Re: [PATCH 1/1] pstore/ram: Check member of buffers during the initialization phase of the pstore

[Kees Cook]

On Fri, Aug 04, 2023 at 04:59:07PM +0800, yunlong xing wrote:
> On Fri, Aug 4, 2023 at 4:10 PM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> >
> > On Tue, Aug 01, 2023 at 02:04:32PM +0800, Yunlong Xing wrote:
> > > From: Enlin Mu <enlin.mu@xxxxxxxxxx>
> > >
> > > The commit 30696378f68a("pstore/ram: Do not treat empty buffers as valid")
> > > would introduce the following issue:
> > >
> > > When finding the buffer_size is zero, it would return directly.However, at
> > > the same time, if the buffer's start is a illegal value, the others would
> > > panic if access the buffer.
> >
> > Which "others" do you mean?
> 
> About “others", You can refer to the following panic call stack:
>  sysdump_panic_event+0x720/0xd38
>  atomic_notifier_call_chain+0x58/0xc0
>  panic+0x1c4/0x6e4
>  die+0x3c0/0x428
>  bug_handler+0x4c/0x9c
>  brk_handler+0x98/0x14c
>  do_debug_exception+0x114/0x2ec
>  el1_dbg+0x18/0xbc
>  usercopy_abort+0x90/0x94
>  __check_object_size+0x17c/0x2c4
>  persistent_ram_update_user+0x50/0x220
>  persistent_ram_write_user+0x354/0x428
>  ramoops_pstore_write_user+0x34/0x50
>  write_pmsg+0x14c/0x26c

I see -- the "start" is corrupted and out of bounds, which leads to
these accesses.

>  do_iter_write+0x1cc/0x2cc
>  vfs_writev+0xf4/0x168
>  do_writev+0xa4/0x200
>  __arm64_sys_writev+0x20/0x2c
>  el0_svc_common+0xc8/0x22c
>  el0_svc_handler+0x1c/0x28
>  el0_svc+0x8/0x100
> >
> > > To avoid these happenning, check if the members are legal during the
> > > initialization phase of the pstore.
> > >
> > > Fixes: 30696378f68a ("pstore/ram: Do not treat empty buffers as valid")
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Signed-off-by: Enlin Mu <enlin.mu@xxxxxxxxxx>
> > > ---
> > >  fs/pstore/ram_core.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/fs/pstore/ram_core.c b/fs/pstore/ram_core.c
> > > index 85aaf0fc6d7d..eb6df190d752 100644
> > > --- a/fs/pstore/ram_core.c
> > > +++ b/fs/pstore/ram_core.c
> > > @@ -519,7 +519,7 @@ static int persistent_ram_post_init(struct persistent_ram_zone *prz, u32 sig,
> > >       sig ^= PERSISTENT_RAM_SIG;
> > >
> > >       if (prz->buffer->sig == sig) {
> > > -             if (buffer_size(prz) == 0) {
> > > +             if (buffer_size(prz) == 0 && buffer_start(prz) == 0) {
> > >                       pr_debug("found existing empty buffer\n");
> > >                       return 0;
> > >               }
> >
> > And in the case of "buffer_size(prz) == 0" but "buffer_start(prz) != 0",
> > this will be caught by:
> >
> >                 if (buffer_size(prz) > prz->buffer_size ||
> >                     buffer_start(prz) > buffer_size(prz)) {
> >                         pr_info("found existing invalid buffer, size %zu, start %zu\n",
> >                                 buffer_size(prz), buffer_start(prz));
> >                         zap = true;
> >                 }
> >
> > i.e. it will be detected and zapped back to a sane state.
> No,This code has no chance of execution because there was a return 0 before it

Right, I meant the behavior with your patch -- with your patch the case
of "size == 0 && start != 0" would be caught by the above check ("start > size")
and zapped back to sanity. (Which is the correct result.)

> >
> > That sounds correct to me, though I wonder if reporting it as an
> > "invalid buffer" is inaccurate? Perhaps we should have a separate case:
> >
> >                 if (buffer_size(prz) == 0) {
> >                         if (buffer_start(prz) == 0)
> >                                 pr_debug("found existing empty buffer\n");
> >                         else {
> >                                 pr_debug("found existing empty buffer with non-zero start\n");
> >                                 zap = true;
> >                         }
> >                 } else if ...
> >
> > What do you think?
> Good, I gree it. For me, it should not return directly while finding
> the buffer_size is zero, We need Check others case.

Right. The only question I have is: how did the "start" get corrupted,
and is that a notable condition? Right now we don't (info-level) log
a size==0 prz since that's an expected state for a regular initialized
prz. So maybe your patch is correct as-is since we'd want to report the
"found existing invalid buffer" case.

> So does the modification method you mentioned require me to resubmit a
> patch or do you need to modify and merge it

I think I'll update the commit log and take this as-is. If the logging
becomes too noisy, we can adjust the case later.

Thanks!

-- 
Kees Cook