Re: [PATCH] obexd: Fix crash of transfer pointer has being freed

[Luiz Augusto von Dentz <luiz.dentz@gmail.com>]

Hi Youwan,

On Fri, Jun 17, 2022 at 3:58 PM Luiz Augusto von Dentz
<luiz.dentz@gmail.com> wrote:
>
> Hi Youwan,
>
> On Wed, Jun 15, 2022 at 5:54 AM Youwan Wang <wangyouwan@uniontech.com> wrote:
> >
> > PC1 -> PC2 in the process of transferring the file,
> > PC2 terminate the reception.
> > It is found that PC1 will receive the Rcvd
> > Forbiden message.
> > transfer_response func is called and go to failed:
> > transfer_complete func is called and go to transfer_free
> >
> > Invalid read of size 8
> >     at 0x129C9B: put_get_data (gobex-transfer.c:138)
> >     by 0x128675: get_body (gobex-packet.c:397)
> >     by 0x128675: g_obex_packet_encode (gobex-packet.c:451)
> >     by 0x1264ED: write_data (gobex.c:461)
> >     by 0x1264ED: write_data (gobex.c:431)
> >     by 0x49B7E97: g_main_context_dispatch ()
> >     by 0x49B8287: ??? ()
> >     by 0x49B8581: g_main_loop_run ()
> >     by 0x120834: main (main.c:322)
> >   Address 0x7300028 is 56 bytes inside a block of size 64 free'd
> >     at 0x48369AB: free (vg_replace_malloc.c:530)
> >     by 0x12A404: transfer_response (gobex-transfer.c:245)
> >     by 0x126B02: handle_response (gobex.c:1128)
> >     by 0x127087: incoming_data (gobex.c:1373)
> >     by 0x49B7E97: g_main_context_dispatch ()
> >     by 0x49B8287: ??? ()
> >     by 0x49B8581: g_main_loop_run ()
> >     by 0x120834: main (main.c:322)
> >   Block was alloc'd at
> >     at 0x4837B65: calloc (vg_replace_malloc.c:752)
> >     by 0x49BD9D8: g_malloc0 ()
> >     by 0x129A49: transfer_new (gobex-transfer.c:258)
> >     by 0x12A612: g_obex_put_req_pkt (gobex-transfer.c:282)
> >     by 0x12A612: g_obex_put_req_pkt (gobex-transfer.c:271)
> >     by 0x145862: transfer_start_put (transfer.c:880)
> >     by 0x145862: obc_transfer_start (transfer.c:911)
> >     by 0x13B487: session_process_transfer (session.c:836)
> >     by 0x13C128: session_process_queue (session.c:887)
> >     by 0x13C128: session_process_queue (session.c:872)
> >     by 0x13C18F: session_process (session.c:820)
> >     by 0x49B7E97: g_main_context_dispatch (in /usr/lib/x86_64-linux-gnu/)
> >     by 0x49B8287: ??? (in /usr/lib/x86_64-linux-gnu/)
> >     by 0x49B8581: g_main_loop_run (in /usr/lib/x86_64-linux-gnu/)
> >     by 0x120834: main (main.c:322)
> > ---
> >  gobex/gobex-transfer.c | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> >
> > diff --git a/gobex/gobex-transfer.c b/gobex/gobex-transfer.c
> > index c94d018b2..8e97fc73e 100644
> > --- a/gobex/gobex-transfer.c
> > +++ b/gobex/gobex-transfer.c
> > @@ -197,6 +197,14 @@ static void transfer_response(GObex *obex, GError *err, GObexPacket *rsp,
> >         }
> >
> >         rspcode = g_obex_packet_get_operation(rsp, &final);
> > +       if (rspcode == G_OBEX_RSP_FORBIDDEN) {
> > +               err = g_error_new(G_OBEX_ERROR, rspcode, "%s",
> > +                                               g_obex_strerror(rspcode));
> > +               g_obex_debug(G_OBEX_DEBUG_ERROR, "%s", err->message);
> > +               g_error_free(err);
> > +               return;
> > +       }
>
> This is not quite right, in fact the if statement below already treat
> such errors, we might be missing something like final:
>
> /* Reset if final so it can no longer be cancelled */
> if (final_rsp)
> obex->pending_req = NULL;
>
> This would be a problem on the remote implementation which is
> indicating that perhaps it is just a temporary failure, anyway then:
>
> id = transfer->req_id;
> transfer->req_id = 0;
>
> This resets the req_id so on transfer_free the following code is not run:
>
> if (transfer->req_id > 0)
> g_obex_cancel_req(transfer->obex, transfer->req_id, TRUE);
>
> So either we have to restore the transfer->req_id = id or not reset it
> to 0, so perhaps something like this works better:
>
> diff --git a/gobex/gobex-transfer.c b/gobex/gobex-transfer.c
> index c94d018b2..8fae65d8d 100644
> --- a/gobex/gobex-transfer.c
> +++ b/gobex/gobex-transfer.c
> @@ -200,6 +200,11 @@ static void transfer_response(GObex *obex, GError
> *err, GObexPacket *rsp,
>         if (rspcode != G_OBEX_RSP_SUCCESS && rspcode != G_OBEX_RSP_CONTINUE) {
>                 err = g_error_new(G_OBEX_ERROR, rspcode, "%s",
>                                                 g_obex_strerror(rspcode));
> +               /* If not marked as final response but transfer_complete gonna
> +                * be called restore the req_id so it can be properly cancelled
> +                */
> +               if (!final)
> +                       transport->req_id = id;
>                 goto failed;
>         }

You haven't responded about this one, I believe there are quite a few
patches of yours that are trying to circumvent this very issue.

> > +
> >         if (rspcode != G_OBEX_RSP_SUCCESS && rspcode != G_OBEX_RSP_CONTINUE) {
> >                 err = g_error_new(G_OBEX_ERROR, rspcode, "%s",
> >                                                 g_obex_strerror(rspcode));
> > --
> > 2.20.1
> >
> >
> >
>
>
> --
> Luiz Augusto von Dentz



-- 
Luiz Augusto von Dentz