n_gsm: race between ld close and gsmtty open
authorChao Bi <chao.bi@intel.com>
Thu, 17 Oct 2013 07:08:27 +0000 (15:08 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 25 Nov 2013 16:52:53 +0000 (08:52 -0800)
commitc284ee2cf12b55fa8496b2d098bf0938688f1c1c
tree5f553255c2b1dd33f6dcab7434ea2d7673bbb71b
parentf3014127ada32f51ce0baf0bee4c6324a601ef59
n_gsm: race between ld close and gsmtty open

ttyA has ld associated to n_gsm, when ttyA is closing, it triggers
to release gsmttyB's ld data dlci[B], then race would happen if gsmttyB
is opening in parallel.

Here are race cases we found recently in test:

CASE #1
====================================================================
releasing dlci[B] race with gsmtty_install(gsmttyB), then panic
in gsmtty_open(gsmttyB), as below:

 tty_release(ttyA)                  tty_open(gsmttyB)
     |                                   |
   -----                           gsmtty_install(gsmttyB)
     |                                   |
   -----                    gsm_dlci_alloc(gsmttyB) => alloc dlci[B]
 tty_ldisc_release(ttyA)               -----
     |                                   |
 gsm_dlci_release(dlci[B])             -----
     |                                   |
 gsm_dlci_free(dlci[B])                -----
     |                                   |
   -----                           gsmtty_open(gsmttyB)

 gsmtty_open()
 {
     struct gsm_dlci *dlci = tty->driver_data; => here it uses dlci[B]
     ...
 }

 In gsmtty_open(gsmttyA), it uses dlci[B] which was release, so hit a panic.
=====================================================================

CASE #2
=====================================================================
releasing dlci[0] race with gsmtty_install(gsmttyB), then panic
in gsmtty_open(), as below:

 tty_release(ttyA)                  tty_open(gsmttyB)
     |                                   |
   -----                           gsmtty_install(gsmttyB)
     |                                   |
   -----                    gsm_dlci_alloc(gsmttyB) => alloc dlci[B]
     |                                   |
   -----                         gsmtty_open(gsmttyB) fail
     |                                   |
   -----                           tty_release(gsmttyB)
     |                                   |
   -----                           gsmtty_close(gsmttyB)
     |                                   |
   -----                        gsmtty_detach_dlci(dlci[B])
     |                                   |
   -----                             dlci_put(dlci[B])
     |                                   |
 tty_ldisc_release(ttyA)               -----
     |                                   |
 gsm_dlci_release(dlci[0])             -----
     |                                   |
 gsm_dlci_free(dlci[0])                -----
     |                                   |
   -----                             dlci_put(dlci[0])

 In gsmtty_detach_dlci(dlci[B]), it tries to use dlci[0] which was released,
 then hit panic.
=====================================================================

IMHO, n_gsm tty operations would refer released ldisc,  as long as
gsm_dlci_release() has chance to release ldisc data when some gsmtty operations
are not completed..

This patch is try to avoid it by:

1) in n_gsm driver, use a global gsm spin lock to avoid gsm_dlci_release() run in
parallel with gsmtty_install();

2) Increase dlci's ref count in gsmtty_install() instead of in gsmtty_open(), the
purpose is to prevent gsm_dlci_release() releasing dlci after gsmtty_install()
allocats dlci but before gsmtty_open increases dlci's ref count;

3) Decrease dlci's ref count in gsmtty_remove(), which is a tty framework api, and
this is the opposite process of step 2).

Signed-off-by: Chao Bi <chao.bi@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/tty/n_gsm.c