Skip to contentSkip to conversation

bpftool: Remove BPF_OBJ_NAME_LEN restriction when looking up bpf program by name

PATCH Superseded Manu Bretelle 2 participants 4 messages Jul 31, 2022 – Aug 1, 2022

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 bpftool: Remove BPF_OBJ_NAME_LEN restriction when looking up bpf program by name

bpftool was limiting the length of names to BPF_OBJ_NAME_LEN in prog_parse fds.

Since commit b662000aff84 ("bpftool: Adding support for BTF program names") we can get the full program name from BTF.

This patch removes the restriction of name length when running bpftool prog show name ${name}.

Test:
Tested against some internal program names that were longer than BPF_OBJ_NAME_LEN, here a redacted example of what was ran to test.

    # previous behaviour
    $ sudo bpftool prog show name some_long_program_name
    Error: can't parse name
    # with the patch
    $ sudo ./bpftool prog show name some_long_program_name
    123456789: tracing  name some_long_program_name  tag taghexa  gpl ....
    ...
    ...
    ...
    # too long
    sudo ./bpftool prog show name $(python3 -c 'print("A"*128)')
    Error: can't parse name
    # not too long but no match
    $ sudo ./bpftool prog show name $(python3 -c 'print("A"*127)')

---

v1 -> v2:
* Fix commit message to follow patch submission guidelines
* use strncmp instead of strcmp
* reintroduce arg length check against MAX_PROG_FULL_NAME

 tools/bpf/bpftool/common.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
tools/bpf/bpftool/common.c +12 -3
index 067e9ea59e3b..3ea747b3b194 100644
@@ -722,6 +722,7 @@ print_all_levels(__maybe_unused enum libbpf_print_level level,
 
 static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
 {
+	char prog_name[MAX_PROG_FULL_NAME];
 	unsigned int id = 0;
 	int fd, nb_fds = 0;
 	void *tmp;
@@ -754,12 +755,20 @@ static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
 			goto err_close_fd;
 		}
 
-		if ((tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) ||
-		    (!tag && strncmp(nametag, info.name, BPF_OBJ_NAME_LEN))) {
+		if (tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) {
 			close(fd);
 			continue;
 		}
 
+		if (!tag) {
+			get_prog_full_name(&info, fd, prog_name,
+				sizeof(prog_name));
+			if (strncmp(nametag, prog_name, sizeof(prog_name))) {
+				close(fd);
+				continue;
+			}
+		}
+
 		if (nb_fds > 0) {
 			tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
 			if (!tmp) {
@@ -820,7 +829,7 @@ int prog_parse_fds(int *argc, char ***argv, int **fds)
 		NEXT_ARGP();
 
 		name = **argv;
-		if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
+		if (strlen(name) > MAX_PROG_FULL_NAME - 1) {
 			p_err("can't parse name");
 			return -1;
 		}
Signature · 1 line
2.30.2
Reviewed-by on v2 1/1
#1 lore raw
Quoted 57 lines · Manu Bretelle

bpftool was limiting the length of names to BPF_OBJ_NAME_LEN in prog_parse fds.

Since commit b662000aff84 ("bpftool: Adding support for BTF program names") we can get the full program name from BTF.

This patch removes the restriction of name length when running bpftool prog show name ${name}.

Test:
Tested against some internal program names that were longer than BPF_OBJ_NAME_LEN, here a redacted example of what was ran to test.

    # previous behaviour
    $ sudo bpftool prog show name some_long_program_name
    Error: can't parse name
    # with the patch
    $ sudo ./bpftool prog show name some_long_program_name
    123456789: tracing  name some_long_program_name  tag taghexa  gpl ....
    ...
    ...
    ...
    # too long
    sudo ./bpftool prog show name $(python3 -c 'print("A"*128)')
    Error: can't parse name
    # not too long but no match
    $ sudo ./bpftool prog show name $(python3 -c 'print("A"*127)')

---

v1 -> v2:
* Fix commit message to follow patch submission guidelines
* use strncmp instead of strcmp
* reintroduce arg length check against MAX_PROG_FULL_NAME

 tools/bpf/bpftool/common.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
tools/bpf/bpftool/common.c +1 -0
index 067e9ea59e3b..3ea747b3b194 100644
@@ -722,6 +722,7 @@ print_all_levels(__maybe_unused enum libbpf_print_level level,
 
 static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
 {
+	char prog_name[MAX_PROG_FULL_NAME];
 	unsigned int id = 0;
 	int fd, nb_fds = 0;
 	void *tmp;
@@ -754,12 +755,20 @@ static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
 			goto err_close_fd;
 		}
 

tools/bpf/bpftool/common.c static int prog_fd_by_nametag(void *nametag, int **fds, bool tag) · line 765

Quoted 10 lines
tools/bpf/bpftool/common.c +4 -2
-		if ((tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) ||
-		    (!tag && strncmp(nametag, info.name, BPF_OBJ_NAME_LEN))) {
+		if (tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) {
 			close(fd);
 			continue;
 		}
 
+		if (!tag) {
+			get_prog_full_name(&info, fd, prog_name,
+				sizeof(prog_name));

Nit: This line should be aligned with the opening parenthesis from the line above, checkpatch.pl complains about it. Probably not worth sending a new version just for that, though.

Quoted 10 lines
tools/bpf/bpftool/common.c +6 -0
+			if (strncmp(nametag, prog_name, sizeof(prog_name))) {
+				close(fd);
+				continue;
+			}
+		}
+
 		if (nb_fds > 0) {
 			tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
 			if (!tmp) {

tools/bpf/bpftool/common.c int prog_parse_fds(int *argc, char ***argv, int **fds) · line 835

Quoted 8 lines
tools/bpf/bpftool/common.c +1 -1
@@ -820,7 +829,7 @@ int prog_parse_fds(int *argc, char ***argv, int **fds)
 		NEXT_ARGP();
 
 		name = **argv;
-		if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
+		if (strlen(name) > MAX_PROG_FULL_NAME - 1) {
 			p_err("can't parse name");
 			return -1;
 		}

Looks good, thank you!

#2 lore raw
Quoted 66 lines · Quentin Monnet
Quoted 64 lines · Manu Bretelle

bpftool was limiting the length of names to BPF_OBJ_NAME_LEN in prog_parse fds.

Since commit b662000aff84 ("bpftool: Adding support for BTF program names") we can get the full program name from BTF.

This patch removes the restriction of name length when running bpftool prog show name ${name}.

Test:
Tested against some internal program names that were longer than BPF_OBJ_NAME_LEN, here a redacted example of what was ran to test.

    # previous behaviour
    $ sudo bpftool prog show name some_long_program_name
    Error: can't parse name
    # with the patch
    $ sudo ./bpftool prog show name some_long_program_name
    123456789: tracing  name some_long_program_name  tag taghexa  gpl ....
    ...
    ...
    ...
    # too long
    sudo ./bpftool prog show name $(python3 -c 'print("A"*128)')
    Error: can't parse name
    # not too long but no match
    $ sudo ./bpftool prog show name $(python3 -c 'print("A"*127)')

---

v1 -> v2:
* Fix commit message to follow patch submission guidelines
* use strncmp instead of strcmp
* reintroduce arg length check against MAX_PROG_FULL_NAME

 tools/bpf/bpftool/common.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
tools/bpf/bpftool/common.c +2 -2
index 067e9ea59e3b..3ea747b3b194 100644
@@ -722,6 +722,7 @@ print_all_levels(__maybe_unused enum libbpf_print_level level,
 
 static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
 {
+     char prog_name[MAX_PROG_FULL_NAME];
      unsigned int id = 0;
      int fd, nb_fds = 0;
      void *tmp;
@@ -754,12 +755,20 @@ static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
                      goto err_close_fd;
              }
 
-             if ((tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) ||
-                 (!tag && strncmp(nametag, info.name, BPF_OBJ_NAME_LEN))) {
+             if (tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) {
                      close(fd);
                      continue;
              }
 
Quoted 7 lines

tools/bpf/bpftool/common.c static int prog_fd_by_nametag(void *nametag, int **fds, bool tag) · line 765

Quoted 3 lines
tools/bpf/bpftool/common.c +3 -0
+             if (!tag) {
+                     get_prog_full_name(&info, fd, prog_name,
+                             sizeof(prog_name));

Nit: This line should be aligned with the opening parenthesis from the line above, checkpatch.pl complains about it. Probably not worth sending a new version just for that, though.

Yeah, I saw that on patchwork. For some reason, the checkpatch.pl version I had from bpf-next tree did not catch this. Originally, I was getting an error because it was more than 75 char long. Eventually found out that shiftwidth should have been set to 8 (mine was 4).
I am happy to provide a corrected version if you want, this is really just a matter of a minute now that I have the right vim indentation setting.

Quoted 23 lines
Quoted 11 lines
tools/bpf/bpftool/common.c +6 -0
+                     if (strncmp(nametag, prog_name, sizeof(prog_name))) {
+                             close(fd);
+                             continue;
+                     }
+             }
+
              if (nb_fds > 0) {
                      tmp = realloc(*fds, (nb_fds + 1) * sizeof(int));
                      if (!tmp) {

tools/bpf/bpftool/common.c int prog_parse_fds(int *argc, char ***argv, int **fds) · line 835

Quoted 8 lines
tools/bpf/bpftool/common.c +1 -1
@@ -820,7 +829,7 @@ int prog_parse_fds(int *argc, char ***argv, int **fds)
              NEXT_ARGP();
 
              name = **argv;
-             if (strlen(name) > BPF_OBJ_NAME_LEN - 1) {
+             if (strlen(name) > MAX_PROG_FULL_NAME - 1) {
                      p_err("can't parse name");
                      return -1;
              }

Looks good, thank you!

#3 lore raw
Quoted 75 lines · Manu Bretelle
Quoted 66 lines · Quentin Monnet
Quoted 64 lines · Manu Bretelle

bpftool was limiting the length of names to BPF_OBJ_NAME_LEN in prog_parse fds.

Since commit b662000aff84 ("bpftool: Adding support for BTF program names") we can get the full program name from BTF.

This patch removes the restriction of name length when running bpftool prog show name ${name}.

Test:
Tested against some internal program names that were longer than BPF_OBJ_NAME_LEN, here a redacted example of what was ran to test.

    # previous behaviour
    $ sudo bpftool prog show name some_long_program_name
    Error: can't parse name
    # with the patch
    $ sudo ./bpftool prog show name some_long_program_name
    123456789: tracing  name some_long_program_name  tag taghexa  gpl ....
    ...
    ...
    ...
    # too long
    sudo ./bpftool prog show name $(python3 -c 'print("A"*128)')
    Error: can't parse name
    # not too long but no match
    $ sudo ./bpftool prog show name $(python3 -c 'print("A"*127)')

---

v1 -> v2:
* Fix commit message to follow patch submission guidelines
* use strncmp instead of strcmp
* reintroduce arg length check against MAX_PROG_FULL_NAME

 tools/bpf/bpftool/common.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
tools/bpf/bpftool/common.c +2 -2
index 067e9ea59e3b..3ea747b3b194 100644
@@ -722,6 +722,7 @@ print_all_levels(__maybe_unused enum libbpf_print_level level,
 
 static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
 {
+     char prog_name[MAX_PROG_FULL_NAME];
      unsigned int id = 0;
      int fd, nb_fds = 0;
      void *tmp;
@@ -754,12 +755,20 @@ static int prog_fd_by_nametag(void *nametag, int **fds, bool tag)
                      goto err_close_fd;
              }
 
-             if ((tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) ||
-                 (!tag && strncmp(nametag, info.name, BPF_OBJ_NAME_LEN))) {
+             if (tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) {
                      close(fd);
                      continue;
              }
 
Quoted 7 lines

tools/bpf/bpftool/common.c static int prog_fd_by_nametag(void *nametag, int **fds, bool tag) · line 765

Quoted 3 lines
tools/bpf/bpftool/common.c +3 -0
+             if (!tag) {
+                     get_prog_full_name(&info, fd, prog_name,
+                             sizeof(prog_name));

Nit: This line should be aligned with the opening parenthesis from the line above, checkpatch.pl complains about it. Probably not worth sending a new version just for that, though.

Quoted 2 lines

Yeah, I saw that on patchwork. For some reason, the checkpatch.pl version I had from bpf-next tree did not catch this.

It's because it's a low-level issue: a “check” for checkpatch, not a “warning” or an “error”. Checkpatch will only report this if you run it with "--strict", which the CI does.

Quoted 6 lines

Originally, I was getting an error because it was more than 75 char long. Eventually found out that shiftwidth should have been set to 8 (mine was 4).
I am happy to provide a corrected version if you want, this is really just a matter of a minute now that I have the right vim indentation setting.

OK let's do this. Please keep my Reviewed-by on v3.