net: prevent skb corruption on frag list segmentation
authorPaolo Abeni <pabeni@redhat.com>
Fri, 7 Jul 2023 08:11:10 +0000 (10:11 +0200)
committerDavid S. Miller <davem@davemloft.net>
Sat, 8 Jul 2023 09:03:26 +0000 (10:03 +0100)
Ian reported several skb corruptions triggered by rx-gro-list,
collecting different oops alike:

[   62.624003] BUG: kernel NULL pointer dereference, address: 00000000000000c0
[   62.631083] #PF: supervisor read access in kernel mode
[   62.636312] #PF: error_code(0x0000) - not-present page
[   62.641541] PGD 0 P4D 0
[   62.644174] Oops: 0000 [#1] PREEMPT SMP NOPTI
[   62.648629] CPU: 1 PID: 913 Comm: napi/eno2-79 Not tainted 6.4.0 #364
[   62.655162] Hardware name: Supermicro Super Server/A2SDi-12C-HLN4F, BIOS 1.7a 10/13/2022
[   62.663344] RIP: 0010:__udp_gso_segment (./include/linux/skbuff.h:2858
./include/linux/udp.h:23 net/ipv4/udp_offload.c:228 net/ipv4/udp_offload.c:261
net/ipv4/udp_offload.c:277)
[   62.687193] RSP: 0018:ffffbd3a83b4f868 EFLAGS: 00010246
[   62.692515] RAX: 00000000000000ce RBX: 0000000000000000 RCX: 0000000000000000
[   62.699743] RDX: ffffa124def8a000 RSI: 0000000000000079 RDI: ffffa125952a14d4
[   62.706970] RBP: ffffa124def8a000 R08: 0000000000000022 R09: 00002000001558c9
[   62.714199] R10: 0000000000000000 R11: 00000000be554639 R12: 00000000000000e2
[   62.721426] R13: ffffa125952a1400 R14: ffffa125952a1400 R15: 00002000001558c9
[   62.728654] FS:  0000000000000000(0000) GS:ffffa127efa40000(0000)
knlGS:0000000000000000
[   62.736852] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   62.742702] CR2: 00000000000000c0 CR3: 00000001034b0000 CR4: 00000000003526e0
[   62.749948] Call Trace:
[   62.752498]  <TASK>
[   62.779267] inet_gso_segment (net/ipv4/af_inet.c:1398)
[   62.787605] skb_mac_gso_segment (net/core/gro.c:141)
[   62.791906] __skb_gso_segment (net/core/dev.c:3403 (discriminator 2))
[   62.800492] validate_xmit_skb (./include/linux/netdevice.h:4862
net/core/dev.c:3659)
[   62.804695] validate_xmit_skb_list (net/core/dev.c:3710)
[   62.809158] sch_direct_xmit (net/sched/sch_generic.c:330)
[   62.813198] __dev_queue_xmit (net/core/dev.c:3805 net/core/dev.c:4210)
net/netfilter/core.c:626)
[   62.821093] br_dev_queue_push_xmit (net/bridge/br_forward.c:55)
[   62.825652] maybe_deliver (net/bridge/br_forward.c:193)
[   62.829420] br_flood (net/bridge/br_forward.c:233)
[   62.832758] br_handle_frame_finish (net/bridge/br_input.c:215)
[   62.837403] br_handle_frame (net/bridge/br_input.c:298
net/bridge/br_input.c:416)
[   62.851417] __netif_receive_skb_core.constprop.0 (net/core/dev.c:5387)
[   62.866114] __netif_receive_skb_list_core (net/core/dev.c:5570)
[   62.871367] netif_receive_skb_list_internal (net/core/dev.c:5638
net/core/dev.c:5727)
[   62.876795] napi_complete_done (./include/linux/list.h:37
./include/net/gro.h:434 ./include/net/gro.h:429 net/core/dev.c:6067)
[   62.881004] ixgbe_poll (drivers/net/ethernet/intel/ixgbe/ixgbe_main.c:3191)
[   62.893534] __napi_poll (net/core/dev.c:6498)
[   62.897133] napi_threaded_poll (./include/linux/netpoll.h:89
net/core/dev.c:6640)
[   62.905276] kthread (kernel/kthread.c:379)
[   62.913435] ret_from_fork (arch/x86/entry/entry_64.S:314)
[   62.917119]  </TASK>

In the critical scenario, rx-gro-list GRO-ed packets are fed, via a
bridge, both to the local input path and to an egress device (tun).

The segmentation of such packets unsafely writes to the cloned skbs
with shared heads.

This change addresses the issue by uncloning as needed the
to-be-segmented skbs.

Reported-by: Ian Kumlien <ian.kumlien@gmail.com>
Tested-by: Ian Kumlien <ian.kumlien@gmail.com>
Fixes: 3a1296a38d0c ("net: Support GRO/GSO fraglist chaining.")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/core/skbuff.c

index 6c5915e..a298992 100644 (file)
@@ -4261,6 +4261,11 @@ struct sk_buff *skb_segment_list(struct sk_buff *skb,
 
        skb_push(skb, -skb_network_offset(skb) + offset);
 
+       /* Ensure the head is writeable before touching the shared info */
+       err = skb_unclone(skb, GFP_ATOMIC);
+       if (err)
+               goto err_linearize;
+
        skb_shinfo(skb)->frag_list = NULL;
 
        while (list_skb) {