[PATCH dwarves v8 0/3] pahole: Inject kfunc decl tags into BTF

[PATCH dwarves v8 0/3] pahole: Inject kfunc decl tags into BTF

[Daniel Xu]

This patchset teaches pahole to parse symbols in .BTF_ids section in
vmlinux and discover exported kfuncs. Pahole then takes the list of
kfuncs and injects a BTF_KIND_DECL_TAG for each kfunc.

Example of encoding:

        $ bpftool btf dump file .tmp_vmlinux.btf | rg "DECL_TAG 'bpf_kfunc'" | wc -l
        121

        $ bpftool btf dump file .tmp_vmlinux.btf | rg 56337
        [56337] FUNC 'bpf_ct_change_timeout' type_id=56336 linkage=static
        [127861] DECL_TAG 'bpf_kfunc' type_id=56337 component_idx=-1

This enables downstream users and tools to dynamically discover which
kfuncs are available on a system by parsing vmlinux or module BTF, both
available in /sys/kernel/btf.

This feature is enabled with --btf_features=decl_tag,decl_tag_kfuncs.

=== Changelog ===

Changes from v7:
* Fix/support detached BTF encoding

Changes from v6:
* Rebase and add decl_tag_kfuncs as default feature

Changes from v5:
* Add gobuffer__sort() helper
* Use strstarts() instead of strncmp()
* Use uint64_t instead of size_t
* Add clarifying comments for get_func_name()

Changes from v4:
* Update man page with decl_tag_kfuncs feature
* Fix release mode build warnings
* Add elf_getshrstrndx() error checking
* Disable tagging if decl_tag feature is off
* Fix malformed func name handling

Changes from v3:
* Guard kfunc tagging behind feature flag
* Use struct btf_id_set8 definition
* Remove unnecessary member from btf_encoder
* Fix code styling

Changes from v2:
* More reliably detect kfunc membership in set8 by tracking set addr ranges
* Rename some variables/functions to be more clear about kfunc vs func

Changes from v1:
* Fix resource leaks
* Fix callee -> caller typo
* Rename btf_decl_tag from kfunc -> bpf_kfunc
* Only grab btf_id_set funcs tagged kfunc
* Presort btf func list

Daniel Xu (3):
  pahole: Save input filename separate from output
  pahole: Add --btf_feature=decl_tag_kfuncs feature
  pahole: Inject kfunc decl tags into BTF

 btf_encoder.c      | 377 +++++++++++++++++++++++++++++++++++++++++++++
 dwarves.h          |   1 +
 man-pages/pahole.1 |   1 +
 pahole.c           |   1 +
 4 files changed, 380 insertions(+)

-- 
2.44.0

[PATCH dwarves v8 1/3] pahole: Save input filename separate from output

[Daniel Xu]

During detached BTF encoding, the input file is not necessarily the same
as the output file. So save them separately. This matters when we need
to look at the input file again, such as for kfunc tagging.

Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
---
 btf_encoder.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/btf_encoder.c b/btf_encoder.c
index 19e9d90..5ffaf5d 100644
--- a/btf_encoder.c
+++ b/btf_encoder.c
@@ -64,6 +64,7 @@ struct btf_encoder {
 	struct btf        *btf;
 	struct cu         *cu;
 	struct gobuffer   percpu_secinfo;
+	const char	  *source_filename;
 	const char	  *filename;
 	struct elf_symtab *symtab;
 	uint32_t	  type_id_off;
@@ -1648,6 +1649,7 @@ struct btf_encoder *btf_encoder__new(struct cu *cu, const char *detached_filenam
 
 	if (encoder) {
 		encoder->raw_output = detached_filename != NULL;
+		encoder->source_filename = strdup(cu->filename);
 		encoder->filename = strdup(encoder->raw_output ? detached_filename : cu->filename);
 		if (encoder->filename == NULL)
 			goto out_delete;
@@ -1730,6 +1732,7 @@ void btf_encoder__delete(struct btf_encoder *encoder)
 	btf_encoders__delete(encoder);
 	__gobuffer__delete(&encoder->percpu_secinfo);
 	zfree(&encoder->filename);
+	zfree(&encoder->source_filename);
 	btf__free(encoder->btf);
 	encoder->btf = NULL;
 	elf_symtab__delete(encoder->symtab);
-- 
2.44.0

[PATCH dwarves v8 2/3] pahole: Add --btf_feature=decl_tag_kfuncs feature

[Daniel Xu]

Add a feature flag to guard tagging of kfuncs. The next commit will
implement the actual tagging.

Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
---
 btf_encoder.c      | 2 ++
 dwarves.h          | 1 +
 man-pages/pahole.1 | 1 +
 pahole.c           | 1 +
 4 files changed, 5 insertions(+)

diff --git a/btf_encoder.c b/btf_encoder.c
index 5ffaf5d..f0ef20a 100644
--- a/btf_encoder.c
+++ b/btf_encoder.c
@@ -76,6 +76,7 @@ struct btf_encoder {
 			  verbose,
 			  force,
 			  gen_floats,
+			  tag_kfuncs,
 			  is_rel;
 	uint32_t	  array_index_id;
 	struct {
@@ -1661,6 +1662,7 @@ struct btf_encoder *btf_encoder__new(struct cu *cu, const char *detached_filenam
 		encoder->force		 = conf_load->btf_encode_force;
 		encoder->gen_floats	 = conf_load->btf_gen_floats;
 		encoder->skip_encoding_vars = conf_load->skip_encoding_btf_vars;
+		encoder->tag_kfuncs	 = conf_load->btf_decl_tag_kfuncs;
 		encoder->verbose	 = verbose;
 		encoder->has_index_type  = false;
 		encoder->need_index_type = false;
diff --git a/dwarves.h b/dwarves.h
index dd35a4e..7d566b6 100644
--- a/dwarves.h
+++ b/dwarves.h
@@ -94,6 +94,7 @@ struct conf_load {
 	bool			btf_gen_floats;
 	bool			btf_encode_force;
 	bool			reproducible_build;
+	bool			btf_decl_tag_kfuncs;
 	uint8_t			hashtable_bits;
 	uint8_t			max_hashtable_bits;
 	uint16_t		kabi_prefix_len;
diff --git a/man-pages/pahole.1 b/man-pages/pahole.1
index e3c58e0..4769b58 100644
--- a/man-pages/pahole.1
+++ b/man-pages/pahole.1
@@ -308,6 +308,7 @@ Encode BTF using the specified feature list, or specify 'default' for all standa
 	                   in some CUs and not others, or when the same
 	                   function name has inconsistent BTF descriptions
 	                   in different CUs.
+	decl_tag_kfuncs    Inject a BTF_KIND_DECL_TAG for each discovered kfunc.
 .fi
 
 Supported non-standard features (not enabled for 'default')
diff --git a/pahole.c b/pahole.c
index 750b847..954498d 100644
--- a/pahole.c
+++ b/pahole.c
@@ -1289,6 +1289,7 @@ struct btf_feature {
 	BTF_DEFAULT_FEATURE(enum64, skip_encoding_btf_enum64, true),
 	BTF_DEFAULT_FEATURE(optimized_func, btf_gen_optimized, false),
 	BTF_DEFAULT_FEATURE(consistent_func, skip_encoding_btf_inconsistent_proto, false),
+	BTF_DEFAULT_FEATURE(decl_tag_kfuncs, btf_decl_tag_kfuncs, false),
 	BTF_NON_DEFAULT_FEATURE(reproducible_build, reproducible_build, false),
 };
 
-- 
2.44.0

[PATCH dwarves v8 3/3] pahole: Inject kfunc decl tags into BTF

[Daniel Xu]

This commit teaches pahole to parse symbols in .BTF_ids section in
vmlinux and discover exported kfuncs. Pahole then takes the list of
kfuncs and injects a BTF_KIND_DECL_TAG for each kfunc.

Example of encoding:

        $ bpftool btf dump file .tmp_vmlinux.btf | rg "DECL_TAG 'bpf_kfunc'" | wc -l
        121

        $ bpftool btf dump file .tmp_vmlinux.btf | rg 56337
        [56337] FUNC 'bpf_ct_change_timeout' type_id=56336 linkage=static
        [127861] DECL_TAG 'bpf_kfunc' type_id=56337 component_idx=-1

This enables downstream users and tools to dynamically discover which
kfuncs are available on a system by parsing vmlinux or module BTF, both
available in /sys/kernel/btf.

This feature is enabled with --btf_features=decl_tag,decl_tag_kfuncs.

Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Tested-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
---
 btf_encoder.c | 372 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 372 insertions(+)

diff --git a/btf_encoder.c b/btf_encoder.c
index f0ef20a..02f0cbb 100644
--- a/btf_encoder.c
+++ b/btf_encoder.c
@@ -34,6 +34,21 @@
 #include <pthread.h>
 
 #define BTF_ENCODER_MAX_PROTO	512
+#define BTF_IDS_SECTION		".BTF_ids"
+#define BTF_ID_FUNC_PFX		"__BTF_ID__func__"
+#define BTF_ID_SET8_PFX		"__BTF_ID__set8__"
+#define BTF_SET8_KFUNCS		(1 << 0)
+#define BTF_KFUNC_TYPE_TAG	"bpf_kfunc"
+
+/* Adapted from include/linux/btf_ids.h */
+struct btf_id_set8 {
+        uint32_t cnt;
+        uint32_t flags;
+        struct {
+                uint32_t id;
+                uint32_t flags;
+        } pairs[];
+};
 
 /* state used to do later encoding of saved functions */
 struct btf_encoder_state {
@@ -76,6 +91,7 @@ struct btf_encoder {
 			  verbose,
 			  force,
 			  gen_floats,
+			  skip_encoding_decl_tag,
 			  tag_kfuncs,
 			  is_rel;
 	uint32_t	  array_index_id;
@@ -95,6 +111,17 @@ struct btf_encoder {
 	} functions;
 };
 
+struct btf_func {
+	const char *name;
+	int	    type_id;
+};
+
+/* Half open interval representing range of addresses containing kfuncs */
+struct btf_kfunc_set_range {
+	uint64_t start;
+	uint64_t end;
+};
+
 static LIST_HEAD(encoders);
 static pthread_mutex_t encoders__lock = PTHREAD_MUTEX_INITIALIZER;
 
@@ -1364,8 +1391,343 @@ out:
 	return err;
 }
 
+/* Returns if `sym` points to a kfunc set */
+static int is_sym_kfunc_set(GElf_Sym *sym, const char *name, Elf_Data *idlist, size_t idlist_addr)
+{
+	void *ptr = idlist->d_buf;
+	struct btf_id_set8 *set;
+	int off;
+
+	/* kfuncs are only found in BTF_SET8's */
+	if (!strstarts(name, BTF_ID_SET8_PFX))
+		return false;
+
+	off = sym->st_value - idlist_addr;
+	if (off >= idlist->d_size) {
+		fprintf(stderr, "%s: symbol '%s' out of bounds\n", __func__, name);
+		return false;
+	}
+
+	/* Check the set8 flags to see if it was marked as kfunc */
+	set = ptr + off;
+	return set->flags & BTF_SET8_KFUNCS;
+}
+
+/*
+ * Parse BTF_ID symbol and return the func name.
+ *
+ * Returns:
+ *	Caller-owned string containing func name if successful.
+ *	NULL if !func or on error.
+ */
+static char *get_func_name(const char *sym)
+{
+	char *func, *end;
+
+	/* Example input: __BTF_ID__func__vfs_close__1
+	 *
+	 * The goal is to strip the prefix and suffix such that we only
+	 * return vfs_close.
+	 */
+
+	if (!strstarts(sym, BTF_ID_FUNC_PFX))
+		return NULL;
+
+	/* Strip prefix and handle malformed input such as  __BTF_ID__func___ */
+	func = strdup(sym + sizeof(BTF_ID_FUNC_PFX) - 1);
+	if (strlen(func) < 2) {
+                free(func);
+                return NULL;
+        }
+
+	/* Strip suffix */
+	end = strrchr(func, '_');
+	if (!end || *(end - 1) != '_') {
+		free(func);
+		return NULL;
+	}
+	*(end - 1) = '\0';
+
+	return func;
+}
+
+static int btf_func_cmp(const void *_a, const void *_b)
+{
+	const struct btf_func *a = _a;
+	const struct btf_func *b = _b;
+
+	return strcmp(a->name, b->name);
+}
+
+/*
+ * Collects all functions described in BTF.
+ * Returns non-zero on error.
+ */
+static int btf_encoder__collect_btf_funcs(struct btf_encoder *encoder, struct gobuffer *funcs)
+{
+	struct btf *btf = encoder->btf;
+	int nr_types, type_id;
+	int err = -1;
+
+	/* First collect all the func entries into an array */
+	nr_types = btf__type_cnt(btf);
+	for (type_id = 1; type_id < nr_types; type_id++) {
+		const struct btf_type *type;
+		struct btf_func func = {};
+		const char *name;
+
+		type = btf__type_by_id(btf, type_id);
+		if (!type) {
+			fprintf(stderr, "%s: malformed BTF, can't resolve type for ID %d\n",
+				__func__, type_id);
+			err = -EINVAL;
+			goto out;
+		}
+
+		if (!btf_is_func(type))
+			continue;
+
+		name = btf__name_by_offset(btf, type->name_off);
+		if (!name) {
+			fprintf(stderr, "%s: malformed BTF, can't resolve name for ID %d\n",
+				__func__, type_id);
+			err = -EINVAL;
+			goto out;
+		}
+
+		func.name = name;
+		func.type_id = type_id;
+		err = gobuffer__add(funcs, &func, sizeof(func));
+		if (err < 0)
+			goto out;
+	}
+
+	/* Now that we've collected funcs, sort them by name */
+	gobuffer__sort(funcs, sizeof(struct btf_func), btf_func_cmp);
+
+	err = 0;
+out:
+	return err;
+}
+
+static int btf_encoder__tag_kfunc(struct btf_encoder *encoder, struct gobuffer *funcs, const char *kfunc)
+{
+	struct btf_func key = { .name = kfunc };
+	struct btf *btf = encoder->btf;
+	struct btf_func *target;
+	const void *base;
+	unsigned int cnt;
+	int err = -1;
+
+	base = gobuffer__entries(funcs);
+	cnt = gobuffer__nr_entries(funcs);
+	target = bsearch(&key, base, cnt, sizeof(key), btf_func_cmp);
+	if (!target) {
+		fprintf(stderr, "%s: failed to find kfunc '%s' in BTF\n", __func__, kfunc);
+		goto out;
+	}
+
+	/* Note we are unconditionally adding the btf_decl_tag even
+	 * though vmlinux may already contain btf_decl_tags for kfuncs.
+	 * We are ok to do this b/c we will later btf__dedup() to remove
+	 * any duplicates.
+	 */
+	err = btf__add_decl_tag(btf, BTF_KFUNC_TYPE_TAG, target->type_id, -1);
+	if (err < 0) {
+		fprintf(stderr, "%s: failed to insert kfunc decl tag for '%s': %d\n",
+			__func__, kfunc, err);
+		goto out;
+	}
+
+	err = 0;
+out:
+	return err;
+}
+
+static int btf_encoder__tag_kfuncs(struct btf_encoder *encoder)
+{
+	const char *filename = encoder->source_filename;
+	struct gobuffer btf_kfunc_ranges = {};
+	struct gobuffer btf_funcs = {};
+	Elf_Data *symbols = NULL;
+	Elf_Data *idlist = NULL;
+	Elf_Scn *symscn = NULL;
+	int symbols_shndx = -1;
+	size_t idlist_addr = 0;
+	int fd = -1, err = -1;
+	int idlist_shndx = -1;
+	size_t strtabidx = 0;
+	Elf_Scn *scn = NULL;
+	Elf *elf = NULL;
+	GElf_Shdr shdr;
+	size_t strndx;
+	char *secname;
+	int nr_syms;
+	int i = 0;
+
+	fd = open(filename, O_RDONLY);
+	if (fd < 0) {
+		fprintf(stderr, "Cannot open %s\n", filename);
+		goto out;
+	}
+
+	if (elf_version(EV_CURRENT) == EV_NONE) {
+		elf_error("Cannot set libelf version");
+		goto out;
+	}
+
+	elf = elf_begin(fd, ELF_C_READ, NULL);
+	if (elf == NULL) {
+		elf_error("Cannot update ELF file");
+		goto out;
+	}
+
+	/* Locate symbol table and .BTF_ids sections */
+	if (elf_getshdrstrndx(elf, &strndx) < 0)
+		goto out;
+
+	while ((scn = elf_nextscn(elf, scn)) != NULL) {
+		Elf_Data *data;
+
+		i++;
+		if (!gelf_getshdr(scn, &shdr)) {
+			elf_error("Failed to get ELF section(%d) hdr", i);
+			goto out;
+		}
+
+		secname = elf_strptr(elf, strndx, shdr.sh_name);
+		if (!secname) {
+			elf_error("Failed to get ELF section(%d) hdr name", i);
+			goto out;
+		}
+
+		data = elf_getdata(scn, 0);
+		if (!data) {
+			elf_error("Failed to get ELF section(%d) data", i);
+			goto out;
+		}
+
+		if (shdr.sh_type == SHT_SYMTAB) {
+			symbols_shndx = i;
+			symscn = scn;
+			symbols = data;
+			strtabidx = shdr.sh_link;
+		} else if (!strcmp(secname, BTF_IDS_SECTION)) {
+			idlist_shndx = i;
+			idlist_addr = shdr.sh_addr;
+			idlist = data;
+		}
+	}
+
+	/* Cannot resolve symbol or .BTF_ids sections. Nothing to do. */
+	if (symbols_shndx == -1 || idlist_shndx == -1) {
+		err = 0;
+		goto out;
+	}
+
+	if (!gelf_getshdr(symscn, &shdr)) {
+		elf_error("Failed to get ELF symbol table header");
+		goto out;
+	}
+	nr_syms = shdr.sh_size / shdr.sh_entsize;
+
+	err = btf_encoder__collect_btf_funcs(encoder, &btf_funcs);
+	if (err) {
+		fprintf(stderr, "%s: failed to collect BTF funcs\n", __func__);
+		goto out;
+	}
+
+	/* First collect all kfunc set ranges.
+	 *
+	 * Note we choose not to sort these ranges and accept a linear
+	 * search when doing lookups. Reasoning is that the number of
+	 * sets is ~O(100) and not worth the additional code to optimize.
+	 */
+	for (i = 0; i < nr_syms; i++) {
+		struct btf_kfunc_set_range range = {};
+		const char *name;
+		GElf_Sym sym;
+
+		if (!gelf_getsym(symbols, i, &sym)) {
+			elf_error("Failed to get ELF symbol(%d)", i);
+			goto out;
+		}
+
+		if (sym.st_shndx != idlist_shndx)
+			continue;
+
+		name = elf_strptr(elf, strtabidx, sym.st_name);
+		if (!is_sym_kfunc_set(&sym, name, idlist, idlist_addr))
+			continue;
+
+		range.start = sym.st_value;
+		range.end = sym.st_value + sym.st_size;
+		gobuffer__add(&btf_kfunc_ranges, &range, sizeof(range));
+	}
+
+	/* Now inject BTF with kfunc decl tag for detected kfuncs */
+	for (i = 0; i < nr_syms; i++) {
+		const struct btf_kfunc_set_range *ranges;
+		unsigned int ranges_cnt;
+		char *func, *name;
+		GElf_Sym sym;
+		bool found;
+		int err;
+		int j;
+
+		if (!gelf_getsym(symbols, i, &sym)) {
+			elf_error("Failed to get ELF symbol(%d)", i);
+			goto out;
+		}
+
+		if (sym.st_shndx != idlist_shndx)
+			continue;
+
+		name = elf_strptr(elf, strtabidx, sym.st_name);
+		func = get_func_name(name);
+		if (!func)
+			continue;
+
+		/* Check if function belongs to a kfunc set */
+		ranges = gobuffer__entries(&btf_kfunc_ranges);
+		ranges_cnt = gobuffer__nr_entries(&btf_kfunc_ranges);
+		found = false;
+		for (j = 0; j < ranges_cnt; j++) {
+			size_t addr = sym.st_value;
+
+			if (ranges[j].start <= addr && addr < ranges[j].end) {
+				found = true;
+				break;
+			}
+		}
+		if (!found) {
+			free(func);
+			continue;
+		}
+
+		err = btf_encoder__tag_kfunc(encoder, &btf_funcs, func);
+		if (err) {
+			fprintf(stderr, "%s: failed to tag kfunc '%s'\n", __func__, func);
+			free(func);
+			goto out;
+		}
+		free(func);
+	}
+
+	err = 0;
+out:
+	__gobuffer__delete(&btf_funcs);
+	__gobuffer__delete(&btf_kfunc_ranges);
+	if (elf)
+		elf_end(elf);
+	if (fd != -1)
+		close(fd);
+	return err;
+}
+
 int btf_encoder__encode(struct btf_encoder *encoder)
 {
+	bool should_tag_kfuncs;
 	int err;
 
 	/* for single-threaded case, saved funcs are added here */
@@ -1378,6 +1740,15 @@ int btf_encoder__encode(struct btf_encoder *encoder)
 	if (btf__type_cnt(encoder->btf) == 1)
 		return 0;
 
+	/* Note vmlinux may already contain btf_decl_tag's for kfuncs. So
+	 * take care to call this before btf_dedup().
+	 */
+	should_tag_kfuncs = encoder->tag_kfuncs && !encoder->skip_encoding_decl_tag;
+	if (should_tag_kfuncs && btf_encoder__tag_kfuncs(encoder)) {
+		fprintf(stderr, "%s: failed to tag kfuncs!\n", __func__);
+		return -1;
+	}
+
 	if (btf__dedup(encoder->btf, NULL)) {
 		fprintf(stderr, "%s: btf__dedup failed!\n", __func__);
 		return -1;
@@ -1662,6 +2033,7 @@ struct btf_encoder *btf_encoder__new(struct cu *cu, const char *detached_filenam
 		encoder->force		 = conf_load->btf_encode_force;
 		encoder->gen_floats	 = conf_load->btf_gen_floats;
 		encoder->skip_encoding_vars = conf_load->skip_encoding_btf_vars;
+		encoder->skip_encoding_decl_tag	 = conf_load->skip_encoding_btf_decl_tag;
 		encoder->tag_kfuncs	 = conf_load->btf_decl_tag_kfuncs;
 		encoder->verbose	 = verbose;
 		encoder->has_index_type  = false;
-- 
2.44.0

Re: [PATCH dwarves v8 3/3] pahole: Inject kfunc decl tags into BTF

[Eduard Zingerman]

On Thu, 2024-04-25 at 18:28 -0600, Daniel Xu wrote:
> This commit teaches pahole to parse symbols in .BTF_ids section in
> vmlinux and discover exported kfuncs. Pahole then takes the list of
> kfuncs and injects a BTF_KIND_DECL_TAG for each kfunc.
> 
> Example of encoding:
> 
>         $ bpftool btf dump file .tmp_vmlinux.btf | rg "DECL_TAG 'bpf_kfunc'" | wc -l
>         121
> 
>         $ bpftool btf dump file .tmp_vmlinux.btf | rg 56337
>         [56337] FUNC 'bpf_ct_change_timeout' type_id=56336 linkage=static
>         [127861] DECL_TAG 'bpf_kfunc' type_id=56337 component_idx=-1
> 
> This enables downstream users and tools to dynamically discover which
> kfuncs are available on a system by parsing vmlinux or module BTF, both
> available in /sys/kernel/btf.
> 
> This feature is enabled with --btf_features=decl_tag,decl_tag_kfuncs.

I tried to double-check results produced by this patch and found that
decl_tag for one kfunc is missing, namely, the following function:

[66020] FUNC 'update_socket_protocol' type_id=66018 linkage=static

And it is present in symbols table (15 is a number of the .BTF_ids section):

60433: ffffffff8293a7fc     4 OBJECT  LOCAL  DEFAULT   15 __BTF_ID__func__update_socket_protocol__78624

Interestingly, this is the last symbol printed for the section.
I'll try to debug this issue.

Re: [PATCH dwarves v8 3/3] pahole: Inject kfunc decl tags into BTF

[Eduard Zingerman]

On Fri, 2024-04-26 at 12:47 -0700, Eduard Zingerman wrote:
> On Thu, 2024-04-25 at 18:28 -0600, Daniel Xu wrote:
> > This commit teaches pahole to parse symbols in .BTF_ids section in
> > vmlinux and discover exported kfuncs. Pahole then takes the list of
> > kfuncs and injects a BTF_KIND_DECL_TAG for each kfunc.
> > 
> > Example of encoding:
> > 
> >         $ bpftool btf dump file .tmp_vmlinux.btf | rg "DECL_TAG 'bpf_kfunc'" | wc -l
> >         121
> > 
> >         $ bpftool btf dump file .tmp_vmlinux.btf | rg 56337
> >         [56337] FUNC 'bpf_ct_change_timeout' type_id=56336 linkage=static
> >         [127861] DECL_TAG 'bpf_kfunc' type_id=56337 component_idx=-1
> > 
> > This enables downstream users and tools to dynamically discover which
> > kfuncs are available on a system by parsing vmlinux or module BTF, both
> > available in /sys/kernel/btf.
> > 
> > This feature is enabled with --btf_features=decl_tag,decl_tag_kfuncs.
> 
> I tried to double-check results produced by this patch and found that
> decl_tag for one kfunc is missing, namely, the following function:
> 
> [66020] FUNC 'update_socket_protocol' type_id=66018 linkage=static
> 
> And it is present in symbols table (15 is a number of the .BTF_ids section):
> 
> 60433: ffffffff8293a7fc     4 OBJECT  LOCAL  DEFAULT   15 __BTF_ID__func__update_socket_protocol__78624
> 
> Interestingly, this is the last symbol printed for the section.
> I'll try to debug this issue.

Nevermind, the 'update_socket_protocol' is not a kfunc, just a member of set8.
Sorry for the noise.

Re: [PATCH dwarves v8 3/3] pahole: Inject kfunc decl tags into BTF

[Eduard Zingerman]

On Thu, 2024-04-25 at 18:28 -0600, Daniel Xu wrote:
> This commit teaches pahole to parse symbols in .BTF_ids section in
> vmlinux and discover exported kfuncs. Pahole then takes the list of
> kfuncs and injects a BTF_KIND_DECL_TAG for each kfunc.
> 
> Example of encoding:
> 
>         $ bpftool btf dump file .tmp_vmlinux.btf | rg "DECL_TAG 'bpf_kfunc'" | wc -l
>         121
> 
>         $ bpftool btf dump file .tmp_vmlinux.btf | rg 56337
>         [56337] FUNC 'bpf_ct_change_timeout' type_id=56336 linkage=static
>         [127861] DECL_TAG 'bpf_kfunc' type_id=56337 component_idx=-1
> 
> This enables downstream users and tools to dynamically discover which
> kfuncs are available on a system by parsing vmlinux or module BTF, both
> available in /sys/kernel/btf.
> 
> This feature is enabled with --btf_features=decl_tag,decl_tag_kfuncs.
> 
> Acked-by: Jiri Olsa <jolsa@kernel.org>
> Tested-by: Jiri Olsa <jolsa@kernel.org>
> Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
> Tested-by: Alan Maguire <alan.maguire@oracle.com>
> Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
> ---

I tested this patch-set on current master with Makefile.btf modified
to have --btf_features=+decl_tag_kfuncs, the tests are passing.

Acked-by: Eduard Zingerman <eddyz87@gmail.com>
(But please fix get_func_name...)

[...]

> +static char *get_func_name(const char *sym)
> +{
> +	char *func, *end;
> +
> +	/* Example input: __BTF_ID__func__vfs_close__1
> +	 *
> +	 * The goal is to strip the prefix and suffix such that we only
> +	 * return vfs_close.
> +	 */
> +
> +	if (!strstarts(sym, BTF_ID_FUNC_PFX))
> +		return NULL;
> +
> +	/* Strip prefix and handle malformed input such as  __BTF_ID__func___ */
> +	func = strdup(sym + sizeof(BTF_ID_FUNC_PFX) - 1);
> +	if (strlen(func) < 2) {
> +                free(func);
> +                return NULL;
> +        }
> +
> +	/* Strip suffix */
> +	end = strrchr(func, '_');
> +	if (!end || *(end - 1) != '_') {

Sorry, I'm complaining about this silly function again...
This will do an invalid read for input like "__BTF_ID__func___a":
- 'func' would be a result of strdup("_a");
- 'end' would point to the first character of 'func';
- 'end - 1' would point outside of 'func'.

Here is a repro with valgrind report:
https://gist.github.com/eddyz87/dfd653ada6584b7b7563fbfe66355eae

> +		free(func);
> +		return NULL;
> +	}
> +	*(end - 1) = '\0';
> +
> +	return func;
> +}

[...]

Re: [PATCH dwarves v8 3/3] pahole: Inject kfunc decl tags into BTF

[Daniel Xu]

Hi Eduard,

On Fri, Apr 26, 2024 at 04:47:53PM GMT, Eduard Zingerman wrote:
> On Thu, 2024-04-25 at 18:28 -0600, Daniel Xu wrote:
> > This commit teaches pahole to parse symbols in .BTF_ids section in
> > vmlinux and discover exported kfuncs. Pahole then takes the list of
> > kfuncs and injects a BTF_KIND_DECL_TAG for each kfunc.
> > 
> > Example of encoding:
> > 
> >         $ bpftool btf dump file .tmp_vmlinux.btf | rg "DECL_TAG 'bpf_kfunc'" | wc -l
> >         121
> > 
> >         $ bpftool btf dump file .tmp_vmlinux.btf | rg 56337
> >         [56337] FUNC 'bpf_ct_change_timeout' type_id=56336 linkage=static
> >         [127861] DECL_TAG 'bpf_kfunc' type_id=56337 component_idx=-1
> > 
> > This enables downstream users and tools to dynamically discover which
> > kfuncs are available on a system by parsing vmlinux or module BTF, both
> > available in /sys/kernel/btf.
> > 
> > This feature is enabled with --btf_features=decl_tag,decl_tag_kfuncs.
> > 
> > Acked-by: Jiri Olsa <jolsa@kernel.org>
> > Tested-by: Jiri Olsa <jolsa@kernel.org>
> > Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
> > Tested-by: Alan Maguire <alan.maguire@oracle.com>
> > Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
> > ---
> 
> I tested this patch-set on current master with Makefile.btf modified
> to have --btf_features=+decl_tag_kfuncs, the tests are passing.
> 
> Acked-by: Eduard Zingerman <eddyz87@gmail.com>
> (But please fix get_func_name...)
> 
> [...]
> 
> > +static char *get_func_name(const char *sym)
> > +{
> > +	char *func, *end;
> > +
> > +	/* Example input: __BTF_ID__func__vfs_close__1
> > +	 *
> > +	 * The goal is to strip the prefix and suffix such that we only
> > +	 * return vfs_close.
> > +	 */
> > +
> > +	if (!strstarts(sym, BTF_ID_FUNC_PFX))
> > +		return NULL;
> > +
> > +	/* Strip prefix and handle malformed input such as  __BTF_ID__func___ */
> > +	func = strdup(sym + sizeof(BTF_ID_FUNC_PFX) - 1);
> > +	if (strlen(func) < 2) {
> > +                free(func);
> > +                return NULL;
> > +        }
> > +
> > +	/* Strip suffix */
> > +	end = strrchr(func, '_');
> > +	if (!end || *(end - 1) != '_') {
> 
> Sorry, I'm complaining about this silly function again...
> This will do an invalid read for input like "__BTF_ID__func___a":
> - 'func' would be a result of strdup("_a");
> - 'end' would point to the first character of 'func';
> - 'end - 1' would point outside of 'func'.
> 
> Here is a repro with valgrind report:
> https://gist.github.com/eddyz87/dfd653ada6584b7b7563fbfe66355eae

Ah, yes. Good catch. I think this diff should do it:

diff --git a/btf_encoder.c b/btf_encoder.c
index 02f0cbb..6cb0c8f 100644
--- a/btf_encoder.c
+++ b/btf_encoder.c
@@ -1435,7 +1435,7 @@ static char *get_func_name(const char *sym)

        /* Strip prefix and handle malformed input such as  __BTF_ID__func___ */
        func = strdup(sym + sizeof(BTF_ID_FUNC_PFX) - 1);
-       if (strlen(func) < 2) {
+       if (!strstr(func, "__")) {
                 free(func);
                 return NULL;
         }

I'll respin.

Thanks,
Daniel

Re: [PATCH dwarves v8 0/3] pahole: Inject kfunc decl tags into BTF

[Jiri Olsa]

On Thu, Apr 25, 2024 at 06:28:38PM -0600, Daniel Xu wrote:
> This patchset teaches pahole to parse symbols in .BTF_ids section in
> vmlinux and discover exported kfuncs. Pahole then takes the list of
> kfuncs and injects a BTF_KIND_DECL_TAG for each kfunc.
> 
> Example of encoding:
> 
>         $ bpftool btf dump file .tmp_vmlinux.btf | rg "DECL_TAG 'bpf_kfunc'" | wc -l
>         121
> 
>         $ bpftool btf dump file .tmp_vmlinux.btf | rg 56337
>         [56337] FUNC 'bpf_ct_change_timeout' type_id=56336 linkage=static
>         [127861] DECL_TAG 'bpf_kfunc' type_id=56337 component_idx=-1
> 
> This enables downstream users and tools to dynamically discover which
> kfuncs are available on a system by parsing vmlinux or module BTF, both
> available in /sys/kernel/btf.
> 
> This feature is enabled with --btf_features=decl_tag,decl_tag_kfuncs.
> 
> === Changelog ===
> 
> Changes from v7:
> * Fix/support detached BTF encoding
> 
> Changes from v6:
> * Rebase and add decl_tag_kfuncs as default feature
> 
> Changes from v5:
> * Add gobuffer__sort() helper
> * Use strstarts() instead of strncmp()
> * Use uint64_t instead of size_t
> * Add clarifying comments for get_func_name()
> 
> Changes from v4:
> * Update man page with decl_tag_kfuncs feature
> * Fix release mode build warnings
> * Add elf_getshrstrndx() error checking
> * Disable tagging if decl_tag feature is off
> * Fix malformed func name handling
> 
> Changes from v3:
> * Guard kfunc tagging behind feature flag
> * Use struct btf_id_set8 definition
> * Remove unnecessary member from btf_encoder
> * Fix code styling
> 
> Changes from v2:
> * More reliably detect kfunc membership in set8 by tracking set addr ranges
> * Rename some variables/functions to be more clear about kfunc vs func
> 
> Changes from v1:
> * Fix resource leaks
> * Fix callee -> caller typo
> * Rename btf_decl_tag from kfunc -> bpf_kfunc
> * Only grab btf_id_set funcs tagged kfunc
> * Presort btf func list
> 
> Daniel Xu (3):
>   pahole: Save input filename separate from output
>   pahole: Add --btf_feature=decl_tag_kfuncs feature
>   pahole: Inject kfunc decl tags into BTF

hi,
I tested the latest version and it still works fine for me, fwiw

Tested-by: Jiri Olsa <jolsa@kernel.org>

jirka

> 
>  btf_encoder.c      | 377 +++++++++++++++++++++++++++++++++++++++++++++
>  dwarves.h          |   1 +
>  man-pages/pahole.1 |   1 +
>  pahole.c           |   1 +
>  4 files changed, 380 insertions(+)
> 
> -- 
> 2.44.0
>