Re: [BUG] net: xfrm: Two possible sleep-in-atomic-context bugs

On Fri, Aug 10, 2018 at 10:02:42AM +0800, bai wrote:

The code may sleep in interrupt handler.
xfrm_trans_reinject() is an interrupt handler set in tasklet_init().
The function call paths (from bottom to top) in Linux-4.16 are:

[FUNC] schedule_timeout
net/core/sock.c, 2044: schedule_timeout in sock_wait_for_wmem
net/core/sock.c, 2083: sock_wait_for_wmem in sock_alloc_send_pskb
net/core/sock.c, 2102: sock_alloc_send_pskb in sock_alloc_send_skb
net/ipv6/mcast.c, 1989: sock_alloc_send_skb in igmp6_send

igmp6_send calls sock_alloc_send_skb with 'noblock = 1',
this means that sock_wait_for_wmem is not executed in
sock_alloc_send_pskb.

net/ipv6/mcast.c, 2391: igmp6_send in igmp6_join_group
net/ipv6/mcast.c, 670: igmp6_join_group in igmp6_group_added
net/ipv6/mcast.c, 914: igmp6_group_added in ipv6_dev_mc_inc
net/ipv6/ndisc.c, 379: ipv6_dev_mc_inc in pndisc_constructor
net/core/neighbour.c, 640: [FUNC_PTR]pndisc_constructor in pneigh_lookup
net/ipv6/ip6_output.c, 483: pneigh_lookup in ip6_forward
./include/net/dst.h, 449: [FUNC_PTR]ip6_forward in dst_input
net/ipv6/ip6_input.c, 71: dst_input in ip6_rcv_finish
net/xfrm/xfrm_input.c, 511: [FUNC_PTR]ip6_rcv_finish in xfrm_trans_reinject

[FUNC] kmalloc(GFP_KERNEL)
net/core/neighbour.c, 630: kmalloc in pneigh_lookup
net/ipv6/ip6_output.c, 483: pneigh_lookup in ip6_forward

ip6_forward calls pneigh_lookup with 'creat = 0',
this means that pneigh_lookup does not do the kmalloc.

./include/net/dst.h, 449: [FUNC_PTR]ip6_forward in dst_input
net/ipv6/ip6_input.c, 71: dst_input in ip6_rcv_finish
net/xfrm/xfrm_input.c, 511: [FUNC_PTR]ip6_rcv_finish in xfrm_trans_reinject

Note that [FUNC_PTR] means a function pointer call is used.

I do not find a good way to fix them, so I only report.
These possible bugs are found by my static analysis tool (DSAC) and checked
by my code review.

Both codepaths are ok, maybe you should fix your tool ;-)