On Thu Nov 5, 2020 at 1:32 PM PST, Andrii Nakryiko wrote:
> On Wed, Nov 4, 2020 at 8:51 PM Daniel Xu <dxu@xxxxxxxxx> wrote:
[...]
> > diff --git a/tools/testing/selftests/bpf/progs/test_probe_read_user_str.c b/tools/testing/selftests/bpf/progs/test_probe_read_user_str.c
> > new file mode 100644
> > index 000000000000..41c3e296566e
> > --- /dev/null
> > +++ b/tools/testing/selftests/bpf/progs/test_probe_read_user_str.c
> > @@ -0,0 +1,34 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +
> > +#include <linux/bpf.h>
> > +#include <bpf/bpf_helpers.h>
> > +#include <bpf/bpf_tracing.h>
> > +
> > +#include <sys/types.h>
> > +
> > +struct sys_enter_write_args {
> > + unsigned long long pad;
> > + int syscall_nr;
> > + int pad1; /* 4 byte hole */
>
> I have a hunch that this explicit padding might break on big-endian
> architectures?..
>
> Can you instead include "vmlinux.h" in this file and use struct
> trace_event_raw_sys_enter? you'll just need ctx->args[2] to get that
> buffer pointer.
>
> Alternatively, and it's probably simpler overall would be to just
> provide user-space pointer through global variable:
>
> void *user_ptr;
>
>
> bpf_probe_read_user_str(buf, ..., user_ptr);
>
> From user-space:
>
> skel->bss->user_ptr = &my_userspace_buf;
>
> Full control. You can trigger tracepoint with just an usleep(1), for
> instance.
Yeah, that sounds better. I'll send a v4 with passing a ptr.
Thanks,
Daniel
[...]
