +1023,10 @@ static int __trace_fprobe_create(int argc, const char *argv[])
 	trace_probe_log_set_index(1);
 
 	/* a symbol(or tracepoint) must be specified */
-	symbol = kstrdup(argv[1], GFP_KERNEL);
-	if (!symbol)
-		return -ENOMEM;
+	ret = parse_symbol_and_return(argc, argv, &symbol, &is_return, is_tracepoint);
+	if (ret < 0)
+		goto parse_error;
 
-	tmp = strchr(symbol, '%');
-	if (tmp) {
-		if (!is_tracepoint && !strcmp(tmp, "%return")) {
-			*tmp = '\0';
-			is_return = true;
-		} else {
-			trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
-			goto parse_error;
-		}
-	}
 	if (!is_return && maxactive) {
 		trace_probe_log_set_index(0);
 		trace_probe_log_err(1, BAD_MAXACT_TYPE);
diff --git a/tools/testing/selftests/ftrace/test.d/dynevent/fprobe_syntax_errors.tc b/tools/testing/selftests/ftrace/test.d/dynevent/fprobe_syntax_errors.tc
index 812f5b3f6055..72563b2e0812 100644
--- a/tools/testing/selftests/ftrace/test.d/dynevent/fprobe_syntax_errors.tc
+++ b/tools/testing/selftests/ftrace/test.d/dynevent/fprobe_syntax_errors.tc
@@ -30,11 +30,11 @@ check_error 'f:^ vfs_read'		# NO_EVENT_NAME
 check_error 'f:foo/^12345678901234567890123456789012345678901234567890123456789012345 vfs_read'	# EVENT_TOO_LONG
 check_error 'f:foo/^bar.1 vfs_read'	# BAD_EVENT_NAME
 
-check_error 'f vfs_read ^$retval'	# RETVAL_ON_PROBE
 check_error 'f vfs_read ^$stack10000'	# BAD_STACK_NUM
 
 check_error 'f vfs_read ^$arg10000'	# BAD_ARG_NUM
 
+check_error 'f vfs_read $retval ^$arg1' # BAD_VAR
 check_error 'f vfs_read ^$none_var'	# BAD_VAR
 check_error 'f vfs_read ^'$REG		# BAD_VAR
 

[PATCH v5 8/9] selftests/ftrace: Add BTF fields access testcases

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Add test cases for accessing the data structure fields using BTF info.
This includes the field access from parameters and retval, and accessing
string information.

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
---
Changes in v2:
 - Use '$retval' instead of 'retval'.
 - Add a test that use both '$retval' and '$arg1' for fprobe.
Changes in v3:
 - Change a test case with a numeric value.
 - Add a test case with mixed '.' and '->' operators.
Changes in v5:
 - Check fields access availability.
---
 .../ftrace/test.d/dynevent/add_remove_btfarg.tc    |   20 ++++++++++++++++++++
 .../ftrace/test.d/dynevent/fprobe_syntax_errors.tc |    8 ++++++++
 2 files changed, 28 insertions(+)

diff --git a/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_btfarg.tc b/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_btfarg.tc
index f34b14ef9781..b9c21a81d248 100644
--- a/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_btfarg.tc
+++ b/tools/testing/selftests/ftrace/test.d/dynevent/add_remove_btfarg.tc
@@ -5,6 +5,7 @@
 
 KPROBES=
 FPROBES=
+FIELDS=
 
 if grep -qF "p[:[<group>/][<event>]] <place> [<args>]" README ; then
   KPROBES=yes
@@ -12,6 +13,9 @@ fi
 if grep -qF "f[:[<group>/][<event>]] <func-name>[%return] [<args>]" README ; then
   FPROBES=yes
 fi
+if grep -qF "<argname>[->field[->field|.field...]]" README ; then
+  FIELDS=yes
+fi
 
 if [ -z "$KPROBES" -a -z "$FPROBES" ] ; then
   exit_unsupported
@@ -21,6 +25,9 @@ echo 0 > events/enable
 echo > dynamic_events
 
 TP=kfree
+TP2=kmem_cache_alloc
+TP3=getname_flags
+TP4=sched_wakeup
 
 if [ "$FPROBES" ] ; then
 echo "f:fpevent $TP object" >> dynamic_events
@@ -33,6 +40,7 @@ echo > dynamic_events
 
 echo "f:fpevent $TP "'$arg1' >> dynamic_events
 grep -q "fpevent.*object=object" dynamic_events
+
 echo > dynamic_events
 
 echo "f:fpevent $TP "'$arg*' >> dynamic_events
@@ -45,6 +53,18 @@ fi
 
 echo > dynamic_events
 
+if [ "$FIELDS" ] ; then
+echo "t:tpevent ${TP2} obj_size=s->object_size" >> dynamic_events
+echo "f:fpevent ${TP3}%return path=\$retval->name:string" >> dynamic_events
+echo "t:tpevent2 ${TP4} p->se.group_node.next->prev" >> dynamic_events
+
+grep -q "tpevent .*obj_size=s->object_size" dynamic_events
+grep -q "fpevent.*path=\$retval->name:string" dynamic_events
+grep -q 'tpevent2 .*p->se.group_node.next->prev' dynamic_events
+
+echo > dynamic_events
+fi
+
 if [ "$KPROBES" ] ; then
 echo "p:kpevent $TP object" >> dynamic_events
 grep -q "kpevent.*object=object" dynamic_events
diff --git a/tools/testing/selftests/ftrace/test.d/dynevent/fprobe_syntax_errors.tc b/tools/testing/selftests/ftrace/test.d/dynevent/fprobe_syntax_errors.tc
index 72563b2e0812..20e42c030095 100644
--- a/tools/testing/selftests/ftrace/test.d/dynevent/fprobe_syntax_errors.tc
+++ b/tools/testing/selftests/ftrace/test.d/dynevent/fprobe_syntax_errors.tc
@@ -103,6 +103,14 @@ check_error 'f vfs_read%return ^$arg*'		# NOFENTRY_ARGS
 check_error 'f vfs_read ^hoge'			# NO_BTFARG
 check_error 'f kfree ^$arg10'			# NO_BTFARG (exceed the number of parameters)
 check_error 'f kfree%return ^$retval'		# NO_RETVAL
+
+if grep -qF "<argname>[->field[->field|.field...]]" README ; then
+check_error 'f vfs_read%return $retval->^foo'	# NO_PTR_STRCT
+check_error 'f vfs_read file->^foo'		# NO_BTF_FIELD
+check_error 'f vfs_read file^-.foo'		# BAD_HYPHEN
+check_error 'f vfs_read ^file:string'		# BAD_TYPE4STR
+fi
+
 else
 check_error 'f vfs_read ^$arg*'			# NOSUP_BTFARG
 check_error 't kfree ^$arg*'			# NOSUP_BTFARG

[PATCH v5 9/9] Documentation: tracing: Update fprobe event example with BTF field

[Masami Hiramatsu (Google)]

From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Update fprobe event example with BTF data structure field specification.

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
---
 Changes in v2:
  - Remove 'retval' and use '$retval'.
 Changes in v3:
  - Add description about mixture of '.' and '->' usage.
---
 Documentation/trace/fprobetrace.rst |   64 +++++++++++++++++++++++++----------
 1 file changed, 46 insertions(+), 18 deletions(-)

diff --git a/Documentation/trace/fprobetrace.rst b/Documentation/trace/fprobetrace.rst
index 7297f9478459..8e9bebcf0a2e 100644
--- a/Documentation/trace/fprobetrace.rst
+++ b/Documentation/trace/fprobetrace.rst
@@ -79,9 +79,9 @@ automatically set by the given name. ::
  f:fprobes/myprobe vfs_read count=count pos=pos
 
 It also chooses the fetch type from BTF information. For example, in the above
-example, the ``count`` is unsigned long, and the ``pos`` is a pointer. Thus, both
-are converted to 64bit unsigned long, but only ``pos`` has "%Lx" print-format as
-below ::
+example, the ``count`` is unsigned long, and the ``pos`` is a pointer. Thus,
+both are converted to 64bit unsigned long, but only ``pos`` has "%Lx"
+print-format as below ::
 
  # cat events/fprobes/myprobe/format
  name: myprobe
@@ -105,9 +105,47 @@ is expanded to all function arguments of the function or the tracepoint. ::
  # cat dynamic_events
  f:fprobes/myprobe vfs_read file=file buf=buf count=count pos=pos
 
-BTF also affects the ``$retval``. If user doesn't set any type, the retval type is
-automatically picked from the BTF. If the function returns ``void``, ``$retval``
-is rejected.
+BTF also affects the ``$retval``. If user doesn't set any type, the retval
+type is automatically picked from the BTF. If the function returns ``void``,
+``$retval`` is rejected.
+
+You can access the data fields of a data structure using allow operator ``->``
+(for pointer type) and dot operator ``.`` (for data structure type.)::
+
+# echo 't sched_switch preempt prev_pid=prev->pid next_pid=next->pid' >> dynamic_events
+
+The field access operators, ``->`` and ``.`` can be combined for accessing deeper
+members and other structure members pointed by the member. e.g. ``foo->bar.baz->qux``
+If there is non-name union member, you can directly access it as the C code does.
+For example::
+
+ struct {
+	union {
+	int a;
+	int b;
+	};
+ } *foo;
+
+To access ``a`` and ``b``, use ``foo->a`` and ``foo->b`` in this case.
+
+This data field access is available for the return value via ``$retval``,
+e.g. ``$retval->name``.
+
+For these BTF arguments and fields, ``:string`` and ``:ustring`` change the
+behavior. If these are used for BTF argument or field, it checks whether
+the BTF type of the argument or the data field is ``char *`` or ``char []``,
+or not.  If not, it rejects applying the string types. Also, with the BTF
+support, you don't need a memory dereference operator (``+0(PTR)``) for
+accessing the string pointed by a ``PTR``. It automatically adds the memory
+dereference operator according to the BTF type. e.g. ::
+
+# echo 't sched_switch prev->comm:string' >> dynamic_events
+# echo 'f getname_flags%return $retval->name:string' >> dynamic_events
+
+The ``prev->comm`` is an embedded char array in the data structure, and
+``$retval->name`` is a char pointer in the data structure. But in both
+cases, you can use ``:string`` type to get the string.
+
 
 Usage examples
 --------------
@@ -161,10 +199,10 @@ parameters. This means you can access any field values in the task
 structure pointed by the ``prev`` and ``next`` arguments.
 
 For example, usually ``task_struct::start_time`` is not traced, but with this
-traceprobe event, you can trace it as below.
+traceprobe event, you can trace that field as below.
 ::
 
-  # echo 't sched_switch comm=+1896(next):string start_time=+1728(next):u64' > dynamic_events
+  # echo 't sched_switch comm=next->comm:string next->start_time' > dynamic_events
   # head -n 20 trace | tail
  #           TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
  #              | |         |   |||||     |         |
@@ -176,13 +214,3 @@ traceprobe event, you can trace it as below.
            <idle>-0       [000] d..3.  5606.690317: sched_switch: (__probestub_sched_switch+0x4/0x10) comm="kworker/0:1" usage=1 start_time=137000000
       kworker/0:1-14      [000] d..3.  5606.690339: sched_switch: (__probestub_sched_switch+0x4/0x10) comm="swapper/0" usage=2 start_time=0
            <idle>-0       [000] d..3.  5606.692368: sched_switch: (__probestub_sched_switch+0x4/0x10) comm="kworker/0:1" usage=1 start_time=137000000
-
-Currently, to find the offset of a specific field in the data structure,
-you need to build kernel with debuginfo and run `perf probe` command with
-`-D` option. e.g.
-::
-
- # perf probe -D "__probestub_sched_switch next->comm:string next->start_time"
- p:probe/__probestub_sched_switch __probestub_sched_switch+0 comm=+1896(%cx):string start_time=+1728(%cx):u64
-
-And replace the ``%cx`` with the ``next``.

Re: [PATCH v5 0/9] tracing: Improbe BTF support on probe events

[Masami Hiramatsu]

Hi Steve,

Can you review this series?
I would like to push this to for-next.

Thank you,

On Mon,  7 Aug 2023 11:54:28 +0900
"Masami Hiramatsu (Google)" <mhiramat@kernel.org> wrote:

> Hi,
> 
> Here is the 5th version of series to improve the BTF support on probe events.
> The previous series is here:
> 
> https://lore.kernel.org/all/169078860386.173706.3091034523220945605.stgit@devnote2/
> 
> This version introduces kernel/trace/trace_btf.c to separate the btf generic
> functions. These functions will be moved to btf.c next merge window.
> This fixes the member-search function to return the bit-offset of the
> parent anonymous union/structure. Thus the caller can calculate the real
> bit-offset from the root data structure.
> This also fixes the ftrace selftest issue which fails if the kernel
> supports only BTF args but not support field access.
> 
> This series can be applied on top of "probes/core" branch of
> https://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace.git/
> 
> You can also get this series from:
> 
> git://git.kernel.org/pub/scm/linux/kernel/git/mhiramat/linux.git topic/fprobe-event-ext
> 
> Thank you,
> 
> ---
> 
> Masami Hiramatsu (Google) (9):
>       tracing/probes: Support BTF argument on module functions
>       tracing/probes: Move finding func-proto API and getting func-param API to trace_btf
>       tracing/probes: Add a function to search a member of a struct/union
>       tracing/probes: Support BTF based data structure field access
>       tracing/probes: Support BTF field access from $retval
>       tracing/probes: Add string type check with BTF
>       tracing/fprobe-event: Assume fprobe is a return event by $retval
>       selftests/ftrace: Add BTF fields access testcases
>       Documentation: tracing: Update fprobe event example with BTF field
> 
> 
>  Documentation/trace/fprobetrace.rst                |   64 ++-
>  include/linux/btf.h                                |    1 
>  kernel/bpf/btf.c                                   |    2 
>  kernel/trace/Makefile                              |    1 
>  kernel/trace/trace.c                               |    3 
>  kernel/trace/trace_btf.c                           |  109 ++++
>  kernel/trace/trace_btf.h                           |   11 
>  kernel/trace/trace_eprobe.c                        |    4 
>  kernel/trace/trace_fprobe.c                        |   59 ++
>  kernel/trace/trace_kprobe.c                        |    1 
>  kernel/trace/trace_probe.c                         |  499 +++++++++++++++-----
>  kernel/trace/trace_probe.h                         |   27 +
>  kernel/trace/trace_uprobe.c                        |    1 
>  .../ftrace/test.d/dynevent/add_remove_btfarg.tc    |   20 +
>  .../ftrace/test.d/dynevent/fprobe_syntax_errors.tc |   10 
>  15 files changed, 637 insertions(+), 175 deletions(-)
>  create mode 100644 kernel/trace/trace_btf.c
>  create mode 100644 kernel/trace/trace_btf.h
> 
> --
> Masami Hiramatsu (Google) <mhiramat@kernel.org>


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>

Re: [PATCH v5 0/9] tracing: Improbe BTF support on probe events

[Steven Rostedt]

On Tue, 22 Aug 2023 00:09:39 +0900
Masami Hiramatsu (Google) <mhiramat@kernel.org> wrote:

> Hi Steve,
> 
> Can you review this series?
> I would like to push this to for-next.
> 

I skimmed the patches and played a little with it, but as I've just
started my vacation I will not be able to do a full review before the
next merge window. But I don't want me to be the cause of you not
getting it in.

Feel free to add:

Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org>

to the entire series. I'm sure I'll be playing with it after it gets
in, and we can fix any bugs that come up then.

-- Steve

Re: [PATCH v5 0/9] tracing: Improbe BTF support on probe events

[Masami Hiramatsu]

On Tue, 22 Aug 2023 10:07:16 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:

> On Tue, 22 Aug 2023 00:09:39 +0900
> Masami Hiramatsu (Google) <mhiramat@kernel.org> wrote:
> 
> > Hi Steve,
> > 
> > Can you review this series?
> > I would like to push this to for-next.
> > 
> 
> I skimmed the patches and played a little with it, but as I've just
> started my vacation I will not be able to do a full review before the
> next merge window. But I don't want me to be the cause of you not
> getting it in.
> 
> Feel free to add:
> 
> Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org>
> 
> to the entire series. I'm sure I'll be playing with it after it gets
> in, and we can fix any bugs that come up then.

Thanks! I'll fix a bit about anon_stack thing and push it to probes/for-next.


> 
> -- Steve


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>

Re: [PATCH v5 0/9] tracing: Improbe BTF support on probe events

[Masami Hiramatsu]

On Tue, 22 Aug 2023 10:07:16 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:

> On Tue, 22 Aug 2023 00:09:39 +0900
> Masami Hiramatsu (Google) <mhiramat@kernel.org> wrote:
> 
> > Hi Steve,
> > 
> > Can you review this series?
> > I would like to push this to for-next.
> > 
> 
> I skimmed the patches and played a little with it, but as I've just
> started my vacation I will not be able to do a full review before the
> next merge window. But I don't want me to be the cause of you not
> getting it in.
> 
> Feel free to add:
> 
> Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org>
> 
> to the entire series. I'm sure I'll be playing with it after it gets
> in, and we can fix any bugs that come up then.

Thank you for your Ack and have a good vacation!
I will continue to test it and fix bugs if there is.

Thanks!

> 
> -- Steve


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>