!C99Shell v. 2.0 [PHP 7 Update] [25.02.2019]!

Software: nginx/1.23.4. PHP/5.6.40-65+ubuntu20.04.1+deb.sury.org+1 

uname -a: Linux foro-restaurado-2 5.15.0-1040-oracle #46-Ubuntu SMP Fri Jul 14 21:47:21 UTC 2023
aarch64
 

uid=33(www-data) gid=33(www-data) groups=33(www-data) 

Safe-mode: OFF (not secure)

/usr/src/linux-oracle-6.8-headers-6.8.0-1027/include/linux/   drwxr-xr-x
Free 83.29 GB of 96.73 GB (86.11%)
Home    Back    Forward    UPDIR    Refresh    Search    Buffer    Encoder    Tools    Proc.    FTP brute    Sec.    SQL    PHP-code    Update    Feedback    Self remove    Logout    


Viewing file:     bpf-cgroup.h (18.36 KB)      -rw-r--r--
Select action/file-type:
(+) | (+) | (+) | Code (+) | Session (+) | (+) | SDB (+) | (+) | (+) | (+) | (+) | (+) |
/* SPDX-License-Identifier: GPL-2.0 */
#ifndef _BPF_CGROUP_H
#define _BPF_CGROUP_H

#include <linux/bpf.h>
#include <linux/bpf-cgroup-defs.h>
#include <linux/errno.h>
#include <linux/jump_label.h>
#include <linux/percpu.h>
#include <linux/rbtree.h>
#include <net/sock.h>
#include <uapi/linux/bpf.h>

struct sock;
struct sockaddr;
struct cgroup;
struct sk_buff;
struct bpf_map;
struct bpf_prog;
struct bpf_sock_ops_kern;
struct bpf_cgroup_storage;
struct ctl_table;
struct ctl_table_header;
struct task_struct;

unsigned int __cgroup_bpf_run_lsm_sock(const void *ctx,
                       const struct bpf_insn *insn);
unsigned int __cgroup_bpf_run_lsm_socket(const void *ctx,
                     const struct bpf_insn *insn);
unsigned int __cgroup_bpf_run_lsm_current(const void *ctx,
                      const struct bpf_insn *insn);

#ifdef CONFIG_CGROUP_BPF

#define CGROUP_ATYPE(type) \
    case BPF_##type: return type

static inline enum cgroup_bpf_attach_type
to_cgroup_bpf_attach_type(enum bpf_attach_type attach_type)
{
    switch (attach_type) {
    CGROUP_ATYPE(CGROUP_INET_INGRESS);
    CGROUP_ATYPE(CGROUP_INET_EGRESS);
    CGROUP_ATYPE(CGROUP_INET_SOCK_CREATE);
    CGROUP_ATYPE(CGROUP_SOCK_OPS);
    CGROUP_ATYPE(CGROUP_DEVICE);
    CGROUP_ATYPE(CGROUP_INET4_BIND);
    CGROUP_ATYPE(CGROUP_INET6_BIND);
    CGROUP_ATYPE(CGROUP_INET4_CONNECT);
    CGROUP_ATYPE(CGROUP_INET6_CONNECT);
    CGROUP_ATYPE(CGROUP_UNIX_CONNECT);
    CGROUP_ATYPE(CGROUP_INET4_POST_BIND);
    CGROUP_ATYPE(CGROUP_INET6_POST_BIND);
    CGROUP_ATYPE(CGROUP_UDP4_SENDMSG);
    CGROUP_ATYPE(CGROUP_UDP6_SENDMSG);
    CGROUP_ATYPE(CGROUP_UNIX_SENDMSG);
    CGROUP_ATYPE(CGROUP_SYSCTL);
    CGROUP_ATYPE(CGROUP_UDP4_RECVMSG);
    CGROUP_ATYPE(CGROUP_UDP6_RECVMSG);
    CGROUP_ATYPE(CGROUP_UNIX_RECVMSG);
    CGROUP_ATYPE(CGROUP_GETSOCKOPT);
    CGROUP_ATYPE(CGROUP_SETSOCKOPT);
    CGROUP_ATYPE(CGROUP_INET4_GETPEERNAME);
    CGROUP_ATYPE(CGROUP_INET6_GETPEERNAME);
    CGROUP_ATYPE(CGROUP_UNIX_GETPEERNAME);
    CGROUP_ATYPE(CGROUP_INET4_GETSOCKNAME);
    CGROUP_ATYPE(CGROUP_INET6_GETSOCKNAME);
    CGROUP_ATYPE(CGROUP_UNIX_GETSOCKNAME);
    CGROUP_ATYPE(CGROUP_INET_SOCK_RELEASE);
    default:
        return CGROUP_BPF_ATTACH_TYPE_INVALID;
    }
}

#undef CGROUP_ATYPE

extern struct static_key_false cgroup_bpf_enabled_key[MAX_CGROUP_BPF_ATTACH_TYPE];
#define cgroup_bpf_enabled(atype) static_branch_unlikely(&cgroup_bpf_enabled_key[atype])

#define for_each_cgroup_storage_type(stype) \
    for (stype = 0; stype < MAX_BPF_CGROUP_STORAGE_TYPE; stype++)

struct bpf_cgroup_storage_map;

struct bpf_storage_buffer {
    struct rcu_head rcu;
    char data[];
};

struct bpf_cgroup_storage {
    union {
        struct bpf_storage_buffer *buf;
        void __percpu *percpu_buf;
    };
    struct bpf_cgroup_storage_map *map;
    struct bpf_cgroup_storage_key key;
    struct list_head list_map;
    struct list_head list_cg;
    struct rb_node node;
    struct rcu_head rcu;
};

struct bpf_cgroup_link {
    struct bpf_link link;
    struct cgroup *cgroup;
    enum bpf_attach_type type;
};

struct bpf_prog_list {
    struct hlist_node node;
    struct bpf_prog *prog;
    struct bpf_cgroup_link *link;
    struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE];
};

int cgroup_bpf_inherit(struct cgroup *cgrp);
void cgroup_bpf_offline(struct cgroup *cgrp);

int __cgroup_bpf_run_filter_skb(struct sock *sk,
                struct sk_buff *skb,
                enum cgroup_bpf_attach_type atype);

int __cgroup_bpf_run_filter_sk(struct sock *sk,
                   enum cgroup_bpf_attach_type atype);

int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
                      struct sockaddr *uaddr,
                      int *uaddrlen,
                      enum cgroup_bpf_attach_type atype,
                      void *t_ctx,
                      u32 *flags);

int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
                     struct bpf_sock_ops_kern *sock_ops,
                     enum cgroup_bpf_attach_type atype);

int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
                      short access, enum cgroup_bpf_attach_type atype);

int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
                   struct ctl_table *table, int write,
                   char **buf, size_t *pcount, loff_t *ppos,
                   enum cgroup_bpf_attach_type atype);

int __cgroup_bpf_run_filter_setsockopt(struct sock *sock, int *level,
                       int *optname, sockptr_t optval,
                       int *optlen, char **kernel_optval);

int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
                       int optname, sockptr_t optval,
                       sockptr_t optlen, int max_optlen,
                       int retval);

int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level,
                        int optname, void *optval,
                        int *optlen, int retval);

static inline enum bpf_cgroup_storage_type cgroup_storage_type(
    struct bpf_map *map)
{
    if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
        return BPF_CGROUP_STORAGE_PERCPU;

    return BPF_CGROUP_STORAGE_SHARED;
}

struct bpf_cgroup_storage *
cgroup_storage_lookup(struct bpf_cgroup_storage_map *map,
              void *key, bool locked);
struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog,
                    enum bpf_cgroup_storage_type stype);
void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage);
void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
                 struct cgroup *cgroup,
                 enum bpf_attach_type type);
void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage);
int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux, struct bpf_map *map);

int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key, void *value);
int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key,
                     void *value, u64 flags);

/* Opportunistic check to see whether we have any BPF program attached*/
static inline bool cgroup_bpf_sock_enabled(struct sock *sk,
                       enum cgroup_bpf_attach_type type)
{
    struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
    struct bpf_prog_array *array;

    array = rcu_access_pointer(cgrp->bpf.effective[type]);
    return array != &bpf_empty_prog_array.hdr;
}

/* Wrappers for __cgroup_bpf_run_filter_skb() guarded by cgroup_bpf_enabled. */
#define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb)                  \
({                                          \
    int __ret = 0;                                  \
    if (cgroup_bpf_enabled(CGROUP_INET_INGRESS) &&                  \
        cgroup_bpf_sock_enabled(sk, CGROUP_INET_INGRESS))              \
        __ret = __cgroup_bpf_run_filter_skb(sk, skb,              \
                            CGROUP_INET_INGRESS); \
                                          \
    __ret;                                      \
})

#define BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb)                   \
({                                           \
    int __ret = 0;                                   \
    if (cgroup_bpf_enabled(CGROUP_INET_EGRESS) && sk) {               \
        typeof(sk) __sk = sk_to_full_sk(sk);                   \
        if (sk_fullsock(__sk) && __sk == skb_to_full_sk(skb) &&           \
            cgroup_bpf_sock_enabled(__sk, CGROUP_INET_EGRESS))           \
            __ret = __cgroup_bpf_run_filter_skb(__sk, skb,           \
                              CGROUP_INET_EGRESS); \
    }                                       \
    __ret;                                       \
})

#define BPF_CGROUP_RUN_SK_PROG(sk, atype)                       \
({                                           \
    int __ret = 0;                                   \
    if (cgroup_bpf_enabled(atype)) {                           \
        __ret = __cgroup_bpf_run_filter_sk(sk, atype);               \
    }                                       \
    __ret;                                       \
})

#define BPF_CGROUP_RUN_PROG_INET_SOCK(sk)                       \
    BPF_CGROUP_RUN_SK_PROG(sk, CGROUP_INET_SOCK_CREATE)

#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk)                   \
    BPF_CGROUP_RUN_SK_PROG(sk, CGROUP_INET_SOCK_RELEASE)

#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk)                       \
    BPF_CGROUP_RUN_SK_PROG(sk, CGROUP_INET4_POST_BIND)

#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND(sk)                       \
    BPF_CGROUP_RUN_SK_PROG(sk, CGROUP_INET6_POST_BIND)

#define BPF_CGROUP_RUN_SA_PROG(sk, uaddr, uaddrlen, atype)               \
({                                           \
    int __ret = 0;                                   \
    if (cgroup_bpf_enabled(atype))                           \
        __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, uaddrlen, \
                              atype, NULL, NULL);  \
    __ret;                                       \
})

#define BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, atype, t_ctx)           \
({                                           \
    int __ret = 0;                                   \
    if (cgroup_bpf_enabled(atype))    {                       \
        lock_sock(sk);                               \
        __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, uaddrlen, \
                              atype, t_ctx, NULL); \
        release_sock(sk);                           \
    }                                       \
    __ret;                                       \
})

/* BPF_CGROUP_INET4_BIND and BPF_CGROUP_INET6_BIND can return extra flags
 * via upper bits of return code. The only flag that is supported
 * (at bit position 0) is to indicate CAP_NET_BIND_SERVICE capability check
 * should be bypassed (BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE).
 */
#define BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, uaddrlen, atype, bind_flags) \
({                                           \
    u32 __flags = 0;                               \
    int __ret = 0;                                   \
    if (cgroup_bpf_enabled(atype))    {                       \
        lock_sock(sk);                               \
        __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, uaddrlen, \
                              atype, NULL, &__flags); \
        release_sock(sk);                           \
        if (__flags & BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE)           \
            *bind_flags |= BIND_NO_CAP_NET_BIND_SERVICE;           \
    }                                       \
    __ret;                                       \
})

#define BPF_CGROUP_PRE_CONNECT_ENABLED(sk)                       \
    ((cgroup_bpf_enabled(CGROUP_INET4_CONNECT) ||               \
      cgroup_bpf_enabled(CGROUP_INET6_CONNECT)) &&               \
     (sk)->sk_prot->pre_connect)

#define BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr, uaddrlen)            \
    BPF_CGROUP_RUN_SA_PROG(sk, uaddr, uaddrlen, CGROUP_INET4_CONNECT)

#define BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr, uaddrlen)            \
    BPF_CGROUP_RUN_SA_PROG(sk, uaddr, uaddrlen, CGROUP_INET6_CONNECT)

#define BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, uaddrlen)        \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_INET4_CONNECT, NULL)

#define BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, uaddrlen)        \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_INET6_CONNECT, NULL)

#define BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, uaddr, uaddrlen)        \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_UNIX_CONNECT, NULL)

#define BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, uaddr, uaddrlen, t_ctx)    \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_UDP4_SENDMSG, t_ctx)

#define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, uaddrlen, t_ctx)    \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_UDP6_SENDMSG, t_ctx)

#define BPF_CGROUP_RUN_PROG_UNIX_SENDMSG_LOCK(sk, uaddr, uaddrlen, t_ctx)    \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_UNIX_SENDMSG, t_ctx)

#define BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, uaddr, uaddrlen)        \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_UDP4_RECVMSG, NULL)

#define BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, uaddr, uaddrlen)        \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_UDP6_RECVMSG, NULL)

#define BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk, uaddr, uaddrlen)        \
    BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, CGROUP_UNIX_RECVMSG, NULL)

/* The SOCK_OPS"_SK" macro should be used when sock_ops->sk is not a
 * fullsock and its parent fullsock cannot be traced by
 * sk_to_full_sk().
 *
 * e.g. sock_ops->sk is a request_sock and it is under syncookie mode.
 * Its listener-sk is not attached to the rsk_listener.
 * In this case, the caller holds the listener-sk (unlocked),
 * set its sock_ops->sk to req_sk, and call this SOCK_OPS"_SK" with
 * the listener-sk such that the cgroup-bpf-progs of the
 * listener-sk will be run.
 *
 * Regardless of syncookie mode or not,
 * calling bpf_setsockopt on listener-sk will not make sense anyway,
 * so passing 'sock_ops->sk == req_sk' to the bpf prog is appropriate here.
 */
#define BPF_CGROUP_RUN_PROG_SOCK_OPS_SK(sock_ops, sk)            \
({                                    \
    int __ret = 0;                            \
    if (cgroup_bpf_enabled(CGROUP_SOCK_OPS))            \
        __ret = __cgroup_bpf_run_filter_sock_ops(sk,        \
                             sock_ops,    \
                             CGROUP_SOCK_OPS); \
    __ret;                                \
})

#define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops)                       \
({                                           \
    int __ret = 0;                                   \
    if (cgroup_bpf_enabled(CGROUP_SOCK_OPS) && (sock_ops)->sk) {       \
        typeof(sk) __sk = sk_to_full_sk((sock_ops)->sk);           \
        if (__sk && sk_fullsock(__sk))                       \
            __ret = __cgroup_bpf_run_filter_sock_ops(__sk,           \
                                 sock_ops,     \
                             CGROUP_SOCK_OPS); \
    }                                       \
    __ret;                                       \
})

#define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(atype, major, minor, access)          \
({                                          \
    int __ret = 0;                                  \
    if (cgroup_bpf_enabled(CGROUP_DEVICE))                  \
        __ret = __cgroup_bpf_check_dev_permission(atype, major, minor, \
                              access,          \
                              CGROUP_DEVICE); \
                                          \
    __ret;                                      \
})


#define BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, count, pos)  \
({                                           \
    int __ret = 0;                                   \
    if (cgroup_bpf_enabled(CGROUP_SYSCTL))                   \
        __ret = __cgroup_bpf_run_filter_sysctl(head, table, write,     \
                               buf, count, pos,        \
                               CGROUP_SYSCTL);     \
    __ret;                                       \
})

#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen,   \
                       kernel_optval)                   \
({                                           \
    int __ret = 0;                                   \
    if (cgroup_bpf_enabled(CGROUP_SETSOCKOPT) &&                   \
        cgroup_bpf_sock_enabled(sock, CGROUP_SETSOCKOPT))               \
        __ret = __cgroup_bpf_run_filter_setsockopt(sock, level,           \
                               optname, optval,    \
                               optlen,           \
                               kernel_optval);     \
    __ret;                                       \
})

#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, optlen,   \
                       max_optlen, retval)               \
({                                           \
    int __ret = retval;                               \
    if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT) &&                   \
        cgroup_bpf_sock_enabled(sock, CGROUP_GETSOCKOPT))               \
        if (!(sock)->sk_prot->bpf_bypass_getsockopt ||               \
            !INDIRECT_CALL_INET_1((sock)->sk_prot->bpf_bypass_getsockopt, \
                    tcp_bpf_bypass_getsockopt,           \
                    level, optname))               \
            __ret = __cgroup_bpf_run_filter_getsockopt(           \
                sock, level, optname, optval, optlen,           \
                max_optlen, retval);                   \
    __ret;                                       \
})

#define BPF_CGROUP_RUN_PROG_GETSOCKOPT_KERN(sock, level, optname, optval,      \
                        optlen, retval)               \
({                                           \
    int __ret = retval;                               \
    if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT))                   \
        __ret = __cgroup_bpf_run_filter_getsockopt_kern(           \
            sock, level, optname, optval, optlen, retval);           \
    __ret;                                       \
})

int cgroup_bpf_prog_attach(const union bpf_attr *attr,
               enum bpf_prog_type ptype, struct bpf_prog *prog);
int cgroup_bpf_prog_detach(const union bpf_attr *attr,
               enum bpf_prog_type ptype);
int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
int cgroup_bpf_prog_query(const union bpf_attr *attr,
              union bpf_attr __user *uattr);

const struct bpf_func_proto *
cgroup_common_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog);
const struct bpf_func_proto *
cgroup_current_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog);
#else

static inline int cgroup_bpf_inherit(struct cgroup *cgrp) { return 0; }
static inline void cgroup_bpf_offline(struct cgroup *cgrp) {}

static inline int cgroup_bpf_prog_attach(const union bpf_attr *attr,
                     enum bpf_prog_type ptype,
                     struct bpf_prog *prog)
{
    return -EINVAL;
}

static inline int cgroup_bpf_prog_detach(const union bpf_attr *attr,
                     enum bpf_prog_type ptype)
{
    return -EINVAL;
}

static inline int cgroup_bpf_link_attach(const union bpf_attr *attr,
                     struct bpf_prog *prog)
{
    return -EINVAL;
}

static inline int cgroup_bpf_prog_query(const union bpf_attr *attr,
                    union bpf_attr __user *uattr)
{
    return -EINVAL;
}

static inline const struct bpf_func_proto *
cgroup_common_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
{
    return NULL;
}

static inline const struct bpf_func_proto *
cgroup_current_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
{
    return NULL;
}

static inline int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux,
                        struct bpf_map *map) { return 0; }
static inline struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(
    struct bpf_prog *prog, enum bpf_cgroup_storage_type stype) { return NULL; }
static inline void bpf_cgroup_storage_free(
    struct bpf_cgroup_storage *storage) {}
static inline int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key,
                         void *value) {
    return 0;
}
static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
                    void *key, void *value, u64 flags) {
    return 0;
}

#define cgroup_bpf_enabled(atype) (0)
#define BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, uaddrlen, atype, t_ctx) ({ 0; })
#define BPF_CGROUP_RUN_SA_PROG(sk, uaddr, uaddrlen, atype) ({ 0; })
#define BPF_CGROUP_PRE_CONNECT_ENABLED(sk) (0)
#define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk,skb) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET_EGRESS(sk,skb) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET_SOCK(sk) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, uaddrlen, atype, flags) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND(sk) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr, uaddrlen) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, uaddrlen) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr, uaddrlen) ({ 0; })
#define BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, uaddrlen) ({ 0; })
#define BPF_CGROUP_RUN_PROG_UNIX_CONNECT_LOCK(sk, uaddr, uaddrlen) ({ 0; })
#define BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, uaddr, uaddrlen, t_ctx) ({ 0; })
#define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, uaddrlen, t_ctx) ({ 0; })
#define BPF_CGROUP_RUN_PROG_UNIX_SENDMSG_LOCK(sk, uaddr, uaddrlen, t_ctx) ({ 0; })
#define BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, uaddr, uaddrlen) ({ 0; })
#define BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, uaddr, uaddrlen) ({ 0; })
#define BPF_CGROUP_RUN_PROG_UNIX_RECVMSG_LOCK(sk, uaddr, uaddrlen) ({ 0; })
#define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) ({ 0; })
#define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(atype, major, minor, access) ({ 0; })
#define BPF_CGROUP_RUN_PROG_SYSCTL(head,table,write,buf,count,pos) ({ 0; })
#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, \
                       optlen, max_optlen, retval) ({ retval; })
#define BPF_CGROUP_RUN_PROG_GETSOCKOPT_KERN(sock, level, optname, optval, \
                        optlen, retval) ({ retval; })
#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen, \
                       kernel_optval) ({ 0; })

#define for_each_cgroup_storage_type(stype) for (; false; )

#endif /* CONFIG_CGROUP_BPF */

#endif /* _BPF_CGROUP_H */

:: Command execute ::

Enter:
 
Select:
 

:: Search ::
  - regexp 

:: Upload ::
 
[ Read-Only ]

:: Make Dir ::
 
[ Read-Only ]
:: Make File ::
 
[ Read-Only ]

:: Go Dir ::
 
:: Go File ::
 

--[ c99shell v. 2.0 [PHP 7 Update] [25.02.2019] maintained by HackingTool | HackingTool | Generation time: 0.0039 ]--