"value":"In the Linux kernel, the following vulnerability has been resolved:\n\nbpf: track changes_pkt_data property for global functions\n\nWhen processing calls to certain helpers, verifier invalidates all\npacket pointers in a current state. For example, consider the\nfollowing program:\n\n __attribute__((__noinline__))\n long skb_pull_data(struct __sk_buff *sk, __u32 len)\n {\n return bpf_skb_pull_data(sk, len);\n }\n\n SEC(\"tc\")\n int test_invalidate_checks(struct __sk_buff *sk)\n {\n int *p = (void *)(long)sk->data;\n if ((void *)(p + 1) > (void *)(long)sk->data_end) return TCX_DROP;\n skb_pull_data(sk, 0);\n *p = 42;\n return TCX_PASS;\n }\n\nAfter a call to bpf_skb_pull_data() the pointer 'p' can't be used\nsafely. See function filter.c:bpf_helper_changes_pkt_data() for a list\nof such helpers.\n\nAt the moment verifier invalidates packet pointers when processing\nhelper function calls, and does not traverse global sub-programs when\nprocessing calls to global sub-programs. This means that calls to\nhelpers done from global sub-programs do not invalidate pointers in\nthe caller state. E.g. the program above is unsafe, but is not\nrejected by verifier.\n\nThis commit fixes the omission by computing field\nbpf_subprog_info->changes_pkt_data for each sub-program before main\nverification pass.\nchanges_pkt_data should be set if:\n- subprogram calls helper for which bpf_helper_changes_pkt_data\n returns true;\n- subprogram calls a global function,\n for which bpf_subprog_info->changes_pkt_data should be set.\n\nThe verifier.c:check_cfg() pass is modified to compute this\ninformation. The commit relies on depth first instruction traversal\ndone by check_cfg() and absence of recursive function calls:\n- check_cfg() would eventually visit every call to subprogram S in a\n state when S is fully explored;\n- when S is fully explored:\n - every direct helper call within S is explored\n (and thus changes_pkt_data is set if needed);\n - every call to subprogram S1 called by S was visited with S1 fully\n explored (and thus S inherits changes_pkt_data from S1).\n\nThe downside of such approach is that dead code elimination is not\ntaken into account: if a helper call inside global function is dead\nbecause of current configuration, verifier would conservatively assume\nthat the call occurs for the purpose of the changes_pkt_data\ncomputation."
