]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
bpf: Add minimal bpf() command documentation
authorJoe Stringer <joe@cilium.io>
Tue, 2 Mar 2021 17:19:34 +0000 (09:19 -0800)
committerAlexei Starovoitov <ast@kernel.org>
Fri, 5 Mar 2021 02:39:44 +0000 (18:39 -0800)
Introduce high-level descriptions of the intent and return codes of the
bpf() syscall commands. Subsequent patches may further flesh out the
content to provide a more useful programming reference.

Signed-off-by: Joe Stringer <joe@cilium.io>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20210302171947.2268128-3-joe@cilium.io
include/uapi/linux/bpf.h

index fb16c590e6d92c1a34c9d0d9fbedc98d83f2c128..052bbfe65f774ba39a492b1fbe881ffa1594e852 100644 (file)
@@ -204,6 +204,374 @@ union bpf_iter_link_info {
  *             A new file descriptor (a nonnegative integer), or -1 if an
  *             error occurred (in which case, *errno* is set appropriately).
  *
+ * BPF_OBJ_PIN
+ *     Description
+ *             Pin an eBPF program or map referred by the specified *bpf_fd*
+ *             to the provided *pathname* on the filesystem.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_OBJ_GET
+ *     Description
+ *             Open a file descriptor for the eBPF object pinned to the
+ *             specified *pathname*.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_PROG_ATTACH
+ *     Description
+ *             Attach an eBPF program to a *target_fd* at the specified
+ *             *attach_type* hook.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_PROG_DETACH
+ *     Description
+ *             Detach the eBPF program associated with the *target_fd* at the
+ *             hook specified by *attach_type*. The program must have been
+ *             previously attached using **BPF_PROG_ATTACH**.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_PROG_TEST_RUN
+ *     Description
+ *             Run an eBPF program a number of times against a provided
+ *             program context and return the modified program context and
+ *             duration of the test run.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_PROG_GET_NEXT_ID
+ *     Description
+ *             Fetch the next eBPF program currently loaded into the kernel.
+ *
+ *             Looks for the eBPF program with an id greater than *start_id*
+ *             and updates *next_id* on success. If no other eBPF programs
+ *             remain with ids higher than *start_id*, returns -1 and sets
+ *             *errno* to **ENOENT**.
+ *
+ *     Return
+ *             Returns zero on success. On error, or when no id remains, -1
+ *             is returned and *errno* is set appropriately.
+ *
+ * BPF_MAP_GET_NEXT_ID
+ *     Description
+ *             Fetch the next eBPF map currently loaded into the kernel.
+ *
+ *             Looks for the eBPF map with an id greater than *start_id*
+ *             and updates *next_id* on success. If no other eBPF maps
+ *             remain with ids higher than *start_id*, returns -1 and sets
+ *             *errno* to **ENOENT**.
+ *
+ *     Return
+ *             Returns zero on success. On error, or when no id remains, -1
+ *             is returned and *errno* is set appropriately.
+ *
+ * BPF_PROG_GET_FD_BY_ID
+ *     Description
+ *             Open a file descriptor for the eBPF program corresponding to
+ *             *prog_id*.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_MAP_GET_FD_BY_ID
+ *     Description
+ *             Open a file descriptor for the eBPF map corresponding to
+ *             *map_id*.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_OBJ_GET_INFO_BY_FD
+ *     Description
+ *             Obtain information about the eBPF object corresponding to
+ *             *bpf_fd*.
+ *
+ *             Populates up to *info_len* bytes of *info*, which will be in
+ *             one of the following formats depending on the eBPF object type
+ *             of *bpf_fd*:
+ *
+ *             * **struct bpf_prog_info**
+ *             * **struct bpf_map_info**
+ *             * **struct bpf_btf_info**
+ *             * **struct bpf_link_info**
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_PROG_QUERY
+ *     Description
+ *             Obtain information about eBPF programs associated with the
+ *             specified *attach_type* hook.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_RAW_TRACEPOINT_OPEN
+ *     Description
+ *             Attach an eBPF program to a tracepoint *name* to access kernel
+ *             internal arguments of the tracepoint in their raw form.
+ *
+ *             The *prog_fd* must be a valid file descriptor associated with
+ *             a loaded eBPF program of type **BPF_PROG_TYPE_RAW_TRACEPOINT**.
+ *
+ *             No ABI guarantees are made about the content of tracepoint
+ *             arguments exposed to the corresponding eBPF program.
+ *
+ *             Applying **close**\ (2) to the file descriptor returned by
+ *             **BPF_RAW_TRACEPOINT_OPEN** will delete the map (but see NOTES).
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_BTF_LOAD
+ *     Description
+ *             Verify and load BPF Type Format (BTF) metadata into the kernel,
+ *             returning a new file descriptor associated with the metadata.
+ *             BTF is described in more detail at
+ *             https://www.kernel.org/doc/html/latest/bpf/btf.html.
+ *
+ *             The *btf* parameter must point to valid memory providing
+ *             *btf_size* bytes of BTF binary metadata.
+ *
+ *             The returned file descriptor can be passed to other **bpf**\ ()
+ *             subcommands such as **BPF_PROG_LOAD** or **BPF_MAP_CREATE** to
+ *             associate the BTF with those objects.
+ *
+ *             Similar to **BPF_PROG_LOAD**, **BPF_BTF_LOAD** has optional
+ *             parameters to specify a *btf_log_buf*, *btf_log_size* and
+ *             *btf_log_level* which allow the kernel to return freeform log
+ *             output regarding the BTF verification process.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_BTF_GET_FD_BY_ID
+ *     Description
+ *             Open a file descriptor for the BPF Type Format (BTF)
+ *             corresponding to *btf_id*.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_TASK_FD_QUERY
+ *     Description
+ *             Obtain information about eBPF programs associated with the
+ *             target process identified by *pid* and *fd*.
+ *
+ *             If the *pid* and *fd* are associated with a tracepoint, kprobe
+ *             or uprobe perf event, then the *prog_id* and *fd_type* will
+ *             be populated with the eBPF program id and file descriptor type
+ *             of type **bpf_task_fd_type**. If associated with a kprobe or
+ *             uprobe, the  *probe_offset* and *probe_addr* will also be
+ *             populated. Optionally, if *buf* is provided, then up to
+ *             *buf_len* bytes of *buf* will be populated with the name of
+ *             the tracepoint, kprobe or uprobe.
+ *
+ *             The resulting *prog_id* may be introspected in deeper detail
+ *             using **BPF_PROG_GET_FD_BY_ID** and **BPF_OBJ_GET_INFO_BY_FD**.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_MAP_LOOKUP_AND_DELETE_ELEM
+ *     Description
+ *             Look up an element with the given *key* in the map referred to
+ *             by the file descriptor *fd*, and if found, delete the element.
+ *
+ *             The **BPF_MAP_TYPE_QUEUE** and **BPF_MAP_TYPE_STACK** map types
+ *             implement this command as a "pop" operation, deleting the top
+ *             element rather than one corresponding to *key*.
+ *             The *key* and *key_len* parameters should be zeroed when
+ *             issuing this operation for these map types.
+ *
+ *             This command is only valid for the following map types:
+ *             * **BPF_MAP_TYPE_QUEUE**
+ *             * **BPF_MAP_TYPE_STACK**
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_MAP_FREEZE
+ *     Description
+ *             Freeze the permissions of the specified map.
+ *
+ *             Write permissions may be frozen by passing zero *flags*.
+ *             Upon success, no future syscall invocations may alter the
+ *             map state of *map_fd*. Write operations from eBPF programs
+ *             are still possible for a frozen map.
+ *
+ *             Not supported for maps of type **BPF_MAP_TYPE_STRUCT_OPS**.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_BTF_GET_NEXT_ID
+ *     Description
+ *             Fetch the next BPF Type Format (BTF) object currently loaded
+ *             into the kernel.
+ *
+ *             Looks for the BTF object with an id greater than *start_id*
+ *             and updates *next_id* on success. If no other BTF objects
+ *             remain with ids higher than *start_id*, returns -1 and sets
+ *             *errno* to **ENOENT**.
+ *
+ *     Return
+ *             Returns zero on success. On error, or when no id remains, -1
+ *             is returned and *errno* is set appropriately.
+ *
+ * BPF_MAP_LOOKUP_BATCH
+ *     Description
+ *             Iterate and fetch multiple elements in a map.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_MAP_LOOKUP_AND_DELETE_BATCH
+ *     Description
+ *             Iterate and delete multiple elements in a map.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_MAP_UPDATE_BATCH
+ *     Description
+ *             Iterate and update multiple elements in a map.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_MAP_DELETE_BATCH
+ *     Description
+ *             Iterate and delete multiple elements in a map.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_LINK_CREATE
+ *     Description
+ *             Attach an eBPF program to a *target_fd* at the specified
+ *             *attach_type* hook and return a file descriptor handle for
+ *             managing the link.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_LINK_UPDATE
+ *     Description
+ *             Update the eBPF program in the specified *link_fd* to
+ *             *new_prog_fd*.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_LINK_GET_FD_BY_ID
+ *     Description
+ *             Open a file descriptor for the eBPF Link corresponding to
+ *             *link_id*.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_LINK_GET_NEXT_ID
+ *     Description
+ *             Fetch the next eBPF link currently loaded into the kernel.
+ *
+ *             Looks for the eBPF link with an id greater than *start_id*
+ *             and updates *next_id* on success. If no other eBPF links
+ *             remain with ids higher than *start_id*, returns -1 and sets
+ *             *errno* to **ENOENT**.
+ *
+ *     Return
+ *             Returns zero on success. On error, or when no id remains, -1
+ *             is returned and *errno* is set appropriately.
+ *
+ * BPF_ENABLE_STATS
+ *     Description
+ *             Enable eBPF runtime statistics gathering.
+ *
+ *             Runtime statistics gathering for the eBPF runtime is disabled
+ *             by default to minimize the corresponding performance overhead.
+ *             This command enables statistics globally.
+ *
+ *             Multiple programs may independently enable statistics.
+ *             After gathering the desired statistics, eBPF runtime statistics
+ *             may be disabled again by calling **close**\ (2) for the file
+ *             descriptor returned by this function. Statistics will only be
+ *             disabled system-wide when all outstanding file descriptors
+ *             returned by prior calls for this subcommand are closed.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_ITER_CREATE
+ *     Description
+ *             Create an iterator on top of the specified *link_fd* (as
+ *             previously created using **BPF_LINK_CREATE**) and return a
+ *             file descriptor that can be used to trigger the iteration.
+ *
+ *             If the resulting file descriptor is pinned to the filesystem
+ *             using  **BPF_OBJ_PIN**, then subsequent **read**\ (2) syscalls
+ *             for that path will trigger the iterator to read kernel state
+ *             using the eBPF program attached to *link_fd*.
+ *
+ *     Return
+ *             A new file descriptor (a nonnegative integer), or -1 if an
+ *             error occurred (in which case, *errno* is set appropriately).
+ *
+ * BPF_LINK_DETACH
+ *     Description
+ *             Forcefully detach the specified *link_fd* from its
+ *             corresponding attachment point.
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
+ * BPF_PROG_BIND_MAP
+ *     Description
+ *             Bind a map to the lifetime of an eBPF program.
+ *
+ *             The map identified by *map_fd* is bound to the program
+ *             identified by *prog_fd* and only released when *prog_fd* is
+ *             released. This may be used in cases where metadata should be
+ *             associated with a program which otherwise does not contain any
+ *             references to the map (for example, embedded in the eBPF
+ *             program instructions).
+ *
+ *     Return
+ *             Returns zero on success. On error, -1 is returned and *errno*
+ *             is set appropriately.
+ *
  * NOTES
  *     eBPF objects (maps and programs) can be shared between processes.
  *     For example, after **fork**\ (2), the child inherits file descriptors