build-sys: add new kernel bpf.h drop-in
authorDaniel Mack <daniel@zonque.org>
Fri, 28 Oct 2016 15:37:54 +0000 (17:37 +0200)
committerLennart Poettering <lennart@poettering.net>
Fri, 22 Sep 2017 13:24:54 +0000 (15:24 +0200)
The defines we need are pretty comprehensive and new, hence copy in the
full header from the kernel.

src/shared/linux/bpf.h [new file with mode: 0644]
src/shared/linux/bpf_common.h [new file with mode: 0644]
src/shared/linux/libbpf.h [new file with mode: 0644]

diff --git a/src/shared/linux/bpf.h b/src/shared/linux/bpf.h
new file mode 100644 (file)
index 0000000..8477b44
--- /dev/null
@@ -0,0 +1,673 @@
+/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of version 2 of the GNU General Public
+ * License as published by the Free Software Foundation.
+ */
+#ifndef __LINUX_BPF_H__
+#define __LINUX_BPF_H__
+
+#include <linux/types.h>
+#include <linux/bpf_common.h>
+
+/* Extended instruction set based on top of classic BPF */
+
+/* instruction classes */
+#define BPF_ALU64      0x07    /* alu mode in double word width */
+
+/* ld/ldx fields */
+#define BPF_DW         0x18    /* double word */
+#define BPF_XADD       0xc0    /* exclusive add */
+
+/* alu/jmp fields */
+#define BPF_MOV                0xb0    /* mov reg to reg */
+#define BPF_ARSH       0xc0    /* sign extending arithmetic shift right */
+
+/* change endianness of a register */
+#define BPF_END                0xd0    /* flags for endianness conversion: */
+#define BPF_TO_LE      0x00    /* convert to little-endian */
+#define BPF_TO_BE      0x08    /* convert to big-endian */
+#define BPF_FROM_LE    BPF_TO_LE
+#define BPF_FROM_BE    BPF_TO_BE
+
+#define BPF_JNE                0x50    /* jump != */
+#define BPF_JSGT       0x60    /* SGT is signed '>', GT in x86 */
+#define BPF_JSGE       0x70    /* SGE is signed '>=', GE in x86 */
+#define BPF_CALL       0x80    /* function call */
+#define BPF_EXIT       0x90    /* function return */
+
+/* Register numbers */
+enum {
+        BPF_REG_0 = 0,
+        BPF_REG_1,
+        BPF_REG_2,
+        BPF_REG_3,
+        BPF_REG_4,
+        BPF_REG_5,
+        BPF_REG_6,
+        BPF_REG_7,
+        BPF_REG_8,
+        BPF_REG_9,
+        BPF_REG_10,
+        __MAX_BPF_REG,
+};
+
+/* BPF has 10 general purpose 64-bit registers and stack frame. */
+#define MAX_BPF_REG    __MAX_BPF_REG
+
+struct bpf_insn {
+        __u8   code;           /* opcode */
+        __u8   dst_reg:4;      /* dest register */
+        __u8   src_reg:4;      /* source register */
+        __s16  off;            /* signed offset */
+        __s32  imm;            /* signed immediate constant */
+};
+
+/* Key of an a BPF_MAP_TYPE_LPM_TRIE entry */
+struct bpf_lpm_trie_key {
+        __u32  prefixlen;      /* up to 32 for AF_INET, 128 for AF_INET6 */
+        __u8   data[0];        /* Arbitrary size */
+};
+
+/* BPF syscall commands, see bpf(2) man-page for details. */
+enum bpf_cmd {
+        BPF_MAP_CREATE,
+        BPF_MAP_LOOKUP_ELEM,
+        BPF_MAP_UPDATE_ELEM,
+        BPF_MAP_DELETE_ELEM,
+        BPF_MAP_GET_NEXT_KEY,
+        BPF_PROG_LOAD,
+        BPF_OBJ_PIN,
+        BPF_OBJ_GET,
+        BPF_PROG_ATTACH,
+        BPF_PROG_DETACH,
+        BPF_PROG_TEST_RUN,
+};
+
+enum bpf_map_type {
+        BPF_MAP_TYPE_UNSPEC,
+        BPF_MAP_TYPE_HASH,
+        BPF_MAP_TYPE_ARRAY,
+        BPF_MAP_TYPE_PROG_ARRAY,
+        BPF_MAP_TYPE_PERF_EVENT_ARRAY,
+        BPF_MAP_TYPE_PERCPU_HASH,
+        BPF_MAP_TYPE_PERCPU_ARRAY,
+        BPF_MAP_TYPE_STACK_TRACE,
+        BPF_MAP_TYPE_CGROUP_ARRAY,
+        BPF_MAP_TYPE_LRU_HASH,
+        BPF_MAP_TYPE_LRU_PERCPU_HASH,
+        BPF_MAP_TYPE_LPM_TRIE,
+        BPF_MAP_TYPE_ARRAY_OF_MAPS,
+        BPF_MAP_TYPE_HASH_OF_MAPS,
+};
+
+enum bpf_prog_type {
+        BPF_PROG_TYPE_UNSPEC,
+        BPF_PROG_TYPE_SOCKET_FILTER,
+        BPF_PROG_TYPE_KPROBE,
+        BPF_PROG_TYPE_SCHED_CLS,
+        BPF_PROG_TYPE_SCHED_ACT,
+        BPF_PROG_TYPE_TRACEPOINT,
+        BPF_PROG_TYPE_XDP,
+        BPF_PROG_TYPE_PERF_EVENT,
+        BPF_PROG_TYPE_CGROUP_SKB,
+        BPF_PROG_TYPE_CGROUP_SOCK,
+        BPF_PROG_TYPE_LWT_IN,
+        BPF_PROG_TYPE_LWT_OUT,
+        BPF_PROG_TYPE_LWT_XMIT,
+};
+
+enum bpf_attach_type {
+        BPF_CGROUP_INET_INGRESS,
+        BPF_CGROUP_INET_EGRESS,
+        BPF_CGROUP_INET_SOCK_CREATE,
+        __MAX_BPF_ATTACH_TYPE
+};
+
+#define MAX_BPF_ATTACH_TYPE __MAX_BPF_ATTACH_TYPE
+
+/* If BPF_F_ALLOW_OVERRIDE flag is used in BPF_PROG_ATTACH command
+ * to the given target_fd cgroup the descendent cgroup will be able to
+ * override effective bpf program that was inherited from this cgroup
+ */
+#define BPF_F_ALLOW_OVERRIDE   (1U << 0)
+
+/* If BPF_F_STRICT_ALIGNMENT is used in BPF_PROG_LOAD command, the
+ * verifier will perform strict alignment checking as if the kernel
+ * has been built with CONFIG_EFFICIENT_UNALIGNED_ACCESS not set,
+ * and NET_IP_ALIGN defined to 2.
+ */
+#define BPF_F_STRICT_ALIGNMENT (1U << 0)
+
+#define BPF_PSEUDO_MAP_FD      1
+
+/* flags for BPF_MAP_UPDATE_ELEM command */
+#define BPF_ANY                0 /* create new element or update existing */
+#define BPF_NOEXIST    1 /* create new element if it didn't exist */
+#define BPF_EXIST      2 /* update existing element */
+
+#define BPF_F_NO_PREALLOC      (1U << 0)
+/* Instead of having one common LRU list in the
+ * BPF_MAP_TYPE_LRU_[PERCPU_]HASH map, use a percpu LRU list
+ * which can scale and perform better.
+ * Note, the LRU nodes (including free nodes) cannot be moved
+ * across different LRU lists.
+ */
+#define BPF_F_NO_COMMON_LRU    (1U << 1)
+
+union bpf_attr {
+        struct { /* anonymous struct used by BPF_MAP_CREATE command */
+                __u32  map_type;       /* one of enum bpf_map_type */
+                __u32  key_size;       /* size of key in bytes */
+                __u32  value_size;     /* size of value in bytes */
+                __u32  max_entries;    /* max number of entries in a map */
+                __u32  map_flags;      /* prealloc or not */
+                __u32  inner_map_fd;   /* fd pointing to the inner map */
+        };
+
+        struct { /* anonymous struct used by BPF_MAP_*_ELEM commands */
+                __u32          map_fd;
+                __aligned_u64  key;
+                union {
+                        __aligned_u64 value;
+                        __aligned_u64 next_key;
+                };
+                __u64          flags;
+        };
+
+        struct { /* anonymous struct used by BPF_PROG_LOAD command */
+                __u32          prog_type;      /* one of enum bpf_prog_type */
+                __u32          insn_cnt;
+                __aligned_u64  insns;
+                __aligned_u64  license;
+                __u32          log_level;      /* verbosity level of verifier */
+                __u32          log_size;       /* size of user buffer */
+                __aligned_u64  log_buf;        /* user supplied buffer */
+                __u32          kern_version;   /* checked when prog_type=kprobe */
+                __u32          prog_flags;
+        };
+
+        struct { /* anonymous struct used by BPF_OBJ_* commands */
+                __aligned_u64  pathname;
+                __u32          bpf_fd;
+        };
+
+        struct { /* anonymous struct used by BPF_PROG_ATTACH/DETACH commands */
+                __u32          target_fd;      /* container object to attach to */
+                __u32          attach_bpf_fd;  /* eBPF program to attach */
+                __u32          attach_type;
+                __u32          attach_flags;
+        };
+
+        struct { /* anonymous struct used by BPF_PROG_TEST_RUN command */
+                __u32          prog_fd;
+                __u32          retval;
+                __u32          data_size_in;
+                __u32          data_size_out;
+                __aligned_u64  data_in;
+                __aligned_u64  data_out;
+                __u32          repeat;
+                __u32          duration;
+        } test;
+} __attribute__((aligned(8)));
+
+/* BPF helper function descriptions:
+ *
+ * void *bpf_map_lookup_elem(&map, &key)
+ *     Return: Map value or NULL
+ *
+ * int bpf_map_update_elem(&map, &key, &value, flags)
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_map_delete_elem(&map, &key)
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_probe_read(void *dst, int size, void *src)
+ *     Return: 0 on success or negative error
+ *
+ * u64 bpf_ktime_get_ns(void)
+ *     Return: current ktime
+ *
+ * int bpf_trace_printk(const char *fmt, int fmt_size, ...)
+ *     Return: length of buffer written or negative error
+ *
+ * u32 bpf_prandom_u32(void)
+ *     Return: random value
+ *
+ * u32 bpf_raw_smp_processor_id(void)
+ *     Return: SMP processor ID
+ *
+ * int bpf_skb_store_bytes(skb, offset, from, len, flags)
+ *     store bytes into packet
+ *     @skb: pointer to skb
+ *     @offset: offset within packet from skb->mac_header
+ *     @from: pointer where to copy bytes from
+ *     @len: number of bytes to store into packet
+ *     @flags: bit 0 - if true, recompute skb->csum
+ *             other bits - reserved
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_l3_csum_replace(skb, offset, from, to, flags)
+ *     recompute IP checksum
+ *     @skb: pointer to skb
+ *     @offset: offset within packet where IP checksum is located
+ *     @from: old value of header field
+ *     @to: new value of header field
+ *     @flags: bits 0-3 - size of header field
+ *             other bits - reserved
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_l4_csum_replace(skb, offset, from, to, flags)
+ *     recompute TCP/UDP checksum
+ *     @skb: pointer to skb
+ *     @offset: offset within packet where TCP/UDP checksum is located
+ *     @from: old value of header field
+ *     @to: new value of header field
+ *     @flags: bits 0-3 - size of header field
+ *             bit 4 - is pseudo header
+ *             other bits - reserved
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_tail_call(ctx, prog_array_map, index)
+ *     jump into another BPF program
+ *     @ctx: context pointer passed to next program
+ *     @prog_array_map: pointer to map which type is BPF_MAP_TYPE_PROG_ARRAY
+ *     @index: index inside array that selects specific program to run
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_clone_redirect(skb, ifindex, flags)
+ *     redirect to another netdev
+ *     @skb: pointer to skb
+ *     @ifindex: ifindex of the net device
+ *     @flags: bit 0 - if set, redirect to ingress instead of egress
+ *             other bits - reserved
+ *     Return: 0 on success or negative error
+ *
+ * u64 bpf_get_current_pid_tgid(void)
+ *     Return: current->tgid << 32 | current->pid
+ *
+ * u64 bpf_get_current_uid_gid(void)
+ *     Return: current_gid << 32 | current_uid
+ *
+ * int bpf_get_current_comm(char *buf, int size_of_buf)
+ *     stores current->comm into buf
+ *     Return: 0 on success or negative error
+ *
+ * u32 bpf_get_cgroup_classid(skb)
+ *     retrieve a proc's classid
+ *     @skb: pointer to skb
+ *     Return: classid if != 0
+ *
+ * int bpf_skb_vlan_push(skb, vlan_proto, vlan_tci)
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_skb_vlan_pop(skb)
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_skb_get_tunnel_key(skb, key, size, flags)
+ * int bpf_skb_set_tunnel_key(skb, key, size, flags)
+ *     retrieve or populate tunnel metadata
+ *     @skb: pointer to skb
+ *     @key: pointer to 'struct bpf_tunnel_key'
+ *     @size: size of 'struct bpf_tunnel_key'
+ *     @flags: room for future extensions
+ *     Return: 0 on success or negative error
+ *
+ * u64 bpf_perf_event_read(&map, index)
+ *     Return: Number events read or error code
+ *
+ * int bpf_redirect(ifindex, flags)
+ *     redirect to another netdev
+ *     @ifindex: ifindex of the net device
+ *     @flags: bit 0 - if set, redirect to ingress instead of egress
+ *             other bits - reserved
+ *     Return: TC_ACT_REDIRECT
+ *
+ * u32 bpf_get_route_realm(skb)
+ *     retrieve a dst's tclassid
+ *     @skb: pointer to skb
+ *     Return: realm if != 0
+ *
+ * int bpf_perf_event_output(ctx, map, index, data, size)
+ *     output perf raw sample
+ *     @ctx: struct pt_regs*
+ *     @map: pointer to perf_event_array map
+ *     @index: index of event in the map
+ *     @data: data on stack to be output as raw data
+ *     @size: size of data
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_get_stackid(ctx, map, flags)
+ *     walk user or kernel stack and return id
+ *     @ctx: struct pt_regs*
+ *     @map: pointer to stack_trace map
+ *     @flags: bits 0-7 - numer of stack frames to skip
+ *             bit 8 - collect user stack instead of kernel
+ *             bit 9 - compare stacks by hash only
+ *             bit 10 - if two different stacks hash into the same stackid
+ *                      discard old
+ *             other bits - reserved
+ *     Return: >= 0 stackid on success or negative error
+ *
+ * s64 bpf_csum_diff(from, from_size, to, to_size, seed)
+ *     calculate csum diff
+ *     @from: raw from buffer
+ *     @from_size: length of from buffer
+ *     @to: raw to buffer
+ *     @to_size: length of to buffer
+ *     @seed: optional seed
+ *     Return: csum result or negative error code
+ *
+ * int bpf_skb_get_tunnel_opt(skb, opt, size)
+ *     retrieve tunnel options metadata
+ *     @skb: pointer to skb
+ *     @opt: pointer to raw tunnel option data
+ *     @size: size of @opt
+ *     Return: option size
+ *
+ * int bpf_skb_set_tunnel_opt(skb, opt, size)
+ *     populate tunnel options metadata
+ *     @skb: pointer to skb
+ *     @opt: pointer to raw tunnel option data
+ *     @size: size of @opt
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_skb_change_proto(skb, proto, flags)
+ *     Change protocol of the skb. Currently supported is v4 -> v6,
+ *     v6 -> v4 transitions. The helper will also resize the skb. eBPF
+ *     program is expected to fill the new headers via skb_store_bytes
+ *     and lX_csum_replace.
+ *     @skb: pointer to skb
+ *     @proto: new skb->protocol type
+ *     @flags: reserved
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_skb_change_type(skb, type)
+ *     Change packet type of skb.
+ *     @skb: pointer to skb
+ *     @type: new skb->pkt_type type
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_skb_under_cgroup(skb, map, index)
+ *     Check cgroup2 membership of skb
+ *     @skb: pointer to skb
+ *     @map: pointer to bpf_map in BPF_MAP_TYPE_CGROUP_ARRAY type
+ *     @index: index of the cgroup in the bpf_map
+ *     Return:
+ *       == 0 skb failed the cgroup2 descendant test
+ *       == 1 skb succeeded the cgroup2 descendant test
+ *        < 0 error
+ *
+ * u32 bpf_get_hash_recalc(skb)
+ *     Retrieve and possibly recalculate skb->hash.
+ *     @skb: pointer to skb
+ *     Return: hash
+ *
+ * u64 bpf_get_current_task(void)
+ *     Returns current task_struct
+ *     Return: current
+ *
+ * int bpf_probe_write_user(void *dst, void *src, int len)
+ *     safely attempt to write to a location
+ *     @dst: destination address in userspace
+ *     @src: source address on stack
+ *     @len: number of bytes to copy
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_current_task_under_cgroup(map, index)
+ *     Check cgroup2 membership of current task
+ *     @map: pointer to bpf_map in BPF_MAP_TYPE_CGROUP_ARRAY type
+ *     @index: index of the cgroup in the bpf_map
+ *     Return:
+ *       == 0 current failed the cgroup2 descendant test
+ *       == 1 current succeeded the cgroup2 descendant test
+ *        < 0 error
+ *
+ * int bpf_skb_change_tail(skb, len, flags)
+ *     The helper will resize the skb to the given new size, to be used f.e.
+ *     with control messages.
+ *     @skb: pointer to skb
+ *     @len: new skb length
+ *     @flags: reserved
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_skb_pull_data(skb, len)
+ *     The helper will pull in non-linear data in case the skb is non-linear
+ *     and not all of len are part of the linear section. Only needed for
+ *     read/write with direct packet access.
+ *     @skb: pointer to skb
+ *     @len: len to make read/writeable
+ *     Return: 0 on success or negative error
+ *
+ * s64 bpf_csum_update(skb, csum)
+ *     Adds csum into skb->csum in case of CHECKSUM_COMPLETE.
+ *     @skb: pointer to skb
+ *     @csum: csum to add
+ *     Return: csum on success or negative error
+ *
+ * void bpf_set_hash_invalid(skb)
+ *     Invalidate current skb->hash.
+ *     @skb: pointer to skb
+ *
+ * int bpf_get_numa_node_id()
+ *     Return: Id of current NUMA node.
+ *
+ * int bpf_skb_change_head()
+ *     Grows headroom of skb and adjusts MAC header offset accordingly.
+ *     Will extends/reallocae as required automatically.
+ *     May change skb data pointer and will thus invalidate any check
+ *     performed for direct packet access.
+ *     @skb: pointer to skb
+ *     @len: length of header to be pushed in front
+ *     @flags: Flags (unused for now)
+ *     Return: 0 on success or negative error
+ *
+ * int bpf_xdp_adjust_head(xdp_md, delta)
+ *     Adjust the xdp_md.data by delta
+ *     @xdp_md: pointer to xdp_md
+ *     @delta: An positive/negative integer to be added to xdp_md.data
+ *     Return: 0 on success or negative on error
+ *
+ * int bpf_probe_read_str(void *dst, int size, const void *unsafe_ptr)
+ *     Copy a NUL terminated string from unsafe address. In case the string
+ *     length is smaller than size, the target is not padded with further NUL
+ *     bytes. In case the string length is larger than size, just count-1
+ *     bytes are copied and the last byte is set to NUL.
+ *     @dst: destination address
+ *     @size: maximum number of bytes to copy, including the trailing NUL
+ *     @unsafe_ptr: unsafe address
+ *     Return:
+ *       > 0 length of the string including the trailing NUL on success
+ *       < 0 error
+ *
+ * u64 bpf_get_socket_cookie(skb)
+ *     Get the cookie for the socket stored inside sk_buff.
+ *     @skb: pointer to skb
+ *     Return: 8 Bytes non-decreasing number on success or 0 if the socket
+ *     field is missing inside sk_buff
+ *
+ * u32 bpf_get_socket_uid(skb)
+ *     Get the owner uid of the socket stored inside sk_buff.
+ *     @skb: pointer to skb
+ *     Return: uid of the socket owner on success or overflowuid if failed.
+ */
+#define __BPF_FUNC_MAPPER(FN)          \
+        FN(unspec),                    \
+        FN(map_lookup_elem),           \
+        FN(map_update_elem),           \
+        FN(map_delete_elem),           \
+        FN(probe_read),                        \
+        FN(ktime_get_ns),              \
+        FN(trace_printk),              \
+        FN(get_prandom_u32),           \
+        FN(get_smp_processor_id),      \
+        FN(skb_store_bytes),           \
+        FN(l3_csum_replace),           \
+        FN(l4_csum_replace),           \
+        FN(tail_call),                 \
+        FN(clone_redirect),            \
+        FN(get_current_pid_tgid),      \
+        FN(get_current_uid_gid),       \
+        FN(get_current_comm),          \
+        FN(get_cgroup_classid),                \
+        FN(skb_vlan_push),             \
+        FN(skb_vlan_pop),              \
+        FN(skb_get_tunnel_key),                \
+        FN(skb_set_tunnel_key),                \
+        FN(perf_event_read),           \
+        FN(redirect),                  \
+        FN(get_route_realm),           \
+        FN(perf_event_output),         \
+        FN(skb_load_bytes),            \
+        FN(get_stackid),               \
+        FN(csum_diff),                 \
+        FN(skb_get_tunnel_opt),                \
+        FN(skb_set_tunnel_opt),                \
+        FN(skb_change_proto),          \
+        FN(skb_change_type),           \
+        FN(skb_under_cgroup),          \
+        FN(get_hash_recalc),           \
+        FN(get_current_task),          \
+        FN(probe_write_user),          \
+        FN(current_task_under_cgroup), \
+        FN(skb_change_tail),           \
+        FN(skb_pull_data),             \
+        FN(csum_update),               \
+        FN(set_hash_invalid),          \
+        FN(get_numa_node_id),          \
+        FN(skb_change_head),           \
+        FN(xdp_adjust_head),           \
+        FN(probe_read_str),            \
+        FN(get_socket_cookie),         \
+        FN(get_socket_uid),
+
+/* integer value in 'imm' field of BPF_CALL instruction selects which helper
+ * function eBPF program intends to call
+ */
+#define __BPF_ENUM_FN(x) BPF_FUNC_ ## x
+enum bpf_func_id {
+        __BPF_FUNC_MAPPER(__BPF_ENUM_FN)
+        __BPF_FUNC_MAX_ID,
+};
+#undef __BPF_ENUM_FN
+
+/* All flags used by eBPF helper functions, placed here. */
+
+/* BPF_FUNC_skb_store_bytes flags. */
+#define BPF_F_RECOMPUTE_CSUM           (1ULL << 0)
+#define BPF_F_INVALIDATE_HASH          (1ULL << 1)
+
+/* BPF_FUNC_l3_csum_replace and BPF_FUNC_l4_csum_replace flags.
+ * First 4 bits are for passing the header field size.
+ */
+#define BPF_F_HDR_FIELD_MASK           0xfULL
+
+/* BPF_FUNC_l4_csum_replace flags. */
+#define BPF_F_PSEUDO_HDR               (1ULL << 4)
+#define BPF_F_MARK_MANGLED_0           (1ULL << 5)
+#define BPF_F_MARK_ENFORCE             (1ULL << 6)
+
+/* BPF_FUNC_clone_redirect and BPF_FUNC_redirect flags. */
+#define BPF_F_INGRESS                  (1ULL << 0)
+
+/* BPF_FUNC_skb_set_tunnel_key and BPF_FUNC_skb_get_tunnel_key flags. */
+#define BPF_F_TUNINFO_IPV6             (1ULL << 0)
+
+/* BPF_FUNC_get_stackid flags. */
+#define BPF_F_SKIP_FIELD_MASK          0xffULL
+#define BPF_F_USER_STACK               (1ULL << 8)
+#define BPF_F_FAST_STACK_CMP           (1ULL << 9)
+#define BPF_F_REUSE_STACKID            (1ULL << 10)
+
+/* BPF_FUNC_skb_set_tunnel_key flags. */
+#define BPF_F_ZERO_CSUM_TX             (1ULL << 1)
+#define BPF_F_DONT_FRAGMENT            (1ULL << 2)
+
+/* BPF_FUNC_perf_event_output and BPF_FUNC_perf_event_read flags. */
+#define BPF_F_INDEX_MASK               0xffffffffULL
+#define BPF_F_CURRENT_CPU              BPF_F_INDEX_MASK
+/* BPF_FUNC_perf_event_output for sk_buff input context. */
+#define BPF_F_CTXLEN_MASK              (0xfffffULL << 32)
+
+/* user accessible mirror of in-kernel sk_buff.
+ * new fields can only be added to the end of this structure
+ */
+struct __sk_buff {
+        __u32 len;
+        __u32 pkt_type;
+        __u32 mark;
+        __u32 queue_mapping;
+        __u32 protocol;
+        __u32 vlan_present;
+        __u32 vlan_tci;
+        __u32 vlan_proto;
+        __u32 priority;
+        __u32 ingress_ifindex;
+        __u32 ifindex;
+        __u32 tc_index;
+        __u32 cb[5];
+        __u32 hash;
+        __u32 tc_classid;
+        __u32 data;
+        __u32 data_end;
+        __u32 napi_id;
+};
+
+struct bpf_tunnel_key {
+        __u32 tunnel_id;
+        union {
+                __u32 remote_ipv4;
+                __u32 remote_ipv6[4];
+        };
+        __u8 tunnel_tos;
+        __u8 tunnel_ttl;
+        __u16 tunnel_ext;
+        __u32 tunnel_label;
+};
+
+/* Generic BPF return codes which all BPF program types may support.
+ * The values are binary compatible with their TC_ACT_* counter-part to
+ * provide backwards compatibility with existing SCHED_CLS and SCHED_ACT
+ * programs.
+ *
+ * XDP is handled seprately, see XDP_*.
+ */
+enum bpf_ret_code {
+        BPF_OK = 0,
+        /* 1 reserved */
+        BPF_DROP = 2,
+        /* 3-6 reserved */
+        BPF_REDIRECT = 7,
+        /* >127 are reserved for prog type specific return codes */
+};
+
+struct bpf_sock {
+        __u32 bound_dev_if;
+        __u32 family;
+        __u32 type;
+        __u32 protocol;
+};
+
+#define XDP_PACKET_HEADROOM 256
+
+/* User return codes for XDP prog type.
+ * A valid XDP program must return one of these defined values. All other
+ * return codes are reserved for future use. Unknown return codes will result
+ * in packet drop.
+ */
+enum xdp_action {
+        XDP_ABORTED = 0,
+        XDP_DROP,
+        XDP_PASS,
+        XDP_TX,
+};
+
+/* user accessible metadata for XDP packet hook
+ * new fields must be added to the end of this structure
+ */
+struct xdp_md {
+        __u32 data;
+        __u32 data_end;
+};
+
+#endif /* __LINUX_BPF_H__ */
diff --git a/src/shared/linux/bpf_common.h b/src/shared/linux/bpf_common.h
new file mode 100644 (file)
index 0000000..afe7433
--- /dev/null
@@ -0,0 +1,55 @@
+#ifndef __LINUX_BPF_COMMON_H__
+#define __LINUX_BPF_COMMON_H__
+
+/* Instruction classes */
+#define BPF_CLASS(code) ((code) & 0x07)
+#define                BPF_LD          0x00
+#define                BPF_LDX         0x01
+#define                BPF_ST          0x02
+#define                BPF_STX         0x03
+#define                BPF_ALU         0x04
+#define                BPF_JMP         0x05
+#define                BPF_RET         0x06
+#define                BPF_MISC        0x07
+
+/* ld/ldx fields */
+#define BPF_SIZE(code)  ((code) & 0x18)
+#define                BPF_W           0x00
+#define                BPF_H           0x08
+#define                BPF_B           0x10
+#define BPF_MODE(code)  ((code) & 0xe0)
+#define                BPF_IMM         0x00
+#define                BPF_ABS         0x20
+#define                BPF_IND         0x40
+#define                BPF_MEM         0x60
+#define                BPF_LEN         0x80
+#define                BPF_MSH         0xa0
+
+/* alu/jmp fields */
+#define BPF_OP(code)    ((code) & 0xf0)
+#define                BPF_ADD         0x00
+#define                BPF_SUB         0x10
+#define                BPF_MUL         0x20
+#define                BPF_DIV         0x30
+#define                BPF_OR          0x40
+#define                BPF_AND         0x50
+#define                BPF_LSH         0x60
+#define                BPF_RSH         0x70
+#define                BPF_NEG         0x80
+#define                BPF_MOD         0x90
+#define                BPF_XOR         0xa0
+
+#define                BPF_JA          0x00
+#define                BPF_JEQ         0x10
+#define                BPF_JGT         0x20
+#define                BPF_JGE         0x30
+#define                BPF_JSET        0x40
+#define BPF_SRC(code)   ((code) & 0x08)
+#define                BPF_K           0x00
+#define                BPF_X           0x08
+
+#ifndef BPF_MAXINSNS
+#define BPF_MAXINSNS 4096
+#endif
+
+#endif /* __LINUX_BPF_COMMON_H__ */
diff --git a/src/shared/linux/libbpf.h b/src/shared/linux/libbpf.h
new file mode 100644 (file)
index 0000000..1989e3a
--- /dev/null
@@ -0,0 +1,198 @@
+/* eBPF mini library */
+#ifndef __LIBBPF_H
+#define __LIBBPF_H
+
+#include <linux/bpf.h>
+
+struct bpf_insn;
+
+/* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
+
+#define BPF_ALU64_REG(OP, DST, SRC)                            \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ALU64 | BPF_OP(OP) | BPF_X,       \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = 0,                                    \
+                .imm   = 0 })
+
+#define BPF_ALU32_REG(OP, DST, SRC)                            \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ALU | BPF_OP(OP) | BPF_X,         \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = 0,                                    \
+                .imm   = 0 })
+
+/* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
+
+#define BPF_ALU64_IMM(OP, DST, IMM)                            \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ALU64 | BPF_OP(OP) | BPF_K,       \
+                .dst_reg = DST,                                        \
+                .src_reg = 0,                                  \
+                .off   = 0,                                    \
+                .imm   = IMM })
+
+#define BPF_ALU32_IMM(OP, DST, IMM)                            \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ALU | BPF_OP(OP) | BPF_K,         \
+                .dst_reg = DST,                                        \
+                .src_reg = 0,                                  \
+                .off   = 0,                                    \
+                .imm   = IMM })
+
+/* Short form of mov, dst_reg = src_reg */
+
+#define BPF_MOV64_REG(DST, SRC)                                        \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ALU64 | BPF_MOV | BPF_X,          \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = 0,                                    \
+                .imm   = 0 })
+
+#define BPF_MOV32_REG(DST, SRC)                                        \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ALU | BPF_MOV | BPF_X,            \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = 0,                                    \
+                .imm   = 0 })
+
+/* Short form of mov, dst_reg = imm32 */
+
+#define BPF_MOV64_IMM(DST, IMM)                                        \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ALU64 | BPF_MOV | BPF_K,          \
+                .dst_reg = DST,                                        \
+                .src_reg = 0,                                  \
+                .off   = 0,                                    \
+                .imm   = IMM })
+
+#define BPF_MOV32_IMM(DST, IMM)                                        \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ALU | BPF_MOV | BPF_K,            \
+                .dst_reg = DST,                                        \
+                .src_reg = 0,                                  \
+                .off   = 0,                                    \
+                .imm   = IMM })
+
+/* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
+#define BPF_LD_IMM64(DST, IMM)                                 \
+        BPF_LD_IMM64_RAW(DST, 0, IMM)
+
+#define BPF_LD_IMM64_RAW(DST, SRC, IMM)                                \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_LD | BPF_DW | BPF_IMM,            \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = 0,                                    \
+                .imm   = (__u32) (IMM) }),                     \
+        ((struct bpf_insn) {                                   \
+                .code  = 0, /* zero is reserved opcode */      \
+                .dst_reg = 0,                                  \
+                .src_reg = 0,                                  \
+                .off   = 0,                                    \
+                .imm   = ((__u64) (IMM)) >> 32 })
+
+#ifndef BPF_PSEUDO_MAP_FD
+# define BPF_PSEUDO_MAP_FD     1
+#endif
+
+/* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
+#define BPF_LD_MAP_FD(DST, MAP_FD)                             \
+        BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
+
+
+/* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
+
+#define BPF_LD_ABS(SIZE, IMM)                                  \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS,    \
+                .dst_reg = 0,                                  \
+                .src_reg = 0,                                  \
+                .off   = 0,                                    \
+                .imm   = IMM })
+
+/* Memory load, dst_reg = *(uint *) (src_reg + off16) */
+
+#define BPF_LDX_MEM(SIZE, DST, SRC, OFF)                       \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM,   \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = OFF,                                  \
+                .imm   = 0 })
+
+/* Memory store, *(uint *) (dst_reg + off16) = src_reg */
+
+#define BPF_STX_MEM(SIZE, DST, SRC, OFF)                       \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM,   \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = OFF,                                  \
+                .imm   = 0 })
+
+/* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
+
+#define BPF_STX_XADD(SIZE, DST, SRC, OFF)                      \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD,  \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = OFF,                                  \
+                .imm   = 0 })
+
+/* Memory store, *(uint *) (dst_reg + off16) = imm32 */
+
+#define BPF_ST_MEM(SIZE, DST, OFF, IMM)                                \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM,    \
+                .dst_reg = DST,                                        \
+                .src_reg = 0,                                  \
+                .off   = OFF,                                  \
+                .imm   = IMM })
+
+/* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
+
+#define BPF_JMP_REG(OP, DST, SRC, OFF)                         \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_JMP | BPF_OP(OP) | BPF_X,         \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = OFF,                                  \
+                .imm   = 0 })
+
+/* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
+
+#define BPF_JMP_IMM(OP, DST, IMM, OFF)                         \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_JMP | BPF_OP(OP) | BPF_K,         \
+                .dst_reg = DST,                                        \
+                .src_reg = 0,                                  \
+                .off   = OFF,                                  \
+                .imm   = IMM })
+
+/* Raw code statement block */
+
+#define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM)                 \
+        ((struct bpf_insn) {                                   \
+                .code  = CODE,                                 \
+                .dst_reg = DST,                                        \
+                .src_reg = SRC,                                        \
+                .off   = OFF,                                  \
+                .imm   = IMM })
+
+/* Program exit */
+
+#define BPF_EXIT_INSN()                                                \
+        ((struct bpf_insn) {                                   \
+                .code  = BPF_JMP | BPF_EXIT,                   \
+                .dst_reg = 0,                                  \
+                .src_reg = 0,                                  \
+                .off   = 0,                                    \
+                .imm   = 0 })
+
+#endif