On 29.05.19 12:37, Greg KH wrote: > On Wed, May 29, 2019 at 12:25:39PM +0200, Stefan Bader wrote: >> From: Jiri Wiesner <jwiesner@xxxxxxxx> >> >> The *_frag_reasm() functions are susceptible to miscalculating the byte >> count of packet fragments in case the truesize of a head buffer changes. >> The truesize member may be changed by the call to skb_unclone(), leaving >> the fragment memory limit counter unbalanced even if all fragments are >> processed. This miscalculation goes unnoticed as long as the network >> namespace which holds the counter is not destroyed. >> >> Should an attempt be made to destroy a network namespace that holds an >> unbalanced fragment memory limit counter the cleanup of the namespace >> never finishes. The thread handling the cleanup gets stuck in >> inet_frags_exit_net() waiting for the percpu counter to reach zero. The >> thread is usually in running state with a stacktrace similar to: >> >> PID: 1073 TASK: ffff880626711440 CPU: 1 COMMAND: "kworker/u48:4" >> #5 [ffff880621563d48] _raw_spin_lock at ffffffff815f5480 >> #6 [ffff880621563d48] inet_evict_bucket at ffffffff8158020b >> #7 [ffff880621563d80] inet_frags_exit_net at ffffffff8158051c >> #8 [ffff880621563db0] ops_exit_list at ffffffff814f5856 >> #9 [ffff880621563dd8] cleanup_net at ffffffff814f67c0 >> #10 [ffff880621563e38] process_one_work at ffffffff81096f14 >> >> It is not possible to create new network namespaces, and processes >> that call unshare() end up being stuck in uninterruptible sleep state >> waiting to acquire the net_mutex. >> >> The bug was observed in the IPv6 netfilter code by Per Sundstrom. >> I thank him for his analysis of the problem. The parts of this patch >> that apply to IPv4 and IPv6 fragment reassembly are preemptive measures. >> >> Signed-off-by: Jiri Wiesner <jwiesner@xxxxxxxx> >> Reported-by: Per Sundstrom <per.sundstrom@xxxxxxxxxx> >> Acked-by: Peter Oskolkov <posk@xxxxxxxxxx> >> Signed-off-by: David S. Miller <davem@xxxxxxxxxxxxx> >> >> (backported from commit ebaf39e6032faf77218220707fc3fa22487784e0) >> [smb: context adjustments in net/ipv6/netfilter/nf_conntrack_reasm.c] >> Signed-off-by: Stefan Bader <stefan.bader@xxxxxxxxxxxxx> > > I can't take a patch for 4.4.y that is not in 4.9.y as anyone upgrading > kernel versions would have a regression :( > > Can you also provide a backport of the needed patches for 4.9.y for this > issue so I can take these? It turns out that I cannot provide patches for 4.9.y because those are not needed there. Patches #1 and #2 of the list I did do not explicitly appear. However patch #3 does and it might be possible to implicitly do the changes of the other two by adjusting the removal of the functions from the old locations and doing the additions unmodified. And the final patch for pulling the skb from the list is included in 4.9.y by backports for using rbtrees in ipv6, too. In 4.9.y however the skb_get() still needs to be dropped. Sasha did not apply it in the end, maybe partially because of my warning that this was not enough in 4.4.y. So I think there are two options for 4.4.y which I would defer to the net-devs to decide: - either also backport the patches to use rbtrees in ipv6 to 4.4.y (including use of inet_frag_pull_head() in ip6_expire_frag_queue() and dropping the skb_get() there. - or some limited change to ip6_expire_frag_queue(). Probably only doing the part not related to rbtrees. This would not require patch #3 as pre-req. Patches #1 and #2 might be considered separately. Those would be unrelated to the crash in ip6_expire-frag_queue() but could fix other issues (not liking the sound of net namespace teardown possibly getting stuck). For 4.9.y, please re-consider picking "ip6: fix skb leak in ip6frag_expire_frag_queue()". Depending on checks down the path of icmpv6_send() it might be a crash, too. In that case it should be enough as inet_frag_pull_head() takes the skb off the queue, so it won't get double freed when releasing the whole queue. -Stefan > > thanks, > > greg k-h >
Attachment:
signature.asc
Description: OpenPGP digital signature
