We have been seeing occasional deadlocks on pernet_ops_rwsem since
September in NIPA. The stuck task was usually modprobe (often loading
a driver like ipvlan), trying to take the lock as a Writer.
lockdep does not track readers for rwsems so the read wasn't obvious
from the reports.
On closer inspection the Reader holding the lock was conntrack looping
forever in nf_conntrack_cleanup_net_list(). Based on past experience
with occasional NIPA crashes I looked thru the tests which run before
the crash and noticed that the crash follows ip_defrag.sh. An immediate
red flag. Scouring thru (de)fragmentation queues reveals skbs sitting
around, holding conntrack references.
The problem is that since conntrack depends on nf_defrag_ipv6,
nf_defrag_ipv6 will load first. Since nf_defrag_ipv6 loads first its
netns exit hooks run _after_ conntrack's netns exit hook.
Flush all fragment queue SKBs during fqdir_pre_exit() to release
conntrack references before conntrack cleanup runs. Also flush
the queues in timer expiry handlers when they discover fqdir->dead
is set, in case packet sneaks in while we're running the pre_exit
flush.
The commit under Fixes is not exactly the culprit, but I think
previously the timer firing would eventually unblock the spinning
conntrack.
Fixes: d5dd88794a ("inet: fix various use-after-free in defrags units")
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251207010942.1672972-4-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In the following patch, we no longer assume inet_frag_kill()
callers own a reference.
Consuming two refcounts from inet_frag_kill() would lead in UAF.
Propagate the pointer to the refs that will be consumed later
by the final inet_frag_putn() call.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250312082250.1803501-4-edumazet@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
mono_delivery_time was added to check if skb->tstamp has delivery
time in mono clock base (i.e. EDT) otherwise skb->tstamp has
timestamp in ingress and delivery_time at egress.
Renaming the bitfield from mono_delivery_time to tstamp_type is for
extensibilty for other timestamps such as userspace timestamp
(i.e. SO_TXTIME) set via sock opts.
As we are renaming the mono_delivery_time to tstamp_type, it makes
sense to start assigning tstamp_type based on enum defined
in this commit.
Earlier we used bool arg flag to check if the tstamp is mono in
function skb_set_delivery_time, Now the signature of the functions
accepts tstamp_type to distinguish between mono and real time.
Also skb_set_delivery_type_by_clockid is a new function which accepts
clockid to determine the tstamp_type.
In future tstamp_type:1 can be extended to support userspace timestamp
by increasing the bitfield.
Signed-off-by: Abhishek Chauhan <quic_abchauha@quicinc.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Martin KaFai Lau <martin.lau@kernel.org>
Link: https://lore.kernel.org/r/20240509211834.3235191-2-quic_abchauha@quicinc.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
This will, after the next patch, hold only the core
drop reasons and minimal infrastructure. Fix a small
kernel-doc issue while at it, to avoid the move
triggering a checker.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Used to track skbs freed after a timeout happened
in a reassmbly unit.
Passing a @reason argument to inet_frag_rbtree_purge()
allows to use correct consumed status for frags
that have been successfully re-assembled.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch adds missing includes to headers under include/net.
All these problems are currently masked by the existing users
including the missing dependency before the broken header.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
A latter patch will postpone the delivery_time clearing until the stack
knows the skb is being delivered locally. That will allow other kernel
forwarding path (e.g. ip[6]_forward) to keep the delivery_time also.
An earlier attempt was to do skb_clear_delivery_time() in
ip_local_deliver() and ip6_input(). The discussion [0] requested
to move it one step later into ip_local_deliver_finish()
and ip6_input_finish() so that the delivery_time can be kept
for the ip_vs forwarding path also.
To do that, this patch also needs to take care of the (rcv) timestamp
usecase in ip_is_fragment(). It needs to expect delivery_time in
the skb->tstamp, so it needs to save the mono_delivery_time bit in
inet_frag_queue such that the delivery_time (if any) can be restored
in the final defragmented skb.
[Note that it will only happen when the locally generated skb is looping
from egress to ingress over a virtual interface (e.g. veth, loopback...),
skb->tstamp may have the delivery time before it is known that it will
be delivered locally and received by another sk.]
[0]: https://lore.kernel.org/netdev/ca728d81-80e8-3767-d5e-d44f6ad96e43@ssi.bg/
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Both fields can be read/written without synchronization,
add proper accessors and documentation.
Fixes: d5dd88794a ("inet: fix various use-after-free in defrags units")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On a few of our systems, I found frequent 'unshare(CLONE_NEWNET)' calls
make the number of active slab objects including 'sock_inode_cache' type
rapidly and continuously increase. As a result, memory pressure occurs.
In more detail, I made an artificial reproducer that resembles the
workload that we found the problem and reproduce the problem faster. It
merely repeats 'unshare(CLONE_NEWNET)' 50,000 times in a loop. It takes
about 2 minutes. On 40 CPU cores / 70GB DRAM machine, the available
memory continuously reduced in a fast speed (about 120MB per second,
15GB in total within the 2 minutes). Note that the issue don't
reproduce on every machine. On my 6 CPU cores machine, the problem
didn't reproduce.
'cleanup_net()' and 'fqdir_work_fn()' are functions that deallocate the
relevant memory objects. They are asynchronously invoked by the work
queues and internally use 'rcu_barrier()' to ensure safe destructions.
'cleanup_net()' works in a batched maneer in a single thread worker,
while 'fqdir_work_fn()' works for each 'fqdir_exit()' call in the
'system_wq'. Therefore, 'fqdir_work_fn()' called frequently under the
workload and made the contention for 'rcu_barrier()' high. In more
detail, the global mutex, 'rcu_state.barrier_mutex' became the
bottleneck.
This commit avoids such contention by doing the 'rcu_barrier()' and
subsequent lightweight works in a batched manner, as similar to that of
'cleanup_net()'. The fqdir hashtable destruction, which is done before
the 'rcu_barrier()', is still allowed to run in parallel for fast
processing, but this commit makes it to use a dedicated work queue
instead of the 'system_wq', to make sure that the number of threads is
bounded.
Signed-off-by: SeongJae Park <sjpark@amazon.de>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20201211112405.31158-1-sjpark@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Before commit d4289fcc9b ("net: IP6 defrag: use rbtrees for IPv6
defrag"), a netperf UDP_STREAM test[0] using big IPv6 datagrams (thus
generating many fragments) and running over an IPsec tunnel, reported
more than 6Gbps throughput. After that patch, the same test gets only
9Mbps when receiving on a be2net nic (driver can make a big difference
here, for example, ixgbe doesn't seem to be affected).
By reusing the IPv4 defragmentation code, IPv6 lost fragment coalescing
(IPv4 fragment coalescing was dropped by commit 14fe22e334 ("Revert
"ipv4: use skb coalescing in defragmentation"")).
Without fragment coalescing, be2net runs out of Rx ring entries and
starts to drop frames (ethtool reports rx_drops_no_frags errors). Since
the netperf traffic is only composed of UDP fragments, any lost packet
prevents reassembly of the full datagram. Therefore, fragments which
have no possibility to ever get reassembled pile up in the reassembly
queue, until the memory accounting exeeds the threshold. At that point
no fragment is accepted anymore, which effectively discards all
netperf traffic.
When reassembly timeout expires, some stale fragments are removed from
the reassembly queue, so a few packets can be received, reassembled
and delivered to the netperf receiver. But the nic still drops frames
and soon the reassembly queue gets filled again with stale fragments.
These long time frames where no datagram can be received explain why
the performance drop is so significant.
Re-introducing fragment coalescing is enough to get the initial
performances again (6.6Gbps with be2net): driver doesn't drop frames
anymore (no more rx_drops_no_frags errors) and the reassembly engine
works at full speed.
This patch is quite conservative and only coalesces skbs for local
IPv4 and IPv6 delivery (in order to avoid changing skb geometry when
forwarding). Coalescing could be extended in the future if need be, as
more scenarios would probably benefit from it.
[0]: Test configuration
Sender:
ip xfrm policy flush
ip xfrm state flush
ip xfrm state add src fc00:1::1 dst fc00:2::1 proto esp spi 0x1000 aead 'rfc4106(gcm(aes))' 0x0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b 96 mode transport sel src fc00:1::1 dst fc00:2::1
ip xfrm policy add src fc00:1::1 dst fc00:2::1 dir in tmpl src fc00:1::1 dst fc00:2::1 proto esp mode transport action allow
ip xfrm state add src fc00:2::1 dst fc00:1::1 proto esp spi 0x1001 aead 'rfc4106(gcm(aes))' 0x0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b 96 mode transport sel src fc00:2::1 dst fc00:1::1
ip xfrm policy add src fc00:2::1 dst fc00:1::1 dir out tmpl src fc00:2::1 dst fc00:1::1 proto esp mode transport action allow
netserver -D -L fc00:2::1
Receiver:
ip xfrm policy flush
ip xfrm state flush
ip xfrm state add src fc00:2::1 dst fc00:1::1 proto esp spi 0x1001 aead 'rfc4106(gcm(aes))' 0x0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b 96 mode transport sel src fc00:2::1 dst fc00:1::1
ip xfrm policy add src fc00:2::1 dst fc00:1::1 dir in tmpl src fc00:2::1 dst fc00:1::1 proto esp mode transport action allow
ip xfrm state add src fc00:1::1 dst fc00:2::1 proto esp spi 0x1000 aead 'rfc4106(gcm(aes))' 0x0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b0b 96 mode transport sel src fc00:1::1 dst fc00:2::1
ip xfrm policy add src fc00:1::1 dst fc00:2::1 dir out tmpl src fc00:1::1 dst fc00:2::1 proto esp mode transport action allow
netperf -H fc00:2::1 -f k -P 0 -L fc00:1::1 -l 60 -t UDP_STREAM -I 99,5 -i 5,5 -T5,5 -6
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The linux-next commit "inet: fix various use-after-free in defrags
units" [1] introduced compilation warnings,
./include/net/inet_frag.h:117:1: warning: 'inline' is not at beginning
of declaration [-Wold-style-declaration]
static void inline fqdir_pre_exit(struct fqdir *fqdir)
^~~~~~
In file included from ./include/net/netns/ipv4.h:10,
from ./include/net/net_namespace.h:20,
from ./include/linux/netdevice.h:38,
from ./include/linux/icmpv6.h:13,
from ./include/linux/ipv6.h:86,
from ./include/net/ipv6.h:12,
from ./include/rdma/ib_verbs.h:51,
from ./include/linux/mlx5/device.h:37,
from ./include/linux/mlx5/driver.h:51,
from
drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c:37:
[1] https://lore.kernel.org/netdev/20190618180900.88939-3-edumazet@google.com/
Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: David S. Miller <davem@davemloft.net>
fqdir_init() is not fast path and is getting bigger.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
syszbot found an interesting use-after-free [1] happening
while IPv4 fragment rhashtable was destroyed at netns dismantle.
While no insertions can possibly happen at the time a dismantling
netns is destroying this rhashtable, timers can still fire and
attempt to remove elements from this rhashtable.
This is forbidden, since rhashtable_free_and_destroy() has
no synchronization against concurrent inserts and deletes.
Add a new fqdir->dead flag so that timers do not attempt
a rhashtable_remove_fast() operation.
We also have to respect an RCU grace period before starting
the rhashtable_free_and_destroy() from process context,
thus we use rcu_work infrastructure.
This is a refinement of a prior rough attempt to fix this bug :
https://marc.info/?l=linux-netdev&m=153845936820900&w=2
Since the rhashtable cleanup is now deferred to a work queue,
netns dismantles should be slightly faster.
[1]
BUG: KASAN: use-after-free in __read_once_size include/linux/compiler.h:194 [inline]
BUG: KASAN: use-after-free in rhashtable_last_table+0x162/0x180 lib/rhashtable.c:212
Read of size 8 at addr ffff8880a6497b70 by task kworker/0:0/5
CPU: 0 PID: 5 Comm: kworker/0:0 Not tainted 5.2.0-rc1+ #2
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Workqueue: events rht_deferred_worker
Call Trace:
__dump_stack lib/dump_stack.c:77 [inline]
dump_stack+0x172/0x1f0 lib/dump_stack.c:113
print_address_description.cold+0x7c/0x20d mm/kasan/report.c:188
__kasan_report.cold+0x1b/0x40 mm/kasan/report.c:317
kasan_report+0x12/0x20 mm/kasan/common.c:614
__asan_report_load8_noabort+0x14/0x20 mm/kasan/generic_report.c:132
__read_once_size include/linux/compiler.h:194 [inline]
rhashtable_last_table+0x162/0x180 lib/rhashtable.c:212
rht_deferred_worker+0x111/0x2030 lib/rhashtable.c:411
process_one_work+0x989/0x1790 kernel/workqueue.c:2269
worker_thread+0x98/0xe40 kernel/workqueue.c:2415
kthread+0x354/0x420 kernel/kthread.c:255
ret_from_fork+0x24/0x30 arch/x86/entry/entry_64.S:352
Allocated by task 32687:
save_stack+0x23/0x90 mm/kasan/common.c:71
set_track mm/kasan/common.c:79 [inline]
__kasan_kmalloc mm/kasan/common.c:489 [inline]
__kasan_kmalloc.constprop.0+0xcf/0xe0 mm/kasan/common.c:462
kasan_kmalloc+0x9/0x10 mm/kasan/common.c:503
__do_kmalloc_node mm/slab.c:3620 [inline]
__kmalloc_node+0x4e/0x70 mm/slab.c:3627
kmalloc_node include/linux/slab.h:590 [inline]
kvmalloc_node+0x68/0x100 mm/util.c:431
kvmalloc include/linux/mm.h:637 [inline]
kvzalloc include/linux/mm.h:645 [inline]
bucket_table_alloc+0x90/0x480 lib/rhashtable.c:178
rhashtable_init+0x3f4/0x7b0 lib/rhashtable.c:1057
inet_frags_init_net include/net/inet_frag.h:109 [inline]
ipv4_frags_init_net+0x182/0x410 net/ipv4/ip_fragment.c:683
ops_init+0xb3/0x410 net/core/net_namespace.c:130
setup_net+0x2d3/0x740 net/core/net_namespace.c:316
copy_net_ns+0x1df/0x340 net/core/net_namespace.c:439
create_new_namespaces+0x400/0x7b0 kernel/nsproxy.c:107
unshare_nsproxy_namespaces+0xc2/0x200 kernel/nsproxy.c:206
ksys_unshare+0x440/0x980 kernel/fork.c:2692
__do_sys_unshare kernel/fork.c:2760 [inline]
__se_sys_unshare kernel/fork.c:2758 [inline]
__x64_sys_unshare+0x31/0x40 kernel/fork.c:2758
do_syscall_64+0xfd/0x680 arch/x86/entry/common.c:301
entry_SYSCALL_64_after_hwframe+0x49/0xbe
Freed by task 7:
save_stack+0x23/0x90 mm/kasan/common.c:71
set_track mm/kasan/common.c:79 [inline]
__kasan_slab_free+0x102/0x150 mm/kasan/common.c:451
kasan_slab_free+0xe/0x10 mm/kasan/common.c:459
__cache_free mm/slab.c:3432 [inline]
kfree+0xcf/0x220 mm/slab.c:3755
kvfree+0x61/0x70 mm/util.c:460
bucket_table_free+0x69/0x150 lib/rhashtable.c:108
rhashtable_free_and_destroy+0x165/0x8b0 lib/rhashtable.c:1155
inet_frags_exit_net+0x3d/0x50 net/ipv4/inet_fragment.c:152
ipv4_frags_exit_net+0x73/0x90 net/ipv4/ip_fragment.c:695
ops_exit_list.isra.0+0xaa/0x150 net/core/net_namespace.c:154
cleanup_net+0x3fb/0x960 net/core/net_namespace.c:553
process_one_work+0x989/0x1790 kernel/workqueue.c:2269
worker_thread+0x98/0xe40 kernel/workqueue.c:2415
kthread+0x354/0x420 kernel/kthread.c:255
ret_from_fork+0x24/0x30 arch/x86/entry/entry_64.S:352
The buggy address belongs to the object at ffff8880a6497b40
which belongs to the cache kmalloc-1k of size 1024
The buggy address is located 48 bytes inside of
1024-byte region [ffff8880a6497b40, ffff8880a6497f40)
The buggy address belongs to the page:
page:ffffea0002992580 refcount:1 mapcount:0 mapping:ffff8880aa400ac0 index:0xffff8880a64964c0 compound_mapcount: 0
flags: 0x1fffc0000010200(slab|head)
raw: 01fffc0000010200 ffffea0002916e88 ffffea000218fe08 ffff8880aa400ac0
raw: ffff8880a64964c0 ffff8880a6496040 0000000100000005 0000000000000000
page dumped because: kasan: bad access detected
Memory state around the buggy address:
ffff8880a6497a00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
ffff8880a6497a80: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
>ffff8880a6497b00: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
^
ffff8880a6497b80: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
ffff8880a6497c00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
Fixes: 648700f76b ("inet: frags: use rhashtables for reassembly units")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Following patch will add rcu grace period before fqdir
rhashtable destruction, so we need to dynamically allocate
fqdir structures to not force expensive synchronize_rcu() calls
in netns dismantle path.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
fqdir will soon be dynamically allocated.
We need to reach the struct net pointer from fqdir,
so add it, and replace the various container_of() constructs
by direct access to the new field.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
And pass an extra parameter, since we will soon
dynamically allocate fqdir structures.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
1) struct netns_frags is renamed to struct fqdir
This structure is really holding many frag queues in a hash table.
2) (struct inet_frag_queue)->net field is renamed to fqdir
since net is generally associated to a 'struct net' pointer
in networking stack.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that all users of struct inet_frag_queue have been converted
to use 'rb_fragments', remove the unused 'fragments' field.
Build with `make allyesconfig` succeeded. ip_defrag selftest passed.
Signed-off-by: Peter Oskolkov <posk@google.com>
Acked-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is a refactoring patch: without changing runtime behavior,
it moves rbtree-related code from IPv4-specific files/functions
into .h/.c defrag files shared with IPv6 defragmentation code.
Signed-off-by: Peter Oskolkov <posk@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Florian Westphal <fw@strlen.de>
Cc: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jakub Kicinski
Eric Dumazet
Abhishek Chauhan
Randy Dunlap
Johannes Berg
Martin KaFai Lau
SeongJae Park
Guillaume Nault
Qian Cai
Peter Oskolkov