virtio_net: add optional flow dissection in virtio_net_hdr_to_skb
Subsystems (10)
v5 cover virtio_net: add optional flow dissection in virtio_net_hdr_to_skb
From: Tanner Love <[email protected]>
First patch extends the flow dissector BPF program type to accept pointer to virtio-net header.
Second patch uses this feature to add optional flow dissection in virtio_net_hdr_to_skb(). This allows admins to define permitted packets more strictly, for example dropping deprecated UDP_UFO packets.
Third patch extends kselftest to cover this feature.
Tanner Love (3):
net: flow_dissector: extend bpf flow dissector support with vnet hdr
virtio_net: add optional flow dissection in virtio_net_hdr_to_skb selftests/net: amend bpf flow dissector prog to do vnet hdr validation
drivers/net/bonding/bond_main.c | 2 +- include/linux/skbuff.h | 35 ++- include/linux/virtio_net.h | 25 ++- include/uapi/linux/bpf.h | 2 + kernel/bpf/verifier.c | 50 ++++- net/bpf/test_run.c | 2 +- net/core/filter.c | 26 +++ net/core/flow_dissector.c | 21 +- net/core/sysctl_net_core.c | 9 + tools/include/uapi/linux/bpf.h | 2 + tools/testing/selftests/bpf/progs/bpf_flow.c | 208 ++++++++++++++---- .../selftests/bpf/test_flow_dissector.c | 181 +++++++++++++-- .../selftests/bpf/test_flow_dissector.sh | 19 ++ 13 files changed, 495 insertions(+), 87 deletions(-)
Signature · 1 line
2.32.0.272.g935e593368-goog
PATCH v5 1/3 net: flow_dissector: extend bpf flow dissector support with vnet hdr
From: Tanner Love <[email protected]>
Amend the bpf flow dissector program type to be able to process virtio-net headers. Do this to enable bpf flow dissector programs to perform virtio-net header validation. The next patch in this series will add a flow dissection hook in virtio_net_hdr_to_skb and make use of this extended functionality. That commit message has more background on the use case.
Add two new members to struct bpf_flow_keys: a pointer to struct virtio_net_hdr, and vhdr_is_little_endian. The latter is required to inform the BPF program of the endianness of the virtio-net header fields, to handle the case of a version 1+ header on a big endian machine.
Changes
v5:
- Use PTR_TO_BTF_ID_OR_NULL instead of defining new PTR_TO_VNET_HDR_OR_NULL
- Make check_flow_keys_access() disallow writes to keys->vhdr
- Make check_flow_keys_access() check loading keys->vhdr is in
sizeof(__u64)
- Use BPF_REG_AX instead of BPF_REG_TMP as scratch reg
- Describe parameter vhdr_is_little_endian in __skb_flow_dissect documentation
v4:
- Add virtio_net_hdr pointer to struct bpf_flow_keys
- Add vhdr_is_little_endian to struct bpf_flow_keys
v2:
- Describe parameter vhdr in __skb_flow_dissect documentation
--- drivers/net/bonding/bond_main.c | 2 +- include/linux/skbuff.h | 35 ++++++++++++++++++----- include/uapi/linux/bpf.h | 2 ++ kernel/bpf/verifier.c | 50 ++++++++++++++++++++++++++++++--- net/bpf/test_run.c | 2 +- net/core/filter.c | 26 +++++++++++++++++ net/core/flow_dissector.c | 18 +++++++++--- tools/include/uapi/linux/bpf.h | 2 ++ 8 files changed, 120 insertions(+), 17 deletions(-)
drivers/net/bonding/bond_main.c +1 -1
@@ -3554,7 +3554,7 @@ static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,case BOND_XMIT_POLICY_ENCAP34 :memset (fk,0 ,sizeof (*fk));return __skb_flow_dissect (NULL , skb, &flow_keys_bonding, - fk,NULL ,0 ,0 ,0 ,0 ); + fk,NULL ,0 ,0 ,0 ,0 ,NULL , false);default :break ; }
include/linux/skbuff.h +28 -7
@@ -1314,21 +1314,27 @@ void skb_flow_dissector_init(struct flow_dissector *flow_dissector, unsigned int key_count);struct bpf_flow_dissector ; +struct virtio_net_hdr ;bool bpf_flow_dissect (struct bpf_prog *prog,struct bpf_flow_dissector *ctx, -__be16 proto,int nhoff,int hlen,unsigned int flags); +__be16 proto,int nhoff,int hlen,unsigned int flags, +const struct virtio_net_hdr *vhdr, +bool vhdr_is_little_endian);bool __skb_flow_dissect (const struct net *net,const struct sk_buff *skb,struct flow_dissector *flow_dissector,void *target_container,const void *data, -__be16 proto,int nhoff,int hlen,unsigned int flags); +__be16 proto,int nhoff,int hlen,unsigned int flags, +const struct virtio_net_hdr *vhdr, +bool vhdr_is_little_endian);static inline bool skb_flow_dissect (const struct sk_buff *skb,struct flow_dissector *flow_dissector,void *target_container,unsigned int flags) {return __skb_flow_dissect (NULL , skb, flow_dissector, - target_container,NULL ,0 ,0 ,0 , flags); + target_container,NULL ,0 ,0 ,0 , flags,NULL , + false); } static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
@@ -1337,7 +1343,22 @@ static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, {memset (flow,0 ,sizeof (*flow));return __skb_flow_dissect (NULL , skb, &flow_keys_dissector, - flow,NULL ,0 ,0 ,0 , flags); + flow,NULL ,0 ,0 ,0 , flags,NULL , false); +} + +static inline bool +__skb_flow_dissect_flow_keys_basic (const struct net *net, +const struct sk_buff *skb, +struct flow_keys_basic *flow, +const void *data,__be16 proto, +int nhoff,int hlen,unsigned int flags, +const struct virtio_net_hdr *vhdr, +bool vhdr_is_little_endian) +{ +memset (flow,0 ,sizeof (*flow)); +return __skb_flow_dissect (net, skb, &flow_keys_basic_dissector, flow, + data, proto, nhoff, hlen, flags, vhdr, + vhdr_is_little_endian); } static inline bool
@@ -1347,9 +1368,9 @@ skb_flow_dissect_flow_keys_basic(const struct net *net, const void *data, __be16 proto, int nhoff, int hlen, unsigned int flags) { - memset(flow, 0, sizeof(*flow)); -return __skb_flow_dissect (net, skb, &flow_keys_basic_dissector, flow, - data, proto, nhoff, hlen, flags); + return __skb_flow_dissect_flow_keys_basic(net, skb, flow, data, proto, +nhoff ,hlen ,flags ,NULL , +false ); } void skb_flow_dissect_meta(const struct sk_buff *skb,
include/uapi/linux/bpf.h +2 -0
@@ -6017,6 +6017,8 @@ struct bpf_flow_keys { };__u32 flags;__be32 flow_label; + __bpf_md_ptr(const struct virtio_net_hdr *, vhdr); +__u8 vhdr_is_little_endian; };struct bpf_func_info {
kernel/bpf/verifier.c +46 -4
80 lines not on this page · raw message
net/bpf/test_run.c +1 -1
@@ -797,7 +797,7 @@ int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,bpf_test_timer_enter (&t);do { retval =bpf_flow_dissect (prog, &ctx, eth->h_proto,ETH_HLEN , - size, flags); + size, flags,NULL , false); }while (bpf_test_timer_continue (&t, repeat, &ret, &duration));bpf_test_timer_leave (&t);
net/core/filter.c +26 -0
@@ -8358,6 +8358,8 @@ static bool flow_dissector_is_valid_access(int off, int size,return false; info->reg_type =PTR_TO_FLOW_KEYS ;return true; + case bpf_ctx_range(struct __sk_buff, len): +return size == size_default;default :return false; }
@@ -8390,6 +8392,30 @@ static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, si->dst_reg, si->src_reg, offsetof(struct bpf_flow_dissector, flow_keys));break ; + +case offsetof(struct __sk_buff , len): + *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, skb), + si->dst_reg, si->src_reg, + offsetof(struct bpf_flow_dissector , skb)); + *insn++ =BPF_JMP_IMM (BPF_JNE , si->dst_reg,0 ,4 ); +/* bpf_flow_dissector->skb == NULL */ +/* dst_reg = bpf_flow_dissector->data_end */ + *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), + si->dst_reg, si->src_reg, + offsetof(struct bpf_flow_dissector , data_end)); +/* AX = bpf_flow_dissector->data */ + *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), +BPF_REG_AX , si->src_reg, + offsetof(struct bpf_flow_dissector , data)); +/* dst_reg -= bpf_flow_dissector->data */ + *insn++ =BPF_ALU64_REG (BPF_SUB , si->dst_reg,BPF_REG_AX ); + *insn++ =BPF_JMP_A (1 ); +/* bpf_flow_dissector->skb != NULL */ +/* bpf_flow_dissector->skb->len */ + *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), + si->dst_reg, si->dst_reg, + offsetof(struct sk_buff , len)); +break ; }return insn - insn_buf;
net/core/flow_dissector.c +14 -4
@@ -28,6 +28,7 @@#include <scsi/fc/fc_fcoe.h> #include <uapi/linux/batadv_packet.h> #include <linux/bpf.h> +#include <linux/virtio_net.h> #if IS_ENABLED (CONFIG_NF_CONNTRACK )#include <net/netfilter/nf_conntrack_core.h> #include <net/netfilter/nf_conntrack_labels.h>
@@ -864,7 +865,9 @@ static void __skb_flow_bpf_to_target(const struct bpf_flow_keys *flow_keys, }bool bpf_flow_dissect (struct bpf_prog *prog,struct bpf_flow_dissector *ctx, -__be16 proto,int nhoff,int hlen,unsigned int flags) +__be16 proto,int nhoff,int hlen,unsigned int flags, +const struct virtio_net_hdr *vhdr, +bool vhdr_is_little_endian) {struct bpf_flow_keys *flow_keys = ctx->flow_keys;u32 result;
@@ -874,6 +877,8 @@ bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, flow_keys->n_proto = proto; flow_keys->nhoff = nhoff; flow_keys->thoff = flow_keys->nhoff; + flow_keys->vhdr = vhdr; + flow_keys->vhdr_is_little_endian = vhdr_is_little_endian;BUILD_BUG_ON ((int )BPF_FLOW_DISSECTOR_F_PARSE_1ST_FRAG != (int )FLOW_DISSECTOR_F_PARSE_1ST_FRAG );
@@ -904,6 +909,8 @@ bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,* @hlen: packet header length, if @data is NULL use skb_headlen(skb) * @flags: flags that control the dissection process, e.g. * FLOW_DISSECTOR_F_STOP_AT_ENCAP. +* @vhdr: virtio_net_header to include in kernel context for BPF flow dissector +* @vhdr_is_little_endian: whether virtio_net_hdr fields are little endian * * The function will try to retrieve individual keys into target specified * by flow_dissector from either the skbuff or a raw buffer specified by the
@@ -915,7 +922,9 @@ bool __skb_flow_dissect(const struct net *net,const struct sk_buff *skb, struct flow_dissector *flow_dissector, void *target_container, const void *data, - __be16 proto, int nhoff, int hlen, unsigned int flags) + __be16 proto, int nhoff, int hlen, unsigned int flags, + const struct virtio_net_hdr *vhdr, + bool vhdr_is_little_endian) { struct flow_dissector_key_control *key_control;struct flow_dissector_key_basic *key_basic;
@@ -1012,7 +1021,8 @@ bool __skb_flow_dissect(const struct net *net, prog =READ_ONCE (run_array->items[0 ].prog); ret =bpf_flow_dissect (prog, &ctx, n_proto, nhoff, - hlen, flags); + hlen, flags, vhdr, + vhdr_is_little_endian);__skb_flow_bpf_to_target (&flow_keys, flow_dissector, target_container);rcu_read_unlock ();
@@ -1610,7 +1620,7 @@ u32 __skb_get_hash_symmetric(const struct sk_buff *skb)memset (&keys,0 ,sizeof (keys));__skb_flow_dissect (NULL , skb, &flow_keys_dissector_symmetric, &keys,NULL ,0 ,0 ,0 , -FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL ); +FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL ,NULL , false);return __flow_hash_from_keys (&keys, &hashrnd); }
tools/include/uapi/linux/bpf.h +2 -0
@@ -6017,6 +6017,8 @@ struct bpf_flow_keys { };__u32 flags;__be32 flow_label; + __bpf_md_ptr(const struct virtio_net_hdr *, vhdr); +__u8 vhdr_is_little_endian; };struct bpf_func_info {
Signature · 1 line
2.32.0.272.g935e593368-goog
PATCH v5 3/3 selftests/net: amend bpf flow dissector prog to do vnet hdr validation
In reply to Tanner LoveFrom: Tanner Love <[email protected]>
Change the BPF flow dissector program to perform various checks on the virtio_net_hdr fields after doing flow dissection.
Amend test_flow_dissector.(c|sh) to add test cases that inject packets with reasonable or unreasonable virtio-net headers and assert that bad packets are dropped and good packets are not. Do this via packet socket; the kernel executes tpacket_snd, which enters virtio_net_hdr_to_skb, where flow dissection / vnet header validation occurs.
Changes
v4:
- Read virtio_net_hdr pointer from struct bpf_flow_keys
- Add vnet header endianness logic to BPF program
--- tools/testing/selftests/bpf/progs/bpf_flow.c | 208 ++++++++++++++---- .../selftests/bpf/test_flow_dissector.c | 181 +++++++++++++-- .../selftests/bpf/test_flow_dissector.sh | 19 ++ 3 files changed, 342 insertions(+), 66 deletions(-)
tools/testing/selftests/bpf/progs/bpf_flow.c +164 -44
@@ -13,6 +13,7 @@#include <linux/tcp.h> #include <linux/udp.h> #include <linux/if_packet.h> +#include <linux/virtio_net.h> #include <sys/socket.h> #include <linux/if_tunnel.h> #include <linux/mpls.h>
@@ -71,15 +72,141 @@ struct { __type(value, struct bpf_flow_keys); } last_dissection SEC(".maps"); -static __always_inline int export_flow_keys(struct bpf_flow_keys *keys, +static __always_inline __u16 __virtio16_to_cpu(bool little_endian, +__virtio16 val) +{ +if (little_endian) +return __le16_to_cpu ((__le16 )val); +else +return __be16_to_cpu ((__be16 )val); +} + +/* Drops invalid virtio-net headers */ +static __always_inline int validate_vnet_hdr(const struct bpf_flow_keys *keys, +__u32 skb_len) +{ +__u16 gso_type, hdr_len, gso_size, csum_start, csum_offset; +const struct virtio_net_hdr *vhdr = keys->vhdr; + +if (!vhdr) +return BPF_OK ; + + gso_type =__virtio16_to_cpu (keys->vhdr_is_little_endian, + vhdr->gso_type); + hdr_len =__virtio16_to_cpu (keys->vhdr_is_little_endian, + vhdr->hdr_len); + gso_size =__virtio16_to_cpu (keys->vhdr_is_little_endian, + vhdr->gso_size); + csum_start =__virtio16_to_cpu (keys->vhdr_is_little_endian, + vhdr->csum_start); + csum_offset =__virtio16_to_cpu (keys->vhdr_is_little_endian, + vhdr->csum_offset); + +/* Check gso */ +if (gso_type !=VIRTIO_NET_HDR_GSO_NONE ) { +if (!(vhdr->flags &VIRTIO_NET_HDR_F_NEEDS_CSUM )) +return BPF_DROP ; + +if (keys->is_encap) +return BPF_DROP ; + +switch (gso_type & ~VIRTIO_NET_HDR_GSO_ECN ) { +case VIRTIO_NET_HDR_GSO_TCPV4 : +if (keys->addr_proto !=ETH_P_IP || + keys->ip_proto !=IPPROTO_TCP ) +return BPF_DROP ; + +if (gso_size >= skb_len - keys->thoff - +sizeof (struct tcphdr )) +return BPF_DROP ; + +break ; +case VIRTIO_NET_HDR_GSO_TCPV6 : +if (keys->addr_proto !=ETH_P_IPV6 || + keys->ip_proto !=IPPROTO_TCP ) +return BPF_DROP ; + +if (gso_size >= skb_len - keys->thoff - +sizeof (struct tcphdr )) +return BPF_DROP ; + +break ; +case VIRTIO_NET_HDR_GSO_UDP : +if (keys->ip_proto !=IPPROTO_UDP ) +return BPF_DROP ; + +if (gso_size >= skb_len - keys->thoff - +sizeof (struct udphdr )) +return BPF_DROP ; + +break ; +default : +return BPF_DROP ; + } + } + +/* Check hdr_len */ +if (hdr_len) { +switch (keys->ip_proto) { +case IPPROTO_TCP : +if (hdr_len != keys->thoff +sizeof (struct tcphdr )) +return BPF_DROP ; + +break ; +case IPPROTO_UDP : +if (hdr_len != keys->thoff +sizeof (struct udphdr )) +return BPF_DROP ; + +break ; + } + } + +/* Check csum */ +if (vhdr->flags &VIRTIO_NET_HDR_F_NEEDS_CSUM ) { +if (keys->addr_proto !=ETH_P_IP && + keys->addr_proto !=ETH_P_IPV6 ) +return BPF_DROP ; + +if (csum_start != keys->thoff) +return BPF_DROP ; + +switch (keys->ip_proto) { +case IPPROTO_TCP : +if (csum_offset != offsetof(struct tcphdr , check)) +return BPF_DROP ; + +break ; +case IPPROTO_UDP : +if (csum_offset != offsetof(struct udphdr , check)) +return BPF_DROP ; + +break ; +default : +return BPF_DROP ; + } + } + +return BPF_OK ; +} + +/* Common steps to perform regardless of where protocol parsing finishes: +* 1. store flow keys in map +* 2. if parse result is BPF_OK, parse the vnet hdr if present +* 3. return the parse result +*/ +static __always_inline int parse_epilogue(struct __sk_buff *skb,int ret) { +const struct bpf_flow_keys *keys = skb->flow_keys;__u32 key = (__u32)(keys->sport) <<16 | keys->dport;struct bpf_flow_keys val;memcpy (&val, keys,sizeof (val));bpf_map_update_elem (&last_dissection, &key, &val,BPF_ANY ); -return ret; + +if (ret !=BPF_OK ) +return ret; +return validate_vnet_hdr (keys, skb->len); }#define IPV6_FLOWLABEL_MASK __bpf_constant_htonl(0x000FFFFF)
@@ -114,8 +241,6 @@ static __always_inline void *bpf_flow_dissect_get_header(struct __sk_buff *skb,/* Dispatches on ETHERTYPE */ static __always_inline int parse_eth_proto(struct __sk_buff *skb, __be16 proto) { -struct bpf_flow_keys *keys = skb->flow_keys; -switch (proto) {case bpf_htons (ETH_P_IP ):bpf_tail_call_static (skb, &jmp_table,IP );
@@ -131,12 +256,10 @@ static __always_inline int parse_eth_proto(struct __sk_buff *skb, __be16 proto)case bpf_htons (ETH_P_8021AD ):bpf_tail_call_static (skb, &jmp_table,VLAN );break ; -default : -/* Protocol not supported */ -return export_flow_keys (keys,BPF_DROP ); } -return export_flow_keys (keys,BPF_DROP ); +/* Protocol not supported */ +return parse_epilogue (skb,BPF_DROP ); }SEC ("flow_dissector" )
@@ -162,28 +285,28 @@ static __always_inline int parse_ip_proto(struct __sk_buff *skb, __u8 proto)case IPPROTO_ICMP : icmp =bpf_flow_dissect_get_header (skb,sizeof (*icmp), &_icmp);if (!icmp) -return export_flow_keys (keys,BPF_DROP ); -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_DROP ); +return parse_epilogue (skb,BPF_OK );case IPPROTO_IPIP : keys->is_encap = true;if (keys->flags &BPF_FLOW_DISSECTOR_F_STOP_AT_ENCAP ) -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK );return parse_eth_proto (skb,bpf_htons (ETH_P_IP ));case IPPROTO_IPV6 : keys->is_encap = true;if (keys->flags &BPF_FLOW_DISSECTOR_F_STOP_AT_ENCAP ) -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK );return parse_eth_proto (skb,bpf_htons (ETH_P_IPV6 ));case IPPROTO_GRE : gre =bpf_flow_dissect_get_header (skb,sizeof (*gre), &_gre);if (!gre) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP );if (bpf_htons (gre->flags &GRE_VERSION ))/* Only inspect standard GRE packets with version 0 */ -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK ); keys->thoff +=sizeof (*gre);/* Step over GRE Flags and Proto */ if (GRE_IS_CSUM (gre->flags))
@@ -195,13 +318,13 @@ static __always_inline int parse_ip_proto(struct __sk_buff *skb, __u8 proto) keys->is_encap = true;if (keys->flags &BPF_FLOW_DISSECTOR_F_STOP_AT_ENCAP ) -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK );if (gre->proto ==bpf_htons (ETH_P_TEB )) { eth =bpf_flow_dissect_get_header (skb,sizeof (*eth), &_eth);if (!eth) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->thoff +=sizeof (*eth);
@@ -212,37 +335,35 @@ static __always_inline int parse_ip_proto(struct __sk_buff *skb, __u8 proto)case IPPROTO_TCP : tcp =bpf_flow_dissect_get_header (skb,sizeof (*tcp), &_tcp);if (!tcp) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP );if (tcp->doff <5 ) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP );if ((__u8 *)tcp + (tcp->doff <<2 ) > data_end) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->sport = tcp->source; keys->dport = tcp->dest; -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK );case IPPROTO_UDP :case IPPROTO_UDPLITE : udp =bpf_flow_dissect_get_header (skb,sizeof (*udp), &_udp);if (!udp) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->sport = udp->source; keys->dport = udp->dest; -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK );default : -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); } -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); } static __always_inline int parse_ipv6_proto(struct __sk_buff *skb, __u8 nexthdr) { -struct bpf_flow_keys *keys = skb->flow_keys; -switch (nexthdr) {case IPPROTO_HOPOPTS :case IPPROTO_DSTOPTS :
@@ -255,7 +376,7 @@ static __always_inline int parse_ipv6_proto(struct __sk_buff *skb, __u8 nexthdr)return parse_ip_proto (skb, nexthdr); } -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); }PROG (IP )(struct __sk_buff *skb)
@@ -268,11 +389,11 @@ PROG(IP)(struct __sk_buff *skb) iph =bpf_flow_dissect_get_header (skb,sizeof (*iph), &_iph);if (!iph) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP );/* IP header cannot be smaller than 20 bytes */ if (iph->ihl <5 ) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->addr_proto =ETH_P_IP ; keys->ipv4_src = iph->saddr;
@@ -281,7 +402,7 @@ PROG(IP)(struct __sk_buff *skb) keys->thoff += iph->ihl <<2 ;if (data + keys->thoff > data_end) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP );if (iph->frag_off &bpf_htons (IP_MF |IP_OFFSET )) { keys->is_frag = true;
@@ -302,7 +423,7 @@ PROG(IP)(struct __sk_buff *skb) }if (done) -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK );return parse_ip_proto (skb, iph->protocol); }
@@ -314,7 +435,7 @@ PROG(IPV6)(struct __sk_buff *skb) ip6h =bpf_flow_dissect_get_header (skb,sizeof (*ip6h), &_ip6h);if (!ip6h) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->addr_proto =ETH_P_IPV6 ;memcpy (&keys->ipv6_src, &ip6h->saddr,2 *sizeof (ip6h->saddr));
@@ -324,7 +445,7 @@ PROG(IPV6)(struct __sk_buff *skb) keys->flow_label =ip6_flowlabel (ip6h);if (keys->flags &BPF_FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL ) -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK );return parse_ipv6_proto (skb, ip6h->nexthdr); }
@@ -336,7 +457,7 @@ PROG(IPV6OP)(struct __sk_buff *skb) ip6h =bpf_flow_dissect_get_header (skb,sizeof (*ip6h), &_ip6h);if (!ip6h) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP );/* hlen is in 8-octets and does not include the first 8 bytes * of the header
@@ -354,7 +475,7 @@ PROG(IPV6FR)(struct __sk_buff *skb) fragh =bpf_flow_dissect_get_header (skb,sizeof (*fragh), &_fragh);if (!fragh) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->thoff +=sizeof (*fragh); keys->is_frag = true;
@@ -367,9 +488,9 @@ PROG(IPV6FR)(struct __sk_buff *skb)* explicitly asked for. */ if (!(keys->flags &BPF_FLOW_DISSECTOR_F_PARSE_1ST_FRAG )) -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK ); }else { -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK ); }return parse_ipv6_proto (skb, fragh->nexthdr);
@@ -377,14 +498,13 @@ PROG(IPV6FR)(struct __sk_buff *skb)PROG (MPLS )(struct __sk_buff *skb) { -struct bpf_flow_keys *keys = skb->flow_keys;struct mpls_label *mpls, _mpls; mpls =bpf_flow_dissect_get_header (skb,sizeof (*mpls), &_mpls);if (!mpls) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); -return export_flow_keys (keys,BPF_OK ); +return parse_epilogue (skb,BPF_OK ); }PROG (VLAN )(struct __sk_buff *skb)
@@ -396,10 +516,10 @@ PROG(VLAN)(struct __sk_buff *skb)if (keys->n_proto ==bpf_htons (ETH_P_8021AD )) { vlan =bpf_flow_dissect_get_header (skb,sizeof (*vlan), &_vlan);if (!vlan) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP );if (vlan->h_vlan_encapsulated_proto !=bpf_htons (ETH_P_8021Q )) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->nhoff +=sizeof (*vlan); keys->thoff +=sizeof (*vlan);
@@ -407,14 +527,14 @@ PROG(VLAN)(struct __sk_buff *skb) vlan =bpf_flow_dissect_get_header (skb,sizeof (*vlan), &_vlan);if (!vlan) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->nhoff +=sizeof (*vlan); keys->thoff +=sizeof (*vlan);/* Only allow 8021AD + 8021Q double tagging and no triple tagging.*/ if (vlan->h_vlan_encapsulated_proto ==bpf_htons (ETH_P_8021AD ) || vlan->h_vlan_encapsulated_proto ==bpf_htons (ETH_P_8021Q )) -return export_flow_keys (keys,BPF_DROP ); +return parse_epilogue (skb,BPF_DROP ); keys->n_proto = vlan->h_vlan_encapsulated_proto;return parse_eth_proto (skb, vlan->h_vlan_encapsulated_proto);
tools/testing/selftests/bpf/test_flow_dissector.c +159 -22
343 lines not on this page · raw message
tools/testing/selftests/bpf/test_flow_dissector.sh +19 -0
@@ -51,6 +51,9 @@ if [[ -z $(ip netns identify $$) ]]; then echo "Skipping root flow dissector test, bpftool not found" >&2 fi + orig_flow_dissect_sysctl=$(</proc/sys/net/core/flow_dissect_vnet_hdr) + sysctl net.core.flow_dissect_vnet_hdr=1 + # Run the rest of the tests in a net namespace. ../net/in_netns.sh "$0" "$@" err=$(( $err + $? ))
@@ -61,6 +64,7 @@ if [[ -z $(ip netns identify $$) ]]; then echo "selftests: $TESTNAME [FAILED]"; fi + sysctl net.core.flow_dissect_vnet_hdr=$orig_flow_dissect_sysctl exit $err fi
@@ -165,4 +169,19 @@ tc filter add dev lo parent ffff: protocol ipv6 pref 1337 flower ip_proto \ # Send 10 IPv6/UDP packets from port 10. Filter should not drop any. ./test_flow_dissector -i 6 -f 10 + +echo "Testing virtio-net header validation..." +echo "Testing valid vnet headers. Should *not* be dropped." +./with_addr.sh ./test_flow_dissector -i 4 -D 192.168.0.1 -S 1.1.1.1 -p -v +echo "Testing partial transport header. Should be dropped." +./with_addr.sh ./test_flow_dissector -i 4 -D 192.168.0.1 -S 1.1.1.1 -p -v -U -E +echo "Testing valid vnet gso spec. Should *not* be dropped." +./with_addr.sh ./test_flow_dissector -i 4 -D 192.168.0.1 -S 1.1.1.1 -p -v -g -c -l 8000 +echo "Testing invalid vnet gso size. Should be dropped." +./with_addr.sh ./test_flow_dissector -i 4 -D 192.168.0.1 -S 1.1.1.1 -p -v -g -c -l 100 -E +echo "Testing invalid vnet header len. Should be dropped." +./with_addr.sh ./test_flow_dissector -i 4 -D 192.168.0.1 -S 1.1.1.1 -p -v -L -E +echo "Testing vnet gso without csum. Should be dropped." +./with_addr.sh ./test_flow_dissector -i 4 -D 192.168.0.1 -S 1.1.1.1 -p -v -g -l 8000 -E + exit 0
Signature · 1 line
2.32.0.272.g935e593368-goog
PATCH v5 2/3 virtio_net: add optional flow dissection in virtio_net_hdr_to_skb
In reply to Tanner LoveFrom: Tanner Love <[email protected]>
Syzkaller bugs have resulted from loose specification of virtio_net_hdr[1]. Enable execution of a BPF flow dissector program in virtio_net_hdr_to_skb to validate the vnet header and drop bad input.
Introduce a new sysctl net.core.flow_dissect_vnet_hdr controlling a static key to decide whether to perform flow dissection. When the key is false, virtio_net_hdr_to_skb computes as before.
A permissive specification of vnet headers is part of the ABI. Some applications now depend on it. Still, many of these packets are bogus. Give admins the option to interpret behavior more strictly. For instance, verifying that a VIRTIO_NET_HDR_GSO_TCPV6 header matches a packet with unencapsulated IPv6/TCP without extension headers, with payload length exceeding gso_size and hdr_len exactly at TCP payload offset.
BPF flow dissection implements protocol parsing in an safe way. And is configurable, so can be as pedantic as the workload allows (e.g., dropping UFO altogether).
Vnet_header flow dissection is *not* a substitute for fixing bugs when reported. But even if not enabled continuously, offers a quick path to mitigating vulnerabilities.
[1] https://syzkaller.appspot.com/bug?id=b419a5ca95062664fe1a60b764621eb4526e2cd0
Changes
v4:
- Expand commit message with rationale for bpf flow dissector based implementation
v3:
- Move sysctl_flow_dissect_vnet_hdr_key definition to flow_dissector.c to fix CONFIG_SYSCTL warning when building UML
--- include/linux/virtio_net.h | 25 +++++++++++++++++++++---- net/core/flow_dissector.c | 3 +++ net/core/sysctl_net_core.c | 9 +++++++++ 3 files changed, 33 insertions(+), 4 deletions(-)
include/linux/virtio_net.h +21 -4
@@ -25,10 +25,13 @@ static inline int virtio_net_hdr_set_proto(struct sk_buff *skb,return 0 ; } +DECLARE_STATIC_KEY_FALSE (sysctl_flow_dissect_vnet_hdr_key); +static inline int virtio_net_hdr_to_skb (struct sk_buff *skb,const struct virtio_net_hdr *hdr,bool little_endian) { +struct flow_keys_basic keys;unsigned int gso_type =0 ;unsigned int thlen =0 ;unsigned int p_off =0 ;
@@ -78,13 +81,24 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb, p_off =skb_transport_offset (skb) + thlen;if (!pskb_may_pull (skb, p_off))return -EINVAL ; - }else { + } + +/* BPF flow dissection for optional strict validation. +* +* Admins can define permitted packets more strictly, such as dropping +* deprecated UDP_UFO packets and requiring skb->protocol to be non-zero +* and matching packet headers. +*/ +if (static_branch_unlikely (&sysctl_flow_dissect_vnet_hdr_key) && + !__skb_flow_dissect_flow_keys_basic (NULL , skb, &keys,NULL ,0 ,0 ,0 , +0 , hdr, little_endian)) +return -EINVAL ; + +if (!(hdr->flags &VIRTIO_NET_HDR_F_NEEDS_CSUM )) {/* gso packets without NEEDS_CSUM do not set transport_offset. * probe and drop if does not match one of the above types. */ if (gso_type && skb->network_header) { -struct flow_keys_basic keys; -if (!skb->protocol) {__be16 protocol =dev_parse_header_protocol (skb);
@@ -92,8 +106,11 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb,if (protocol && protocol != skb->protocol)return -EINVAL ; } + retry: -if (!skb_flow_dissect_flow_keys_basic (NULL , skb, &keys, +/* only if flow dissection not already done */ +if (!static_branch_unlikely (&sysctl_flow_dissect_vnet_hdr_key) && + !skb_flow_dissect_flow_keys_basic (NULL , skb, &keys,NULL ,0 ,0 ,0 ,0 )) {/* UFO does not specify ipv4 or 6: try both */
net/core/flow_dissector.c +3 -0
@@ -35,6 +35,9 @@#endif #include <linux/bpf-netns.h> +DEFINE_STATIC_KEY_FALSE (sysctl_flow_dissect_vnet_hdr_key); +EXPORT_SYMBOL (sysctl_flow_dissect_vnet_hdr_key); +static void dissector_set_key (struct flow_dissector *flow_dissector,enum flow_dissector_key_id key_id) {
net/core/sysctl_net_core.c +9 -0
@@ -36,6 +36,8 @@ static int net_msg_warn; /* Unused, but still a sysctl */int sysctl_fb_tunnels_only_for_init_net__read_mostly =0 ;EXPORT_SYMBOL (sysctl_fb_tunnels_only_for_init_net); +DECLARE_STATIC_KEY_FALSE (sysctl_flow_dissect_vnet_hdr_key); +/* 0 - Keep current behavior: * IPv4: inherit all current settings from init_net * IPv6: reset all settings to default
@@ -580,6 +582,13 @@ static struct ctl_table net_core_table[] = { .extra1 = SYSCTL_ONE, .extra2 = &int_3600, }, + { + .procname = "flow_dissect_vnet_hdr", + .data = &sysctl_flow_dissect_vnet_hdr_key.key, + .maxlen = sizeof(sysctl_flow_dissect_vnet_hdr_key), + .mode = 0644, + .proc_handler = proc_do_static_key, + }, { } };
Signature · 1 line
2.32.0.272.g935e593368-goog
Re: [PATCH net-next v5 1/3] net: flow_dissector: extend bpf flow dissector support with vnet hdr
In reply to Tanner LoveOn Thu, Jun 10, 2021 at 02:38:51PM -0400, Tanner Love wrote:
[ ... ]
kernel/bpf/verifier.c static int check_flow_keys_access(struct bpf_verifier_env *env, int off, · line 3387
Quoted 13 lines
static int check_flow_keys_access(struct
bpf_verifier_env*env, int off,kernel/bpf/verifier.c +4
-1- int size) + int size, enum bpf_access_type t) {if (size <0 || off <0 || (u64 )off + size >sizeof (struct bpf_flow_keys )) {@@ -3381,6 +3382,35 @@ static int check_flow_keys_access(struct bpf_verifier_env *env, int off, off, size);return -EACCES ; } + +switch (off) { +case offsetof(struct bpf_flow_keys , vhdr):
It is not enough to stop writing to keys->vhdr. e.g. what if off is offsetof(struct bpf_flow_keys, vhdr) + 1?
Will it break your existing use case if align access (off % size != 0) is enforced now? Take a look at bpf_skb_is_valid_access() in filter.c. Otherwise, another way is needed here.
A nit. It is a good chance to move the new BTF_ID_LIST_SINGLE and most of the check_flow_keys_access() to filter.c. Take a look at check_sock_access().
Quoted 42 lines
kernel/bpf/verifier.c +28
-0+if (t ==BPF_WRITE ) { +verbose (env, +"invalid write to flow keys off=%d size=%d\n" , + off, size); +return -EACCES ; + } + +if (size !=sizeof (__u64)) { +verbose (env, +"invalid access to flow keys off=%d size=%d\n" , + off, size); +return -EACCES ; + } + +break ; +case offsetof(struct bpf_flow_keys , vhdr_is_little_endian): +if (t ==BPF_WRITE ) { +verbose (env, +"invalid write to flow keys off=%d size=%d\n" , + off, size); +return -EACCES ; + } + +break ; + } +return 0 ; }@@ -4053,6 +4083,8 @@ static int check_stack_access_within_bounds(return err; } +BTF_ID_LIST_SINGLE (bpf_flow_dissector_btf_ids, struct, virtio_net_hdr); +/* check whether memory at (regno + off) is accessible for t = (read | write) * if t==write, value_regno is a register which value is stored into memory * if t==read, value_regno is a register which will receive the value from memory @@ -4217,9 +4249,19 @@ static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regnreturn -EACCES ; }
kernel/bpf/verifier.c static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn · line 4258
Quoted 10 lines
kernel/bpf/verifier.c +7
-3- err =check_flow_keys_access (env, off, size); -if (!err && t ==BPF_READ && value_regno >=0 ) -mark_reg_unknown (env, regs, value_regno); + err =check_flow_keys_access (env, off, size, t); +if (!err && t ==BPF_READ && value_regno >=0 ) { +if (off == offsetof(struct bpf_flow_keys , vhdr)) { +mark_reg_known_zero (env, regs, value_regno); + regs[value_regno].type =PTR_TO_BTF_ID_OR_NULL ; + regs[value_regno].btf = btf_vmlinux; + regs[value_regno].btf_id = bpf_flow_dissector_btf_ids[0 ];
It needs to check "!bpf_flow_dissector_btf_ids[0]" in case CONFIG_DEBUG_INFO_BTF is not enabled. Take a look at the RET_PTR_TO_BTF_ID case in verifier.c.
Re: [PATCH net-next v5 1/3] net: flow_dissector: extend bpf flow dissector support with vnet hdr
In reply to a message not in the archiveQuoted 10 lines · Tanner Love
Quoted 4 lines
A nit. It is a good chance to move the new
BTF_ID_LIST_SINGLEand most of thecheck_flow_keys_access()tofilter.c. Take a look atcheck_sock_access().It's not clear to me why it's preferable to move most of the
check_flow_keys_access()tofilter.c. In particular, the part of your comment that I don't understand is the "most of" part. Why would we want to separate the flow-keys-access-checking logic into two separate functions? Thanks
Right, actually, the whole function can be moved.
I found it easier to follow from flow_dissector_is_valid_access() to flow_keys's access check without jumping around between two different files.
Quoted 6 lines
Additionally, it seems that we don't actually need the
BTF_ID_LIST_SINGLE. Can we not, instead, set regs[value_regno].btf_id tobtf_find_by_name_kind(btf_vmlinux, "virtio_net_hdr", BTF_KIND_STRUCT)? (And we'll check that that value is not <= 0.)
BTF_ID_LIST_SINGLE is resolved during compilation. btf_find_by_name_kind() will be repeatedly finding the btf_id during runtime. It is not like a killer but still unnecessary.
Re: [PATCH net-next v5 1/3] net: flow_dissector: extend bpf flow dissector support with vnet hdr
Quoted 11 lines · Martin KaFai Lau
Quoted 10 lines · Tanner Love
Quoted 4 lines
A nit. It is a good chance to move the new
BTF_ID_LIST_SINGLEand most of thecheck_flow_keys_access()tofilter.c. Take a look atcheck_sock_access().It's not clear to me why it's preferable to move most of the
check_flow_keys_access()tofilter.c. In particular, the part of your comment that I don't understand is the "most of" part. Why would we want to separate the flow-keys-access-checking logic into two separate functions? Thanks
Quoted 4 lines
Right, actually, the whole function can be moved.
I found it easier to follow fromflow_dissector_is_valid_access()to flow_keys's access check without jumping around between two different files.
The verifier verbose() logs can be kept in verifier.c though. I think all the -EACCES cases can be consolidated to one verbose() call in check_mem_access() under PTR_TO_FLOW_KEYS.