bpftool: Register struct_ops with a link.
PATCH v2 1/2 bpftool: Register struct_ops with a link.
You can include an optional path after specifying the object name for the 'struct_ops register' subcommand.
Since the commit 226bc6ae6405 ("Merge branch 'Transit between BPF TCP congestion controls.'") has been accepted, it is now possible to create a link for a struct_ops. This can be done by defining a struct_ops in SEC(".struct_ops.link") to make libbpf returns a real link. If we don't pin the links before leaving bpftool, they will disappear. To instruct bpftool to pin the links in a directory with the names of the maps, we need to provide the path of that directory.
--- tools/bpf/bpftool/common.c | 14 +++++++ tools/bpf/bpftool/main.h | 3 ++ tools/bpf/bpftool/prog.c | 13 ------ tools/bpf/bpftool/struct_ops.c | 76 ++++++++++++++++++++++++++++------ 4 files changed, 80 insertions(+), 26 deletions(-)
tools/bpf/bpftool/common.c +14 -0
@@ -1091,3 +1091,17 @@ const char *bpf_attach_type_input_str(enum bpf_attach_type t)default :return libbpf_bpf_attach_type_str (t); } } + +int pathname_concat (char *buf,int buf_sz,const char *path, +const char *name) +{ +int len; + + len =snprintf (buf, buf_sz,"%s/%s" , path, name); +if (len <0 ) +return -EINVAL ; +if (len >= buf_sz) +return -ENAMETOOLONG ; + +return 0 ; +}
tools/bpf/bpftool/main.h +3 -0
@@ -262,4 +262,7 @@ static inline bool hashmap__empty(struct hashmap *map) return map ? hashmap__size(map) == 0 : true; } +int pathname_concat (char *buf,int buf_sz,const char *path, +const char *name); +#endif
tools/bpf/bpftool/prog.c +0 -13
@@ -1472,19 +1472,6 @@ auto_attach_program(struct bpf_program *prog, const char *path)return err; } -static int pathname_concat (char *buf,size_t buf_sz,const char *path,const char *name) -{ -int len; - - len =snprintf (buf, buf_sz,"%s/%s" , path, name); -if (len <0 ) -return -EINVAL ; -if ((size_t )len >= buf_sz) -return -ENAMETOOLONG ; - -return 0 ; -} -static int auto_attach_programs (struct bpf_object *obj,const char *path) {
tools/bpf/bpftool/struct_ops.c +63 -13
112 lines not on this page · raw message
Signature · 1 line
2.34.1
PATCH v2 2/2 bpftool: Update doc to explain struct_ops register subcommand.
The "struct_ops register" subcommand now allows for an optional *LINK_DIR* to be included. This specifies the directory path where bpftool will pin struct_ops links with the same name as their corresponding map names.
--- tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst +7 -3
@@ -51,10 +51,14 @@ DESCRIPTION for the given struct_ops. Otherwise, it dumps all struct_ops currently existing in the system. - **bpftool struct_ops register** *OBJ* + **bpftool struct_ops register** *OBJ* [*LINK_DIR*] Register bpf struct_ops from *OBJ*. All struct_ops under - the ELF section ".struct_ops" will be registered to - its kernel subsystem. + the ELF section ".struct_ops" and ".struct_ops.link" will + be registered to its kernel subsystem. For each + struct_ops in the ".struct_ops.link" section, a link + will be created. You can give *LINK_DIR* to provide a + directory path where these links will be pinned with the + same name as their corresponding map name. **bpftool struct_ops unregister** *STRUCT_OPS_MAP* Unregister the *STRUCT_OPS_MAP* from the kernel subsystem.
Signature · 1 line
2.34.1
Re: [PATCH bpf-next v2 2/2] bpftool: Update doc to explain struct_ops register subcommand.
Quoted 22 lines · Kui-Feng Lee
The "
struct_opsregister" subcommand now allows for an optional *LINK_DIR* to be included. This specifies the directory path where bpftool will pinstruct_opslinks with the same name as their corresponding map names.Signed-off-by Kui-Feng Lee <[email protected]>--- tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst +1
-2@@ -51,10 +51,14 @@ DESCRIPTION for the given struct_ops. Otherwise, it dumps all struct_ops currently existing in the system. - **bpftool struct_ops register** *OBJ* + **bpftool struct_ops register** *OBJ* [*LINK_DIR*] Register bpf struct_ops from *OBJ*. All struct_ops under - the ELF section ".struct_ops" will be registered to
tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst DESCRIPTION · line 64
Quoted 10 lines
tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst +6
-1- its kernel subsystem. + the ELF section ".struct_ops" and ".struct_ops.link" will + be registered to its kernel subsystem. For each + struct_ops in the ".struct_ops.link" section, a link + will be created. You can give *LINK_DIR* to provide a + directory path where these links will be pinned with the + same name as their corresponding map name. **bpftool struct_ops unregister** *STRUCT_OPS_MAP* Unregister the *STRUCT_OPS_MAP* from the kernel subsystem.
Thanks! Since there are some nits to address on the first patch anyway, would you mind updating the command summary earlier in this file as well, please? In section "STRUCT_OPS COMMANDS".
Thanks,
Quentin
Re: [PATCH bpf-next v2 2/2] bpftool: Update doc to explain struct_ops register subcommand.
Quoted 31 lines · Quentin Monnet
Quoted 30 lines · Kui-Feng Lee
The "
struct_opsregister" subcommand now allows for an optional *LINK_DIR* to be included. This specifies the directory path where bpftool will pinstruct_opslinks with the same name as their corresponding map names.Signed-off-by Kui-Feng Lee <[email protected]>--- tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst +7
-3@@ -51,10 +51,14 @@ DESCRIPTION for the given struct_ops. Otherwise, it dumps all struct_ops currently existing in the system. - **bpftool struct_ops register** *OBJ* + **bpftool struct_ops register** *OBJ* [*LINK_DIR*] Register bpf struct_ops from *OBJ*. All struct_ops under - the ELF section ".struct_ops" will be registered to - its kernel subsystem. + the ELF section ".struct_ops" and ".struct_ops.link" will + be registered to its kernel subsystem. For each + struct_ops in the ".struct_ops.link" section, a link + will be created. You can give *LINK_DIR* to provide a + directory path where these links will be pinned with the + same name as their corresponding map name.
Quoted 6 lines
tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst DESCRIPTION · line 64
Quoted 2 lines
tools/bpf/bpftool/Documentation/bpftool-struct_ops.rst +0
-0**bpftool struct_ops unregister** *STRUCT_OPS_MAP* Unregister the *STRUCT_OPS_MAP* from the kernel subsystem.Thanks! Since there are some nits to address on the first patch anyway, would you mind updating the command summary earlier in this file as well, please? In section "
STRUCT_OPSCOMMANDS".
Sure!
Quoted 3 lines
Thanks,
Quentin
Quoted 12 lines · Kui-Feng Lee
You can include an optional path after specifying the object name for the '
struct_opsregister' subcommand.Since the commit
226bc6ae6405("Merge branch 'Transit between BPF TCP congestion controls.'") has been accepted, it is now possible to create a link for astruct_ops. This can be done by defining astruct_opsinSEC(".struct_ops.link")to make libbpf returns a real link. If we don't pin the links before leaving bpftool, they will disappear. To instruct bpftool to pin the links in a directory with the names of the maps, we need to provide the path of that directory.
Quoted 1 line
Signed-off-by Kui-Feng Lee <[email protected]>
Right I'd forgotten we could register several struct_ops at once with the command. OK then, makes sense to pass a directory and pin with the existing names in that case.
This patch looks all good, apart from a few indent nitpicks.
Quoted 76 lines
--- tools/bpf/bpftool/common.c | 14 +++++++ tools/bpf/bpftool/main.h | 3 ++ tools/bpf/bpftool/prog.c | 13 ------ tools/bpf/bpftool/struct_ops.c | 76 ++++++++++++++++++++++++++++------ 4 files changed, 80 insertions(+), 26 deletions(-)tools/bpf/bpftool/common.c +14
-0@@ -1091,3 +1091,17 @@ const char *bpf_attach_type_input_str(enum bpf_attach_type t)default :return libbpf_bpf_attach_type_str (t); } } + +int pathname_concat (char *buf,int buf_sz,const char *path, +const char *name) +{ +int len; + + len =snprintf (buf, buf_sz,"%s/%s" , path, name); +if (len <0 ) +return -EINVAL ; +if (len >= buf_sz) +return -ENAMETOOLONG ; + +return 0 ; +}tools/bpf/bpftool/main.h +3
-0@@ -262,4 +262,7 @@ static inline bool hashmap__empty(struct hashmap *map) return map ? hashmap__size(map) == 0 : true; } +int pathname_concat (char *buf,int buf_sz,const char *path, +const char *name); +#endif tools/bpf/bpftool/prog.c +0
-13@@ -1472,19 +1472,6 @@ auto_attach_program(struct bpf_program *prog, const char *path)return err; } -static int pathname_concat (char *buf,size_t buf_sz,const char *path,const char *name) -{ -int len; - - len =snprintf (buf, buf_sz,"%s/%s" , path, name); -if (len <0 ) -return -EINVAL ; -if ((size_t )len >= buf_sz) -return -ENAMETOOLONG ; - -return 0 ; -} -static int auto_attach_programs (struct bpf_object *obj,const char *path) {tools/bpf/bpftool/struct_ops.c +3
-0@@ -475,21 +475,48 @@ static int do_unregister(int argc, char **argv)return cmd_retval (&res, true); } +static int pin_link (struct bpf_link *link,const char *pindir, +const char *name) +{
tools/bpf/bpftool/struct_ops.c static int do_unregister(int argc, char **argv) · line 490
Quoted 10 lines
tools/bpf/bpftool/struct_ops.c +10
-0+char pinfile[PATH_MAX ]; +int err; + + err =pathname_concat (pinfile,sizeof (pinfile), pindir, name); +if (err) +return -1 ; + + err =bpf_link__pin (link, pinfile); +if (err) +return -1 ;
"return bpf_link__pin(link, pinfile);" would work as well. But I don't mind much.
Quoted 45 lines
tools/bpf/bpftool/struct_ops.c +15
-9+ +return 0 ; +} +static int do_register (int argc,char **argv) {LIBBPF_OPTS (bpf_object_open_opts, open_opts); +__u32 link_info_len =sizeof (struct bpf_link_info ); +struct bpf_link_info link_info = {};struct bpf_map_info info = {};__u32 info_len =sizeof (info);int nr_errs =0 , nr_maps =0 ; +const char *linkdir =NULL ;struct bpf_object *obj;struct bpf_link *link;struct bpf_map *map;const char *file; -if (argc !=1 ) +if (argc !=1 && argc !=2 )usage (); file =GET_ARG (); +if (argc ==1 ) + linkdir =GET_ARG (); + +if (linkdir &&mount_bpffs_for_pin (linkdir)) { +p_err ("can't mount bpffs for pinning" ); +return -1 ; + }if (verifier_logs)/* log_level1 + log_level2 + stats, but not stable UAPI */ @@ -519,21 +546,44 @@ static int do_register(int argc, char **argv) } nr_maps++; -bpf_link__disconnect (link); -bpf_link__destroy (link); - -if (!bpf_map_get_info_by_fd (bpf_map__fd (map), &info, - &info_len)) - p_info("Registered %s %s id %u", - get_kern_struct_ops_name(&info), - bpf_map__name(map),
tools/bpf/bpftool/struct_ops.c static int do_register(int argc, char **argv) · line 555
Quoted 10 lines
tools/bpf/bpftool/struct_ops.c +3
-3- info.id); - else +if (bpf_map_get_info_by_fd (bpf_map__fd (map), &info, + &info_len)) {/* Not p_err. The struct_ops was attached * successfully. */ p_info ("Registered %s but can't find id: %s" , - bpf_map__name(map), strerror(errno)); +bpf_map__name (map),strerror (errno));
I think the indent was correct and this change results from your previous version with "p_err()"?
tools/bpf/bpftool/struct_ops.c static int do_register(int argc, char **argv) · line 566
Quoted 11 lines
tools/bpf/bpftool/struct_ops.c +11
-0+ goto clean_link; + } +if (!(bpf_map__map_flags (map) &BPF_F_LINK )) { +p_info ("Registered %s %s id %u" , +get_kern_struct_ops_name (&info), + info.name, + info.id); + goto clean_link; + } + if (bpf_link_get_info_by_fd(bpf_link__fd(link), + &link_info,
Please fix the indent.
Quoted 35 lines
tools/bpf/bpftool/struct_ops.c +21
-2+ &link_info_len)) { +p_err ("Registered %s but can't find link id: %s" , +bpf_map__name (map),strerror (errno)); + nr_errs++; + goto clean_link; + } +if (linkdir &&pin_link (link, linkdir, info.name)) { +p_err ("can't pin link %u for %s: %s" , + link_info.id, info.name, +strerror (errno)); + nr_errs++; + goto clean_link; + } +p_info ("Registered %s %s map id %u link id %u" , +get_kern_struct_ops_name (&info), + info.name, info.id, link_info.id); + +clean_link: +bpf_link__disconnect (link); +bpf_link__destroy (link); }bpf_object__close (obj);@@ -562,7 +612,7 @@ static int do_help(int argc, char **argv)fprintf (stderr,"Usage: %1$s %2$s { show | list } [STRUCT_OPS_MAP]\n" " %1$s %2$s dump [STRUCT_OPS_MAP]\n" - " %1$s %2$s register OBJ\n" +" %1$s %2$s register OBJ [LINK_DIR]\n" " %1$s %2$s unregister STRUCT_OPS_MAP\n" " %1$s %2$s help\n" "\n" --2.34.1
Quoted 94 lines · Quentin Monnet
Quoted 12 lines · Kui-Feng Lee
You can include an optional path after specifying the object name for the '
struct_opsregister' subcommand.Since the commit
226bc6ae6405("Merge branch 'Transit between BPF TCP congestion controls.'") has been accepted, it is now possible to create a link for astruct_ops. This can be done by defining astruct_opsinSEC(".struct_ops.link")to make libbpf returns a real link. If we don't pin the links before leaving bpftool, they will disappear. To instruct bpftool to pin the links in a directory with the names of the maps, we need to provide the path of that directory.Quoted 1 line
Signed-off-by Kui-Feng Lee <[email protected]>Right I'd forgotten we could register several
struct_opsat once with the command. OK then, makes sense to pass a directory and pin with the existing names in that case.This patch looks all good, apart from a few indent nitpicks.
Quoted 73 lines
--- tools/bpf/bpftool/common.c | 14 +++++++ tools/bpf/bpftool/main.h | 3 ++ tools/bpf/bpftool/prog.c | 13 ------ tools/bpf/bpftool/struct_ops.c | 76 ++++++++++++++++++++++++++++------ 4 files changed, 80 insertions(+), 26 deletions(-)tools/bpf/bpftool/common.c +14
-0@@ -1091,3 +1091,17 @@ const char *bpf_attach_type_input_str(enum bpf_attach_type t)default :return libbpf_bpf_attach_type_str (t); } } + +int pathname_concat (char *buf,int buf_sz,const char *path, +const char *name) +{ +int len; + + len =snprintf (buf, buf_sz,"%s/%s" , path, name); +if (len <0 ) +return -EINVAL ; +if (len >= buf_sz) +return -ENAMETOOLONG ; + +return 0 ; +}tools/bpf/bpftool/main.h +3
-0@@ -262,4 +262,7 @@ static inline bool hashmap__empty(struct hashmap *map) return map ? hashmap__size(map) == 0 : true; } +int pathname_concat (char *buf,int buf_sz,const char *path, +const char *name); +#endif tools/bpf/bpftool/prog.c +0
-13@@ -1472,19 +1472,6 @@ auto_attach_program(struct bpf_program *prog, const char *path)return err; } -static int pathname_concat (char *buf,size_t buf_sz,const char *path,const char *name) -{ -int len; - - len =snprintf (buf, buf_sz,"%s/%s" , path, name); -if (len <0 ) -return -EINVAL ; -if ((size_t )len >= buf_sz) -return -ENAMETOOLONG ; - -return 0 ; -} -static int auto_attach_programs (struct bpf_object *obj,const char *path) {tools/bpf/bpftool/struct_ops.c +0
-0@@ -475,21 +475,48 @@ static int do_unregister(int argc, char **argv)return cmd_retval (&res, true); }
Quoted 16 lines
tools/bpf/bpftool/struct_ops.c static int do_unregister(int argc, char **argv) · line 490
Quoted 13 lines
tools/bpf/bpftool/struct_ops.c +13
-0+static int pin_link (struct bpf_link *link,const char *pindir, +const char *name) +{ +char pinfile[PATH_MAX ]; +int err; + + err =pathname_concat (pinfile,sizeof (pinfile), pindir, name); +if (err) +return -1 ; + + err =bpf_link__pin (link, pinfile); +if (err) +return -1 ;"return
bpf_link__pin(link, pinfile);" would work as well. But I don't mind much.
I will change to this.
Quoted 51 lines
Quoted 51 lines
tools/bpf/bpftool/struct_ops.c +17
-11+ +return 0 ; +} +static int do_register (int argc,char **argv) {LIBBPF_OPTS (bpf_object_open_opts, open_opts); +__u32 link_info_len =sizeof (struct bpf_link_info ); +struct bpf_link_info link_info = {};struct bpf_map_info info = {};__u32 info_len =sizeof (info);int nr_errs =0 , nr_maps =0 ; +const char *linkdir =NULL ;struct bpf_object *obj;struct bpf_link *link;struct bpf_map *map;const char *file; -if (argc !=1 ) +if (argc !=1 && argc !=2 )usage (); file =GET_ARG (); +if (argc ==1 ) + linkdir =GET_ARG (); + +if (linkdir &&mount_bpffs_for_pin (linkdir)) { +p_err ("can't mount bpffs for pinning" ); +return -1 ; + }if (verifier_logs)/* log_level1 + log_level2 + stats, but not stable UAPI */ @@ -519,21 +546,44 @@ static int do_register(int argc, char **argv) } nr_maps++; -bpf_link__disconnect (link); -bpf_link__destroy (link); - -if (!bpf_map_get_info_by_fd (bpf_map__fd (map), &info, - &info_len)) -p_info ("Registered %s %s id %u" , -get_kern_struct_ops_name (&info), -bpf_map__name (map), - info.id); -else +if (bpf_map_get_info_by_fd (bpf_map__fd (map), &info, + &info_len)) {/* Not p_err. The struct_ops was attached
Quoted 8 lines
tools/bpf/bpftool/struct_ops.c static int do_register(int argc, char **argv) · line 555
Quoted 5 lines
tools/bpf/bpftool/struct_ops.c +1
-1* successfully. */ p_info ("Registered %s but can't find id: %s" , -bpf_map__name (map),strerror (errno)); +bpf_map__name (map),strerror (errno));I think the indent was correct and this change results from your previous version with "
p_err()"?
Yes, it is my bad!
Quoted 13 lines
tools/bpf/bpftool/struct_ops.c static int do_register(int argc, char **argv) · line 566
Quoted 12 lines
tools/bpf/bpftool/struct_ops.c +11
-0+ goto clean_link; + } +if (!(bpf_map__map_flags (map) &BPF_F_LINK )) { +p_info ("Registered %s %s id %u" , +get_kern_struct_ops_name (&info), + info.name, + info.id); + goto clean_link; + } + if (bpf_link_get_info_by_fd(bpf_link__fd(link), + &link_info,
Quoted 1 line
Please fix the indent.
Sure
Quoted 36 lines
Quoted 36 lines
tools/bpf/bpftool/struct_ops.c +21
-2+ &link_info_len)) { +p_err ("Registered %s but can't find link id: %s" , +bpf_map__name (map),strerror (errno)); + nr_errs++; + goto clean_link; + } +if (linkdir &&pin_link (link, linkdir, info.name)) { +p_err ("can't pin link %u for %s: %s" , + link_info.id, info.name, +strerror (errno)); + nr_errs++; + goto clean_link; + } +p_info ("Registered %s %s map id %u link id %u" , +get_kern_struct_ops_name (&info), + info.name, info.id, link_info.id); + +clean_link: +bpf_link__disconnect (link); +bpf_link__destroy (link); }bpf_object__close (obj);@@ -562,7 +612,7 @@ static int do_help(int argc, char **argv)fprintf (stderr,"Usage: %1$s %2$s { show | list } [STRUCT_OPS_MAP]\n" " %1$s %2$s dump [STRUCT_OPS_MAP]\n" - " %1$s %2$s register OBJ\n" +" %1$s %2$s register OBJ [LINK_DIR]\n" " %1$s %2$s unregister STRUCT_OPS_MAP\n" " %1$s %2$s help\n" "\n" --2.34.1