net: drop bad gso csum_start and offset in virtio_net_hdr
Subsystems (6)
PATCH v2 1/1 net: drop bad gso csum_start and offset in virtio_net_hdr
From: Willem de Bruijn <[email protected]>
Tighten csum_start and csum_offset checks in virtio_net_hdr_to_skb for GSO packets.
The function already checks that a checksum requested with VIRTIO_NET_HDR_F_NEEDS_CSUM is in skb linear. But for GSO packets this might not hold for segs after segmentation.
Syzkaller demonstrated to reach this warning in skb_checksum_help
offset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))
By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]
The geometry of the bad input packet at tcp_gso_segment:
[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536
ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value from gso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks to SKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.
GSO packet are almost always CHECKSUM_PARTIAL. USO packets may be CHECKSUM_NONE since commit 10154dbded6d6 ("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test for ip_summed == CHECKSUM_PARTIAL first.
This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
---
v1->v2
- skb_transport_header instead of skb->transport_header (edumazet@)
- typo: migitate -> mitigate
--- include/linux/virtio_net.h | 16 +++++----------- net/ipv4/tcp_offload.c | 3 +++ net/ipv4/udp_offload.c | 4 ++++ 3 files changed, 12 insertions(+), 11 deletions(-)
include/linux/virtio_net.h +5 -11
@@ -56,7 +56,6 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb,unsigned int thlen =0 ;unsigned int p_off =0 ;unsigned int ip_proto; -u64 ret, remainder, gso_size;if (hdr->gso_type !=VIRTIO_NET_HDR_GSO_NONE ) {switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN ) {
@@ -99,16 +98,6 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb,u32 off =__virtio16_to_cpu (little_endian, hdr->csum_offset);u32 needed = start +max_t (u32, thlen, off +sizeof (__sum16)); -if (hdr->gso_size) { - gso_size =__virtio16_to_cpu (little_endian, hdr->gso_size); - ret =div64_u64_rem (skb->len, gso_size, &remainder); -if (!(ret && (hdr->gso_size > needed) && - ((remainder > needed) || (remainder ==0 )))) { -return -EINVAL ; - } -skb_shinfo (skb)->tx_flags |=SKBFL_SHARED_FRAG ; - } -if (!pskb_may_pull (skb, needed))return -EINVAL ;
@@ -182,6 +171,11 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb,if (gso_type !=SKB_GSO_UDP_L4 )return -EINVAL ;break ; +case SKB_GSO_TCPV4 : +case SKB_GSO_TCPV6 : +if (skb->csum_offset != offsetof(struct tcphdr , check)) +return -EINVAL ; +break ; }/* Kernel has a special handling for GSO_BY_FRAGS. */
net/ipv4/tcp_offload.c +3 -0
@@ -140,6 +140,9 @@ struct sk_buff *tcp_gso_segment(struct sk_buff *skb,if (thlen <sizeof (*th)) goto out; +if (unlikely (skb_checksum_start (skb) !=skb_transport_header (skb))) + goto out; +if (!pskb_may_pull (skb, thlen)) goto out;
net/ipv4/udp_offload.c +4 -0
@@ -278,6 +278,10 @@ struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,if (gso_skb->len <=sizeof (*uh) + mss)return ERR_PTR (-EINVAL ); +if (unlikely (skb_checksum_start (gso_skb) != +skb_transport_header (gso_skb))) +return ERR_PTR (-EINVAL ); +if (skb_gso_ok (gso_skb, features |NETIF_F_GSO_ROBUST )) {/* Packet is from an untrusted source, reset gso_segs. */ skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
Signature · 1 line
2.46.0.rc1.232.g9752f9e123-goog
This patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:
Quoted 10 lines · On Mon, 29 Jul 2024 16:10:12 -0400 you wrote:
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.[...]
Here is the summary with links:
- [net,v2] net: drop bad gso csum_start and offset in virtio_net_hdr
https://git.kernel.org/netdev/net/c/89add40066f9
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>
Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---
this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit 30b03f2a0592eee1267298298eac9dd655f55ab2
Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700
udp: Fall back to software USO if IPv6 extension headers are present
and
commit b128ed5ab27330deeeaf51ea8bb69f1442a96f7f
Author: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700
udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see
...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying
...
the "bad gso" message keeps repeating and init is not executed.
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
virtio_is_little_endian(vi->vdev))) {
net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
dev->name, hdr->hdr.gso_type,
hdr->hdr.gso_size);
goto frame_err;
}Type 1 is VIRTIO_NET_HDR_GSO_TCPV4
Most likely this application is inserting a packet with flag VIRTIO_NET_HDR_F_NEEDS_CSUM and a wrong csum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other virtio_net_hdr fields and the packet length.
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.
Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.
This happens on the virtio-net receive path, so the bad data is received from the hypervisor.
From what I gather that arm64 fvp (Fixed Virtual Platforms) hypervisor is closed source?
Disabling GRO on this device will likely work as temporary workaround.
What we can do is instead of dropping packets to correct their offset:
case SKB_GSO_TCPV4:
case SKB_GSO_TCPV6:
- if (skb->csum_offset != offsetof(struct tcphdr, check))
- return -EINVAL;
+ if (skb->csum_offset != offsetof(struct tcphdr, check)) {
+ DEBUG_NET_WARN_ON_ONCE(1);
+ skb->csum_offset = offsetof(struct tcphdr, check);
+ }
If the issue is in csum_offset. If the new csum_start check fails, that won't help.
It would be helpful to see these values at this point, e.g., with skb_dump(KERN_INFO, skb, false);
Michael S. Tsirkin
maintainerQuoted 181 lines · Willem de Bruijn
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.This happens on the virtio-net receive path, so the bad data is received from the hypervisor.
Quoted 1 line
From what I gather that arm64 fvp (Fixed Virtual Platforms) hypervisor
is closed source?
Disabling GRO on this device will likely work as temporary workaround.
What we can do is instead of dropping packets to correct their offset:
case
SKB_GSO_TCPV4:
caseSKB_GSO_TCPV6:(quoted diff) +4
-2- if (skb->csum_offset != offsetof(struct tcphdr, check)) - return -EINVAL; + if (skb->csum_offset != offsetof(struct tcphdr, check)) { + DEBUG_NET_WARN_ON_ONCE(1); + skb->csum_offset = offsetof(struct tcphdr, check); + }
Quoted 5 lines
If the issue is in
csum_offset. If the newcsum_startcheck fails, that won't help.It would be helpful to see these values at this point, e.g., with
skb_dump(KERN_INFO, skb, false);
It's an iteresting question whether when VIRTIO_NET_F_GUEST_HDRLEN is not negotiated, csum_offset can be relied upon for GRO.
Jason, WDYT?
Signature · 1 line
MST
Quoted 186 lines · Michael S. Tsirkin
Quoted 184 lines · Willem de Bruijn
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.This happens on the virtio-net receive path, so the bad data is received from the hypervisor.
Quoted 1 line
From what I gather that arm64 fvp (Fixed Virtual Platforms) hypervisor
is closed source?
Disabling GRO on this device will likely work as temporary workaround.
What we can do is instead of dropping packets to correct their offset:
case
SKB_GSO_TCPV4:
caseSKB_GSO_TCPV6:(quoted diff) +4
-2- if (skb->csum_offset != offsetof(struct tcphdr, check)) - return -EINVAL; + if (skb->csum_offset != offsetof(struct tcphdr, check)) { + DEBUG_NET_WARN_ON_ONCE(1); + skb->csum_offset = offsetof(struct tcphdr, check); + }If the issue is in
csum_offset. If the newcsum_startcheck fails, that won't help.
Quoted 8 lines
Quoted 2 lines
It would be helpful to see these values at this point, e.g., with
skb_dump(KERN_INFO, skb, false);It's an iteresting question whether when
VIRTIO_NET_F_GUEST_HDRLENis not negotiated,csum_offsetcan be relied upon for GRO.Jason, WDYT?
I don't see how it connects. GUEST_HDRLEN is about transmission but not for receiving, current Linux driver doesn't use hdrlen at all so hardened csum_offset looks like a must.
And we have
"""
If one of the VIRTIO_NET_F_GUEST_TSO4, TSO6, UFO, USO4 or USO6 options have been negotiated, the driver MAY use hdr_len only as a hint about the transport header size. The driver MUST NOT rely on hdr_len to be correct. Note: This is due to various bugs in implementations. """
Thanks
Quoted 4 lines
--
MST
Quoted 200 lines · Jason Wang
Quoted 186 lines · Michael S. Tsirkin
Quoted 184 lines · Willem de Bruijn
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.This happens on the virtio-net receive path, so the bad data is received from the hypervisor.
Quoted 1 line
From what I gather that arm64 fvp (Fixed Virtual Platforms) hypervisor
is closed source?
Disabling GRO on this device will likely work as temporary workaround.
What we can do is instead of dropping packets to correct their offset:
case
SKB_GSO_TCPV4:
caseSKB_GSO_TCPV6:(quoted diff) +4
-2- if (skb->csum_offset != offsetof(struct tcphdr, check)) - return -EINVAL; + if (skb->csum_offset != offsetof(struct tcphdr, check)) { + DEBUG_NET_WARN_ON_ONCE(1); + skb->csum_offset = offsetof(struct tcphdr, check); + }If the issue is in
csum_offset. If the newcsum_startcheck fails, that won't help.Quoted 8 lines
Quoted 2 lines
It would be helpful to see these values at this point, e.g., with
skb_dump(KERN_INFO, skb, false);It's an iteresting question whether when
VIRTIO_NET_F_GUEST_HDRLENis not negotiated,csum_offsetcan be relied upon for GRO.Jason, WDYT?
I don't see how it connects.
GUEST_HDRLENis about transmission but not for receiving, current Linux driver doesn't use hdrlen at all so hardenedcsum_offsetlooks like a must.
Quoted 8 lines
And we have
"""
If one of theVIRTIO_NET_F_GUEST_TSO4, TSO6, UFO, USO4 or USO6 options have been negotiated, the driver MAY usehdr_lenonly as a hint about the transport header size. The driver MUST NOT rely onhdr_lento be correct. Note: This is due to various bugs in implementations. """
I think I made a mistake in assuming that virtio_net_hdr_to_skb is only used in the tx path, and that the GSO flags thus imply GSO, which requires CHECKSUM_PARTIAL.
In virtnet_receive, this is the rx path and those flags imply GRO. That can use CHECKSUM_UNNECESSARY, as virtnet does:
if (flags & VIRTIO_NET_HDR_F_DATA_VALID)
skb->ip_summed = CHECKSUM_UNNECESSARY;
So I guess VIRTIO_NET_HDR_GSO_* without VIRTIO_NET_HDR_F_DATA_VALID would be wrong on rx.
But the new check
if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
[...]
case SKB_GSO_TCPV4:
case SKB_GSO_TCPV6:
if (skb->csum_offset != offsetof(struct tcphdr, check))
return -EINVAL;should be limited to callers of virtio_net_hdr_to_skb on the tx/GSO path.
Looking what the cleanest/minimal patch is to accomplish that.
Quoted 235 lines · Willem de Bruijn
Quoted 200 lines · Jason Wang
Quoted 186 lines · Michael S. Tsirkin
Quoted 184 lines · Willem de Bruijn
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.This happens on the virtio-net receive path, so the bad data is received from the hypervisor.
Quoted 1 line
From what I gather that arm64 fvp (Fixed Virtual Platforms) hypervisor
is closed source?
Disabling GRO on this device will likely work as temporary workaround.
What we can do is instead of dropping packets to correct their offset:
case
SKB_GSO_TCPV4:
caseSKB_GSO_TCPV6:(quoted diff) +4
-2- if (skb->csum_offset != offsetof(struct tcphdr, check)) - return -EINVAL; + if (skb->csum_offset != offsetof(struct tcphdr, check)) { + DEBUG_NET_WARN_ON_ONCE(1); + skb->csum_offset = offsetof(struct tcphdr, check); + }If the issue is in
csum_offset. If the newcsum_startcheck fails, that won't help.Quoted 8 lines
Quoted 2 lines
It would be helpful to see these values at this point, e.g., with
skb_dump(KERN_INFO, skb, false);It's an iteresting question whether when
VIRTIO_NET_F_GUEST_HDRLENis not negotiated,csum_offsetcan be relied upon for GRO.Jason, WDYT?
I don't see how it connects.
GUEST_HDRLENis about transmission but not for receiving, current Linux driver doesn't use hdrlen at all so hardenedcsum_offsetlooks like a must.Quoted 8 lines
And we have
"""
If one of theVIRTIO_NET_F_GUEST_TSO4, TSO6, UFO, USO4 or USO6 options have been negotiated, the driver MAY usehdr_lenonly as a hint about the transport header size. The driver MUST NOT rely onhdr_lento be correct. Note: This is due to various bugs in implementations. """I think I made a mistake in assuming that
virtio_net_hdr_to_skbis only used in the tx path, and that the GSO flags thus imply GSO, which requiresCHECKSUM_PARTIAL.In
virtnet_receive, this is the rx path and those flags imply GRO. That can useCHECKSUM_UNNECESSARY, as virtnet does:if (flags &
VIRTIO_NET_HDR_F_DATA_VALID)
skb->ip_summed=CHECKSUM_UNNECESSARY;So I guess VIRTIO_NET_HDR_GSO_* without
VIRTIO_NET_HDR_F_DATA_VALIDwould be wrong on rx.But the new check
if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { [...] case SKB_GSO_TCPV4: case SKB_GSO_TCPV6: if (skb->csum_offset != offsetof(struct tcphdr, check)) return -EINVAL;
Quoted 4 lines
should be limited to callers of
virtio_net_hdr_to_skbon the tx/GSO path.Looking what the cleanest/minimal patch is to accomplish that.
virtio_net_hdr_to_skb() translates virtio-net header to skb metadata, so it's RX. For TX the helper should be virtio_net_hdr_from_skb() which translates skb metadata to virtio hdr.
Thanks
Quoted 237 lines · Jason Wang
Quoted 235 lines · Willem de Bruijn
Quoted 200 lines · Jason Wang
Quoted 186 lines · Michael S. Tsirkin
Quoted 184 lines · Willem de Bruijn
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.This happens on the virtio-net receive path, so the bad data is received from the hypervisor.
Quoted 1 line
From what I gather that arm64 fvp (Fixed Virtual Platforms) hypervisor
is closed source?
Disabling GRO on this device will likely work as temporary workaround.
What we can do is instead of dropping packets to correct their offset:
case
SKB_GSO_TCPV4:
caseSKB_GSO_TCPV6:(quoted diff) +4
-2- if (skb->csum_offset != offsetof(struct tcphdr, check)) - return -EINVAL; + if (skb->csum_offset != offsetof(struct tcphdr, check)) { + DEBUG_NET_WARN_ON_ONCE(1); + skb->csum_offset = offsetof(struct tcphdr, check); + }If the issue is in
csum_offset. If the newcsum_startcheck fails, that won't help.Quoted 8 lines
Quoted 2 lines
It would be helpful to see these values at this point, e.g., with
skb_dump(KERN_INFO, skb, false);It's an iteresting question whether when
VIRTIO_NET_F_GUEST_HDRLENis not negotiated,csum_offsetcan be relied upon for GRO.Jason, WDYT?
I don't see how it connects.
GUEST_HDRLENis about transmission but not for receiving, current Linux driver doesn't use hdrlen at all so hardenedcsum_offsetlooks like a must.Quoted 8 lines
And we have
"""
If one of theVIRTIO_NET_F_GUEST_TSO4, TSO6, UFO, USO4 or USO6 options have been negotiated, the driver MAY usehdr_lenonly as a hint about the transport header size. The driver MUST NOT rely onhdr_lento be correct. Note: This is due to various bugs in implementations. """I think I made a mistake in assuming that
virtio_net_hdr_to_skbis only used in the tx path, and that the GSO flags thus imply GSO, which requiresCHECKSUM_PARTIAL.In
virtnet_receive, this is the rx path and those flags imply GRO. That can useCHECKSUM_UNNECESSARY, as virtnet does:if (flags &
VIRTIO_NET_HDR_F_DATA_VALID)
skb->ip_summed=CHECKSUM_UNNECESSARY;So I guess VIRTIO_NET_HDR_GSO_* without
VIRTIO_NET_HDR_F_DATA_VALIDwould be wrong on rx.But the new check
if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { [...] case SKB_GSO_TCPV4: case SKB_GSO_TCPV6: if (skb->csum_offset != offsetof(struct tcphdr, check)) return -EINVAL;
Quoted 8 lines
Quoted 4 lines
should be limited to callers of
virtio_net_hdr_to_skbon the tx/GSO path.Looking what the cleanest/minimal patch is to accomplish that.
virtio_net_hdr_to_skb()translates virtio-net header to skb metadata, so it's RX. For TX the helper should bevirtio_net_hdr_from_skb()which translates skb metadata to virtio hdr.
virtio_net_hdr_to_skb is used by PF_PACKET, tun and tap when injecting a packet into the egress path.
Quoted 241 lines · Willem de Bruijn
Quoted 239 lines · Jason Wang
Quoted 237 lines · Willem de Bruijn
Quoted 200 lines · Jason Wang
Quoted 186 lines · Michael S. Tsirkin
Quoted 184 lines · Willem de Bruijn
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.This happens on the virtio-net receive path, so the bad data is received from the hypervisor.
Quoted 1 line
From what I gather that arm64 fvp (Fixed Virtual Platforms) hypervisor
is closed source?
Disabling GRO on this device will likely work as temporary workaround.
What we can do is instead of dropping packets to correct their offset:
case
SKB_GSO_TCPV4:
caseSKB_GSO_TCPV6:(quoted diff) +4
-2- if (skb->csum_offset != offsetof(struct tcphdr, check)) - return -EINVAL; + if (skb->csum_offset != offsetof(struct tcphdr, check)) { + DEBUG_NET_WARN_ON_ONCE(1); + skb->csum_offset = offsetof(struct tcphdr, check); + }If the issue is in
csum_offset. If the newcsum_startcheck fails, that won't help.Quoted 8 lines
Quoted 2 lines
It would be helpful to see these values at this point, e.g., with
skb_dump(KERN_INFO, skb, false);It's an iteresting question whether when
VIRTIO_NET_F_GUEST_HDRLENis not negotiated,csum_offsetcan be relied upon for GRO.Jason, WDYT?
I don't see how it connects.
GUEST_HDRLENis about transmission but not for receiving, current Linux driver doesn't use hdrlen at all so hardenedcsum_offsetlooks like a must.Quoted 8 lines
And we have
"""
If one of theVIRTIO_NET_F_GUEST_TSO4, TSO6, UFO, USO4 or USO6 options have been negotiated, the driver MAY usehdr_lenonly as a hint about the transport header size. The driver MUST NOT rely onhdr_lento be correct. Note: This is due to various bugs in implementations. """I think I made a mistake in assuming that
virtio_net_hdr_to_skbis only used in the tx path, and that the GSO flags thus imply GSO, which requiresCHECKSUM_PARTIAL.In
virtnet_receive, this is the rx path and those flags imply GRO. That can useCHECKSUM_UNNECESSARY, as virtnet does:if (flags &
VIRTIO_NET_HDR_F_DATA_VALID)
skb->ip_summed=CHECKSUM_UNNECESSARY;So I guess VIRTIO_NET_HDR_GSO_* without
VIRTIO_NET_HDR_F_DATA_VALIDwould be wrong on rx.But the new check
if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { [...] case SKB_GSO_TCPV4: case SKB_GSO_TCPV6: if (skb->csum_offset != offsetof(struct tcphdr, check)) return -EINVAL;should be limited to callers of
virtio_net_hdr_to_skbon the tx/GSO path.
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
Looking what the cleanest/minimal patch is to accomplish that.
virtio_net_hdr_to_skb()translates virtio-net header to skb metadata, so it's RX. For TX the helper should bevirtio_net_hdr_from_skb()which translates skb metadata to virtio hdr.
virtio_net_hdr_to_skbis used byPF_PACKET, tun and tap
Exactly.
Quoted 1 line
when injecting a packet into the egress path.
For tuntap it's still the RX path. For PF_PACEKT and macvtap, it's the tx.
Maybe a new parameter to virtio_net_hdr_to_skb()?
Thanks
Quoted 1 line
Quoted 25 lines
Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
So I guess VIRTIO_NET_HDR_GSO_* without
VIRTIO_NET_HDR_F_DATA_VALIDwould be wrong on rx.But the new check
if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { [...] case SKB_GSO_TCPV4: case SKB_GSO_TCPV6: if (skb->csum_offset != offsetof(struct tcphdr, check)) return -EINVAL;should be limited to callers of
virtio_net_hdr_to_skbon the tx/GSO path.Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
Looking what the cleanest/minimal patch is to accomplish that.
virtio_net_hdr_to_skb()translates virtio-net header to skb metadata, so it's RX. For TX the helper should bevirtio_net_hdr_from_skb()which translates skb metadata to virtio hdr.
virtio_net_hdr_to_skbis used byPF_PACKET, tun and tap
Quoted 7 lines
Exactly.
Quoted 1 line
when injecting a packet into the egress path.
For tuntap it's still the RX path. For
PF_PACEKTand macvtap, it's the tx.Maybe a new parameter to
virtio_net_hdr_to_skb()?
This is the most straightforward approach. But requires changse to all callers, in a patch targeting all the stable branches.
I'd prefer if we can detect ingress vs egress directly.
Based on ip_summed, pkt_type, is_skb_wmem or so. But so far have not found a suitable condition.
I noticed something else: as you point out TUN is ingress. Unlike virtnet_receive, it does not set ip_summed to CHECKSUM_UNNECESSARY if VIRTIO_NET_HDR_F_DATA_VALID. It probably should. GRO expects packets to have had their integrity verified. CHECKSUM_NONE on ingress is not correct for GRO.
And also related: no GRO should be generated by a device unless VIRTIO_NET_HDR_F_DATA_VALID is also passed? I have to check the spec if it says anything about this.
Quoted 29 lines · Willem de Bruijn
Quoted 29 lines
Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
So I guess VIRTIO_NET_HDR_GSO_* without
VIRTIO_NET_HDR_F_DATA_VALIDwould be wrong on rx.But the new check
if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { [...] case SKB_GSO_TCPV4: case SKB_GSO_TCPV6: if (skb->csum_offset != offsetof(struct tcphdr, check)) return -EINVAL;should be limited to callers of
virtio_net_hdr_to_skbon the tx/GSO path.Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
Looking what the cleanest/minimal patch is to accomplish that.
virtio_net_hdr_to_skb()translates virtio-net header to skb metadata, so it's RX. For TX the helper should bevirtio_net_hdr_from_skb()which translates skb metadata to virtio hdr.
virtio_net_hdr_to_skbis used byPF_PACKET, tun and tapExactly.
Quoted 1 line
when injecting a packet into the egress path.
Quoted 8 lines
Quoted 3 lines
For tuntap it's still the RX path. For
PF_PACEKTand macvtap, it's the tx.Maybe a new parameter to
virtio_net_hdr_to_skb()?This is the most straightforward approach. But requires changse to all callers, in a patch targeting all the stable branches.
I'd prefer if we can detect ingress vs egress directly.
Not doing this, because both on ingress and egress the allowed ip_summed types are more relaxed than I imagined.
Let's just make the check more narrow to avoid such false positives.
GRO indeed allows CHECKSUM_NONE.
But TSO also accepts packets that are not CHECKSUM_PARTIAL, and will fix up csum_start/csum_off. In tcp4_gso_segment:
if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
const struct iphdr *iph = ip_hdr(skb);
struct tcphdr *th = tcp_hdr(skb);
/* Set up checksum pseudo header, usually expect stack to
* have done this already.
*/
th->check = 0;
skb->ip_summed = CHECKSUM_PARTIAL;
__tcp_v4_send_check(skb, iph->saddr, iph->daddr);
}With __tcp_v4_send_check:
void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
{
struct tcphdr *th = tcp_hdr(skb);
th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
skb->csum_start = skb_transport_header(skb) - skb->head;
skb->csum_offset = offsetof(struct tcphdr, check);
} That means that we can relax the check on input from userspace to bad CHECKSUM_PARTIAL input:
@@ -173,7 +173,8 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb,
break;
case SKB_GSO_TCPV4:
case SKB_GSO_TCPV6:
- if (skb->csum_offset != offsetof(struct tcphdr, check))
+ if (skb->ip_summed == CHECKSUM_PARTIAL &&
+ skb->csum_offset != offsetof(struct tcphdr, check))
return -EINVAL;I've verified that this test still catches the bad packet from the syzkaller report in the Link in the commit.
Quoted 12 lines
Based on
ip_summed,pkt_type,is_skb_wmemor so. But so far have not found a suitable condition.I noticed something else: as you point out TUN is ingress. Unlike
virtnet_receive, it does not setip_summedtoCHECKSUM_UNNECESSARYifVIRTIO_NET_HDR_F_DATA_VALID. It probably should. GRO expects packets to have had their integrity verified.CHECKSUM_NONEon ingress is not correct for GRO.And also related: no GRO should be generated by a device unless
VIRTIO_NET_HDR_F_DATA_VALIDis also passed? I have to check the spec if it says anything about this.
Quoted 85 lines · Willem de Bruijn
Quoted 29 lines · Willem de Bruijn
Quoted 29 lines
Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
So I guess VIRTIO_NET_HDR_GSO_* without
VIRTIO_NET_HDR_F_DATA_VALIDwould be wrong on rx.But the new check
if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { [...] case SKB_GSO_TCPV4: case SKB_GSO_TCPV6: if (skb->csum_offset != offsetof(struct tcphdr, check)) return -EINVAL;should be limited to callers of
virtio_net_hdr_to_skbon the tx/GSO path.Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
Looking what the cleanest/minimal patch is to accomplish that.
virtio_net_hdr_to_skb()translates virtio-net header to skb metadata, so it's RX. For TX the helper should bevirtio_net_hdr_from_skb()which translates skb metadata to virtio hdr.
virtio_net_hdr_to_skbis used byPF_PACKET, tun and tapExactly.
Quoted 1 line
when injecting a packet into the egress path.
Quoted 8 lines
Quoted 3 lines
For tuntap it's still the RX path. For
PF_PACEKTand macvtap, it's the tx.Maybe a new parameter to
virtio_net_hdr_to_skb()?This is the most straightforward approach. But requires changse to all callers, in a patch targeting all the stable branches.
I'd prefer if we can detect ingress vs egress directly.
Not doing this, because both on ingress and egress the allowed
ip_summedtypes are more relaxed than I imagined.Let's just make the check more narrow to avoid such false positives.
GRO indeed allows
CHECKSUM_NONE.But TSO also accepts packets that are not
CHECKSUM_PARTIAL, and will fix up csum_start/csum_off. Intcp4_gso_segment:if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) { const struct iphdr *iph = ip_hdr(skb); struct tcphdr *th = tcp_hdr(skb); /* Set up checksum pseudo header, usually expect stack to * have done this already. */ th->check = 0; skb->ip_summed = CHECKSUM_PARTIAL; __tcp_v4_send_check(skb, iph->saddr, iph->daddr); }With
__tcp_v4_send_check:void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr) { struct tcphdr *th = tcp_hdr(skb); th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); skb->csum_start = skb_transport_header(skb) - skb->head; skb->csum_offset = offsetof(struct tcphdr, check); }That means that we can relax the check on input from userspace to bad
CHECKSUM_PARTIALinput:(quoted diff) +2
-1@@ -173,7 +173,8 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb, break; case SKB_GSO_TCPV4: case SKB_GSO_TCPV6: - if (skb->csum_offset != offsetof(struct tcphdr, check)) + if (skb->ip_summed == CHECKSUM_PARTIAL && + skb->csum_offset != offsetof(struct tcphdr, check)) return -EINVAL;
Quoted 2 lines
I've verified that this test still catches the bad packet from the syzkaller report in the Link in the commit.
Sent: https://lore.kernel.org/netdev/[email protected]/T/#u
Quoted 8 lines
Quoted 8 lines
Based on
ip_summed,pkt_type,is_skb_wmemor so. But so far have not found a suitable condition.I noticed something else: as you point out TUN is ingress. Unlike
virtnet_receive, it does not setip_summedtoCHECKSUM_UNNECESSARYifVIRTIO_NET_HDR_F_DATA_VALID. It probably should. GRO expects packets to have had their integrity verified.CHECKSUM_NONEon ingress is not correct for GRO.
Actually CHECKSUM_NONE is allowed. It just triggers software checksum validation.
Tun by default does not use GRO, only if enabling IFF_NAPI.
If a packet arrives at GRO with CHECKSUM_PARTIAL, then its checksum is assumed valid, per __skb_gro_checksum_validate_needed. So that would be one way for tun users today to get efficient GRO.
Quoted 3 lines
Quoted 3 lines
And also related: no GRO should be generated by a device unless
VIRTIO_NET_HDR_F_DATA_VALIDis also passed? I have to check the spec if it says anything about this.
Given that GRO handles !CHECKSUM_UNNECESSARY, probably no need to require VIRTIO_NET_HDR_F_DATA_VALID with virtio GRO either.
Quoted 181 lines · Willem de Bruijn
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.This happens on the virtio-net receive path, so the bad data is received from the hypervisor.
Quoted 1 line
From what I gather that arm64 fvp (Fixed Virtual Platforms) hypervisor
is closed source?
Disabling GRO on this device will likely work as temporary workaround.
What we can do is instead of dropping packets to correct their offset:
case
SKB_GSO_TCPV4:
caseSKB_GSO_TCPV6:(quoted diff) +4
-2- if (skb->csum_offset != offsetof(struct tcphdr, check)) - return -EINVAL; + if (skb->csum_offset != offsetof(struct tcphdr, check)) { + DEBUG_NET_WARN_ON_ONCE(1); + skb->csum_offset = offsetof(struct tcphdr, check); + }
Quoted 5 lines
If the issue is in
csum_offset. If the newcsum_startcheck fails, that won't help.It would be helpful to see these values at this point, e.g., with
skb_dump(KERN_INFO, skb, false);
thanks for the quick response.
i sent this issue on my last day at the company so i won't be able to help with this any more, sorry.
fvp is closed source but has freely available binaries for x86_64 glibc based linux systems (behind registration and license agreements) so in principle the issue can be reproduced outside of arm but using fvp is not obvious.
hopefully somebody at arm can pick it up or at least report this thread to the fvp team internally.
Quoted 7 lines · Szabolcs Nagy
fvp is closed source but has freely available binaries for
x86_64glibc based linux systems (behind registration and license agreements) so in principle the issue can be reproduced outside of arm but using fvp is not obvious.hopefully somebody at arm can pick it up or at least report this thread to the fvp team internally.
FWIW there's a tool called shrinkwrap which makes it quite a lot easier to get going:
https://gitlab.arm.com/tooling/shrinkwrap
though since the models are very flexibile valid configurations that people see issues with aren't always covered by shrinkwrap.
Quoted 10 lines · Mark Brown
Quoted 7 lines · Szabolcs Nagy
fvp is closed source but has freely available binaries for
x86_64glibc based linux systems (behind registration and license agreements) so in principle the issue can be reproduced outside of arm but using fvp is not obvious.hopefully somebody at arm can pick it up or at least report this thread to the fvp team internally.
Quoted 7 lines
FWIW there's a tool called shrinkwrap which makes it quite a lot easier to get going:
https://gitlab.arm.com/tooling/shrinkwrap
though since the models are very flexibile valid configurations that people see issues with aren't always covered by shrinkwrap.
It is fairly trivial to change the default config and use virtio-net to reproduce this issue. If anyone tries the above tool, they can apply below diff and should be able to reproduce the issue.
--
Regards,
Sudeep
config/FVP_Base_RevC-2xAEMvA-base.yaml +4 -2
@@ -39,8 +39,10 @@ description: >- # Networking. By default use user-space networking, mapping port 22 in the # FVP to a user-specified port on the host (see rtvar:LOCAL_NET_PORT). This # enables ssh. - -C bp.smsc_91c111.enabled: 1 - -C bp.hostbridge.userNetworking: 1 + -C bp.smsc_91c111.enabled: 0 + -C bp.hostbridge.userNetworking: 0 + -C bp.virtio_net.enabled: 1 + -C bp.virtio_net.hostbridge.userNetworking: 1 -C bp.hostbridge.userNetPorts: ${rtvar:LOCAL_NET_PORT}=22 # FVP Performance tweaks.
Quoted 10 lines · Mark Brown
Quoted 7 lines · Szabolcs Nagy
fvp is closed source but has freely available binaries for
x86_64glibc based linux systems (behind registration and license agreements) so in principle the issue can be reproduced outside of arm but using fvp is not obvious.hopefully somebody at arm can pick it up or at least report this thread to the fvp team internally.
Quoted 7 lines
FWIW there's a tool called shrinkwrap which makes it quite a lot easier to get going:
https://gitlab.arm.com/tooling/shrinkwrap
though since the models are very flexibile valid configurations that people see issues with aren't always covered by shrinkwrap.
Thanks both.
From what I gather this affects any device passing packets with GRO, so should not be limited to fvp even.
Quoted 156 lines · Willem de Bruijn
Quoted 128 lines · Szabolcs Nagy
Quoted 68 lines · Willem de Bruijn
From: Willem de Bruijn <[email protected]>
Tighten
csum_startandcsum_offsetchecks invirtio_net_hdr_to_skbfor GSO packets.The function already checks that a checksum requested with
VIRTIO_NET_HDR_F_NEEDS_CSUMis in skb linear. But for GSO packets this might not hold for segs after segmentation.Syzkaller demonstrated to reach this warning in
skb_checksum_helpoffset = skb_checksum_start_offset(skb); ret = -EINVAL; if (WARN_ON_ONCE(offset >= skb_headlen(skb)))By injecting a TSO packet:
WARNING: CPU: 1 PID: 3539 at net/core/dev.c:3284 skb_checksum_help+0x3d0/0x5b0 ip_do_fragment+0x209/0x1b20 net/ipv4/ip_output.c:774 ip_finish_output_gso net/ipv4/ip_output.c:279 [inline] __ip_finish_output+0x2bd/0x4b0 net/ipv4/ip_output.c:301 iptunnel_xmit+0x50c/0x930 net/ipv4/ip_tunnel_core.c:82 ip_tunnel_xmit+0x2296/0x2c70 net/ipv4/ip_tunnel.c:813 __gre_xmit net/ipv4/ip_gre.c:469 [inline] ipgre_xmit+0x759/0xa60 net/ipv4/ip_gre.c:661 __netdev_start_xmit include/linux/netdevice.h:4850 [inline] netdev_start_xmit include/linux/netdevice.h:4864 [inline] xmit_one net/core/dev.c:3595 [inline] dev_hard_start_xmit+0x261/0x8c0 net/core/dev.c:3611 __dev_queue_xmit+0x1b97/0x3c90 net/core/dev.c:4261 packet_snd net/packet/af_packet.c:3073 [inline]The geometry of the bad input packet at
tcp_gso_segment:[ 52.003050][ T8403] skb len=12202 headroom=244 headlen=12093 tailroom=0 [ 52.003050][ T8403] mac=(168,24) mac_len=24 net=(192,52) trans=244 [ 52.003050][ T8403] shinfo(txflags=0 nr_frags=1 gso(size=1552 type=3 segs=0)) [ 52.003050][ T8403] csum(0x60000c7 start=199 offset=1536ip_summed=3 complete_sw=0 valid=0 level=0)
Mitigate with stricter input validation.
csum_offset: for GSO packets, deduce the correct value fromgso_type. This is already done for USO. Extend it to TSO. Let UFO be:
udp[46]_ufo_fragment ignores these fields and always computes the checksum in software.
csum_start: finding the real offset requires parsing to the transport header. Do not add a parser, use existing segmentation parsing. Thanks toSKB_GSO_DODGY, that also catches bad packets that are hw offloaded. Again test both TSO and USO. Do not test UFO for the above reason, and do not test UDP tunnel offload.GSO packet are almost always
CHECKSUM_PARTIAL. USO packets may beCHECKSUM_NONEsince commit10154dbded6d6("udp: Allow GSO transmit from devices with no checksum offload"), but then still these fields are initialized correctly in udp4_hwcsum/udp6_hwcsum_outgoing. So no need to test forip_summed==CHECKSUM_PARTIALfirst.This revises an existing fix mentioned in the Fixes tag, which broke small packets with GSO offload, as detected by kselftests.
Fixes e269d79c7d35 ("net: missing check virtio")Signed-off-by Willem de Bruijn <[email protected]>Quoted 6 lines
---
v1->v2
-skb_transport_headerinstead ofskb->transport_header(edumazet@)
- typo: migitate -> mitigate
---this breaks booting from nfs root on an arm64 fvp model for me.
i see two fixup commits
commit
30b03f2a0592eee1267298298eac9dd655f55ab2Author: Jakub Sitnicki <[email protected]> AuthorDate: 2024-08-08 11:56:22 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-09 21:58:08 -0700udp: Fall back to software USO if IPv6 extension headers are present
and
commit
b128ed5ab27330deeeaf51ea8bb69f1442a96f7fAuthor: Felix Fietkau <[email protected]> AuthorDate: 2024-08-19 17:06:21 +0200 Commit: Jakub Kicinski <[email protected]> CommitDate: 2024-08-21 17:15:05 -0700udp: fix receiving fraglist GSO packets
but they don't fix the issue for me,
at the boot console i see...
[ 3.686846] Sending DHCP requests ., OK [ 3.687302] IP-Config: Got DHCP answer from 172.20.51.254, my address is 172.20.51.1 [ 3.687423] IP-Config: Complete: [ 3.687482] device=eth0, hwaddr=ea:0d:79:71:af:cd, ipaddr=172.20.51.1, mask=255.255.255.0, gw=172.20.51.254 [ 3.687631] host=172.20.51.1, domain=, nis-domain=(none) [ 3.687719] bootserver=172.20.51.254, rootserver=10.2.80.41, rootpath= [ 3.687771] nameserver0=172.20.51.254, nameserver1=172.20.51.252, nameserver2=172.20.51.251 [ 3.689075] clk: Disabling unused clocks [ 3.689167] PM: genpd: Disabling unused power domains [ 3.689258] ALSA device list: [ 3.689330] No soundcards found. [ 3.716297] VFS: Mounted root (nfs4 filesystem) on device 0:24. [ 3.716843] devtmpfs: mounted [ 3.734352] Freeing unused kernel memory: 10112K [ 3.735178] Run /sbin/init as init process [ 3.743770] eth0: bad gso: type: 1, size: 1440 [ 3.744186] eth0: bad gso: type: 1, size: 1440 ... [ 154.610991] eth0: bad gso: type: 1, size: 1440 [ 185.330941] nfs: server 10.2.80.41 not responding, still trying...
the "bad gso" message keeps repeating and init is not executed.
Quoted 7 lines
if i revert the 3 patches above on 6.11-rc6 then init runs without "bad gso" error.
this affects testing the arm64-gcs patches on top of 6.11-rc3 and 6.11-rc6
not sure if this is an fvp or kernel bug.
Thanks for the report, sorry that you're encountering this breakage.
Makes sense that this commit introduced it
if (virtio_net_hdr_to_skb(skb, &hdr->hdr, virtio_is_little_endian(vi->vdev))) { net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", dev->name, hdr->hdr.gso_type, hdr->hdr.gso_size); goto frame_err; }Type 1 is
VIRTIO_NET_HDR_GSO_TCPV4Most likely this application is inserting a packet with flag
VIRTIO_NET_HDR_F_NEEDS_CSUMand a wrongcsum_start. Or is requesting TSO without checksum offload at all. In which case the kernel goes out of its way to find the right offset, but may fail.
Quoted 4 lines
Which nfs-client is this? I'd like to take a look at the sourcecode.
Unfortunately the kernel warning lacks a few useful pieces of data, such as the other
virtio_net_hdrfields and the packet length.
fwiw, this is not nfs specific: if you boot a system with this kernel, them most network operations will fail (e.g. scp-ing something on this system or trying to curl or wget something from within) with the above mentioned error eth0: bad gso: type: 1, size: 1440
Kind regards,
Yury
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.