On Wed, 2023-04-26 at 13:47 +0200, Alexandr Nedvedicky wrote:
> Hello,
> 
> </snip>
> On Wed, Apr 26, 2023 at 11:37:58AM +1000, David Gwynne wrote:
> > >  fail:
> > > -       if (flags & FWRITE)
> > > -               rw_exit_write(&pfioctl_rw);
> > > -       else
> > > -               rw_exit_read(&pfioctl_rw);
> > > +       rw_exit_write(&pfioctl_rw);
> > 
> > i dont think having the open mode flags affect whether you take a shared
> > or exclusive lock makes sense anyway, so im happy with these bits.
> 
>     while updating my diff I've noticed pfclose() needs
>     same change: dropping 'flags & FWRITE' test. This is
>     fixed i new diff below.
> 
> </snip>
> > 
> >         int unit = minor(dev);
> > 
> >         if (unit & ((1 << CLONE_SHIFT) - 1))
> >                 return (ENXIO);
> > 
> > this has some ties into the second issue, which is that we shouldn't
> > use the pid as an identifier for state across syscalls like this
> > in the kernel. the reason is that userland could be weird or buggy
> > (or hostile) and fork in between creating a transaction and ending
> > a transaction, but it will still have the fd it from from open(/dev/pf).
> > instead, we should be using the whole dev_t or the minor number,
> > as long as it includes the bits that the cloning infrastructure
> > uses, as the identifier.
> 
>     in new diff I'm using a minor(dev) to identify transaction owner.
> > 
> > i would also check that the process^Wminor number that created a
> > transaction is the same when looking up a transaction in pf_find_trans.
> 
>     the pf_find_trans() is a dead code in diff below as it is
>     not being used there. Just to make sure I got things right
>     so I can update '3/3' diff in stack accordingly. Let's assume
>     snippet below comes from pfioctl() function
> 
> int                                                                           
>                                                                               
>                       
>                                                 
> pfioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct proc *p) {
> ...
>         t = pf_find_trans(ticket);
>         if (t == NULL)
>                 return (ENXIO);
> 
>         KASSERT(t->pft_unit == minor(dev));
> 
>     is that kind of check in KASSET() something you have on your mind?
>     perhaps I can trade KASSERT() to regular code:
> 
>         if (t->pft_unit != minor(dev))
>                 return (EPERM);
> 
> 
> thank you for pointers to bpf.c updated diff is below.
> 
> regards
> sashan
> 
> --------8<---------------8<---------------8<------------------8<--------
> diff --git a/sys/net/pf_ioctl.c b/sys/net/pf_ioctl.c
> index 1141069dcf6..b9904c545c5 100644
> --- a/sys/net/pf_ioctl.c
> +++ b/sys/net/pf_ioctl.c
> @@ -117,6 +117,11 @@ void                        pf_qid_unref(u_int16_t);
>  int                     pf_states_clr(struct pfioc_state_kill *);
>  int                     pf_states_get(struct pfioc_states *);
>  
> +struct pf_trans                *pf_open_trans(uint32_t);
> +struct pf_trans                *pf_find_trans(uint64_t);
> +void                    pf_free_trans(struct pf_trans *);
> +void                    pf_rollback_trans(struct pf_trans *);
> +
>  struct pf_rule          pf_default_rule, pf_default_rule_new;
>  
>  struct {
> @@ -168,6 +173,8 @@ int                  pf_rtlabel_add(struct pf_addr_wrap 
> *);
>  void                    pf_rtlabel_remove(struct pf_addr_wrap *);
>  void                    pf_rtlabel_copyout(struct pf_addr_wrap *);
>  
> +uint64_t trans_ticket = 1;
> +LIST_HEAD(, pf_trans)  pf_ioctl_trans = LIST_HEAD_INITIALIZER(pf_trans);
>  
>  void
>  pfattach(int num)
> @@ -293,6 +300,28 @@ pfopen(dev_t dev, int flags, int fmt, struct proc *p)
>  int
>  pfclose(dev_t dev, int flags, int fmt, struct proc *p)
>  {
> +       struct pf_trans *w, *s;
> +       LIST_HEAD(, pf_trans)   tmp_list;
> +       uint32_t unit = minor(dev);
> +
> +       if (minor(dev) >= 1)
> +               return (ENXIO);

If you want to use the minor dev-id to release the transaction,
the above two lines should go away.


> +
> +       LIST_INIT(&tmp_list);
> +       rw_enter_write(&pfioctl_rw);
> +       LIST_FOREACH_SAFE(w, &pf_ioctl_trans, pft_entry, s) {
> +               if (w->pft_unit == unit) {
> +                       LIST_REMOVE(w, pft_entry);
> +                       LIST_INSERT_HEAD(&tmp_list, w, pft_entry);
> +               }
> +       }
> +       rw_exit_write(&pfioctl_rw);
> +
> +       while ((w = LIST_FIRST(&tmp_list)) != NULL) {
> +               LIST_REMOVE(w, pft_entry);
> +               pf_free_trans(w);
> +       }
> +
>         return (0);
>  }
>  
> @@ -1142,10 +1171,7 @@ pfioctl(dev_t dev, u_long cmd, caddr_t addr, int 
> flags, struct proc *p)
>                         return (EACCES);
>                 }
>  
> -       if (flags & FWRITE)
> -               rw_enter_write(&pfioctl_rw);
> -       else
> -               rw_enter_read(&pfioctl_rw);
> +       rw_enter_write(&pfioctl_rw);
>  
>         switch (cmd) {
>  
> @@ -3022,10 +3048,7 @@ pfioctl(dev_t dev, u_long cmd, caddr_t addr, int 
> flags, struct proc *p)
>                 break;
>         }
>  fail:
> -       if (flags & FWRITE)
> -               rw_exit_write(&pfioctl_rw);
> -       else
> -               rw_exit_read(&pfioctl_rw);
> +       rw_exit_write(&pfioctl_rw);
>  
>         return (error);
>  }
> @@ -3244,3 +3267,52 @@ pf_sysctl(void *oldp, size_t *oldlenp, void *newp, 
> size_t newlen)
>  
>         return sysctl_rdstruct(oldp, oldlenp, newp, &pfs, sizeof(pfs));
>  }
> +
> +struct pf_trans *
> +pf_open_trans(uint32_t unit)
> +{
> +       static uint64_t ticket = 1;
> +       struct pf_trans *t;
> +
> +       rw_assert_wrlock(&pfioctl_rw);
> +
> +       t = malloc(sizeof(*t), M_TEMP, M_WAITOK);
> +       memset(t, 0, sizeof(struct pf_trans));
> +       t->pft_unit = unit;
> +       t->pft_ticket = ticket++;
> +
> +       LIST_INSERT_HEAD(&pf_ioctl_trans, t, pft_entry);
> +
> +       return (t);
> +}
> +
> +struct pf_trans *
> +pf_find_trans(uint64_t ticket)
> +{
> +       struct pf_trans *t;
> +
> +       rw_assert_anylock(&pfioctl_rw);
> +
> +       LIST_FOREACH(t, &pf_ioctl_trans, pft_entry) {
> +               if (t->pft_ticket == ticket)
> +                       break;
> +       }
> +
> +       return (t);
> +}
> +
> +void
> +pf_free_trans(struct pf_trans *t)
> +{
> +       free(t, M_TEMP, sizeof(*t));
> +}
> +
> +void
> +pf_rollback_trans(struct pf_trans *t)
> +{
> +       if (t != NULL) {
> +               rw_assert_wrlock(&pfioctl_rw);
> +               LIST_REMOVE(t, pft_entry);
> +               pf_free_trans(t);
> +       }
> +}
> diff --git a/sys/net/pfvar_priv.h b/sys/net/pfvar_priv.h
> index 38fff6473aa..5af2027733a 100644
> --- a/sys/net/pfvar_priv.h
> +++ b/sys/net/pfvar_priv.h
> @@ -322,6 +322,17 @@ enum {
>  
>  extern struct cpumem *pf_anchor_stack;
>  
> +enum pf_trans_type {
> +       PF_TRANS_NONE,
> +       PF_TRANS_MAX
> +};
> +
> +struct pf_trans {
> +       LIST_ENTRY(pf_trans)    pft_entry;
> +       uint32_t                pft_unit;               /* process id */
> +       uint64_t                pft_ticket;
> +       enum pf_trans_type      pft_type;
> +};
>  extern struct task     pf_purge_task;
>  extern struct timeout  pf_purge_to;
>  
> 

Attachment: smime.p7s
Description: S/MIME cryptographic signature

Reply via email to