[PATCH v2 RESEND 0/4] Drivers: hv: utils: prevent crash when a utility driver is disabled host side

[Vitaly Kuznetsov]

Changes since v2:
- No changes, this is just a resend.

Changes since v1:
- Don't re-introduce memory leak in hvt_op_write(), preserve the error
  code from on_msg() [Dan Carpenter]

Original description:

I'm observing a crash when a utility driver is disabled host side (e.g.
'Guest services' is disabled live) when we have userspace daemon
connected:

[   90.244859] general protection fault: 0000 [#1] SMP
...
[   90.800082] CPU: 3 PID: 473 Comm: hypervfcopyd Not tainted 4.3.0-rc7_netvsc_noalloc+ #1053
...
[   90.800082] Call Trace:
[   90.800082]  [<ffffffff81187008>] __fput+0xc8/0x1f0
[   90.800082]  [<ffffffff8118716e>] ____fput+0xe/0x10
[   90.800082]  [<ffffffff8107e4b3>] task_work_run+0x73/0x90
[   90.800082]  [<ffffffff81066995>] do_exit+0x335/0xa90
[   90.800082]  [<ffffffff8106716f>] do_group_exit+0x3f/0xc0
[   90.800082]  [<ffffffff81071abe>] get_signal+0x25e/0x5e0
[   90.800082]  [<ffffffff81015278>] do_signal+0x28/0x580
[   90.800082]  [<ffffffff81086656>] ? finish_task_switch+0xa6/0x180
[   90.800082]  [<ffffffff81443ebf>] ? __schedule+0x28f/0x870
[   90.800082]  [<ffffffffa01ebbaa>] ? hvt_op_read+0x12a/0x140 [hv_utils]
[   90.800082]  [<ffffffff8109eca0>] ? wake_atomic_t_function+0x70/0x70
...
[   90.800082] RIP  [<ffffffff810d3866>] module_put+0x16/0x90
[   90.800082]  RSP <ffff88003eb1bb88>
[   95.734261] ---[ end trace 0e09af6a6294a668 ]---

The problem is that hvutil_transport_destroy() which does misc_deregister()
freeing the appropriate device is reachable by two paths: module unload
and from util_remove(). While module unload path is protected by .owner in
struct file_operations util_remove() path is not. Freeing the device while
someone holds an open fd for it is a show stopper.

In general, it is not possible to revoke an fd from all users so the only
way to solve the issue is to defer freeing the hvutil_transport structure
asking the daemon to exit gracefully by responding -EBADF to all
operations on unload.

Patch 1 fixes an unrelated issue I spotted, patch 2 renames outmsg_lock to
'lock' as we're gonna use it to protect test-and-set operations on 'mode',
patch 3 introduces HVUTIL_TRANSPORT_DESTROY mode of operation, patch 4
fixes the issue itself.

Patches are rebased on previously sent Olaf's fixes.

Vitaly Kuznetsov (4):
  Drivers: hv: utils: fix memory leak on on_msg() failure
  Drivers: hv: utils: rename outmsg_lock
  Drivers: hv: utils: introduce HVUTIL_TRANSPORT_DESTROY mode
  Drivers: hv: utils: fix crash when device is removed from host side

 drivers/hv/hv_utils_transport.c | 112 +++++++++++++++++++++++++++++++---------
 drivers/hv/hv_utils_transport.h |   3 +-
 2 files changed, 89 insertions(+), 26 deletions(-)

-- 
2.4.3

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/