Re: [PATCH v13 2/4] bpf: added new helper bpf_get_ns_current_pid_tgid

[Carlos Antonio Neira Bustos <cneirabustos@gmail.com>]

On Wed, Oct 09, 2019 at 09:14:42AM -0700, Andrii Nakryiko wrote:
> On Wed, Oct 9, 2019 at 8:27 AM Carlos Neira <cneirabustos@gmail.com> wrote:
> >
> > New bpf helper bpf_get_ns_current_pid_tgid,
> > This helper will return pid and tgid from current task
> > which namespace matches dev_t and inode number provided,
> > this will allows us to instrument a process inside a container.
> >
> > Signed-off-by: Carlos Neira <cneirabustos@gmail.com>
> > ---
> >  include/linux/bpf.h      |  1 +
> >  include/uapi/linux/bpf.h | 22 +++++++++++++++++++-
> >  kernel/bpf/core.c        |  1 +
> >  kernel/bpf/helpers.c     | 43 ++++++++++++++++++++++++++++++++++++++++
> >  kernel/trace/bpf_trace.c |  2 ++
> >  5 files changed, 68 insertions(+), 1 deletion(-)
> >
> > diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> > index 5b9d22338606..231001475504 100644
> > --- a/include/linux/bpf.h
> > +++ b/include/linux/bpf.h
> > @@ -1055,6 +1055,7 @@ extern const struct bpf_func_proto bpf_get_local_storage_proto;
> >  extern const struct bpf_func_proto bpf_strtol_proto;
> >  extern const struct bpf_func_proto bpf_strtoul_proto;
> >  extern const struct bpf_func_proto bpf_tcp_sock_proto;
> > +extern const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto;
> >
> >  /* Shared helpers among cBPF and eBPF. */
> >  void bpf_user_rnd_init_once(void);
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index 77c6be96d676..6ad3f2abf00d 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -2750,6 +2750,19 @@ union bpf_attr {
> >   *             **-EOPNOTSUPP** kernel configuration does not enable SYN cookies
> >   *
> >   *             **-EPROTONOSUPPORT** IP packet version is not 4 or 6
> > + *
> > + * u64 bpf_get_ns_current_pid_tgid(struct *bpf_pidns_info, u32 size)
> 
> Should be:
> 
> struct bpf_pidns_info *nsdata
> 
> > + *     Return
> > + *             0 on success, values for pid and tgid from nsinfo will be as seen
> > + *             from the namespace that matches dev and inum from nsinfo.
> 
> I think its cleaner to have a Description section, explaining that it
> will return pid/tgid in bpf_pidns_info, and then describe exit codes
> in Return section.
> 
> > + *
> > + *             On failure, the returned value is one of the following:
> > + *
> > + *             **-EINVAL** if dev and inum supplied don't match dev_t and inode number
> > + *              with nsfs of current task, or if dev conversion to dev_t lost high bits.
> > + *
> > + *             **-ENOENT** if /proc/self/ns does not exists.
> > + *
> >   */
> >  #define __BPF_FUNC_MAPPER(FN)          \
> >         FN(unspec),                     \
> > @@ -2862,7 +2875,8 @@ union bpf_attr {
> >         FN(sk_storage_get),             \
> >         FN(sk_storage_delete),          \
> >         FN(send_signal),                \
> > -       FN(tcp_gen_syncookie),
> > +       FN(tcp_gen_syncookie),          \
> > +       FN(get_ns_current_pid_tgid),
> >
> >  /* integer value in 'imm' field of BPF_CALL instruction selects which helper
> >   * function eBPF program intends to call
> > @@ -3613,4 +3627,10 @@ struct bpf_sockopt {
> >         __s32   retval;
> >  };
> >
> > +struct bpf_pidns_info {
> > +       __u64 dev;
> > +       __u64 inum;
> 
> seems like conventionally this should be named "ino", this is what
> ns_match calls it, so let's stay consistent.
> 
> > +       __u32 pid;
> > +       __u32 tgid;
> > +};
> 
> So it seems like dev and inum are treated as input parameters, while
> pid/tgid is output parameter, right? Wouldn't it be cleaner to have
> dev and inum as explicit arguments into bpf_get_ns_current_pid_tgid()?
> What's also not great, is that on failure you'll memset this entire
> struct to zero, and user will lose its dev/inum. So in practice you'll
> be keeping dev/inum somewhere else, then constructing and filling in
> this bpf_pidns_info struct every time you need to invoke
> bpf_get_ns_current_pid_tgid.
> 
> Maybe it was discussed already, but IMO feels cleaner to have only
> pid/tgid in bpf_pidns_info and pass dev/inum as direct arguments.
> 
> >  #endif /* _UAPI__LINUX_BPF_H__ */
> > diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> > index 66088a9e9b9e..b2fd5358f472 100644
> > --- a/kernel/bpf/core.c
> > +++ b/kernel/bpf/core.c
> > @@ -2042,6 +2042,7 @@ const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
> >  const struct bpf_func_proto bpf_get_current_comm_proto __weak;
> >  const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
> >  const struct bpf_func_proto bpf_get_local_storage_proto __weak;
> > +const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto __weak;
> >
> >  const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
> >  {
> > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > index 5e28718928ca..78a1ce7726aa 100644
> > --- a/kernel/bpf/helpers.c
> > +++ b/kernel/bpf/helpers.c
> > @@ -11,6 +11,8 @@
> >  #include <linux/uidgid.h>
> >  #include <linux/filter.h>
> >  #include <linux/ctype.h>
> > +#include <linux/pid_namespace.h>
> > +#include <linux/proc_ns.h>
> >
> >  #include "../../lib/kstrtox.h"
> >
> > @@ -487,3 +489,44 @@ const struct bpf_func_proto bpf_strtoul_proto = {
> >         .arg4_type      = ARG_PTR_TO_LONG,
> >  };
> >  #endif
> > +
> > +BPF_CALL_2(bpf_get_ns_current_pid_tgid, struct bpf_pidns_info *, nsdata, u32,
> > +       size)
> > +{
> > +       struct task_struct *task = current;
> > +       struct pid_namespace *pidns;
> > +       int err = -EINVAL;
> > +
> > +       if (unlikely(size != sizeof(struct bpf_pidns_info)))
> > +               goto clear;
> > +
> > +       if ((u64)(dev_t)nsdata->dev != nsdata->dev)
> 
> this seems unlikely() as well :)
> 
> > +               goto clear;
> > +
> > +       if (unlikely(!task))
> > +               goto clear;
> > +
> > +       pidns = task_active_pid_ns(task);
> > +       if (unlikely(!pidns)) {
> > +               err = -ENOENT;
> > +               goto clear;
> > +       }
> > +
> > +       if (!ns_match(&pidns->ns, (dev_t)nsdata->dev, nsdata->inum))
> > +               goto clear;
> > +
> > +       nsdata->pid = task_pid_nr_ns(task, pidns);
> > +       nsdata->tgid = task_tgid_nr_ns(task, pidns);
> > +       return 0;
> > +clear:
> > +       memset((void *)nsdata, 0, (size_t) size);
> > +       return err;
> > +}
> > +
> > +const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
> > +       .func           = bpf_get_ns_current_pid_tgid,
> > +       .gpl_only       = false,
> > +       .ret_type       = RET_INTEGER,
> > +       .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
> 
> So this is a lie, you do expect part of that struct to be initialized.
> One more reason to just split off dev/inum(ino?).
> 
> 
> > +       .arg2_type      = ARG_CONST_SIZE,
> > +};
> > diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> > index 44bd08f2443b..32331a1dcb6d 100644
> > --- a/kernel/trace/bpf_trace.c
> > +++ b/kernel/trace/bpf_trace.c
> > @@ -735,6 +735,8 @@ tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
> >  #endif
> >         case BPF_FUNC_send_signal:
> >                 return &bpf_send_signal_proto;
> > +       case BPF_FUNC_get_ns_current_pid_tgid:
> > +               return &bpf_get_ns_current_pid_tgid_proto;
> >         default:
> >                 return NULL;
> >         }
> > --
> > 2.20.1
> >
Thanks for reviewing this, I'll make the changes you suggest.
I'm not sure about removing dev and ino from struct bpf_pidns_info, I
think is useful to know to which ns pid/tgid belong to using dev and ino
to figure it out. Maybe in the future we could filter bpf_pidns_info
structs by dev/ino, just an idea.  

Bests