On Tue, Jan 09, 2018 at 11:44:05AM -0800, Dan Williams wrote: > On Tue, Jan 9, 2018 at 11:34 AM, Jiri Kosina <jikos@xxxxxxxxxx> wrote: > > On Fri, 5 Jan 2018, Dan Williams wrote: > > > > [ ... snip ... ] > >> Andi Kleen (1): > >> x86, barrier: stop speculation for failed access_ok > >> > >> Dan Williams (13): > >> x86: implement nospec_barrier() > >> [media] uvcvideo: prevent bounds-check bypass via speculative execution > >> carl9170: prevent bounds-check bypass via speculative execution > >> p54: prevent bounds-check bypass via speculative execution > >> qla2xxx: prevent bounds-check bypass via speculative execution > >> cw1200: prevent bounds-check bypass via speculative execution > >> Thermal/int340x: prevent bounds-check bypass via speculative execution > >> ipv6: prevent bounds-check bypass via speculative execution > >> ipv4: prevent bounds-check bypass via speculative execution > >> vfs, fdtable: prevent bounds-check bypass via speculative execution > >> net: mpls: prevent bounds-check bypass via speculative execution > >> udf: prevent bounds-check bypass via speculative execution > >> userns: prevent bounds-check bypass via speculative execution > >> > >> Mark Rutland (4): > >> asm-generic/barrier: add generic nospec helpers > >> Documentation: document nospec helpers > >> arm64: implement nospec_ptr() > >> arm: implement nospec_ptr() > > > > So considering the recent publication of [1], how come we all of a sudden > > don't need the barriers in ___bpf_prog_run(), namely for LD_IMM_DW and > > LDX_MEM_##SIZEOP, and something comparable for eBPF JIT? > > > > Is this going to be handled in eBPF in some other way? > > > > Without that in place, and considering Jann Horn's paper, it would seem > > like PTI doesn't really lock it down fully, right? > > Here is the latest (v3) bpf fix: > > https://patchwork.ozlabs.org/patch/856645/ > > I currently have v2 on my 'nospec' branch and will move that to v3 for > the next update, unless it goes upstream before then. That patch seems specific to CONFIG_BPF_SYSCALL. Is the bpf() syscall the only attack vector? Or are there other ways to run bpf programs that we should be worried about? -- Josh
