Message ID | 20231008052101.144422-6-akihiko.odaki@daynix.com (mailing list archive) |
---|---|
State | Superseded |
Delegated to: | Netdev Maintainers |
Headers | show |
Series | tun: Introduce virtio-net hashing feature | expand |
On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > virtio-net have two usage of hashes: one is RSS and another is hash > reporting. Conventionally the hash calculation was done by the VMM. > However, computing the hash after the queue was chosen defeats the > purpose of RSS. > > Another approach is to use eBPF steering program. This approach has > another downside: it cannot report the calculated hash due to the > restrictive nature of eBPF. > > Introduce the code to compute hashes to the kernel in order to overcome > thse challenges. An alternative solution is to extend the eBPF steering > program so that it will be able to report to the userspace, but it makes > little sense to allow to implement different hashing algorithms with > eBPF since the hash value reported by virtio-net is strictly defined by > the specification. > > The hash value already stored in sk_buff is not used and computed > independently since it may have been computed in a way not conformant > with the specification. > > Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > --- > +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > + .max_indirection_table_length = > + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > + > + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > +}; No need to have explicit capabilities exchange like this? Tun either supports all or none. > case TUNSETSTEERINGEBPF: > - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > + if (IS_ERR(bpf_ret)) > + ret = PTR_ERR(bpf_ret); > + else if (bpf_ret) > + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; Don't make one feature disable another. TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive functions. If one is enabled the other call should fail, with EBUSY for instance. > + case TUNSETVNETHASH: > + len = sizeof(vnet_hash); > + if (copy_from_user(&vnet_hash, argp, len)) { > + ret = -EFAULT; > + break; > + } > + > + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > + !tun_is_little_endian(tun))) || > + vnet_hash.indirection_table_mask >= > + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > + ret = -EINVAL; > + break; > + } > + > + argp = (u8 __user *)argp + len; > + len = (vnet_hash.indirection_table_mask + 1) * 2; > + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > + ret = -EFAULT; > + break; > + } > + > + argp = (u8 __user *)argp + len; > + len = virtio_net_hash_key_length(vnet_hash.types); > + > + if (copy_from_user(vnet_hash_key, argp, len)) { > + ret = -EFAULT; > + break; > + } Probably easier and less error-prone to define a fixed size control struct with the max indirection table size. Btw: please trim the CC: list considerably on future patches.
On 2023/10/09 4:07, Willem de Bruijn wrote: > On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> virtio-net have two usage of hashes: one is RSS and another is hash >> reporting. Conventionally the hash calculation was done by the VMM. >> However, computing the hash after the queue was chosen defeats the >> purpose of RSS. >> >> Another approach is to use eBPF steering program. This approach has >> another downside: it cannot report the calculated hash due to the >> restrictive nature of eBPF. >> >> Introduce the code to compute hashes to the kernel in order to overcome >> thse challenges. An alternative solution is to extend the eBPF steering >> program so that it will be able to report to the userspace, but it makes >> little sense to allow to implement different hashing algorithms with >> eBPF since the hash value reported by virtio-net is strictly defined by >> the specification. >> >> The hash value already stored in sk_buff is not used and computed >> independently since it may have been computed in a way not conformant >> with the specification. >> >> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >> --- > >> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >> + .max_indirection_table_length = >> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >> + >> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >> +}; > > No need to have explicit capabilities exchange like this? Tun either > supports all or none. tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. It is because the flow dissector does not support IPv6 extensions. The specification is also vague, and does not tell how many TLVs should be consumed at most when interpreting destination option header so I chose to avoid adding code for these hash types to the flow dissector. I doubt anyone will complain about it since nobody complains for Linux. I'm also adding this so that we can extend it later. max_indirection_table_length may grow for systems with 128+ CPUs, or types may have other bits for new protocols in the future. > >> case TUNSETSTEERINGEBPF: >> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >> + if (IS_ERR(bpf_ret)) >> + ret = PTR_ERR(bpf_ret); >> + else if (bpf_ret) >> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > > Don't make one feature disable another. > > TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > functions. If one is enabled the other call should fail, with EBUSY > for instance. > >> + case TUNSETVNETHASH: >> + len = sizeof(vnet_hash); >> + if (copy_from_user(&vnet_hash, argp, len)) { >> + ret = -EFAULT; >> + break; >> + } >> + >> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >> + !tun_is_little_endian(tun))) || >> + vnet_hash.indirection_table_mask >= >> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >> + ret = -EINVAL; >> + break; >> + } >> + >> + argp = (u8 __user *)argp + len; >> + len = (vnet_hash.indirection_table_mask + 1) * 2; >> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >> + ret = -EFAULT; >> + break; >> + } >> + >> + argp = (u8 __user *)argp + len; >> + len = virtio_net_hash_key_length(vnet_hash.types); >> + >> + if (copy_from_user(vnet_hash_key, argp, len)) { >> + ret = -EFAULT; >> + break; >> + } > > Probably easier and less error-prone to define a fixed size control > struct with the max indirection table size. I made its size variable because the indirection table and key may grow in the future as I wrote above. > > Btw: please trim the CC: list considerably on future patches. I'll do so in the next version with the TUNSETSTEERINGEBPF change you proposed.
On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/09 4:07, Willem de Bruijn wrote: > > On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> virtio-net have two usage of hashes: one is RSS and another is hash > >> reporting. Conventionally the hash calculation was done by the VMM. > >> However, computing the hash after the queue was chosen defeats the > >> purpose of RSS. > >> > >> Another approach is to use eBPF steering program. This approach has > >> another downside: it cannot report the calculated hash due to the > >> restrictive nature of eBPF. > >> > >> Introduce the code to compute hashes to the kernel in order to overcome > >> thse challenges. An alternative solution is to extend the eBPF steering > >> program so that it will be able to report to the userspace, but it makes > >> little sense to allow to implement different hashing algorithms with > >> eBPF since the hash value reported by virtio-net is strictly defined by > >> the specification. > >> > >> The hash value already stored in sk_buff is not used and computed > >> independently since it may have been computed in a way not conformant > >> with the specification. > >> > >> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >> --- > > > >> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > >> + .max_indirection_table_length = > >> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > >> + > >> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > >> +}; > > > > No need to have explicit capabilities exchange like this? Tun either > > supports all or none. > > tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, > VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. > > It is because the flow dissector does not support IPv6 extensions. The > specification is also vague, and does not tell how many TLVs should be > consumed at most when interpreting destination option header so I chose > to avoid adding code for these hash types to the flow dissector. I doubt > anyone will complain about it since nobody complains for Linux. > > I'm also adding this so that we can extend it later. > max_indirection_table_length may grow for systems with 128+ CPUs, or > types may have other bits for new protocols in the future. > > > > >> case TUNSETSTEERINGEBPF: > >> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >> + if (IS_ERR(bpf_ret)) > >> + ret = PTR_ERR(bpf_ret); > >> + else if (bpf_ret) > >> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > > > > Don't make one feature disable another. > > > > TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > > functions. If one is enabled the other call should fail, with EBUSY > > for instance. > > > >> + case TUNSETVNETHASH: > >> + len = sizeof(vnet_hash); > >> + if (copy_from_user(&vnet_hash, argp, len)) { > >> + ret = -EFAULT; > >> + break; > >> + } > >> + > >> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > >> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > >> + !tun_is_little_endian(tun))) || > >> + vnet_hash.indirection_table_mask >= > >> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > >> + ret = -EINVAL; > >> + break; > >> + } > >> + > >> + argp = (u8 __user *)argp + len; > >> + len = (vnet_hash.indirection_table_mask + 1) * 2; > >> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > >> + ret = -EFAULT; > >> + break; > >> + } > >> + > >> + argp = (u8 __user *)argp + len; > >> + len = virtio_net_hash_key_length(vnet_hash.types); > >> + > >> + if (copy_from_user(vnet_hash_key, argp, len)) { > >> + ret = -EFAULT; > >> + break; > >> + } > > > > Probably easier and less error-prone to define a fixed size control > > struct with the max indirection table size. > > I made its size variable because the indirection table and key may grow > in the future as I wrote above. > > > > > Btw: please trim the CC: list considerably on future patches. > > I'll do so in the next version with the TUNSETSTEERINGEBPF change you > proposed. To be clear: please don't just resubmit with that one change. The skb and cb issues are quite fundamental issues that need to be resolved. I'd like to understand why adjusting the existing BPF feature for this exact purpose cannot be amended to return the key it produced. As you point out, the C flow dissector is insufficient. The BPF flow dissector does not have this problem. The same argument would go for the pre-existing BPF steering program.
On 2023/10/09 5:08, Willem de Bruijn wrote: > On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/09 4:07, Willem de Bruijn wrote: >>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>> reporting. Conventionally the hash calculation was done by the VMM. >>>> However, computing the hash after the queue was chosen defeats the >>>> purpose of RSS. >>>> >>>> Another approach is to use eBPF steering program. This approach has >>>> another downside: it cannot report the calculated hash due to the >>>> restrictive nature of eBPF. >>>> >>>> Introduce the code to compute hashes to the kernel in order to overcome >>>> thse challenges. An alternative solution is to extend the eBPF steering >>>> program so that it will be able to report to the userspace, but it makes >>>> little sense to allow to implement different hashing algorithms with >>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>> the specification. >>>> >>>> The hash value already stored in sk_buff is not used and computed >>>> independently since it may have been computed in a way not conformant >>>> with the specification. >>>> >>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>> --- >>> >>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >>>> + .max_indirection_table_length = >>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >>>> + >>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >>>> +}; >>> >>> No need to have explicit capabilities exchange like this? Tun either >>> supports all or none. >> >> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, >> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. >> >> It is because the flow dissector does not support IPv6 extensions. The >> specification is also vague, and does not tell how many TLVs should be >> consumed at most when interpreting destination option header so I chose >> to avoid adding code for these hash types to the flow dissector. I doubt >> anyone will complain about it since nobody complains for Linux. >> >> I'm also adding this so that we can extend it later. >> max_indirection_table_length may grow for systems with 128+ CPUs, or >> types may have other bits for new protocols in the future. >> >>> >>>> case TUNSETSTEERINGEBPF: >>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>> + if (IS_ERR(bpf_ret)) >>>> + ret = PTR_ERR(bpf_ret); >>>> + else if (bpf_ret) >>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; >>> >>> Don't make one feature disable another. >>> >>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive >>> functions. If one is enabled the other call should fail, with EBUSY >>> for instance. >>> >>>> + case TUNSETVNETHASH: >>>> + len = sizeof(vnet_hash); >>>> + if (copy_from_user(&vnet_hash, argp, len)) { >>>> + ret = -EFAULT; >>>> + break; >>>> + } >>>> + >>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >>>> + !tun_is_little_endian(tun))) || >>>> + vnet_hash.indirection_table_mask >= >>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >>>> + ret = -EINVAL; >>>> + break; >>>> + } >>>> + >>>> + argp = (u8 __user *)argp + len; >>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; >>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >>>> + ret = -EFAULT; >>>> + break; >>>> + } >>>> + >>>> + argp = (u8 __user *)argp + len; >>>> + len = virtio_net_hash_key_length(vnet_hash.types); >>>> + >>>> + if (copy_from_user(vnet_hash_key, argp, len)) { >>>> + ret = -EFAULT; >>>> + break; >>>> + } >>> >>> Probably easier and less error-prone to define a fixed size control >>> struct with the max indirection table size. >> >> I made its size variable because the indirection table and key may grow >> in the future as I wrote above. >> >>> >>> Btw: please trim the CC: list considerably on future patches. >> >> I'll do so in the next version with the TUNSETSTEERINGEBPF change you >> proposed. > > To be clear: please don't just resubmit with that one change. > > The skb and cb issues are quite fundamental issues that need to be resolved. > > I'd like to understand why adjusting the existing BPF feature for this > exact purpose cannot be amended to return the key it produced. eBPF steering program is not designed for this particular problem in my understanding. It was introduced to derive hash values with an understanding of application-specific semantics of packets instead of generic IP/TCP/UDP semantics. This problem is rather different in terms that the hash derivation is strictly defined by virtio-net. I don't think it makes sense to introduce the complexity of BPF when you always run the same code. It can utilize the existing flow dissector and also make it easier to use for the userspace by implementing this in the kernel. > > As you point out, the C flow dissector is insufficient. The BPF flow > dissector does not have this problem. The same argument would go for > the pre-existing BPF steering program. It is possible to extend the C flow dissector just as it is possible to implement a BPF flow dissector. The more serious problem is that virtio-net specification (and Microsoft RSS it follows) does not tell how to implement IPv6 extension support.
On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/09 5:08, Willem de Bruijn wrote: > > On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> On 2023/10/09 4:07, Willem de Bruijn wrote: > >>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>> However, computing the hash after the queue was chosen defeats the > >>>> purpose of RSS. > >>>> > >>>> Another approach is to use eBPF steering program. This approach has > >>>> another downside: it cannot report the calculated hash due to the > >>>> restrictive nature of eBPF. > >>>> > >>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>> program so that it will be able to report to the userspace, but it makes > >>>> little sense to allow to implement different hashing algorithms with > >>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>> the specification. > >>>> > >>>> The hash value already stored in sk_buff is not used and computed > >>>> independently since it may have been computed in a way not conformant > >>>> with the specification. > >>>> > >>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>>> --- > >>> > >>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > >>>> + .max_indirection_table_length = > >>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > >>>> + > >>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > >>>> +}; > >>> > >>> No need to have explicit capabilities exchange like this? Tun either > >>> supports all or none. > >> > >> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, > >> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. > >> > >> It is because the flow dissector does not support IPv6 extensions. The > >> specification is also vague, and does not tell how many TLVs should be > >> consumed at most when interpreting destination option header so I chose > >> to avoid adding code for these hash types to the flow dissector. I doubt > >> anyone will complain about it since nobody complains for Linux. > >> > >> I'm also adding this so that we can extend it later. > >> max_indirection_table_length may grow for systems with 128+ CPUs, or > >> types may have other bits for new protocols in the future. > >> > >>> > >>>> case TUNSETSTEERINGEBPF: > >>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>> + if (IS_ERR(bpf_ret)) > >>>> + ret = PTR_ERR(bpf_ret); > >>>> + else if (bpf_ret) > >>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > >>> > >>> Don't make one feature disable another. > >>> > >>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > >>> functions. If one is enabled the other call should fail, with EBUSY > >>> for instance. > >>> > >>>> + case TUNSETVNETHASH: > >>>> + len = sizeof(vnet_hash); > >>>> + if (copy_from_user(&vnet_hash, argp, len)) { > >>>> + ret = -EFAULT; > >>>> + break; > >>>> + } > >>>> + > >>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > >>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > >>>> + !tun_is_little_endian(tun))) || > >>>> + vnet_hash.indirection_table_mask >= > >>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > >>>> + ret = -EINVAL; > >>>> + break; > >>>> + } > >>>> + > >>>> + argp = (u8 __user *)argp + len; > >>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; > >>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > >>>> + ret = -EFAULT; > >>>> + break; > >>>> + } > >>>> + > >>>> + argp = (u8 __user *)argp + len; > >>>> + len = virtio_net_hash_key_length(vnet_hash.types); > >>>> + > >>>> + if (copy_from_user(vnet_hash_key, argp, len)) { > >>>> + ret = -EFAULT; > >>>> + break; > >>>> + } > >>> > >>> Probably easier and less error-prone to define a fixed size control > >>> struct with the max indirection table size. > >> > >> I made its size variable because the indirection table and key may grow > >> in the future as I wrote above. > >> > >>> > >>> Btw: please trim the CC: list considerably on future patches. > >> > >> I'll do so in the next version with the TUNSETSTEERINGEBPF change you > >> proposed. > > > > To be clear: please don't just resubmit with that one change. > > > > The skb and cb issues are quite fundamental issues that need to be resolved. > > > > I'd like to understand why adjusting the existing BPF feature for this > > exact purpose cannot be amended to return the key it produced. > > eBPF steering program is not designed for this particular problem in my > understanding. It was introduced to derive hash values with an > understanding of application-specific semantics of packets instead of > generic IP/TCP/UDP semantics. > > This problem is rather different in terms that the hash derivation is > strictly defined by virtio-net. I don't think it makes sense to > introduce the complexity of BPF when you always run the same code. > > It can utilize the existing flow dissector and also make it easier to > use for the userspace by implementing this in the kernel. Ok. There does appear to be overlap in functionality. But it might be easier to deploy to just have standard Toeplitz available without having to compile and load an eBPF program. As for the sk_buff and cb[] changes. The first is really not needed. sk_buff simply would not scale if every edge case needs a few bits. For the control block, generally it is not safe to use that across layers. In this case, between qdisc enqueue of a given device and ndo_start_xmit of that device, I suppose it is. Though uncommon. I wonder if there is any precedent. The data will have to be stored in the skb somewhere. A simpler option is just skb->hash? This code would use skb_get_hash, if it would always produce a Toeplitz hash, anyway. > > > > As you point out, the C flow dissector is insufficient. The BPF flow > > dissector does not have this problem. The same argument would go for > > the pre-existing BPF steering program. > It is possible to extend the C flow dissector just as it is possible to > implement a BPF flow dissector. The more serious problem is that > virtio-net specification (and Microsoft RSS it follows) does not tell > how to implement IPv6 extension support.
On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > virtio-net have two usage of hashes: one is RSS and another is hash > reporting. Conventionally the hash calculation was done by the VMM. > However, computing the hash after the queue was chosen defeats the > purpose of RSS. > > Another approach is to use eBPF steering program. This approach has > another downside: it cannot report the calculated hash due to the > restrictive nature of eBPF. > > Introduce the code to compute hashes to the kernel in order to overcome > thse challenges. An alternative solution is to extend the eBPF steering > program so that it will be able to report to the userspace, but it makes > little sense to allow to implement different hashing algorithms with > eBPF since the hash value reported by virtio-net is strictly defined by > the specification. > > The hash value already stored in sk_buff is not used and computed > independently since it may have been computed in a way not conformant > with the specification. > > Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, > } > > if (vnet_hdr_sz) { > - struct virtio_net_hdr gso; > + union { > + struct virtio_net_hdr hdr; > + struct virtio_net_hdr_v1_hash v1_hash_hdr; > + } hdr; > + int ret; > > if (iov_iter_count(iter) < vnet_hdr_sz) > return -EINVAL; > > - if (virtio_net_hdr_from_skb(skb, &gso, > - tun_is_little_endian(tun), true, > - vlan_hlen)) { > + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && > + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && > + skb->tun_vnet_hash) { Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of the set hash ioctl failing otherwise? Such checks should be limited to control path where possible > + vnet_hdr_content_sz = sizeof(hdr.v1_hash_hdr); > + ret = virtio_net_hdr_v1_hash_from_skb(skb, > + &hdr.v1_hash_hdr, > + true, > + vlan_hlen, > + &vnet_hash); > + } else { > + vnet_hdr_content_sz = sizeof(hdr.hdr); > + ret = virtio_net_hdr_from_skb(skb, &hdr.hdr, > + tun_is_little_endian(tun), > + true, vlan_hlen); > + } > +
On 2023/10/09 17:13, Willem de Bruijn wrote: > On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> virtio-net have two usage of hashes: one is RSS and another is hash >> reporting. Conventionally the hash calculation was done by the VMM. >> However, computing the hash after the queue was chosen defeats the >> purpose of RSS. >> >> Another approach is to use eBPF steering program. This approach has >> another downside: it cannot report the calculated hash due to the >> restrictive nature of eBPF. >> >> Introduce the code to compute hashes to the kernel in order to overcome >> thse challenges. An alternative solution is to extend the eBPF steering >> program so that it will be able to report to the userspace, but it makes >> little sense to allow to implement different hashing algorithms with >> eBPF since the hash value reported by virtio-net is strictly defined by >> the specification. >> >> The hash value already stored in sk_buff is not used and computed >> independently since it may have been computed in a way not conformant >> with the specification. >> >> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >> @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, >> } >> >> if (vnet_hdr_sz) { >> - struct virtio_net_hdr gso; >> + union { >> + struct virtio_net_hdr hdr; >> + struct virtio_net_hdr_v1_hash v1_hash_hdr; >> + } hdr; >> + int ret; >> >> if (iov_iter_count(iter) < vnet_hdr_sz) >> return -EINVAL; >> >> - if (virtio_net_hdr_from_skb(skb, &gso, >> - tun_is_little_endian(tun), true, >> - vlan_hlen)) { >> + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && >> + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && >> + skb->tun_vnet_hash) { > > Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of > the set hash ioctl failing otherwise? > > Such checks should be limited to control path where possible There is a potential race since tun->vnet_hash.flags and vnet_hdr_sz are not read at once.
On 2023/10/09 17:04, Willem de Bruijn wrote: > On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/09 5:08, Willem de Bruijn wrote: >>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> On 2023/10/09 4:07, Willem de Bruijn wrote: >>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>> >>>>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>>>> reporting. Conventionally the hash calculation was done by the VMM. >>>>>> However, computing the hash after the queue was chosen defeats the >>>>>> purpose of RSS. >>>>>> >>>>>> Another approach is to use eBPF steering program. This approach has >>>>>> another downside: it cannot report the calculated hash due to the >>>>>> restrictive nature of eBPF. >>>>>> >>>>>> Introduce the code to compute hashes to the kernel in order to overcome >>>>>> thse challenges. An alternative solution is to extend the eBPF steering >>>>>> program so that it will be able to report to the userspace, but it makes >>>>>> little sense to allow to implement different hashing algorithms with >>>>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>>>> the specification. >>>>>> >>>>>> The hash value already stored in sk_buff is not used and computed >>>>>> independently since it may have been computed in a way not conformant >>>>>> with the specification. >>>>>> >>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>>>> --- >>>>> >>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >>>>>> + .max_indirection_table_length = >>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >>>>>> + >>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >>>>>> +}; >>>>> >>>>> No need to have explicit capabilities exchange like this? Tun either >>>>> supports all or none. >>>> >>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, >>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. >>>> >>>> It is because the flow dissector does not support IPv6 extensions. The >>>> specification is also vague, and does not tell how many TLVs should be >>>> consumed at most when interpreting destination option header so I chose >>>> to avoid adding code for these hash types to the flow dissector. I doubt >>>> anyone will complain about it since nobody complains for Linux. >>>> >>>> I'm also adding this so that we can extend it later. >>>> max_indirection_table_length may grow for systems with 128+ CPUs, or >>>> types may have other bits for new protocols in the future. >>>> >>>>> >>>>>> case TUNSETSTEERINGEBPF: >>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>> + if (IS_ERR(bpf_ret)) >>>>>> + ret = PTR_ERR(bpf_ret); >>>>>> + else if (bpf_ret) >>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; >>>>> >>>>> Don't make one feature disable another. >>>>> >>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive >>>>> functions. If one is enabled the other call should fail, with EBUSY >>>>> for instance. >>>>> >>>>>> + case TUNSETVNETHASH: >>>>>> + len = sizeof(vnet_hash); >>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { >>>>>> + ret = -EFAULT; >>>>>> + break; >>>>>> + } >>>>>> + >>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >>>>>> + !tun_is_little_endian(tun))) || >>>>>> + vnet_hash.indirection_table_mask >= >>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >>>>>> + ret = -EINVAL; >>>>>> + break; >>>>>> + } >>>>>> + >>>>>> + argp = (u8 __user *)argp + len; >>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; >>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >>>>>> + ret = -EFAULT; >>>>>> + break; >>>>>> + } >>>>>> + >>>>>> + argp = (u8 __user *)argp + len; >>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); >>>>>> + >>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { >>>>>> + ret = -EFAULT; >>>>>> + break; >>>>>> + } >>>>> >>>>> Probably easier and less error-prone to define a fixed size control >>>>> struct with the max indirection table size. >>>> >>>> I made its size variable because the indirection table and key may grow >>>> in the future as I wrote above. >>>> >>>>> >>>>> Btw: please trim the CC: list considerably on future patches. >>>> >>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you >>>> proposed. >>> >>> To be clear: please don't just resubmit with that one change. >>> >>> The skb and cb issues are quite fundamental issues that need to be resolved. >>> >>> I'd like to understand why adjusting the existing BPF feature for this >>> exact purpose cannot be amended to return the key it produced. >> >> eBPF steering program is not designed for this particular problem in my >> understanding. It was introduced to derive hash values with an >> understanding of application-specific semantics of packets instead of >> generic IP/TCP/UDP semantics. >> >> This problem is rather different in terms that the hash derivation is >> strictly defined by virtio-net. I don't think it makes sense to >> introduce the complexity of BPF when you always run the same code. >> >> It can utilize the existing flow dissector and also make it easier to >> use for the userspace by implementing this in the kernel. > > Ok. There does appear to be overlap in functionality. But it might be > easier to deploy to just have standard Toeplitz available without > having to compile and load an eBPF program. > > As for the sk_buff and cb[] changes. The first is really not needed. > sk_buff simply would not scale if every edge case needs a few bits. An alternative is to move the bit to cb[] and clear it for every code paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. I think we can put the bit in sk_buff for now. We can implement the alternative when we are short of bits. > > For the control block, generally it is not safe to use that across > layers. In this case, between qdisc enqueue of a given device and > ndo_start_xmit of that device, I suppose it is. Though uncommon. I > wonder if there is any precedent. That's one of the reasons modifying qdisc_skb_cb instead of creating another struct embedding it as bpf_skb_data_end and tc_skb_cb do. This clarifies that it is qdisc's responsibility to keep these data intact. > > The data will have to be stored in the skb somewhere. A simpler option > is just skb->hash? This code would use skb_get_hash, if it would > always produce a Toeplitz hash, anyway. We still need tun_vnet_hash_report to identify hash type. skb_get_hash() uses Siphash instead of Toeplitz, and the configuration of Toeplitz relies on tun's internal state. It is still possible to store a hash calculated with tun's own logic to skb->hash, but someone may later overwrite it with __skb_get_hash() though it's unlikely.
On Mon, Oct 9, 2023 at 3:44 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/09 17:13, Willem de Bruijn wrote: > > On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> virtio-net have two usage of hashes: one is RSS and another is hash > >> reporting. Conventionally the hash calculation was done by the VMM. > >> However, computing the hash after the queue was chosen defeats the > >> purpose of RSS. > >> > >> Another approach is to use eBPF steering program. This approach has > >> another downside: it cannot report the calculated hash due to the > >> restrictive nature of eBPF. > >> > >> Introduce the code to compute hashes to the kernel in order to overcome > >> thse challenges. An alternative solution is to extend the eBPF steering > >> program so that it will be able to report to the userspace, but it makes > >> little sense to allow to implement different hashing algorithms with > >> eBPF since the hash value reported by virtio-net is strictly defined by > >> the specification. > >> > >> The hash value already stored in sk_buff is not used and computed > >> independently since it may have been computed in a way not conformant > >> with the specification. > >> > >> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > > > >> @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, > >> } > >> > >> if (vnet_hdr_sz) { > >> - struct virtio_net_hdr gso; > >> + union { > >> + struct virtio_net_hdr hdr; > >> + struct virtio_net_hdr_v1_hash v1_hash_hdr; > >> + } hdr; > >> + int ret; > >> > >> if (iov_iter_count(iter) < vnet_hdr_sz) > >> return -EINVAL; > >> > >> - if (virtio_net_hdr_from_skb(skb, &gso, > >> - tun_is_little_endian(tun), true, > >> - vlan_hlen)) { > >> + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && > >> + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && > >> + skb->tun_vnet_hash) { > > > > Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of > > the set hash ioctl failing otherwise? > > > > Such checks should be limited to control path where possible > > There is a potential race since tun->vnet_hash.flags and vnet_hdr_sz are > not read at once. It should not be possible to downgrade the hdr_sz once v1 is selected.
On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/09 17:04, Willem de Bruijn wrote: > > On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> On 2023/10/09 5:08, Willem de Bruijn wrote: > >>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> On 2023/10/09 4:07, Willem de Bruijn wrote: > >>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>> > >>>>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>>>> However, computing the hash after the queue was chosen defeats the > >>>>>> purpose of RSS. > >>>>>> > >>>>>> Another approach is to use eBPF steering program. This approach has > >>>>>> another downside: it cannot report the calculated hash due to the > >>>>>> restrictive nature of eBPF. > >>>>>> > >>>>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>>>> program so that it will be able to report to the userspace, but it makes > >>>>>> little sense to allow to implement different hashing algorithms with > >>>>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>>>> the specification. > >>>>>> > >>>>>> The hash value already stored in sk_buff is not used and computed > >>>>>> independently since it may have been computed in a way not conformant > >>>>>> with the specification. > >>>>>> > >>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>>>>> --- > >>>>> > >>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > >>>>>> + .max_indirection_table_length = > >>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > >>>>>> + > >>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > >>>>>> +}; > >>>>> > >>>>> No need to have explicit capabilities exchange like this? Tun either > >>>>> supports all or none. > >>>> > >>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, > >>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. > >>>> > >>>> It is because the flow dissector does not support IPv6 extensions. The > >>>> specification is also vague, and does not tell how many TLVs should be > >>>> consumed at most when interpreting destination option header so I chose > >>>> to avoid adding code for these hash types to the flow dissector. I doubt > >>>> anyone will complain about it since nobody complains for Linux. > >>>> > >>>> I'm also adding this so that we can extend it later. > >>>> max_indirection_table_length may grow for systems with 128+ CPUs, or > >>>> types may have other bits for new protocols in the future. > >>>> > >>>>> > >>>>>> case TUNSETSTEERINGEBPF: > >>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>> + if (IS_ERR(bpf_ret)) > >>>>>> + ret = PTR_ERR(bpf_ret); > >>>>>> + else if (bpf_ret) > >>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > >>>>> > >>>>> Don't make one feature disable another. > >>>>> > >>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > >>>>> functions. If one is enabled the other call should fail, with EBUSY > >>>>> for instance. > >>>>> > >>>>>> + case TUNSETVNETHASH: > >>>>>> + len = sizeof(vnet_hash); > >>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { > >>>>>> + ret = -EFAULT; > >>>>>> + break; > >>>>>> + } > >>>>>> + > >>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > >>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > >>>>>> + !tun_is_little_endian(tun))) || > >>>>>> + vnet_hash.indirection_table_mask >= > >>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > >>>>>> + ret = -EINVAL; > >>>>>> + break; > >>>>>> + } > >>>>>> + > >>>>>> + argp = (u8 __user *)argp + len; > >>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; > >>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > >>>>>> + ret = -EFAULT; > >>>>>> + break; > >>>>>> + } > >>>>>> + > >>>>>> + argp = (u8 __user *)argp + len; > >>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); > >>>>>> + > >>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { > >>>>>> + ret = -EFAULT; > >>>>>> + break; > >>>>>> + } > >>>>> > >>>>> Probably easier and less error-prone to define a fixed size control > >>>>> struct with the max indirection table size. > >>>> > >>>> I made its size variable because the indirection table and key may grow > >>>> in the future as I wrote above. > >>>> > >>>>> > >>>>> Btw: please trim the CC: list considerably on future patches. > >>>> > >>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you > >>>> proposed. > >>> > >>> To be clear: please don't just resubmit with that one change. > >>> > >>> The skb and cb issues are quite fundamental issues that need to be resolved. > >>> > >>> I'd like to understand why adjusting the existing BPF feature for this > >>> exact purpose cannot be amended to return the key it produced. > >> > >> eBPF steering program is not designed for this particular problem in my > >> understanding. It was introduced to derive hash values with an > >> understanding of application-specific semantics of packets instead of > >> generic IP/TCP/UDP semantics. > >> > >> This problem is rather different in terms that the hash derivation is > >> strictly defined by virtio-net. I don't think it makes sense to > >> introduce the complexity of BPF when you always run the same code. > >> > >> It can utilize the existing flow dissector and also make it easier to > >> use for the userspace by implementing this in the kernel. > > > > Ok. There does appear to be overlap in functionality. But it might be > > easier to deploy to just have standard Toeplitz available without > > having to compile and load an eBPF program. > > > > As for the sk_buff and cb[] changes. The first is really not needed. > > sk_buff simply would not scale if every edge case needs a few bits. > > An alternative is to move the bit to cb[] and clear it for every code > paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. > > I think we can put the bit in sk_buff for now. We can implement the > alternative when we are short of bits. I disagree. sk_buff fields add a cost to every code path. They cannot be added for every edge case. > > > > > For the control block, generally it is not safe to use that across > > layers. In this case, between qdisc enqueue of a given device and > > ndo_start_xmit of that device, I suppose it is. Though uncommon. I > > wonder if there is any precedent. > > That's one of the reasons modifying qdisc_skb_cb instead of creating > another struct embedding it as bpf_skb_data_end and tc_skb_cb do. This > clarifies that it is qdisc's responsibility to keep these data intact. > > > > > The data will have to be stored in the skb somewhere. A simpler option > > is just skb->hash? This code would use skb_get_hash, if it would > > always produce a Toeplitz hash, anyway. > > We still need tun_vnet_hash_report to identify hash type. > > skb_get_hash() uses Siphash instead of Toeplitz, and the configuration > of Toeplitz relies on tun's internal state. It is still possible to > store a hash calculated with tun's own logic to skb->hash, but someone > may later overwrite it with __skb_get_hash() though it's unlikely. That won't happen in this data path. Fair point on the hash type.
On 2023/10/09 18:57, Willem de Bruijn wrote: > On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/09 17:04, Willem de Bruijn wrote: >>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> On 2023/10/09 5:08, Willem de Bruijn wrote: >>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>> >>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: >>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>> >>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. >>>>>>>> However, computing the hash after the queue was chosen defeats the >>>>>>>> purpose of RSS. >>>>>>>> >>>>>>>> Another approach is to use eBPF steering program. This approach has >>>>>>>> another downside: it cannot report the calculated hash due to the >>>>>>>> restrictive nature of eBPF. >>>>>>>> >>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome >>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering >>>>>>>> program so that it will be able to report to the userspace, but it makes >>>>>>>> little sense to allow to implement different hashing algorithms with >>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>>>>>> the specification. >>>>>>>> >>>>>>>> The hash value already stored in sk_buff is not used and computed >>>>>>>> independently since it may have been computed in a way not conformant >>>>>>>> with the specification. >>>>>>>> >>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>>>>>> --- >>>>>>> >>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >>>>>>>> + .max_indirection_table_length = >>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >>>>>>>> + >>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >>>>>>>> +}; >>>>>>> >>>>>>> No need to have explicit capabilities exchange like this? Tun either >>>>>>> supports all or none. >>>>>> >>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, >>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. >>>>>> >>>>>> It is because the flow dissector does not support IPv6 extensions. The >>>>>> specification is also vague, and does not tell how many TLVs should be >>>>>> consumed at most when interpreting destination option header so I chose >>>>>> to avoid adding code for these hash types to the flow dissector. I doubt >>>>>> anyone will complain about it since nobody complains for Linux. >>>>>> >>>>>> I'm also adding this so that we can extend it later. >>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or >>>>>> types may have other bits for new protocols in the future. >>>>>> >>>>>>> >>>>>>>> case TUNSETSTEERINGEBPF: >>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>> + if (IS_ERR(bpf_ret)) >>>>>>>> + ret = PTR_ERR(bpf_ret); >>>>>>>> + else if (bpf_ret) >>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; >>>>>>> >>>>>>> Don't make one feature disable another. >>>>>>> >>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive >>>>>>> functions. If one is enabled the other call should fail, with EBUSY >>>>>>> for instance. >>>>>>> >>>>>>>> + case TUNSETVNETHASH: >>>>>>>> + len = sizeof(vnet_hash); >>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { >>>>>>>> + ret = -EFAULT; >>>>>>>> + break; >>>>>>>> + } >>>>>>>> + >>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >>>>>>>> + !tun_is_little_endian(tun))) || >>>>>>>> + vnet_hash.indirection_table_mask >= >>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >>>>>>>> + ret = -EINVAL; >>>>>>>> + break; >>>>>>>> + } >>>>>>>> + >>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; >>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >>>>>>>> + ret = -EFAULT; >>>>>>>> + break; >>>>>>>> + } >>>>>>>> + >>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); >>>>>>>> + >>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { >>>>>>>> + ret = -EFAULT; >>>>>>>> + break; >>>>>>>> + } >>>>>>> >>>>>>> Probably easier and less error-prone to define a fixed size control >>>>>>> struct with the max indirection table size. >>>>>> >>>>>> I made its size variable because the indirection table and key may grow >>>>>> in the future as I wrote above. >>>>>> >>>>>>> >>>>>>> Btw: please trim the CC: list considerably on future patches. >>>>>> >>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you >>>>>> proposed. >>>>> >>>>> To be clear: please don't just resubmit with that one change. >>>>> >>>>> The skb and cb issues are quite fundamental issues that need to be resolved. >>>>> >>>>> I'd like to understand why adjusting the existing BPF feature for this >>>>> exact purpose cannot be amended to return the key it produced. >>>> >>>> eBPF steering program is not designed for this particular problem in my >>>> understanding. It was introduced to derive hash values with an >>>> understanding of application-specific semantics of packets instead of >>>> generic IP/TCP/UDP semantics. >>>> >>>> This problem is rather different in terms that the hash derivation is >>>> strictly defined by virtio-net. I don't think it makes sense to >>>> introduce the complexity of BPF when you always run the same code. >>>> >>>> It can utilize the existing flow dissector and also make it easier to >>>> use for the userspace by implementing this in the kernel. >>> >>> Ok. There does appear to be overlap in functionality. But it might be >>> easier to deploy to just have standard Toeplitz available without >>> having to compile and load an eBPF program. >>> >>> As for the sk_buff and cb[] changes. The first is really not needed. >>> sk_buff simply would not scale if every edge case needs a few bits. >> >> An alternative is to move the bit to cb[] and clear it for every code >> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. >> >> I think we can put the bit in sk_buff for now. We can implement the >> alternative when we are short of bits. > > I disagree. sk_buff fields add a cost to every code path. They cannot > be added for every edge case. It only takes an unused bit and does not grow the sk_buff size so I think it has practically no cost for now.
On 2023/10/09 18:54, Willem de Bruijn wrote: > On Mon, Oct 9, 2023 at 3:44 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/09 17:13, Willem de Bruijn wrote: >>> On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>> reporting. Conventionally the hash calculation was done by the VMM. >>>> However, computing the hash after the queue was chosen defeats the >>>> purpose of RSS. >>>> >>>> Another approach is to use eBPF steering program. This approach has >>>> another downside: it cannot report the calculated hash due to the >>>> restrictive nature of eBPF. >>>> >>>> Introduce the code to compute hashes to the kernel in order to overcome >>>> thse challenges. An alternative solution is to extend the eBPF steering >>>> program so that it will be able to report to the userspace, but it makes >>>> little sense to allow to implement different hashing algorithms with >>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>> the specification. >>>> >>>> The hash value already stored in sk_buff is not used and computed >>>> independently since it may have been computed in a way not conformant >>>> with the specification. >>>> >>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>> >>>> @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, >>>> } >>>> >>>> if (vnet_hdr_sz) { >>>> - struct virtio_net_hdr gso; >>>> + union { >>>> + struct virtio_net_hdr hdr; >>>> + struct virtio_net_hdr_v1_hash v1_hash_hdr; >>>> + } hdr; >>>> + int ret; >>>> >>>> if (iov_iter_count(iter) < vnet_hdr_sz) >>>> return -EINVAL; >>>> >>>> - if (virtio_net_hdr_from_skb(skb, &gso, >>>> - tun_is_little_endian(tun), true, >>>> - vlan_hlen)) { >>>> + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && >>>> + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && >>>> + skb->tun_vnet_hash) { >>> >>> Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of >>> the set hash ioctl failing otherwise? >>> >>> Such checks should be limited to control path where possible >> >> There is a potential race since tun->vnet_hash.flags and vnet_hdr_sz are >> not read at once. > > It should not be possible to downgrade the hdr_sz once v1 is selected. I see nothing that prevents shrinking the header size. tun->vnet_hash.flags is read after vnet_hdr_sz so the race can happen even for the case the header size grows though this can be fixed by reordering the two reads.
On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/09 18:57, Willem de Bruijn wrote: > > On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> On 2023/10/09 17:04, Willem de Bruijn wrote: > >>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> On 2023/10/09 5:08, Willem de Bruijn wrote: > >>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>> > >>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: > >>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>> > >>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>>>>>> However, computing the hash after the queue was chosen defeats the > >>>>>>>> purpose of RSS. > >>>>>>>> > >>>>>>>> Another approach is to use eBPF steering program. This approach has > >>>>>>>> another downside: it cannot report the calculated hash due to the > >>>>>>>> restrictive nature of eBPF. > >>>>>>>> > >>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>>>>>> program so that it will be able to report to the userspace, but it makes > >>>>>>>> little sense to allow to implement different hashing algorithms with > >>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>>>>>> the specification. > >>>>>>>> > >>>>>>>> The hash value already stored in sk_buff is not used and computed > >>>>>>>> independently since it may have been computed in a way not conformant > >>>>>>>> with the specification. > >>>>>>>> > >>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>>>>>>> --- > >>>>>>> > >>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > >>>>>>>> + .max_indirection_table_length = > >>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > >>>>>>>> + > >>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > >>>>>>>> +}; > >>>>>>> > >>>>>>> No need to have explicit capabilities exchange like this? Tun either > >>>>>>> supports all or none. > >>>>>> > >>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, > >>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. > >>>>>> > >>>>>> It is because the flow dissector does not support IPv6 extensions. The > >>>>>> specification is also vague, and does not tell how many TLVs should be > >>>>>> consumed at most when interpreting destination option header so I chose > >>>>>> to avoid adding code for these hash types to the flow dissector. I doubt > >>>>>> anyone will complain about it since nobody complains for Linux. > >>>>>> > >>>>>> I'm also adding this so that we can extend it later. > >>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or > >>>>>> types may have other bits for new protocols in the future. > >>>>>> > >>>>>>> > >>>>>>>> case TUNSETSTEERINGEBPF: > >>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>> + if (IS_ERR(bpf_ret)) > >>>>>>>> + ret = PTR_ERR(bpf_ret); > >>>>>>>> + else if (bpf_ret) > >>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > >>>>>>> > >>>>>>> Don't make one feature disable another. > >>>>>>> > >>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > >>>>>>> functions. If one is enabled the other call should fail, with EBUSY > >>>>>>> for instance. > >>>>>>> > >>>>>>>> + case TUNSETVNETHASH: > >>>>>>>> + len = sizeof(vnet_hash); > >>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { > >>>>>>>> + ret = -EFAULT; > >>>>>>>> + break; > >>>>>>>> + } > >>>>>>>> + > >>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > >>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > >>>>>>>> + !tun_is_little_endian(tun))) || > >>>>>>>> + vnet_hash.indirection_table_mask >= > >>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > >>>>>>>> + ret = -EINVAL; > >>>>>>>> + break; > >>>>>>>> + } > >>>>>>>> + > >>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; > >>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > >>>>>>>> + ret = -EFAULT; > >>>>>>>> + break; > >>>>>>>> + } > >>>>>>>> + > >>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); > >>>>>>>> + > >>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { > >>>>>>>> + ret = -EFAULT; > >>>>>>>> + break; > >>>>>>>> + } > >>>>>>> > >>>>>>> Probably easier and less error-prone to define a fixed size control > >>>>>>> struct with the max indirection table size. > >>>>>> > >>>>>> I made its size variable because the indirection table and key may grow > >>>>>> in the future as I wrote above. > >>>>>> > >>>>>>> > >>>>>>> Btw: please trim the CC: list considerably on future patches. > >>>>>> > >>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you > >>>>>> proposed. > >>>>> > >>>>> To be clear: please don't just resubmit with that one change. > >>>>> > >>>>> The skb and cb issues are quite fundamental issues that need to be resolved. > >>>>> > >>>>> I'd like to understand why adjusting the existing BPF feature for this > >>>>> exact purpose cannot be amended to return the key it produced. > >>>> > >>>> eBPF steering program is not designed for this particular problem in my > >>>> understanding. It was introduced to derive hash values with an > >>>> understanding of application-specific semantics of packets instead of > >>>> generic IP/TCP/UDP semantics. > >>>> > >>>> This problem is rather different in terms that the hash derivation is > >>>> strictly defined by virtio-net. I don't think it makes sense to > >>>> introduce the complexity of BPF when you always run the same code. > >>>> > >>>> It can utilize the existing flow dissector and also make it easier to > >>>> use for the userspace by implementing this in the kernel. > >>> > >>> Ok. There does appear to be overlap in functionality. But it might be > >>> easier to deploy to just have standard Toeplitz available without > >>> having to compile and load an eBPF program. > >>> > >>> As for the sk_buff and cb[] changes. The first is really not needed. > >>> sk_buff simply would not scale if every edge case needs a few bits. > >> > >> An alternative is to move the bit to cb[] and clear it for every code > >> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. > >> > >> I think we can put the bit in sk_buff for now. We can implement the > >> alternative when we are short of bits. > > > > I disagree. sk_buff fields add a cost to every code path. They cannot > > be added for every edge case. > > It only takes an unused bit and does not grow the sk_buff size so I > think it has practically no cost for now. The problem is that that thinking leads to death by a thousand cuts. "for now" forces the cost of having to think hard how to avoid growing sk_buff onto the next person. Let's do it right from the start.
On Mon, Oct 9, 2023 at 3:05 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > > > On 2023/10/09 18:54, Willem de Bruijn wrote: > > On Mon, Oct 9, 2023 at 3:44 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> On 2023/10/09 17:13, Willem de Bruijn wrote: > >>> On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>> However, computing the hash after the queue was chosen defeats the > >>>> purpose of RSS. > >>>> > >>>> Another approach is to use eBPF steering program. This approach has > >>>> another downside: it cannot report the calculated hash due to the > >>>> restrictive nature of eBPF. > >>>> > >>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>> program so that it will be able to report to the userspace, but it makes > >>>> little sense to allow to implement different hashing algorithms with > >>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>> the specification. > >>>> > >>>> The hash value already stored in sk_buff is not used and computed > >>>> independently since it may have been computed in a way not conformant > >>>> with the specification. > >>>> > >>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>> > >>>> @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, > >>>> } > >>>> > >>>> if (vnet_hdr_sz) { > >>>> - struct virtio_net_hdr gso; > >>>> + union { > >>>> + struct virtio_net_hdr hdr; > >>>> + struct virtio_net_hdr_v1_hash v1_hash_hdr; > >>>> + } hdr; > >>>> + int ret; > >>>> > >>>> if (iov_iter_count(iter) < vnet_hdr_sz) > >>>> return -EINVAL; > >>>> > >>>> - if (virtio_net_hdr_from_skb(skb, &gso, > >>>> - tun_is_little_endian(tun), true, > >>>> - vlan_hlen)) { > >>>> + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && > >>>> + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && > >>>> + skb->tun_vnet_hash) { > >>> > >>> Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of > >>> the set hash ioctl failing otherwise? > >>> > >>> Such checks should be limited to control path where possible > >> > >> There is a potential race since tun->vnet_hash.flags and vnet_hdr_sz are > >> not read at once. > > > > It should not be possible to downgrade the hdr_sz once v1 is selected. > > I see nothing that prevents shrinking the header size. > > tun->vnet_hash.flags is read after vnet_hdr_sz so the race can happen > even for the case the header size grows though this can be fixed by > reordering the two reads. One option is to fail any control path that tries to re-negotiate header size once this hash option is enabled? There is no practical reason to allow feature re-negotiation at any arbitrary time.
On 2023/10/09 19:07, Willem de Bruijn wrote: > On Mon, Oct 9, 2023 at 3:05 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> >> >> On 2023/10/09 18:54, Willem de Bruijn wrote: >>> On Mon, Oct 9, 2023 at 3:44 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> On 2023/10/09 17:13, Willem de Bruijn wrote: >>>>> On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>> >>>>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>>>> reporting. Conventionally the hash calculation was done by the VMM. >>>>>> However, computing the hash after the queue was chosen defeats the >>>>>> purpose of RSS. >>>>>> >>>>>> Another approach is to use eBPF steering program. This approach has >>>>>> another downside: it cannot report the calculated hash due to the >>>>>> restrictive nature of eBPF. >>>>>> >>>>>> Introduce the code to compute hashes to the kernel in order to overcome >>>>>> thse challenges. An alternative solution is to extend the eBPF steering >>>>>> program so that it will be able to report to the userspace, but it makes >>>>>> little sense to allow to implement different hashing algorithms with >>>>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>>>> the specification. >>>>>> >>>>>> The hash value already stored in sk_buff is not used and computed >>>>>> independently since it may have been computed in a way not conformant >>>>>> with the specification. >>>>>> >>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>>> >>>>>> @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, >>>>>> } >>>>>> >>>>>> if (vnet_hdr_sz) { >>>>>> - struct virtio_net_hdr gso; >>>>>> + union { >>>>>> + struct virtio_net_hdr hdr; >>>>>> + struct virtio_net_hdr_v1_hash v1_hash_hdr; >>>>>> + } hdr; >>>>>> + int ret; >>>>>> >>>>>> if (iov_iter_count(iter) < vnet_hdr_sz) >>>>>> return -EINVAL; >>>>>> >>>>>> - if (virtio_net_hdr_from_skb(skb, &gso, >>>>>> - tun_is_little_endian(tun), true, >>>>>> - vlan_hlen)) { >>>>>> + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && >>>>>> + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && >>>>>> + skb->tun_vnet_hash) { >>>>> >>>>> Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of >>>>> the set hash ioctl failing otherwise? >>>>> >>>>> Such checks should be limited to control path where possible >>>> >>>> There is a potential race since tun->vnet_hash.flags and vnet_hdr_sz are >>>> not read at once. >>> >>> It should not be possible to downgrade the hdr_sz once v1 is selected. >> >> I see nothing that prevents shrinking the header size. >> >> tun->vnet_hash.flags is read after vnet_hdr_sz so the race can happen >> even for the case the header size grows though this can be fixed by >> reordering the two reads. > > One option is to fail any control path that tries to re-negotiate > header size once this hash option is enabled? > > There is no practical reason to allow feature re-negotiation at any > arbitrary time. I think it's a bit awkward interface design since tun allows to reconfigure any of its parameters, but it's certainly possible.
On 2023/10/09 19:06, Willem de Bruijn wrote: > On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/09 18:57, Willem de Bruijn wrote: >>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> On 2023/10/09 17:04, Willem de Bruijn wrote: >>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>> >>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: >>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>> >>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: >>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>> >>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. >>>>>>>>>> However, computing the hash after the queue was chosen defeats the >>>>>>>>>> purpose of RSS. >>>>>>>>>> >>>>>>>>>> Another approach is to use eBPF steering program. This approach has >>>>>>>>>> another downside: it cannot report the calculated hash due to the >>>>>>>>>> restrictive nature of eBPF. >>>>>>>>>> >>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome >>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering >>>>>>>>>> program so that it will be able to report to the userspace, but it makes >>>>>>>>>> little sense to allow to implement different hashing algorithms with >>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>>>>>>>> the specification. >>>>>>>>>> >>>>>>>>>> The hash value already stored in sk_buff is not used and computed >>>>>>>>>> independently since it may have been computed in a way not conformant >>>>>>>>>> with the specification. >>>>>>>>>> >>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>>>>>>>> --- >>>>>>>>> >>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >>>>>>>>>> + .max_indirection_table_length = >>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >>>>>>>>>> + >>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >>>>>>>>>> +}; >>>>>>>>> >>>>>>>>> No need to have explicit capabilities exchange like this? Tun either >>>>>>>>> supports all or none. >>>>>>>> >>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, >>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. >>>>>>>> >>>>>>>> It is because the flow dissector does not support IPv6 extensions. The >>>>>>>> specification is also vague, and does not tell how many TLVs should be >>>>>>>> consumed at most when interpreting destination option header so I chose >>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt >>>>>>>> anyone will complain about it since nobody complains for Linux. >>>>>>>> >>>>>>>> I'm also adding this so that we can extend it later. >>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or >>>>>>>> types may have other bits for new protocols in the future. >>>>>>>> >>>>>>>>> >>>>>>>>>> case TUNSETSTEERINGEBPF: >>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>> + if (IS_ERR(bpf_ret)) >>>>>>>>>> + ret = PTR_ERR(bpf_ret); >>>>>>>>>> + else if (bpf_ret) >>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; >>>>>>>>> >>>>>>>>> Don't make one feature disable another. >>>>>>>>> >>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive >>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY >>>>>>>>> for instance. >>>>>>>>> >>>>>>>>>> + case TUNSETVNETHASH: >>>>>>>>>> + len = sizeof(vnet_hash); >>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { >>>>>>>>>> + ret = -EFAULT; >>>>>>>>>> + break; >>>>>>>>>> + } >>>>>>>>>> + >>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >>>>>>>>>> + !tun_is_little_endian(tun))) || >>>>>>>>>> + vnet_hash.indirection_table_mask >= >>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >>>>>>>>>> + ret = -EINVAL; >>>>>>>>>> + break; >>>>>>>>>> + } >>>>>>>>>> + >>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; >>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >>>>>>>>>> + ret = -EFAULT; >>>>>>>>>> + break; >>>>>>>>>> + } >>>>>>>>>> + >>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); >>>>>>>>>> + >>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { >>>>>>>>>> + ret = -EFAULT; >>>>>>>>>> + break; >>>>>>>>>> + } >>>>>>>>> >>>>>>>>> Probably easier and less error-prone to define a fixed size control >>>>>>>>> struct with the max indirection table size. >>>>>>>> >>>>>>>> I made its size variable because the indirection table and key may grow >>>>>>>> in the future as I wrote above. >>>>>>>> >>>>>>>>> >>>>>>>>> Btw: please trim the CC: list considerably on future patches. >>>>>>>> >>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you >>>>>>>> proposed. >>>>>>> >>>>>>> To be clear: please don't just resubmit with that one change. >>>>>>> >>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. >>>>>>> >>>>>>> I'd like to understand why adjusting the existing BPF feature for this >>>>>>> exact purpose cannot be amended to return the key it produced. >>>>>> >>>>>> eBPF steering program is not designed for this particular problem in my >>>>>> understanding. It was introduced to derive hash values with an >>>>>> understanding of application-specific semantics of packets instead of >>>>>> generic IP/TCP/UDP semantics. >>>>>> >>>>>> This problem is rather different in terms that the hash derivation is >>>>>> strictly defined by virtio-net. I don't think it makes sense to >>>>>> introduce the complexity of BPF when you always run the same code. >>>>>> >>>>>> It can utilize the existing flow dissector and also make it easier to >>>>>> use for the userspace by implementing this in the kernel. >>>>> >>>>> Ok. There does appear to be overlap in functionality. But it might be >>>>> easier to deploy to just have standard Toeplitz available without >>>>> having to compile and load an eBPF program. >>>>> >>>>> As for the sk_buff and cb[] changes. The first is really not needed. >>>>> sk_buff simply would not scale if every edge case needs a few bits. >>>> >>>> An alternative is to move the bit to cb[] and clear it for every code >>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. >>>> >>>> I think we can put the bit in sk_buff for now. We can implement the >>>> alternative when we are short of bits. >>> >>> I disagree. sk_buff fields add a cost to every code path. They cannot >>> be added for every edge case. >> >> It only takes an unused bit and does not grow the sk_buff size so I >> think it has practically no cost for now. > > The problem is that that thinking leads to death by a thousand cuts. > > "for now" forces the cost of having to think hard how to avoid growing > sk_buff onto the next person. Let's do it right from the start. I see. I described an alternative to move the bit to cb[] and clear it in all code paths that leads to ndo_start_xmit() earlier. Does that sound good to you?
On Mon, Oct 9, 2023 at 3:11 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/09 19:07, Willem de Bruijn wrote: > > On Mon, Oct 9, 2023 at 3:05 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> > >> > >> On 2023/10/09 18:54, Willem de Bruijn wrote: > >>> On Mon, Oct 9, 2023 at 3:44 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> On 2023/10/09 17:13, Willem de Bruijn wrote: > >>>>> On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>> > >>>>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>>>> However, computing the hash after the queue was chosen defeats the > >>>>>> purpose of RSS. > >>>>>> > >>>>>> Another approach is to use eBPF steering program. This approach has > >>>>>> another downside: it cannot report the calculated hash due to the > >>>>>> restrictive nature of eBPF. > >>>>>> > >>>>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>>>> program so that it will be able to report to the userspace, but it makes > >>>>>> little sense to allow to implement different hashing algorithms with > >>>>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>>>> the specification. > >>>>>> > >>>>>> The hash value already stored in sk_buff is not used and computed > >>>>>> independently since it may have been computed in a way not conformant > >>>>>> with the specification. > >>>>>> > >>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>>>> > >>>>>> @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, > >>>>>> } > >>>>>> > >>>>>> if (vnet_hdr_sz) { > >>>>>> - struct virtio_net_hdr gso; > >>>>>> + union { > >>>>>> + struct virtio_net_hdr hdr; > >>>>>> + struct virtio_net_hdr_v1_hash v1_hash_hdr; > >>>>>> + } hdr; > >>>>>> + int ret; > >>>>>> > >>>>>> if (iov_iter_count(iter) < vnet_hdr_sz) > >>>>>> return -EINVAL; > >>>>>> > >>>>>> - if (virtio_net_hdr_from_skb(skb, &gso, > >>>>>> - tun_is_little_endian(tun), true, > >>>>>> - vlan_hlen)) { > >>>>>> + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && > >>>>>> + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && > >>>>>> + skb->tun_vnet_hash) { > >>>>> > >>>>> Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of > >>>>> the set hash ioctl failing otherwise? > >>>>> > >>>>> Such checks should be limited to control path where possible > >>>> > >>>> There is a potential race since tun->vnet_hash.flags and vnet_hdr_sz are > >>>> not read at once. > >>> > >>> It should not be possible to downgrade the hdr_sz once v1 is selected. > >> > >> I see nothing that prevents shrinking the header size. > >> > >> tun->vnet_hash.flags is read after vnet_hdr_sz so the race can happen > >> even for the case the header size grows though this can be fixed by > >> reordering the two reads. > > > > One option is to fail any control path that tries to re-negotiate > > header size once this hash option is enabled? > > > > There is no practical reason to allow feature re-negotiation at any > > arbitrary time. > > I think it's a bit awkward interface design since tun allows to > reconfigure any of its parameters, but it's certainly possible. If this would be the only exception to that rule, and this is the only place that needs a datapath check, then it's fine to leave as is. In general, this runtime configurability serves little purpose but to help syzbot exercise code paths no real application would attempt. But I won't ask to diverge from whatever tun already does. We just have to be more careful about the possible races it brings.
On Mon, Oct 9, 2023 at 3:12 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/09 19:06, Willem de Bruijn wrote: > > On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> On 2023/10/09 18:57, Willem de Bruijn wrote: > >>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> On 2023/10/09 17:04, Willem de Bruijn wrote: > >>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>> > >>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: > >>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>> > >>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: > >>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>> > >>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>>>>>>>> However, computing the hash after the queue was chosen defeats the > >>>>>>>>>> purpose of RSS. > >>>>>>>>>> > >>>>>>>>>> Another approach is to use eBPF steering program. This approach has > >>>>>>>>>> another downside: it cannot report the calculated hash due to the > >>>>>>>>>> restrictive nature of eBPF. > >>>>>>>>>> > >>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>>>>>>>> program so that it will be able to report to the userspace, but it makes > >>>>>>>>>> little sense to allow to implement different hashing algorithms with > >>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>>>>>>>> the specification. > >>>>>>>>>> > >>>>>>>>>> The hash value already stored in sk_buff is not used and computed > >>>>>>>>>> independently since it may have been computed in a way not conformant > >>>>>>>>>> with the specification. > >>>>>>>>>> > >>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>>>>>>>>> --- > >>>>>>>>> > >>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > >>>>>>>>>> + .max_indirection_table_length = > >>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > >>>>>>>>>> + > >>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > >>>>>>>>>> +}; > >>>>>>>>> > >>>>>>>>> No need to have explicit capabilities exchange like this? Tun either > >>>>>>>>> supports all or none. > >>>>>>>> > >>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, > >>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. > >>>>>>>> > >>>>>>>> It is because the flow dissector does not support IPv6 extensions. The > >>>>>>>> specification is also vague, and does not tell how many TLVs should be > >>>>>>>> consumed at most when interpreting destination option header so I chose > >>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt > >>>>>>>> anyone will complain about it since nobody complains for Linux. > >>>>>>>> > >>>>>>>> I'm also adding this so that we can extend it later. > >>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or > >>>>>>>> types may have other bits for new protocols in the future. > >>>>>>>> > >>>>>>>>> > >>>>>>>>>> case TUNSETSTEERINGEBPF: > >>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>>>> + if (IS_ERR(bpf_ret)) > >>>>>>>>>> + ret = PTR_ERR(bpf_ret); > >>>>>>>>>> + else if (bpf_ret) > >>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > >>>>>>>>> > >>>>>>>>> Don't make one feature disable another. > >>>>>>>>> > >>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > >>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY > >>>>>>>>> for instance. > >>>>>>>>> > >>>>>>>>>> + case TUNSETVNETHASH: > >>>>>>>>>> + len = sizeof(vnet_hash); > >>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { > >>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>> + break; > >>>>>>>>>> + } > >>>>>>>>>> + > >>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > >>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > >>>>>>>>>> + !tun_is_little_endian(tun))) || > >>>>>>>>>> + vnet_hash.indirection_table_mask >= > >>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > >>>>>>>>>> + ret = -EINVAL; > >>>>>>>>>> + break; > >>>>>>>>>> + } > >>>>>>>>>> + > >>>>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; > >>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > >>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>> + break; > >>>>>>>>>> + } > >>>>>>>>>> + > >>>>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); > >>>>>>>>>> + > >>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { > >>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>> + break; > >>>>>>>>>> + } > >>>>>>>>> > >>>>>>>>> Probably easier and less error-prone to define a fixed size control > >>>>>>>>> struct with the max indirection table size. > >>>>>>>> > >>>>>>>> I made its size variable because the indirection table and key may grow > >>>>>>>> in the future as I wrote above. > >>>>>>>> > >>>>>>>>> > >>>>>>>>> Btw: please trim the CC: list considerably on future patches. > >>>>>>>> > >>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you > >>>>>>>> proposed. > >>>>>>> > >>>>>>> To be clear: please don't just resubmit with that one change. > >>>>>>> > >>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. > >>>>>>> > >>>>>>> I'd like to understand why adjusting the existing BPF feature for this > >>>>>>> exact purpose cannot be amended to return the key it produced. > >>>>>> > >>>>>> eBPF steering program is not designed for this particular problem in my > >>>>>> understanding. It was introduced to derive hash values with an > >>>>>> understanding of application-specific semantics of packets instead of > >>>>>> generic IP/TCP/UDP semantics. > >>>>>> > >>>>>> This problem is rather different in terms that the hash derivation is > >>>>>> strictly defined by virtio-net. I don't think it makes sense to > >>>>>> introduce the complexity of BPF when you always run the same code. > >>>>>> > >>>>>> It can utilize the existing flow dissector and also make it easier to > >>>>>> use for the userspace by implementing this in the kernel. > >>>>> > >>>>> Ok. There does appear to be overlap in functionality. But it might be > >>>>> easier to deploy to just have standard Toeplitz available without > >>>>> having to compile and load an eBPF program. > >>>>> > >>>>> As for the sk_buff and cb[] changes. The first is really not needed. > >>>>> sk_buff simply would not scale if every edge case needs a few bits. > >>>> > >>>> An alternative is to move the bit to cb[] and clear it for every code > >>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. > >>>> > >>>> I think we can put the bit in sk_buff for now. We can implement the > >>>> alternative when we are short of bits. > >>> > >>> I disagree. sk_buff fields add a cost to every code path. They cannot > >>> be added for every edge case. > >> > >> It only takes an unused bit and does not grow the sk_buff size so I > >> think it has practically no cost for now. > > > > The problem is that that thinking leads to death by a thousand cuts. > > > > "for now" forces the cost of having to think hard how to avoid growing > > sk_buff onto the next person. Let's do it right from the start. > > I see. I described an alternative to move the bit to cb[] and clear it > in all code paths that leads to ndo_start_xmit() earlier. Does that > sound good to you? If you use the control block to pass information between __dev_queue_xmit on the tun device and tun_net_xmit, using gso_skb_cb, the field can be left undefined in all non-tun paths. tun_select_queue can initialize. I would still use skb->hash to encode the hash. That hash type of that field is not strictly defined. It can be siphash from ___skb_get_hash or a device hash, which most likely also uses Toeplitz. Then you also don't run into the problem of growing the struct size.
Akihiko Odaki sorry about reposts. Having an email with two "@" in the CC list: xuanzhuo@linux.alibaba.comshuah@kernel.org tripped up mutt's reply-all for me and made it send to you only. I am guessing you meant two addresses: xuanzhuo@linux.alibaba.com and shuah@kernel.org. On Sun, Oct 08, 2023 at 02:20:49PM +0900, Akihiko Odaki wrote: > virtio-net have two usage of hashes: one is RSS and another is hash > reporting. Conventionally the hash calculation was done by the VMM. > However, computing the hash after the queue was chosen defeats the > purpose of RSS. > > Another approach is to use eBPF steering program. This approach has > another downside: it cannot report the calculated hash due to the > restrictive nature of eBPF. > > Introduce the code to compute hashes to the kernel in order to overcome > thse challenges. An alternative solution is to extend the eBPF steering > program so that it will be able to report to the userspace, but it makes > little sense to allow to implement different hashing algorithms with > eBPF since the hash value reported by virtio-net is strictly defined by > the specification. > > The hash value already stored in sk_buff is not used and computed > independently since it may have been computed in a way not conformant > with the specification. > > Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > --- > drivers/net/tun.c | 187 ++++++++++++++++++++++++++++++++---- > include/uapi/linux/if_tun.h | 16 +++ > 2 files changed, 182 insertions(+), 21 deletions(-) > > diff --git a/drivers/net/tun.c b/drivers/net/tun.c > index 89ab9efe522c..561a573cd008 100644 > --- a/drivers/net/tun.c > +++ b/drivers/net/tun.c > @@ -171,6 +171,9 @@ struct tun_prog { > struct bpf_prog *prog; > }; > > +#define TUN_VNET_HASH_MAX_KEY_SIZE 40 > +#define TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH 128 > + where do these come from? > /* Since the socket were moved to tun_file, to preserve the behavior of persist > * device, socket filter, sndbuf and vnet header size were restore when the > * file were attached to a persist device. > @@ -209,6 +212,9 @@ struct tun_struct { > struct tun_prog __rcu *steering_prog; > struct tun_prog __rcu *filter_prog; > struct ethtool_link_ksettings link_ksettings; > + struct tun_vnet_hash vnet_hash; > + u16 vnet_hash_indirection_table[TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH]; > + u32 vnet_hash_key[TUN_VNET_HASH_MAX_KEY_SIZE / 4]; That's quite a lot of data to add in this struct, and will be used by a small minority of users. Are you pushing any hot data out of cache with this? Why not allocate these as needed? > /* init args */ > struct file *file; > struct ifreq *ifr; > @@ -219,6 +225,13 @@ struct veth { > __be16 h_vlan_TCI; > }; > > +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > + .max_indirection_table_length = > + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > + > + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > +}; > + > static void tun_flow_init(struct tun_struct *tun); > static void tun_flow_uninit(struct tun_struct *tun); > > @@ -320,10 +333,16 @@ static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp) > if (get_user(be, argp)) > return -EFAULT; > > - if (be) > + if (be) { > + if (!(tun->flags & TUN_VNET_LE) && > + (tun->vnet_hash.flags & TUN_VNET_HASH_REPORT)) { > + return -EINVAL; > + } > + > tun->flags |= TUN_VNET_BE; > - else > + } else { > tun->flags &= ~TUN_VNET_BE; > + } > > return 0; > } > @@ -558,15 +577,47 @@ static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb) > return ret % numqueues; > } > > +static u16 tun_vnet_select_queue(struct tun_struct *tun, struct sk_buff *skb) > +{ > + u32 value = qdisc_skb_cb(skb)->tun_vnet_hash_value; > + u32 numqueues; > + u32 index; > + u16 queue; > + > + numqueues = READ_ONCE(tun->numqueues); > + if (!numqueues) > + return 0; > + > + index = value & READ_ONCE(tun->vnet_hash.indirection_table_mask); > + queue = READ_ONCE(tun->vnet_hash_indirection_table[index]); > + if (!queue) > + queue = READ_ONCE(tun->vnet_hash.unclassified_queue); Apparently 0 is an illegal queue value? You are making this part of UAPI better document things like this. > + > + return queue % numqueues; > +} > + > static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb, > struct net_device *sb_dev) > { > struct tun_struct *tun = netdev_priv(dev); > + u8 vnet_hash_flags = READ_ONCE(tun->vnet_hash.flags); > + struct virtio_net_hash hash; > u16 ret; > > + if (vnet_hash_flags & (TUN_VNET_HASH_RSS | TUN_VNET_HASH_REPORT)) { > + virtio_net_hash(skb, READ_ONCE(tun->vnet_hash.types), > + tun->vnet_hash_key, &hash); > + What are all these READ_ONCE things doing? E.g. you seem to be very content to have tun->vnet_hash.types read twice apparently? This is volatile which basically breaks all compiler's attempts to optimize code - needs to be used judiciously. > + skb->tun_vnet_hash = true; > + qdisc_skb_cb(skb)->tun_vnet_hash_value = hash.value; > + qdisc_skb_cb(skb)->tun_vnet_hash_report = hash.report; > + } > + > rcu_read_lock(); > if (rcu_dereference(tun->steering_prog)) > ret = tun_ebpf_select_queue(tun, skb); > + else if (vnet_hash_flags & TUN_VNET_HASH_RSS) > + ret = tun_vnet_select_queue(tun, skb); > else > ret = tun_automq_select_queue(tun, skb); > rcu_read_unlock(); > @@ -2088,10 +2139,15 @@ static ssize_t tun_put_user(struct tun_struct *tun, > struct iov_iter *iter) > { > struct tun_pi pi = { 0, skb->protocol }; > + struct virtio_net_hash vnet_hash = { > + .value = qdisc_skb_cb(skb)->tun_vnet_hash_value, > + .report = qdisc_skb_cb(skb)->tun_vnet_hash_report > + }; > ssize_t total; > int vlan_offset = 0; > int vlan_hlen = 0; > int vnet_hdr_sz = 0; > + size_t vnet_hdr_content_sz; > > if (skb_vlan_tag_present(skb)) > vlan_hlen = VLAN_HLEN; > @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, > } > > if (vnet_hdr_sz) { > - struct virtio_net_hdr gso; > + union { > + struct virtio_net_hdr hdr; > + struct virtio_net_hdr_v1_hash v1_hash_hdr; > + } hdr; > + int ret; > > if (iov_iter_count(iter) < vnet_hdr_sz) > return -EINVAL; > > - if (virtio_net_hdr_from_skb(skb, &gso, > - tun_is_little_endian(tun), true, > - vlan_hlen)) { > + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && > + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && > + skb->tun_vnet_hash) { > + vnet_hdr_content_sz = sizeof(hdr.v1_hash_hdr); > + ret = virtio_net_hdr_v1_hash_from_skb(skb, > + &hdr.v1_hash_hdr, > + true, > + vlan_hlen, > + &vnet_hash); > + } else { > + vnet_hdr_content_sz = sizeof(hdr.hdr); > + ret = virtio_net_hdr_from_skb(skb, &hdr.hdr, > + tun_is_little_endian(tun), > + true, vlan_hlen); > + } > + > + if (ret) { > struct skb_shared_info *sinfo = skb_shinfo(skb); > pr_err("unexpected GSO type: " > "0x%x, gso_size %d, hdr_len %d\n", > - sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size), > - tun16_to_cpu(tun, gso.hdr_len)); > + sinfo->gso_type, tun16_to_cpu(tun, hdr.hdr.gso_size), > + tun16_to_cpu(tun, hdr.hdr.hdr_len)); > print_hex_dump(KERN_ERR, "tun: ", > DUMP_PREFIX_NONE, > 16, 1, skb->head, > - min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true); > + min((int)tun16_to_cpu(tun, hdr.hdr.hdr_len), 64), true); > WARN_ON_ONCE(1); > return -EINVAL; > } > > - if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso)) > + if (copy_to_iter(&hdr, vnet_hdr_content_sz, iter) != vnet_hdr_content_sz) > return -EFAULT; > > - iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso)); > + iov_iter_advance(iter, vnet_hdr_sz - vnet_hdr_content_sz); > } > > if (vlan_hlen) { > @@ -3007,24 +3081,27 @@ static int tun_set_queue(struct file *file, struct ifreq *ifr) > return ret; > } > > -static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p, > - void __user *data) > +static struct bpf_prog *tun_set_ebpf(struct tun_struct *tun, > + struct tun_prog __rcu **prog_p, > + void __user *data) > { > struct bpf_prog *prog; > int fd; > + int ret; > > if (copy_from_user(&fd, data, sizeof(fd))) > - return -EFAULT; > + return ERR_PTR(-EFAULT); > > if (fd == -1) { > prog = NULL; > } else { > prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); > if (IS_ERR(prog)) > - return PTR_ERR(prog); > + return prog; > } > > - return __tun_set_ebpf(tun, prog_p, prog); > + ret = __tun_set_ebpf(tun, prog_p, prog); > + return ret ? ERR_PTR(ret) : prog; > } > > /* Return correct value for tun->dev->addr_len based on tun->dev->type. */ > @@ -3082,6 +3159,11 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, > int le; > int ret; > bool do_notify = false; > + struct bpf_prog *bpf_ret; > + struct tun_vnet_hash vnet_hash; > + u16 vnet_hash_indirection_table[TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH]; > + u8 vnet_hash_key[TUN_VNET_HASH_MAX_KEY_SIZE]; > + size_t len; > > if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || > (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) { > @@ -3295,7 +3377,10 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, > ret = -EFAULT; > break; > } > - if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) { > + if (vnet_hdr_sz < > + (int)((tun->vnet_hash.flags & TUN_VNET_HASH_REPORT) ? > + sizeof(struct virtio_net_hdr_v1_hash) : > + sizeof(struct virtio_net_hdr))) { > ret = -EINVAL; > break; > } > @@ -3314,10 +3399,16 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, > ret = -EFAULT; > break; > } > - if (le) > + if (le) { > tun->flags |= TUN_VNET_LE; > - else > + } else { > + if (!tun_legacy_is_little_endian(tun)) { > + ret = -EINVAL; > + break; > + } > + > tun->flags &= ~TUN_VNET_LE; > + } > break; > > case TUNGETVNETBE: > @@ -3360,11 +3451,17 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, > break; > > case TUNSETSTEERINGEBPF: > - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > + if (IS_ERR(bpf_ret)) > + ret = PTR_ERR(bpf_ret); > + else if (bpf_ret) > + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; what is this doing? > break; > > case TUNSETFILTEREBPF: > - ret = tun_set_ebpf(tun, &tun->filter_prog, argp); > + bpf_ret = tun_set_ebpf(tun, &tun->filter_prog, argp); > + if (IS_ERR(bpf_ret)) > + ret = PTR_ERR(bpf_ret); > break; > > case TUNSETCARRIER: > @@ -3382,6 +3479,54 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, > ret = open_related_ns(&net->ns, get_net_ns); > break; > > + case TUNGETVNETHASHCAP: > + if (copy_to_user(argp, &tun_vnet_hash_cap, > + sizeof(tun_vnet_hash_cap))) > + ret = -EFAULT; > + break; > + > + case TUNSETVNETHASH: > + len = sizeof(vnet_hash); > + if (copy_from_user(&vnet_hash, argp, len)) { > + ret = -EFAULT; > + break; > + } what if flags has some bits set you don't know how to handle? should these be ignored as now or cause a failure? these decisions all affect uapi. > + > + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > + !tun_is_little_endian(tun))) || > + vnet_hash.indirection_table_mask >= > + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > + ret = -EINVAL; > + break; > + } Given this is later used to index within an array one has to be very careful about spectre things here, which this code isn't. > + > + argp = (u8 __user *)argp + len; > + len = (vnet_hash.indirection_table_mask + 1) * 2; comment pointer math tricks like this extensively please. > + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > + ret = -EFAULT; > + break; > + } > + > + argp = (u8 __user *)argp + len; > + len = virtio_net_hash_key_length(vnet_hash.types); > + > + if (copy_from_user(vnet_hash_key, argp, len)) { > + ret = -EFAULT; > + break; > + } > + > + tun->vnet_hash = vnet_hash; > + memcpy(tun->vnet_hash_indirection_table, > + vnet_hash_indirection_table, > + (vnet_hash.indirection_table_mask + 1) * 2); > + memcpy(tun->vnet_hash_key, vnet_hash_key, len); > + > + if (vnet_hash.flags & TUN_VNET_HASH_RSS) > + __tun_set_ebpf(tun, &tun->steering_prog, NULL); > + > + break; > + > default: > ret = -EINVAL; > break; > diff --git a/include/uapi/linux/if_tun.h b/include/uapi/linux/if_tun.h > index 287cdc81c939..dc591cd897c8 100644 > --- a/include/uapi/linux/if_tun.h > +++ b/include/uapi/linux/if_tun.h > @@ -61,6 +61,8 @@ > #define TUNSETFILTEREBPF _IOR('T', 225, int) > #define TUNSETCARRIER _IOW('T', 226, int) > #define TUNGETDEVNETNS _IO('T', 227) > +#define TUNGETVNETHASHCAP _IO('T', 228) > +#define TUNSETVNETHASH _IOW('T', 229, unsigned int) > > /* TUNSETIFF ifr flags */ > #define IFF_TUN 0x0001 > @@ -115,4 +117,18 @@ struct tun_filter { > __u8 addr[][ETH_ALEN]; > }; > > +struct tun_vnet_hash_cap { > + __u16 max_indirection_table_length; > + __u32 types; > +}; > + There's hidden padding in this struct - not good, copy will leak kernel info out. > +#define TUN_VNET_HASH_RSS 0x01 > +#define TUN_VNET_HASH_REPORT 0x02 Do you intend to add more flags down the road? How will userspace know what is supported? > +struct tun_vnet_hash { > + __u8 flags; > + __u32 types; > + __u16 indirection_table_mask; > + __u16 unclassified_queue; > +}; > + Padding here too. Best avoided. In any case, document UAPI please. > #endif /* _UAPI__IF_TUN_H */ > -- > 2.42.0
On Mon, Oct 09, 2023 at 05:44:20PM +0900, Akihiko Odaki wrote: > On 2023/10/09 17:13, Willem de Bruijn wrote: > > On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > > > > > virtio-net have two usage of hashes: one is RSS and another is hash > > > reporting. Conventionally the hash calculation was done by the VMM. > > > However, computing the hash after the queue was chosen defeats the > > > purpose of RSS. > > > > > > Another approach is to use eBPF steering program. This approach has > > > another downside: it cannot report the calculated hash due to the > > > restrictive nature of eBPF. > > > > > > Introduce the code to compute hashes to the kernel in order to overcome > > > thse challenges. An alternative solution is to extend the eBPF steering > > > program so that it will be able to report to the userspace, but it makes > > > little sense to allow to implement different hashing algorithms with > > > eBPF since the hash value reported by virtio-net is strictly defined by > > > the specification. > > > > > > The hash value already stored in sk_buff is not used and computed > > > independently since it may have been computed in a way not conformant > > > with the specification. > > > > > > Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > > > > > @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, > > > } > > > > > > if (vnet_hdr_sz) { > > > - struct virtio_net_hdr gso; > > > + union { > > > + struct virtio_net_hdr hdr; > > > + struct virtio_net_hdr_v1_hash v1_hash_hdr; > > > + } hdr; > > > + int ret; > > > > > > if (iov_iter_count(iter) < vnet_hdr_sz) > > > return -EINVAL; > > > > > > - if (virtio_net_hdr_from_skb(skb, &gso, > > > - tun_is_little_endian(tun), true, > > > - vlan_hlen)) { > > > + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && > > > + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && > > > + skb->tun_vnet_hash) { > > > > Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of > > the set hash ioctl failing otherwise? > > > > Such checks should be limited to control path where possible > > There is a potential race since tun->vnet_hash.flags and vnet_hdr_sz are not > read at once. And then it's a complete mess and you get inconsistent behaviour with packets getting sent all over the place, right? So maybe keep a pointer to this struct so it can be changed atomically then. Maybe even something with rcu I donnu.
On 2023/10/09 19:44, Willem de Bruijn wrote: > On Mon, Oct 9, 2023 at 3:12 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/09 19:06, Willem de Bruijn wrote: >>> On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> On 2023/10/09 18:57, Willem de Bruijn wrote: >>>>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>> >>>>>> On 2023/10/09 17:04, Willem de Bruijn wrote: >>>>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>> >>>>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: >>>>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>> >>>>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: >>>>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>> >>>>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. >>>>>>>>>>>> However, computing the hash after the queue was chosen defeats the >>>>>>>>>>>> purpose of RSS. >>>>>>>>>>>> >>>>>>>>>>>> Another approach is to use eBPF steering program. This approach has >>>>>>>>>>>> another downside: it cannot report the calculated hash due to the >>>>>>>>>>>> restrictive nature of eBPF. >>>>>>>>>>>> >>>>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome >>>>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering >>>>>>>>>>>> program so that it will be able to report to the userspace, but it makes >>>>>>>>>>>> little sense to allow to implement different hashing algorithms with >>>>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>>>>>>>>>> the specification. >>>>>>>>>>>> >>>>>>>>>>>> The hash value already stored in sk_buff is not used and computed >>>>>>>>>>>> independently since it may have been computed in a way not conformant >>>>>>>>>>>> with the specification. >>>>>>>>>>>> >>>>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>>>>>>>>>> --- >>>>>>>>>>> >>>>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >>>>>>>>>>>> + .max_indirection_table_length = >>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >>>>>>>>>>>> + >>>>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >>>>>>>>>>>> +}; >>>>>>>>>>> >>>>>>>>>>> No need to have explicit capabilities exchange like this? Tun either >>>>>>>>>>> supports all or none. >>>>>>>>>> >>>>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, >>>>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. >>>>>>>>>> >>>>>>>>>> It is because the flow dissector does not support IPv6 extensions. The >>>>>>>>>> specification is also vague, and does not tell how many TLVs should be >>>>>>>>>> consumed at most when interpreting destination option header so I chose >>>>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt >>>>>>>>>> anyone will complain about it since nobody complains for Linux. >>>>>>>>>> >>>>>>>>>> I'm also adding this so that we can extend it later. >>>>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or >>>>>>>>>> types may have other bits for new protocols in the future. >>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>> case TUNSETSTEERINGEBPF: >>>>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>>>> + if (IS_ERR(bpf_ret)) >>>>>>>>>>>> + ret = PTR_ERR(bpf_ret); >>>>>>>>>>>> + else if (bpf_ret) >>>>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; >>>>>>>>>>> >>>>>>>>>>> Don't make one feature disable another. >>>>>>>>>>> >>>>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive >>>>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY >>>>>>>>>>> for instance. >>>>>>>>>>> >>>>>>>>>>>> + case TUNSETVNETHASH: >>>>>>>>>>>> + len = sizeof(vnet_hash); >>>>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { >>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>> + break; >>>>>>>>>>>> + } >>>>>>>>>>>> + >>>>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >>>>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >>>>>>>>>>>> + !tun_is_little_endian(tun))) || >>>>>>>>>>>> + vnet_hash.indirection_table_mask >= >>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >>>>>>>>>>>> + ret = -EINVAL; >>>>>>>>>>>> + break; >>>>>>>>>>>> + } >>>>>>>>>>>> + >>>>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; >>>>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>> + break; >>>>>>>>>>>> + } >>>>>>>>>>>> + >>>>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); >>>>>>>>>>>> + >>>>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { >>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>> + break; >>>>>>>>>>>> + } >>>>>>>>>>> >>>>>>>>>>> Probably easier and less error-prone to define a fixed size control >>>>>>>>>>> struct with the max indirection table size. >>>>>>>>>> >>>>>>>>>> I made its size variable because the indirection table and key may grow >>>>>>>>>> in the future as I wrote above. >>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> Btw: please trim the CC: list considerably on future patches. >>>>>>>>>> >>>>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you >>>>>>>>>> proposed. >>>>>>>>> >>>>>>>>> To be clear: please don't just resubmit with that one change. >>>>>>>>> >>>>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. >>>>>>>>> >>>>>>>>> I'd like to understand why adjusting the existing BPF feature for this >>>>>>>>> exact purpose cannot be amended to return the key it produced. >>>>>>>> >>>>>>>> eBPF steering program is not designed for this particular problem in my >>>>>>>> understanding. It was introduced to derive hash values with an >>>>>>>> understanding of application-specific semantics of packets instead of >>>>>>>> generic IP/TCP/UDP semantics. >>>>>>>> >>>>>>>> This problem is rather different in terms that the hash derivation is >>>>>>>> strictly defined by virtio-net. I don't think it makes sense to >>>>>>>> introduce the complexity of BPF when you always run the same code. >>>>>>>> >>>>>>>> It can utilize the existing flow dissector and also make it easier to >>>>>>>> use for the userspace by implementing this in the kernel. >>>>>>> >>>>>>> Ok. There does appear to be overlap in functionality. But it might be >>>>>>> easier to deploy to just have standard Toeplitz available without >>>>>>> having to compile and load an eBPF program. >>>>>>> >>>>>>> As for the sk_buff and cb[] changes. The first is really not needed. >>>>>>> sk_buff simply would not scale if every edge case needs a few bits. >>>>>> >>>>>> An alternative is to move the bit to cb[] and clear it for every code >>>>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. >>>>>> >>>>>> I think we can put the bit in sk_buff for now. We can implement the >>>>>> alternative when we are short of bits. >>>>> >>>>> I disagree. sk_buff fields add a cost to every code path. They cannot >>>>> be added for every edge case. >>>> >>>> It only takes an unused bit and does not grow the sk_buff size so I >>>> think it has practically no cost for now. >>> >>> The problem is that that thinking leads to death by a thousand cuts. >>> >>> "for now" forces the cost of having to think hard how to avoid growing >>> sk_buff onto the next person. Let's do it right from the start. >> >> I see. I described an alternative to move the bit to cb[] and clear it >> in all code paths that leads to ndo_start_xmit() earlier. Does that >> sound good to you? > > If you use the control block to pass information between > __dev_queue_xmit on the tun device and tun_net_xmit, using gso_skb_cb, > the field can be left undefined in all non-tun paths. tun_select_queue > can initialize. The problem is that tun_select_queue() is not always called. netdev_core_pick_tx() ensures dev->real_num_tx_queues != 1 before calling it, but this variable may change later and result in a race condition. Another case is that XDP with predefined queue. > > I would still use skb->hash to encode the hash. That hash type of that > field is not strictly defined. It can be siphash from ___skb_get_hash > or a device hash, which most likely also uses Toeplitz. Then you also > don't run into the problem of growing the struct size. I'm concerned exactly because it's not strictly defined. Someone may decide to overwrite it later if we are not cautious enough. qdisc_skb_cb also has sufficient space to contain both of the hash value and type.
On 2023/10/09 20:38, Michael S. Tsirkin wrote: > Akihiko Odaki sorry about reposts. > Having an email with two "@" in the CC list: > xuanzhuo@linux.alibaba.comshuah@kernel.org > tripped up mutt's reply-all for me and made it send to you only. > > I am guessing you meant two addresses: xuanzhuo@linux.alibaba.com > and shuah@kernel.org. Oops, I'll send the future versions with corrected addresses (and fewer CCs). > > > On Sun, Oct 08, 2023 at 02:20:49PM +0900, Akihiko Odaki wrote: >> virtio-net have two usage of hashes: one is RSS and another is hash >> reporting. Conventionally the hash calculation was done by the VMM. >> However, computing the hash after the queue was chosen defeats the >> purpose of RSS. >> >> Another approach is to use eBPF steering program. This approach has >> another downside: it cannot report the calculated hash due to the >> restrictive nature of eBPF. >> >> Introduce the code to compute hashes to the kernel in order to overcome >> thse challenges. An alternative solution is to extend the eBPF steering >> program so that it will be able to report to the userspace, but it makes >> little sense to allow to implement different hashing algorithms with >> eBPF since the hash value reported by virtio-net is strictly defined by >> the specification. >> >> The hash value already stored in sk_buff is not used and computed >> independently since it may have been computed in a way not conformant >> with the specification. >> >> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >> --- >> drivers/net/tun.c | 187 ++++++++++++++++++++++++++++++++---- >> include/uapi/linux/if_tun.h | 16 +++ >> 2 files changed, 182 insertions(+), 21 deletions(-) >> >> diff --git a/drivers/net/tun.c b/drivers/net/tun.c >> index 89ab9efe522c..561a573cd008 100644 >> --- a/drivers/net/tun.c >> +++ b/drivers/net/tun.c >> @@ -171,6 +171,9 @@ struct tun_prog { >> struct bpf_prog *prog; >> }; >> >> +#define TUN_VNET_HASH_MAX_KEY_SIZE 40 >> +#define TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH 128 >> + > > where do these come from? TUN_VNET_HASH_MAX_KEY_SIZE is the total size of IPv6 addresses and TCP/UDP ports. TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH is from the following statement in VIRTIO Version 1.2 clause 5.1.4.1: > The device MUST set rss_max_indirection_table_length to at least 128, > if it offers VIRTIO_NET_F_RSS. > >> /* Since the socket were moved to tun_file, to preserve the behavior of persist >> * device, socket filter, sndbuf and vnet header size were restore when the >> * file were attached to a persist device. >> @@ -209,6 +212,9 @@ struct tun_struct { >> struct tun_prog __rcu *steering_prog; >> struct tun_prog __rcu *filter_prog; >> struct ethtool_link_ksettings link_ksettings; >> + struct tun_vnet_hash vnet_hash; >> + u16 vnet_hash_indirection_table[TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH]; >> + u32 vnet_hash_key[TUN_VNET_HASH_MAX_KEY_SIZE / 4]; > > That's quite a lot of data to add in this struct, and will be used > by a small minority of users. Are you pushing any hot data out of cache > with this? Why not allocate these as needed? I put this here just because it's convenient. It would be better to have a separate structure. > >> /* init args */ >> struct file *file; >> struct ifreq *ifr; >> @@ -219,6 +225,13 @@ struct veth { >> __be16 h_vlan_TCI; >> }; >> >> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >> + .max_indirection_table_length = >> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >> + >> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >> +}; >> + >> static void tun_flow_init(struct tun_struct *tun); >> static void tun_flow_uninit(struct tun_struct *tun); >> >> @@ -320,10 +333,16 @@ static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp) >> if (get_user(be, argp)) >> return -EFAULT; >> >> - if (be) >> + if (be) { >> + if (!(tun->flags & TUN_VNET_LE) && >> + (tun->vnet_hash.flags & TUN_VNET_HASH_REPORT)) { >> + return -EINVAL; >> + } >> + >> tun->flags |= TUN_VNET_BE; >> - else >> + } else { >> tun->flags &= ~TUN_VNET_BE; >> + } >> >> return 0; >> } >> @@ -558,15 +577,47 @@ static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb) >> return ret % numqueues; >> } >> >> +static u16 tun_vnet_select_queue(struct tun_struct *tun, struct sk_buff *skb) >> +{ >> + u32 value = qdisc_skb_cb(skb)->tun_vnet_hash_value; >> + u32 numqueues; >> + u32 index; >> + u16 queue; >> + >> + numqueues = READ_ONCE(tun->numqueues); >> + if (!numqueues) >> + return 0; >> + >> + index = value & READ_ONCE(tun->vnet_hash.indirection_table_mask); >> + queue = READ_ONCE(tun->vnet_hash_indirection_table[index]); >> + if (!queue) >> + queue = READ_ONCE(tun->vnet_hash.unclassified_queue); > > Apparently 0 is an illegal queue value? You are making this part > of UAPI better document things like this. Well, I don't think so. I naively copied this from QEMU's eBPF steering program but I think this should check tun_vnet_hash_report == VIRTIO_NET_HASH_REPORT_NONE. I'll fix it. > >> + >> + return queue % numqueues; >> +} >> + >> static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb, >> struct net_device *sb_dev) >> { >> struct tun_struct *tun = netdev_priv(dev); >> + u8 vnet_hash_flags = READ_ONCE(tun->vnet_hash.flags); >> + struct virtio_net_hash hash; >> u16 ret; >> >> + if (vnet_hash_flags & (TUN_VNET_HASH_RSS | TUN_VNET_HASH_REPORT)) { >> + virtio_net_hash(skb, READ_ONCE(tun->vnet_hash.types), >> + tun->vnet_hash_key, &hash); >> + > > What are all these READ_ONCE things doing? > E.g. you seem to be very content to have tun->vnet_hash.types read twice apparently? > This is volatile which basically breaks all compiler's attempts > to optimize code - needs to be used judiciously. I'll replace them with one RCU dereference once I extract them into a structure. > > > >> + skb->tun_vnet_hash = true; >> + qdisc_skb_cb(skb)->tun_vnet_hash_value = hash.value; >> + qdisc_skb_cb(skb)->tun_vnet_hash_report = hash.report; >> + } >> + >> rcu_read_lock(); >> if (rcu_dereference(tun->steering_prog)) >> ret = tun_ebpf_select_queue(tun, skb); >> + else if (vnet_hash_flags & TUN_VNET_HASH_RSS) >> + ret = tun_vnet_select_queue(tun, skb); >> else >> ret = tun_automq_select_queue(tun, skb); >> rcu_read_unlock(); >> @@ -2088,10 +2139,15 @@ static ssize_t tun_put_user(struct tun_struct *tun, >> struct iov_iter *iter) >> { >> struct tun_pi pi = { 0, skb->protocol }; >> + struct virtio_net_hash vnet_hash = { >> + .value = qdisc_skb_cb(skb)->tun_vnet_hash_value, >> + .report = qdisc_skb_cb(skb)->tun_vnet_hash_report >> + }; >> ssize_t total; >> int vlan_offset = 0; >> int vlan_hlen = 0; >> int vnet_hdr_sz = 0; >> + size_t vnet_hdr_content_sz; >> >> if (skb_vlan_tag_present(skb)) >> vlan_hlen = VLAN_HLEN; >> @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, >> } >> >> if (vnet_hdr_sz) { >> - struct virtio_net_hdr gso; >> + union { >> + struct virtio_net_hdr hdr; >> + struct virtio_net_hdr_v1_hash v1_hash_hdr; >> + } hdr; >> + int ret; >> >> if (iov_iter_count(iter) < vnet_hdr_sz) >> return -EINVAL; >> >> - if (virtio_net_hdr_from_skb(skb, &gso, >> - tun_is_little_endian(tun), true, >> - vlan_hlen)) { >> + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && >> + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && >> + skb->tun_vnet_hash) { >> + vnet_hdr_content_sz = sizeof(hdr.v1_hash_hdr); >> + ret = virtio_net_hdr_v1_hash_from_skb(skb, >> + &hdr.v1_hash_hdr, >> + true, >> + vlan_hlen, >> + &vnet_hash); >> + } else { >> + vnet_hdr_content_sz = sizeof(hdr.hdr); >> + ret = virtio_net_hdr_from_skb(skb, &hdr.hdr, >> + tun_is_little_endian(tun), >> + true, vlan_hlen); >> + } >> + >> + if (ret) { >> struct skb_shared_info *sinfo = skb_shinfo(skb); >> pr_err("unexpected GSO type: " >> "0x%x, gso_size %d, hdr_len %d\n", >> - sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size), >> - tun16_to_cpu(tun, gso.hdr_len)); >> + sinfo->gso_type, tun16_to_cpu(tun, hdr.hdr.gso_size), >> + tun16_to_cpu(tun, hdr.hdr.hdr_len)); >> print_hex_dump(KERN_ERR, "tun: ", >> DUMP_PREFIX_NONE, >> 16, 1, skb->head, >> - min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true); >> + min((int)tun16_to_cpu(tun, hdr.hdr.hdr_len), 64), true); >> WARN_ON_ONCE(1); >> return -EINVAL; >> } >> >> - if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso)) >> + if (copy_to_iter(&hdr, vnet_hdr_content_sz, iter) != vnet_hdr_content_sz) >> return -EFAULT; >> >> - iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso)); >> + iov_iter_advance(iter, vnet_hdr_sz - vnet_hdr_content_sz); >> } >> >> if (vlan_hlen) { >> @@ -3007,24 +3081,27 @@ static int tun_set_queue(struct file *file, struct ifreq *ifr) >> return ret; >> } >> >> -static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p, >> - void __user *data) >> +static struct bpf_prog *tun_set_ebpf(struct tun_struct *tun, >> + struct tun_prog __rcu **prog_p, >> + void __user *data) >> { >> struct bpf_prog *prog; >> int fd; >> + int ret; >> >> if (copy_from_user(&fd, data, sizeof(fd))) >> - return -EFAULT; >> + return ERR_PTR(-EFAULT); >> >> if (fd == -1) { >> prog = NULL; >> } else { >> prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); >> if (IS_ERR(prog)) >> - return PTR_ERR(prog); >> + return prog; >> } >> >> - return __tun_set_ebpf(tun, prog_p, prog); >> + ret = __tun_set_ebpf(tun, prog_p, prog); >> + return ret ? ERR_PTR(ret) : prog; >> } >> >> /* Return correct value for tun->dev->addr_len based on tun->dev->type. */ >> @@ -3082,6 +3159,11 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, >> int le; >> int ret; >> bool do_notify = false; >> + struct bpf_prog *bpf_ret; >> + struct tun_vnet_hash vnet_hash; >> + u16 vnet_hash_indirection_table[TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH]; >> + u8 vnet_hash_key[TUN_VNET_HASH_MAX_KEY_SIZE]; >> + size_t len; >> >> if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || >> (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) { >> @@ -3295,7 +3377,10 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, >> ret = -EFAULT; >> break; >> } >> - if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) { >> + if (vnet_hdr_sz < >> + (int)((tun->vnet_hash.flags & TUN_VNET_HASH_REPORT) ? >> + sizeof(struct virtio_net_hdr_v1_hash) : >> + sizeof(struct virtio_net_hdr))) { >> ret = -EINVAL; >> break; >> } >> @@ -3314,10 +3399,16 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, >> ret = -EFAULT; >> break; >> } >> - if (le) >> + if (le) { >> tun->flags |= TUN_VNET_LE; >> - else >> + } else { >> + if (!tun_legacy_is_little_endian(tun)) { >> + ret = -EINVAL; >> + break; >> + } >> + >> tun->flags &= ~TUN_VNET_LE; >> + } >> break; >> >> case TUNGETVNETBE: >> @@ -3360,11 +3451,17 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, >> break; >> >> case TUNSETSTEERINGEBPF: >> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >> + if (IS_ERR(bpf_ret)) >> + ret = PTR_ERR(bpf_ret); >> + else if (bpf_ret) >> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > > what is this doing? It's unsetting RSS because steering eBPF is incompatible with RSS. Willem proposed to return EBUSY instead so I'll do so in the future. > >> break; >> >> case TUNSETFILTEREBPF: >> - ret = tun_set_ebpf(tun, &tun->filter_prog, argp); >> + bpf_ret = tun_set_ebpf(tun, &tun->filter_prog, argp); >> + if (IS_ERR(bpf_ret)) >> + ret = PTR_ERR(bpf_ret); >> break; >> >> case TUNSETCARRIER: >> @@ -3382,6 +3479,54 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, >> ret = open_related_ns(&net->ns, get_net_ns); >> break; >> >> + case TUNGETVNETHASHCAP: >> + if (copy_to_user(argp, &tun_vnet_hash_cap, >> + sizeof(tun_vnet_hash_cap))) >> + ret = -EFAULT; >> + break; >> + >> + case TUNSETVNETHASH: >> + len = sizeof(vnet_hash); >> + if (copy_from_user(&vnet_hash, argp, len)) { >> + ret = -EFAULT; >> + break; >> + } > > > what if flags has some bits set you don't know how to handle? > should these be ignored as now or cause a failure? > these decisions all affect uapi. Currently they are ignored. > >> + >> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >> + !tun_is_little_endian(tun))) || >> + vnet_hash.indirection_table_mask >= >> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >> + ret = -EINVAL; >> + break; >> + } > > Given this is later used to index within an array one has to > be very careful about spectre things here, which this code isn't. I have totally forgotten that. Thanks for pointing this out. > > >> + >> + argp = (u8 __user *)argp + len; >> + len = (vnet_hash.indirection_table_mask + 1) * 2; > > comment pointer math tricks like this extensively please. Ok, I will. > >> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >> + ret = -EFAULT; >> + break; >> + } >> + >> + argp = (u8 __user *)argp + len; >> + len = virtio_net_hash_key_length(vnet_hash.types); >> + >> + if (copy_from_user(vnet_hash_key, argp, len)) { >> + ret = -EFAULT; >> + break; >> + } >> + >> + tun->vnet_hash = vnet_hash; >> + memcpy(tun->vnet_hash_indirection_table, >> + vnet_hash_indirection_table, >> + (vnet_hash.indirection_table_mask + 1) * 2); >> + memcpy(tun->vnet_hash_key, vnet_hash_key, len); >> + >> + if (vnet_hash.flags & TUN_VNET_HASH_RSS) >> + __tun_set_ebpf(tun, &tun->steering_prog, NULL); >> + >> + break; >> + >> default: >> ret = -EINVAL; >> break; >> diff --git a/include/uapi/linux/if_tun.h b/include/uapi/linux/if_tun.h >> index 287cdc81c939..dc591cd897c8 100644 >> --- a/include/uapi/linux/if_tun.h >> +++ b/include/uapi/linux/if_tun.h >> @@ -61,6 +61,8 @@ >> #define TUNSETFILTEREBPF _IOR('T', 225, int) >> #define TUNSETCARRIER _IOW('T', 226, int) >> #define TUNGETDEVNETNS _IO('T', 227) >> +#define TUNGETVNETHASHCAP _IO('T', 228) >> +#define TUNSETVNETHASH _IOW('T', 229, unsigned int) >> >> /* TUNSETIFF ifr flags */ >> #define IFF_TUN 0x0001 >> @@ -115,4 +117,18 @@ struct tun_filter { >> __u8 addr[][ETH_ALEN]; >> }; >> >> +struct tun_vnet_hash_cap { >> + __u16 max_indirection_table_length; >> + __u32 types; >> +}; >> + > > There's hidden padding in this struct - not good, copy > will leak kernel info out. I'll add an explicit padding. > > > >> +#define TUN_VNET_HASH_RSS 0x01 >> +#define TUN_VNET_HASH_REPORT 0x02 > > > Do you intend to add more flags down the road? > How will userspace know what is supported? I don't intend to do so, but perhaps it may be better to return EINVAL when an unknown flag is set. > >> +struct tun_vnet_hash { >> + __u8 flags; >> + __u32 types; >> + __u16 indirection_table_mask; >> + __u16 unclassified_queue; >> +}; >> + > > Padding here too. Best avoided. Perhaps it may be easier just to grow flags into u32. > > In any case, document UAPI please. Sure, I'll.
On 2023/10/09 20:50, Michael S. Tsirkin wrote: > On Mon, Oct 09, 2023 at 05:44:20PM +0900, Akihiko Odaki wrote: >> On 2023/10/09 17:13, Willem de Bruijn wrote: >>> On Sun, Oct 8, 2023 at 12:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>> reporting. Conventionally the hash calculation was done by the VMM. >>>> However, computing the hash after the queue was chosen defeats the >>>> purpose of RSS. >>>> >>>> Another approach is to use eBPF steering program. This approach has >>>> another downside: it cannot report the calculated hash due to the >>>> restrictive nature of eBPF. >>>> >>>> Introduce the code to compute hashes to the kernel in order to overcome >>>> thse challenges. An alternative solution is to extend the eBPF steering >>>> program so that it will be able to report to the userspace, but it makes >>>> little sense to allow to implement different hashing algorithms with >>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>> the specification. >>>> >>>> The hash value already stored in sk_buff is not used and computed >>>> independently since it may have been computed in a way not conformant >>>> with the specification. >>>> >>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>> >>>> @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, >>>> } >>>> >>>> if (vnet_hdr_sz) { >>>> - struct virtio_net_hdr gso; >>>> + union { >>>> + struct virtio_net_hdr hdr; >>>> + struct virtio_net_hdr_v1_hash v1_hash_hdr; >>>> + } hdr; >>>> + int ret; >>>> >>>> if (iov_iter_count(iter) < vnet_hdr_sz) >>>> return -EINVAL; >>>> >>>> - if (virtio_net_hdr_from_skb(skb, &gso, >>>> - tun_is_little_endian(tun), true, >>>> - vlan_hlen)) { >>>> + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && >>>> + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && >>>> + skb->tun_vnet_hash) { >>> >>> Isn't vnet_hdr_sz guaranteed to be >= hdr.v1_hash_hdr, by virtue of >>> the set hash ioctl failing otherwise? >>> >>> Such checks should be limited to control path where possible >> >> There is a potential race since tun->vnet_hash.flags and vnet_hdr_sz are not >> read at once. > > And then it's a complete mess and you get inconsistent > behaviour with packets getting sent all over the place, right? > So maybe keep a pointer to this struct so it can be > changed atomically then. Maybe even something with rcu I donnu. I think it's a good idea to use RCU for the vnet_hash members, but vnet_hdr_sz is something not specific to vnet_hash so this check will be still necessary.
On Tue, Oct 10, 2023 at 9:52 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/09 19:44, Willem de Bruijn wrote: > > On Mon, Oct 9, 2023 at 3:12 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> On 2023/10/09 19:06, Willem de Bruijn wrote: > >>> On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> On 2023/10/09 18:57, Willem de Bruijn wrote: > >>>>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>> > >>>>>> On 2023/10/09 17:04, Willem de Bruijn wrote: > >>>>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>> > >>>>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: > >>>>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>> > >>>>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: > >>>>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>>>> > >>>>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>>>>>>>>>> However, computing the hash after the queue was chosen defeats the > >>>>>>>>>>>> purpose of RSS. > >>>>>>>>>>>> > >>>>>>>>>>>> Another approach is to use eBPF steering program. This approach has > >>>>>>>>>>>> another downside: it cannot report the calculated hash due to the > >>>>>>>>>>>> restrictive nature of eBPF. > >>>>>>>>>>>> > >>>>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>>>>>>>>>> program so that it will be able to report to the userspace, but it makes > >>>>>>>>>>>> little sense to allow to implement different hashing algorithms with > >>>>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>>>>>>>>>> the specification. > >>>>>>>>>>>> > >>>>>>>>>>>> The hash value already stored in sk_buff is not used and computed > >>>>>>>>>>>> independently since it may have been computed in a way not conformant > >>>>>>>>>>>> with the specification. > >>>>>>>>>>>> > >>>>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>>>>>>>>>>> --- > >>>>>>>>>>> > >>>>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > >>>>>>>>>>>> + .max_indirection_table_length = > >>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > >>>>>>>>>>>> + > >>>>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > >>>>>>>>>>>> +}; > >>>>>>>>>>> > >>>>>>>>>>> No need to have explicit capabilities exchange like this? Tun either > >>>>>>>>>>> supports all or none. > >>>>>>>>>> > >>>>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, > >>>>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. > >>>>>>>>>> > >>>>>>>>>> It is because the flow dissector does not support IPv6 extensions. The > >>>>>>>>>> specification is also vague, and does not tell how many TLVs should be > >>>>>>>>>> consumed at most when interpreting destination option header so I chose > >>>>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt > >>>>>>>>>> anyone will complain about it since nobody complains for Linux. > >>>>>>>>>> > >>>>>>>>>> I'm also adding this so that we can extend it later. > >>>>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or > >>>>>>>>>> types may have other bits for new protocols in the future. > >>>>>>>>>> > >>>>>>>>>>> > >>>>>>>>>>>> case TUNSETSTEERINGEBPF: > >>>>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>>>>>> + if (IS_ERR(bpf_ret)) > >>>>>>>>>>>> + ret = PTR_ERR(bpf_ret); > >>>>>>>>>>>> + else if (bpf_ret) > >>>>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > >>>>>>>>>>> > >>>>>>>>>>> Don't make one feature disable another. > >>>>>>>>>>> > >>>>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > >>>>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY > >>>>>>>>>>> for instance. > >>>>>>>>>>> > >>>>>>>>>>>> + case TUNSETVNETHASH: > >>>>>>>>>>>> + len = sizeof(vnet_hash); > >>>>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { > >>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>> + break; > >>>>>>>>>>>> + } > >>>>>>>>>>>> + > >>>>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > >>>>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > >>>>>>>>>>>> + !tun_is_little_endian(tun))) || > >>>>>>>>>>>> + vnet_hash.indirection_table_mask >= > >>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > >>>>>>>>>>>> + ret = -EINVAL; > >>>>>>>>>>>> + break; > >>>>>>>>>>>> + } > >>>>>>>>>>>> + > >>>>>>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; > >>>>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > >>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>> + break; > >>>>>>>>>>>> + } > >>>>>>>>>>>> + > >>>>>>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); > >>>>>>>>>>>> + > >>>>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { > >>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>> + break; > >>>>>>>>>>>> + } > >>>>>>>>>>> > >>>>>>>>>>> Probably easier and less error-prone to define a fixed size control > >>>>>>>>>>> struct with the max indirection table size. > >>>>>>>>>> > >>>>>>>>>> I made its size variable because the indirection table and key may grow > >>>>>>>>>> in the future as I wrote above. > >>>>>>>>>> > >>>>>>>>>>> > >>>>>>>>>>> Btw: please trim the CC: list considerably on future patches. > >>>>>>>>>> > >>>>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you > >>>>>>>>>> proposed. > >>>>>>>>> > >>>>>>>>> To be clear: please don't just resubmit with that one change. > >>>>>>>>> > >>>>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. > >>>>>>>>> > >>>>>>>>> I'd like to understand why adjusting the existing BPF feature for this > >>>>>>>>> exact purpose cannot be amended to return the key it produced. > >>>>>>>> > >>>>>>>> eBPF steering program is not designed for this particular problem in my > >>>>>>>> understanding. It was introduced to derive hash values with an > >>>>>>>> understanding of application-specific semantics of packets instead of > >>>>>>>> generic IP/TCP/UDP semantics. > >>>>>>>> > >>>>>>>> This problem is rather different in terms that the hash derivation is > >>>>>>>> strictly defined by virtio-net. I don't think it makes sense to > >>>>>>>> introduce the complexity of BPF when you always run the same code. > >>>>>>>> > >>>>>>>> It can utilize the existing flow dissector and also make it easier to > >>>>>>>> use for the userspace by implementing this in the kernel. > >>>>>>> > >>>>>>> Ok. There does appear to be overlap in functionality. But it might be > >>>>>>> easier to deploy to just have standard Toeplitz available without > >>>>>>> having to compile and load an eBPF program. > >>>>>>> > >>>>>>> As for the sk_buff and cb[] changes. The first is really not needed. > >>>>>>> sk_buff simply would not scale if every edge case needs a few bits. > >>>>>> > >>>>>> An alternative is to move the bit to cb[] and clear it for every code > >>>>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. > >>>>>> > >>>>>> I think we can put the bit in sk_buff for now. We can implement the > >>>>>> alternative when we are short of bits. > >>>>> > >>>>> I disagree. sk_buff fields add a cost to every code path. They cannot > >>>>> be added for every edge case. > >>>> > >>>> It only takes an unused bit and does not grow the sk_buff size so I > >>>> think it has practically no cost for now. > >>> > >>> The problem is that that thinking leads to death by a thousand cuts. > >>> > >>> "for now" forces the cost of having to think hard how to avoid growing > >>> sk_buff onto the next person. Let's do it right from the start. > >> > >> I see. I described an alternative to move the bit to cb[] and clear it > >> in all code paths that leads to ndo_start_xmit() earlier. Does that > >> sound good to you? > > > > If you use the control block to pass information between > > __dev_queue_xmit on the tun device and tun_net_xmit, using gso_skb_cb, > > the field can be left undefined in all non-tun paths. tun_select_queue > > can initialize. > > The problem is that tun_select_queue() is not always called. > netdev_core_pick_tx() ensures dev->real_num_tx_queues != 1 before > calling it, but this variable may change later and result in a race > condition. Another case is that XDP with predefined queue. > > > > > I would still use skb->hash to encode the hash. That hash type of that > > field is not strictly defined. It can be siphash from ___skb_get_hash > > or a device hash, which most likely also uses Toeplitz. Then you also > > don't run into the problem of growing the struct size. > > I'm concerned exactly because it's not strictly defined. Someone may > decide to overwrite it later if we are not cautious enough. qdisc_skb_cb > also has sufficient space to contain both of the hash value and type. How about using skb extensions? Thanks >
On 2023/10/10 14:45, Jason Wang wrote: > On Tue, Oct 10, 2023 at 9:52 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/09 19:44, Willem de Bruijn wrote: >>> On Mon, Oct 9, 2023 at 3:12 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> On 2023/10/09 19:06, Willem de Bruijn wrote: >>>>> On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>> >>>>>> On 2023/10/09 18:57, Willem de Bruijn wrote: >>>>>>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>> >>>>>>>> On 2023/10/09 17:04, Willem de Bruijn wrote: >>>>>>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>> >>>>>>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: >>>>>>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>> >>>>>>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: >>>>>>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>>>> >>>>>>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>>>>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. >>>>>>>>>>>>>> However, computing the hash after the queue was chosen defeats the >>>>>>>>>>>>>> purpose of RSS. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Another approach is to use eBPF steering program. This approach has >>>>>>>>>>>>>> another downside: it cannot report the calculated hash due to the >>>>>>>>>>>>>> restrictive nature of eBPF. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome >>>>>>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering >>>>>>>>>>>>>> program so that it will be able to report to the userspace, but it makes >>>>>>>>>>>>>> little sense to allow to implement different hashing algorithms with >>>>>>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>>>>>>>>>>>> the specification. >>>>>>>>>>>>>> >>>>>>>>>>>>>> The hash value already stored in sk_buff is not used and computed >>>>>>>>>>>>>> independently since it may have been computed in a way not conformant >>>>>>>>>>>>>> with the specification. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>>>>>>>>>>>> --- >>>>>>>>>>>>> >>>>>>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >>>>>>>>>>>>>> + .max_indirection_table_length = >>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >>>>>>>>>>>>>> + >>>>>>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >>>>>>>>>>>>>> +}; >>>>>>>>>>>>> >>>>>>>>>>>>> No need to have explicit capabilities exchange like this? Tun either >>>>>>>>>>>>> supports all or none. >>>>>>>>>>>> >>>>>>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, >>>>>>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. >>>>>>>>>>>> >>>>>>>>>>>> It is because the flow dissector does not support IPv6 extensions. The >>>>>>>>>>>> specification is also vague, and does not tell how many TLVs should be >>>>>>>>>>>> consumed at most when interpreting destination option header so I chose >>>>>>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt >>>>>>>>>>>> anyone will complain about it since nobody complains for Linux. >>>>>>>>>>>> >>>>>>>>>>>> I'm also adding this so that we can extend it later. >>>>>>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or >>>>>>>>>>>> types may have other bits for new protocols in the future. >>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>>> case TUNSETSTEERINGEBPF: >>>>>>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>>>>>> + if (IS_ERR(bpf_ret)) >>>>>>>>>>>>>> + ret = PTR_ERR(bpf_ret); >>>>>>>>>>>>>> + else if (bpf_ret) >>>>>>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; >>>>>>>>>>>>> >>>>>>>>>>>>> Don't make one feature disable another. >>>>>>>>>>>>> >>>>>>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive >>>>>>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY >>>>>>>>>>>>> for instance. >>>>>>>>>>>>> >>>>>>>>>>>>>> + case TUNSETVNETHASH: >>>>>>>>>>>>>> + len = sizeof(vnet_hash); >>>>>>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { >>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>> + break; >>>>>>>>>>>>>> + } >>>>>>>>>>>>>> + >>>>>>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >>>>>>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >>>>>>>>>>>>>> + !tun_is_little_endian(tun))) || >>>>>>>>>>>>>> + vnet_hash.indirection_table_mask >= >>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >>>>>>>>>>>>>> + ret = -EINVAL; >>>>>>>>>>>>>> + break; >>>>>>>>>>>>>> + } >>>>>>>>>>>>>> + >>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; >>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>> + break; >>>>>>>>>>>>>> + } >>>>>>>>>>>>>> + >>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); >>>>>>>>>>>>>> + >>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { >>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>> + break; >>>>>>>>>>>>>> + } >>>>>>>>>>>>> >>>>>>>>>>>>> Probably easier and less error-prone to define a fixed size control >>>>>>>>>>>>> struct with the max indirection table size. >>>>>>>>>>>> >>>>>>>>>>>> I made its size variable because the indirection table and key may grow >>>>>>>>>>>> in the future as I wrote above. >>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>> Btw: please trim the CC: list considerably on future patches. >>>>>>>>>>>> >>>>>>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you >>>>>>>>>>>> proposed. >>>>>>>>>>> >>>>>>>>>>> To be clear: please don't just resubmit with that one change. >>>>>>>>>>> >>>>>>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. >>>>>>>>>>> >>>>>>>>>>> I'd like to understand why adjusting the existing BPF feature for this >>>>>>>>>>> exact purpose cannot be amended to return the key it produced. >>>>>>>>>> >>>>>>>>>> eBPF steering program is not designed for this particular problem in my >>>>>>>>>> understanding. It was introduced to derive hash values with an >>>>>>>>>> understanding of application-specific semantics of packets instead of >>>>>>>>>> generic IP/TCP/UDP semantics. >>>>>>>>>> >>>>>>>>>> This problem is rather different in terms that the hash derivation is >>>>>>>>>> strictly defined by virtio-net. I don't think it makes sense to >>>>>>>>>> introduce the complexity of BPF when you always run the same code. >>>>>>>>>> >>>>>>>>>> It can utilize the existing flow dissector and also make it easier to >>>>>>>>>> use for the userspace by implementing this in the kernel. >>>>>>>>> >>>>>>>>> Ok. There does appear to be overlap in functionality. But it might be >>>>>>>>> easier to deploy to just have standard Toeplitz available without >>>>>>>>> having to compile and load an eBPF program. >>>>>>>>> >>>>>>>>> As for the sk_buff and cb[] changes. The first is really not needed. >>>>>>>>> sk_buff simply would not scale if every edge case needs a few bits. >>>>>>>> >>>>>>>> An alternative is to move the bit to cb[] and clear it for every code >>>>>>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. >>>>>>>> >>>>>>>> I think we can put the bit in sk_buff for now. We can implement the >>>>>>>> alternative when we are short of bits. >>>>>>> >>>>>>> I disagree. sk_buff fields add a cost to every code path. They cannot >>>>>>> be added for every edge case. >>>>>> >>>>>> It only takes an unused bit and does not grow the sk_buff size so I >>>>>> think it has practically no cost for now. >>>>> >>>>> The problem is that that thinking leads to death by a thousand cuts. >>>>> >>>>> "for now" forces the cost of having to think hard how to avoid growing >>>>> sk_buff onto the next person. Let's do it right from the start. >>>> >>>> I see. I described an alternative to move the bit to cb[] and clear it >>>> in all code paths that leads to ndo_start_xmit() earlier. Does that >>>> sound good to you? >>> >>> If you use the control block to pass information between >>> __dev_queue_xmit on the tun device and tun_net_xmit, using gso_skb_cb, >>> the field can be left undefined in all non-tun paths. tun_select_queue >>> can initialize. >> >> The problem is that tun_select_queue() is not always called. >> netdev_core_pick_tx() ensures dev->real_num_tx_queues != 1 before >> calling it, but this variable may change later and result in a race >> condition. Another case is that XDP with predefined queue. >> >>> >>> I would still use skb->hash to encode the hash. That hash type of that >>> field is not strictly defined. It can be siphash from ___skb_get_hash >>> or a device hash, which most likely also uses Toeplitz. Then you also >>> don't run into the problem of growing the struct size. >> >> I'm concerned exactly because it's not strictly defined. Someone may >> decide to overwrite it later if we are not cautious enough. qdisc_skb_cb >> also has sufficient space to contain both of the hash value and type. > > How about using skb extensions? I think it will work. I'll try it in the next version.
On Tue, Oct 10, 2023 at 1:51 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/10 14:45, Jason Wang wrote: > > On Tue, Oct 10, 2023 at 9:52 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> On 2023/10/09 19:44, Willem de Bruijn wrote: > >>> On Mon, Oct 9, 2023 at 3:12 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> On 2023/10/09 19:06, Willem de Bruijn wrote: > >>>>> On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>> > >>>>>> On 2023/10/09 18:57, Willem de Bruijn wrote: > >>>>>>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>> > >>>>>>>> On 2023/10/09 17:04, Willem de Bruijn wrote: > >>>>>>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>> > >>>>>>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: > >>>>>>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>>>> > >>>>>>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: > >>>>>>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>>>>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>>>>>>>>>>>> However, computing the hash after the queue was chosen defeats the > >>>>>>>>>>>>>> purpose of RSS. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> Another approach is to use eBPF steering program. This approach has > >>>>>>>>>>>>>> another downside: it cannot report the calculated hash due to the > >>>>>>>>>>>>>> restrictive nature of eBPF. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>>>>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>>>>>>>>>>>> program so that it will be able to report to the userspace, but it makes > >>>>>>>>>>>>>> little sense to allow to implement different hashing algorithms with > >>>>>>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>>>>>>>>>>>> the specification. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> The hash value already stored in sk_buff is not used and computed > >>>>>>>>>>>>>> independently since it may have been computed in a way not conformant > >>>>>>>>>>>>>> with the specification. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>>>>>>>>>>>>> --- > >>>>>>>>>>>>> > >>>>>>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > >>>>>>>>>>>>>> + .max_indirection_table_length = > >>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > >>>>>>>>>>>>>> + > >>>>>>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > >>>>>>>>>>>>>> +}; > >>>>>>>>>>>>> > >>>>>>>>>>>>> No need to have explicit capabilities exchange like this? Tun either > >>>>>>>>>>>>> supports all or none. > >>>>>>>>>>>> > >>>>>>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, > >>>>>>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. > >>>>>>>>>>>> > >>>>>>>>>>>> It is because the flow dissector does not support IPv6 extensions. The > >>>>>>>>>>>> specification is also vague, and does not tell how many TLVs should be > >>>>>>>>>>>> consumed at most when interpreting destination option header so I chose > >>>>>>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt > >>>>>>>>>>>> anyone will complain about it since nobody complains for Linux. > >>>>>>>>>>>> > >>>>>>>>>>>> I'm also adding this so that we can extend it later. > >>>>>>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or > >>>>>>>>>>>> types may have other bits for new protocols in the future. > >>>>>>>>>>>> > >>>>>>>>>>>>> > >>>>>>>>>>>>>> case TUNSETSTEERINGEBPF: > >>>>>>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>>>>>>>> + if (IS_ERR(bpf_ret)) > >>>>>>>>>>>>>> + ret = PTR_ERR(bpf_ret); > >>>>>>>>>>>>>> + else if (bpf_ret) > >>>>>>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > >>>>>>>>>>>>> > >>>>>>>>>>>>> Don't make one feature disable another. > >>>>>>>>>>>>> > >>>>>>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > >>>>>>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY > >>>>>>>>>>>>> for instance. > >>>>>>>>>>>>> > >>>>>>>>>>>>>> + case TUNSETVNETHASH: > >>>>>>>>>>>>>> + len = sizeof(vnet_hash); > >>>>>>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { > >>>>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>>>> + break; > >>>>>>>>>>>>>> + } > >>>>>>>>>>>>>> + > >>>>>>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > >>>>>>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > >>>>>>>>>>>>>> + !tun_is_little_endian(tun))) || > >>>>>>>>>>>>>> + vnet_hash.indirection_table_mask >= > >>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > >>>>>>>>>>>>>> + ret = -EINVAL; > >>>>>>>>>>>>>> + break; > >>>>>>>>>>>>>> + } > >>>>>>>>>>>>>> + > >>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; > >>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > >>>>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>>>> + break; > >>>>>>>>>>>>>> + } > >>>>>>>>>>>>>> + > >>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); > >>>>>>>>>>>>>> + > >>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { > >>>>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>>>> + break; > >>>>>>>>>>>>>> + } > >>>>>>>>>>>>> > >>>>>>>>>>>>> Probably easier and less error-prone to define a fixed size control > >>>>>>>>>>>>> struct with the max indirection table size. > >>>>>>>>>>>> > >>>>>>>>>>>> I made its size variable because the indirection table and key may grow > >>>>>>>>>>>> in the future as I wrote above. > >>>>>>>>>>>> > >>>>>>>>>>>>> > >>>>>>>>>>>>> Btw: please trim the CC: list considerably on future patches. > >>>>>>>>>>>> > >>>>>>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you > >>>>>>>>>>>> proposed. > >>>>>>>>>>> > >>>>>>>>>>> To be clear: please don't just resubmit with that one change. > >>>>>>>>>>> > >>>>>>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. > >>>>>>>>>>> > >>>>>>>>>>> I'd like to understand why adjusting the existing BPF feature for this > >>>>>>>>>>> exact purpose cannot be amended to return the key it produced. > >>>>>>>>>> > >>>>>>>>>> eBPF steering program is not designed for this particular problem in my > >>>>>>>>>> understanding. It was introduced to derive hash values with an > >>>>>>>>>> understanding of application-specific semantics of packets instead of > >>>>>>>>>> generic IP/TCP/UDP semantics. > >>>>>>>>>> > >>>>>>>>>> This problem is rather different in terms that the hash derivation is > >>>>>>>>>> strictly defined by virtio-net. I don't think it makes sense to > >>>>>>>>>> introduce the complexity of BPF when you always run the same code. > >>>>>>>>>> > >>>>>>>>>> It can utilize the existing flow dissector and also make it easier to > >>>>>>>>>> use for the userspace by implementing this in the kernel. > >>>>>>>>> > >>>>>>>>> Ok. There does appear to be overlap in functionality. But it might be > >>>>>>>>> easier to deploy to just have standard Toeplitz available without > >>>>>>>>> having to compile and load an eBPF program. > >>>>>>>>> > >>>>>>>>> As for the sk_buff and cb[] changes. The first is really not needed. > >>>>>>>>> sk_buff simply would not scale if every edge case needs a few bits. > >>>>>>>> > >>>>>>>> An alternative is to move the bit to cb[] and clear it for every code > >>>>>>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. > >>>>>>>> > >>>>>>>> I think we can put the bit in sk_buff for now. We can implement the > >>>>>>>> alternative when we are short of bits. > >>>>>>> > >>>>>>> I disagree. sk_buff fields add a cost to every code path. They cannot > >>>>>>> be added for every edge case. > >>>>>> > >>>>>> It only takes an unused bit and does not grow the sk_buff size so I > >>>>>> think it has practically no cost for now. > >>>>> > >>>>> The problem is that that thinking leads to death by a thousand cuts. > >>>>> > >>>>> "for now" forces the cost of having to think hard how to avoid growing > >>>>> sk_buff onto the next person. Let's do it right from the start. > >>>> > >>>> I see. I described an alternative to move the bit to cb[] and clear it > >>>> in all code paths that leads to ndo_start_xmit() earlier. Does that > >>>> sound good to you? > >>> > >>> If you use the control block to pass information between > >>> __dev_queue_xmit on the tun device and tun_net_xmit, using gso_skb_cb, > >>> the field can be left undefined in all non-tun paths. tun_select_queue > >>> can initialize. > >> > >> The problem is that tun_select_queue() is not always called. > >> netdev_core_pick_tx() ensures dev->real_num_tx_queues != 1 before > >> calling it, but this variable may change later and result in a race > >> condition. Another case is that XDP with predefined queue. > >> > >>> > >>> I would still use skb->hash to encode the hash. That hash type of that > >>> field is not strictly defined. It can be siphash from ___skb_get_hash > >>> or a device hash, which most likely also uses Toeplitz. Then you also > >>> don't run into the problem of growing the struct size. > >> > >> I'm concerned exactly because it's not strictly defined. Someone may > >> decide to overwrite it later if we are not cautious enough. qdisc_skb_cb > >> also has sufficient space to contain both of the hash value and type. > > > > How about using skb extensions? > > I think it will work. I'll try it in the next version. Btw, I still think using eBPF for hash might be better. Though the hashing rule is defined in the spec, it may be extended in the future. For example, several extensions has been proposed: 1) RSS context 2) encapsulated packet hashing Thanks >
On 2023/10/10 15:00, Jason Wang wrote: > On Tue, Oct 10, 2023 at 1:51 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/10 14:45, Jason Wang wrote: >>> On Tue, Oct 10, 2023 at 9:52 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> On 2023/10/09 19:44, Willem de Bruijn wrote: >>>>> On Mon, Oct 9, 2023 at 3:12 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>> >>>>>> On 2023/10/09 19:06, Willem de Bruijn wrote: >>>>>>> On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>> >>>>>>>> On 2023/10/09 18:57, Willem de Bruijn wrote: >>>>>>>>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>> >>>>>>>>>> On 2023/10/09 17:04, Willem de Bruijn wrote: >>>>>>>>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>> >>>>>>>>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: >>>>>>>>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>>>> >>>>>>>>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: >>>>>>>>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>>>>>>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. >>>>>>>>>>>>>>>> However, computing the hash after the queue was chosen defeats the >>>>>>>>>>>>>>>> purpose of RSS. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Another approach is to use eBPF steering program. This approach has >>>>>>>>>>>>>>>> another downside: it cannot report the calculated hash due to the >>>>>>>>>>>>>>>> restrictive nature of eBPF. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome >>>>>>>>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering >>>>>>>>>>>>>>>> program so that it will be able to report to the userspace, but it makes >>>>>>>>>>>>>>>> little sense to allow to implement different hashing algorithms with >>>>>>>>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>>>>>>>>>>>>>> the specification. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> The hash value already stored in sk_buff is not used and computed >>>>>>>>>>>>>>>> independently since it may have been computed in a way not conformant >>>>>>>>>>>>>>>> with the specification. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>>>>>>>>>>>>>> --- >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >>>>>>>>>>>>>>>> + .max_indirection_table_length = >>>>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >>>>>>>>>>>>>>>> +}; >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> No need to have explicit capabilities exchange like this? Tun either >>>>>>>>>>>>>>> supports all or none. >>>>>>>>>>>>>> >>>>>>>>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, >>>>>>>>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. >>>>>>>>>>>>>> >>>>>>>>>>>>>> It is because the flow dissector does not support IPv6 extensions. The >>>>>>>>>>>>>> specification is also vague, and does not tell how many TLVs should be >>>>>>>>>>>>>> consumed at most when interpreting destination option header so I chose >>>>>>>>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt >>>>>>>>>>>>>> anyone will complain about it since nobody complains for Linux. >>>>>>>>>>>>>> >>>>>>>>>>>>>> I'm also adding this so that we can extend it later. >>>>>>>>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or >>>>>>>>>>>>>> types may have other bits for new protocols in the future. >>>>>>>>>>>>>> >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> case TUNSETSTEERINGEBPF: >>>>>>>>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>>>>>>>> + if (IS_ERR(bpf_ret)) >>>>>>>>>>>>>>>> + ret = PTR_ERR(bpf_ret); >>>>>>>>>>>>>>>> + else if (bpf_ret) >>>>>>>>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Don't make one feature disable another. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive >>>>>>>>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY >>>>>>>>>>>>>>> for instance. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> + case TUNSETVNETHASH: >>>>>>>>>>>>>>>> + len = sizeof(vnet_hash); >>>>>>>>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { >>>>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>>>> + break; >>>>>>>>>>>>>>>> + } >>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >>>>>>>>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >>>>>>>>>>>>>>>> + !tun_is_little_endian(tun))) || >>>>>>>>>>>>>>>> + vnet_hash.indirection_table_mask >= >>>>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >>>>>>>>>>>>>>>> + ret = -EINVAL; >>>>>>>>>>>>>>>> + break; >>>>>>>>>>>>>>>> + } >>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; >>>>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >>>>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>>>> + break; >>>>>>>>>>>>>>>> + } >>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); >>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { >>>>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>>>> + break; >>>>>>>>>>>>>>>> + } >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Probably easier and less error-prone to define a fixed size control >>>>>>>>>>>>>>> struct with the max indirection table size. >>>>>>>>>>>>>> >>>>>>>>>>>>>> I made its size variable because the indirection table and key may grow >>>>>>>>>>>>>> in the future as I wrote above. >>>>>>>>>>>>>> >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Btw: please trim the CC: list considerably on future patches. >>>>>>>>>>>>>> >>>>>>>>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you >>>>>>>>>>>>>> proposed. >>>>>>>>>>>>> >>>>>>>>>>>>> To be clear: please don't just resubmit with that one change. >>>>>>>>>>>>> >>>>>>>>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. >>>>>>>>>>>>> >>>>>>>>>>>>> I'd like to understand why adjusting the existing BPF feature for this >>>>>>>>>>>>> exact purpose cannot be amended to return the key it produced. >>>>>>>>>>>> >>>>>>>>>>>> eBPF steering program is not designed for this particular problem in my >>>>>>>>>>>> understanding. It was introduced to derive hash values with an >>>>>>>>>>>> understanding of application-specific semantics of packets instead of >>>>>>>>>>>> generic IP/TCP/UDP semantics. >>>>>>>>>>>> >>>>>>>>>>>> This problem is rather different in terms that the hash derivation is >>>>>>>>>>>> strictly defined by virtio-net. I don't think it makes sense to >>>>>>>>>>>> introduce the complexity of BPF when you always run the same code. >>>>>>>>>>>> >>>>>>>>>>>> It can utilize the existing flow dissector and also make it easier to >>>>>>>>>>>> use for the userspace by implementing this in the kernel. >>>>>>>>>>> >>>>>>>>>>> Ok. There does appear to be overlap in functionality. But it might be >>>>>>>>>>> easier to deploy to just have standard Toeplitz available without >>>>>>>>>>> having to compile and load an eBPF program. >>>>>>>>>>> >>>>>>>>>>> As for the sk_buff and cb[] changes. The first is really not needed. >>>>>>>>>>> sk_buff simply would not scale if every edge case needs a few bits. >>>>>>>>>> >>>>>>>>>> An alternative is to move the bit to cb[] and clear it for every code >>>>>>>>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. >>>>>>>>>> >>>>>>>>>> I think we can put the bit in sk_buff for now. We can implement the >>>>>>>>>> alternative when we are short of bits. >>>>>>>>> >>>>>>>>> I disagree. sk_buff fields add a cost to every code path. They cannot >>>>>>>>> be added for every edge case. >>>>>>>> >>>>>>>> It only takes an unused bit and does not grow the sk_buff size so I >>>>>>>> think it has practically no cost for now. >>>>>>> >>>>>>> The problem is that that thinking leads to death by a thousand cuts. >>>>>>> >>>>>>> "for now" forces the cost of having to think hard how to avoid growing >>>>>>> sk_buff onto the next person. Let's do it right from the start. >>>>>> >>>>>> I see. I described an alternative to move the bit to cb[] and clear it >>>>>> in all code paths that leads to ndo_start_xmit() earlier. Does that >>>>>> sound good to you? >>>>> >>>>> If you use the control block to pass information between >>>>> __dev_queue_xmit on the tun device and tun_net_xmit, using gso_skb_cb, >>>>> the field can be left undefined in all non-tun paths. tun_select_queue >>>>> can initialize. >>>> >>>> The problem is that tun_select_queue() is not always called. >>>> netdev_core_pick_tx() ensures dev->real_num_tx_queues != 1 before >>>> calling it, but this variable may change later and result in a race >>>> condition. Another case is that XDP with predefined queue. >>>> >>>>> >>>>> I would still use skb->hash to encode the hash. That hash type of that >>>>> field is not strictly defined. It can be siphash from ___skb_get_hash >>>>> or a device hash, which most likely also uses Toeplitz. Then you also >>>>> don't run into the problem of growing the struct size. >>>> >>>> I'm concerned exactly because it's not strictly defined. Someone may >>>> decide to overwrite it later if we are not cautious enough. qdisc_skb_cb >>>> also has sufficient space to contain both of the hash value and type. >>> >>> How about using skb extensions? >> >> I think it will work. I'll try it in the next version. > > Btw, I still think using eBPF for hash might be better. > > Though the hashing rule is defined in the spec, it may be extended in > the future. For example, several extensions has been proposed: > > 1) RSS context > 2) encapsulated packet hashing Looking at the proposals, I'm now more inclined to extend the BPF steering program. Yuri, who wrote the RFC patches to extend the BPF steering program, also raised an concern that it may become hard to implement virtio-net extensions in the future. It is much easier to deploy a new BPF program to support extensions since it will be included in QEMU and can be deployed at once without concerning other kernel stuff. I was still not sure how likely such an extension will emerge especially when the hardware RSS capability is not evolving for a decade or so. But those proposals show that there are more demands of new features for virtio-net.
On Tue, Oct 10, 2023 at 2:19 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > > On 2023/10/10 15:00, Jason Wang wrote: > > On Tue, Oct 10, 2023 at 1:51 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >> > >> On 2023/10/10 14:45, Jason Wang wrote: > >>> On Tue, Oct 10, 2023 at 9:52 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>> > >>>> On 2023/10/09 19:44, Willem de Bruijn wrote: > >>>>> On Mon, Oct 9, 2023 at 3:12 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>> > >>>>>> On 2023/10/09 19:06, Willem de Bruijn wrote: > >>>>>>> On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>> > >>>>>>>> On 2023/10/09 18:57, Willem de Bruijn wrote: > >>>>>>>>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>> > >>>>>>>>>> On 2023/10/09 17:04, Willem de Bruijn wrote: > >>>>>>>>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>>>> > >>>>>>>>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: > >>>>>>>>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: > >>>>>>>>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: > >>>>>>>>>>>>>>>> > >>>>>>>>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash > >>>>>>>>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. > >>>>>>>>>>>>>>>> However, computing the hash after the queue was chosen defeats the > >>>>>>>>>>>>>>>> purpose of RSS. > >>>>>>>>>>>>>>>> > >>>>>>>>>>>>>>>> Another approach is to use eBPF steering program. This approach has > >>>>>>>>>>>>>>>> another downside: it cannot report the calculated hash due to the > >>>>>>>>>>>>>>>> restrictive nature of eBPF. > >>>>>>>>>>>>>>>> > >>>>>>>>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome > >>>>>>>>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering > >>>>>>>>>>>>>>>> program so that it will be able to report to the userspace, but it makes > >>>>>>>>>>>>>>>> little sense to allow to implement different hashing algorithms with > >>>>>>>>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by > >>>>>>>>>>>>>>>> the specification. > >>>>>>>>>>>>>>>> > >>>>>>>>>>>>>>>> The hash value already stored in sk_buff is not used and computed > >>>>>>>>>>>>>>>> independently since it may have been computed in a way not conformant > >>>>>>>>>>>>>>>> with the specification. > >>>>>>>>>>>>>>>> > >>>>>>>>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> > >>>>>>>>>>>>>>>> --- > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { > >>>>>>>>>>>>>>>> + .max_indirection_table_length = > >>>>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, > >>>>>>>>>>>>>>>> + > >>>>>>>>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES > >>>>>>>>>>>>>>>> +}; > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>> No need to have explicit capabilities exchange like this? Tun either > >>>>>>>>>>>>>>> supports all or none. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, > >>>>>>>>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> It is because the flow dissector does not support IPv6 extensions. The > >>>>>>>>>>>>>> specification is also vague, and does not tell how many TLVs should be > >>>>>>>>>>>>>> consumed at most when interpreting destination option header so I chose > >>>>>>>>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt > >>>>>>>>>>>>>> anyone will complain about it since nobody complains for Linux. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> I'm also adding this so that we can extend it later. > >>>>>>>>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or > >>>>>>>>>>>>>> types may have other bits for new protocols in the future. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>>> case TUNSETSTEERINGEBPF: > >>>>>>>>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); > >>>>>>>>>>>>>>>> + if (IS_ERR(bpf_ret)) > >>>>>>>>>>>>>>>> + ret = PTR_ERR(bpf_ret); > >>>>>>>>>>>>>>>> + else if (bpf_ret) > >>>>>>>>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>> Don't make one feature disable another. > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive > >>>>>>>>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY > >>>>>>>>>>>>>>> for instance. > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>>> + case TUNSETVNETHASH: > >>>>>>>>>>>>>>>> + len = sizeof(vnet_hash); > >>>>>>>>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { > >>>>>>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>>>>>> + break; > >>>>>>>>>>>>>>>> + } > >>>>>>>>>>>>>>>> + > >>>>>>>>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && > >>>>>>>>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || > >>>>>>>>>>>>>>>> + !tun_is_little_endian(tun))) || > >>>>>>>>>>>>>>>> + vnet_hash.indirection_table_mask >= > >>>>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { > >>>>>>>>>>>>>>>> + ret = -EINVAL; > >>>>>>>>>>>>>>>> + break; > >>>>>>>>>>>>>>>> + } > >>>>>>>>>>>>>>>> + > >>>>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; > >>>>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { > >>>>>>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>>>>>> + break; > >>>>>>>>>>>>>>>> + } > >>>>>>>>>>>>>>>> + > >>>>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; > >>>>>>>>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); > >>>>>>>>>>>>>>>> + > >>>>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { > >>>>>>>>>>>>>>>> + ret = -EFAULT; > >>>>>>>>>>>>>>>> + break; > >>>>>>>>>>>>>>>> + } > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>> Probably easier and less error-prone to define a fixed size control > >>>>>>>>>>>>>>> struct with the max indirection table size. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> I made its size variable because the indirection table and key may grow > >>>>>>>>>>>>>> in the future as I wrote above. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>> Btw: please trim the CC: list considerably on future patches. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you > >>>>>>>>>>>>>> proposed. > >>>>>>>>>>>>> > >>>>>>>>>>>>> To be clear: please don't just resubmit with that one change. > >>>>>>>>>>>>> > >>>>>>>>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. > >>>>>>>>>>>>> > >>>>>>>>>>>>> I'd like to understand why adjusting the existing BPF feature for this > >>>>>>>>>>>>> exact purpose cannot be amended to return the key it produced. > >>>>>>>>>>>> > >>>>>>>>>>>> eBPF steering program is not designed for this particular problem in my > >>>>>>>>>>>> understanding. It was introduced to derive hash values with an > >>>>>>>>>>>> understanding of application-specific semantics of packets instead of > >>>>>>>>>>>> generic IP/TCP/UDP semantics. > >>>>>>>>>>>> > >>>>>>>>>>>> This problem is rather different in terms that the hash derivation is > >>>>>>>>>>>> strictly defined by virtio-net. I don't think it makes sense to > >>>>>>>>>>>> introduce the complexity of BPF when you always run the same code. > >>>>>>>>>>>> > >>>>>>>>>>>> It can utilize the existing flow dissector and also make it easier to > >>>>>>>>>>>> use for the userspace by implementing this in the kernel. > >>>>>>>>>>> > >>>>>>>>>>> Ok. There does appear to be overlap in functionality. But it might be > >>>>>>>>>>> easier to deploy to just have standard Toeplitz available without > >>>>>>>>>>> having to compile and load an eBPF program. > >>>>>>>>>>> > >>>>>>>>>>> As for the sk_buff and cb[] changes. The first is really not needed. > >>>>>>>>>>> sk_buff simply would not scale if every edge case needs a few bits. > >>>>>>>>>> > >>>>>>>>>> An alternative is to move the bit to cb[] and clear it for every code > >>>>>>>>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. > >>>>>>>>>> > >>>>>>>>>> I think we can put the bit in sk_buff for now. We can implement the > >>>>>>>>>> alternative when we are short of bits. > >>>>>>>>> > >>>>>>>>> I disagree. sk_buff fields add a cost to every code path. They cannot > >>>>>>>>> be added for every edge case. > >>>>>>>> > >>>>>>>> It only takes an unused bit and does not grow the sk_buff size so I > >>>>>>>> think it has practically no cost for now. > >>>>>>> > >>>>>>> The problem is that that thinking leads to death by a thousand cuts. > >>>>>>> > >>>>>>> "for now" forces the cost of having to think hard how to avoid growing > >>>>>>> sk_buff onto the next person. Let's do it right from the start. > >>>>>> > >>>>>> I see. I described an alternative to move the bit to cb[] and clear it > >>>>>> in all code paths that leads to ndo_start_xmit() earlier. Does that > >>>>>> sound good to you? > >>>>> > >>>>> If you use the control block to pass information between > >>>>> __dev_queue_xmit on the tun device and tun_net_xmit, using gso_skb_cb, > >>>>> the field can be left undefined in all non-tun paths. tun_select_queue > >>>>> can initialize. > >>>> > >>>> The problem is that tun_select_queue() is not always called. > >>>> netdev_core_pick_tx() ensures dev->real_num_tx_queues != 1 before > >>>> calling it, but this variable may change later and result in a race > >>>> condition. Another case is that XDP with predefined queue. > >>>> > >>>>> > >>>>> I would still use skb->hash to encode the hash. That hash type of that > >>>>> field is not strictly defined. It can be siphash from ___skb_get_hash > >>>>> or a device hash, which most likely also uses Toeplitz. Then you also > >>>>> don't run into the problem of growing the struct size. > >>>> > >>>> I'm concerned exactly because it's not strictly defined. Someone may > >>>> decide to overwrite it later if we are not cautious enough. qdisc_skb_cb > >>>> also has sufficient space to contain both of the hash value and type. > >>> > >>> How about using skb extensions? > >> > >> I think it will work. I'll try it in the next version. > > > > Btw, I still think using eBPF for hash might be better. > > > > Though the hashing rule is defined in the spec, it may be extended in > > the future. For example, several extensions has been proposed: > > > > 1) RSS context > > 2) encapsulated packet hashing > > Looking at the proposals, I'm now more inclined to extend the BPF > steering program. Just to make sure we are at the same page. If the eBPF program needs to access skb extensions, it would not be a steering program anymore (not a filter). Or do you mean it is a dedicated eBPF program that calculates the hash? > > Yuri, who wrote the RFC patches to extend the BPF steering program, also > raised an concern that it may become hard to implement virtio-net > extensions in the future. It is much easier to deploy a new BPF program > to support extensions since it will be included in QEMU and can be > deployed at once without concerning other kernel stuff. > > I was still not sure how likely such an extension will emerge especially > when the hardware RSS capability is not evolving for a decade or so. But > those proposals show that there are more demands of new features for > virtio-net. It's not only the RSS, if you track virtio development, flow directors are also being proposed. Thanks >
On 2023/10/11 12:18, Jason Wang wrote: > On Tue, Oct 10, 2023 at 2:19 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >> >> On 2023/10/10 15:00, Jason Wang wrote: >>> On Tue, Oct 10, 2023 at 1:51 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>> >>>> On 2023/10/10 14:45, Jason Wang wrote: >>>>> On Tue, Oct 10, 2023 at 9:52 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>> >>>>>> On 2023/10/09 19:44, Willem de Bruijn wrote: >>>>>>> On Mon, Oct 9, 2023 at 3:12 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>> >>>>>>>> On 2023/10/09 19:06, Willem de Bruijn wrote: >>>>>>>>> On Mon, Oct 9, 2023 at 3:02 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>> >>>>>>>>>> On 2023/10/09 18:57, Willem de Bruijn wrote: >>>>>>>>>>> On Mon, Oct 9, 2023 at 3:57 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>> >>>>>>>>>>>> On 2023/10/09 17:04, Willem de Bruijn wrote: >>>>>>>>>>>>> On Sun, Oct 8, 2023 at 3:46 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>>>> >>>>>>>>>>>>>> On 2023/10/09 5:08, Willem de Bruijn wrote: >>>>>>>>>>>>>>> On Sun, Oct 8, 2023 at 10:04 PM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> On 2023/10/09 4:07, Willem de Bruijn wrote: >>>>>>>>>>>>>>>>> On Sun, Oct 8, 2023 at 7:22 AM Akihiko Odaki <akihiko.odaki@daynix.com> wrote: >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> virtio-net have two usage of hashes: one is RSS and another is hash >>>>>>>>>>>>>>>>>> reporting. Conventionally the hash calculation was done by the VMM. >>>>>>>>>>>>>>>>>> However, computing the hash after the queue was chosen defeats the >>>>>>>>>>>>>>>>>> purpose of RSS. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Another approach is to use eBPF steering program. This approach has >>>>>>>>>>>>>>>>>> another downside: it cannot report the calculated hash due to the >>>>>>>>>>>>>>>>>> restrictive nature of eBPF. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Introduce the code to compute hashes to the kernel in order to overcome >>>>>>>>>>>>>>>>>> thse challenges. An alternative solution is to extend the eBPF steering >>>>>>>>>>>>>>>>>> program so that it will be able to report to the userspace, but it makes >>>>>>>>>>>>>>>>>> little sense to allow to implement different hashing algorithms with >>>>>>>>>>>>>>>>>> eBPF since the hash value reported by virtio-net is strictly defined by >>>>>>>>>>>>>>>>>> the specification. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> The hash value already stored in sk_buff is not used and computed >>>>>>>>>>>>>>>>>> independently since it may have been computed in a way not conformant >>>>>>>>>>>>>>>>>> with the specification. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> >>>>>>>>>>>>>>>>>> --- >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { >>>>>>>>>>>>>>>>>> + .max_indirection_table_length = >>>>>>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, >>>>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>>>> + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES >>>>>>>>>>>>>>>>>> +}; >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> No need to have explicit capabilities exchange like this? Tun either >>>>>>>>>>>>>>>>> supports all or none. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> tun does not support VIRTIO_NET_RSS_HASH_TYPE_IP_EX, >>>>>>>>>>>>>>>> VIRTIO_NET_RSS_HASH_TYPE_TCP_EX, and VIRTIO_NET_RSS_HASH_TYPE_UDP_EX. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> It is because the flow dissector does not support IPv6 extensions. The >>>>>>>>>>>>>>>> specification is also vague, and does not tell how many TLVs should be >>>>>>>>>>>>>>>> consumed at most when interpreting destination option header so I chose >>>>>>>>>>>>>>>> to avoid adding code for these hash types to the flow dissector. I doubt >>>>>>>>>>>>>>>> anyone will complain about it since nobody complains for Linux. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> I'm also adding this so that we can extend it later. >>>>>>>>>>>>>>>> max_indirection_table_length may grow for systems with 128+ CPUs, or >>>>>>>>>>>>>>>> types may have other bits for new protocols in the future. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> case TUNSETSTEERINGEBPF: >>>>>>>>>>>>>>>>>> - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>>>>>>>>>> + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); >>>>>>>>>>>>>>>>>> + if (IS_ERR(bpf_ret)) >>>>>>>>>>>>>>>>>> + ret = PTR_ERR(bpf_ret); >>>>>>>>>>>>>>>>>> + else if (bpf_ret) >>>>>>>>>>>>>>>>>> + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Don't make one feature disable another. >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> TUNSETSTEERINGEBPF and TUNSETVNETHASH are mutually exclusive >>>>>>>>>>>>>>>>> functions. If one is enabled the other call should fail, with EBUSY >>>>>>>>>>>>>>>>> for instance. >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> + case TUNSETVNETHASH: >>>>>>>>>>>>>>>>>> + len = sizeof(vnet_hash); >>>>>>>>>>>>>>>>>> + if (copy_from_user(&vnet_hash, argp, len)) { >>>>>>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>>>>>> + break; >>>>>>>>>>>>>>>>>> + } >>>>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>>>> + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && >>>>>>>>>>>>>>>>>> + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || >>>>>>>>>>>>>>>>>> + !tun_is_little_endian(tun))) || >>>>>>>>>>>>>>>>>> + vnet_hash.indirection_table_mask >= >>>>>>>>>>>>>>>>>> + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { >>>>>>>>>>>>>>>>>> + ret = -EINVAL; >>>>>>>>>>>>>>>>>> + break; >>>>>>>>>>>>>>>>>> + } >>>>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>>>>>>>>>> + len = (vnet_hash.indirection_table_mask + 1) * 2; >>>>>>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { >>>>>>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>>>>>> + break; >>>>>>>>>>>>>>>>>> + } >>>>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>>>> + argp = (u8 __user *)argp + len; >>>>>>>>>>>>>>>>>> + len = virtio_net_hash_key_length(vnet_hash.types); >>>>>>>>>>>>>>>>>> + >>>>>>>>>>>>>>>>>> + if (copy_from_user(vnet_hash_key, argp, len)) { >>>>>>>>>>>>>>>>>> + ret = -EFAULT; >>>>>>>>>>>>>>>>>> + break; >>>>>>>>>>>>>>>>>> + } >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Probably easier and less error-prone to define a fixed size control >>>>>>>>>>>>>>>>> struct with the max indirection table size. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> I made its size variable because the indirection table and key may grow >>>>>>>>>>>>>>>> in the future as I wrote above. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Btw: please trim the CC: list considerably on future patches. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> I'll do so in the next version with the TUNSETSTEERINGEBPF change you >>>>>>>>>>>>>>>> proposed. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> To be clear: please don't just resubmit with that one change. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> The skb and cb issues are quite fundamental issues that need to be resolved. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> I'd like to understand why adjusting the existing BPF feature for this >>>>>>>>>>>>>>> exact purpose cannot be amended to return the key it produced. >>>>>>>>>>>>>> >>>>>>>>>>>>>> eBPF steering program is not designed for this particular problem in my >>>>>>>>>>>>>> understanding. It was introduced to derive hash values with an >>>>>>>>>>>>>> understanding of application-specific semantics of packets instead of >>>>>>>>>>>>>> generic IP/TCP/UDP semantics. >>>>>>>>>>>>>> >>>>>>>>>>>>>> This problem is rather different in terms that the hash derivation is >>>>>>>>>>>>>> strictly defined by virtio-net. I don't think it makes sense to >>>>>>>>>>>>>> introduce the complexity of BPF when you always run the same code. >>>>>>>>>>>>>> >>>>>>>>>>>>>> It can utilize the existing flow dissector and also make it easier to >>>>>>>>>>>>>> use for the userspace by implementing this in the kernel. >>>>>>>>>>>>> >>>>>>>>>>>>> Ok. There does appear to be overlap in functionality. But it might be >>>>>>>>>>>>> easier to deploy to just have standard Toeplitz available without >>>>>>>>>>>>> having to compile and load an eBPF program. >>>>>>>>>>>>> >>>>>>>>>>>>> As for the sk_buff and cb[] changes. The first is really not needed. >>>>>>>>>>>>> sk_buff simply would not scale if every edge case needs a few bits. >>>>>>>>>>>> >>>>>>>>>>>> An alternative is to move the bit to cb[] and clear it for every code >>>>>>>>>>>> paths that lead to ndo_start_xmit(), but I'm worried that it is error-prone. >>>>>>>>>>>> >>>>>>>>>>>> I think we can put the bit in sk_buff for now. We can implement the >>>>>>>>>>>> alternative when we are short of bits. >>>>>>>>>>> >>>>>>>>>>> I disagree. sk_buff fields add a cost to every code path. They cannot >>>>>>>>>>> be added for every edge case. >>>>>>>>>> >>>>>>>>>> It only takes an unused bit and does not grow the sk_buff size so I >>>>>>>>>> think it has practically no cost for now. >>>>>>>>> >>>>>>>>> The problem is that that thinking leads to death by a thousand cuts. >>>>>>>>> >>>>>>>>> "for now" forces the cost of having to think hard how to avoid growing >>>>>>>>> sk_buff onto the next person. Let's do it right from the start. >>>>>>>> >>>>>>>> I see. I described an alternative to move the bit to cb[] and clear it >>>>>>>> in all code paths that leads to ndo_start_xmit() earlier. Does that >>>>>>>> sound good to you? >>>>>>> >>>>>>> If you use the control block to pass information between >>>>>>> __dev_queue_xmit on the tun device and tun_net_xmit, using gso_skb_cb, >>>>>>> the field can be left undefined in all non-tun paths. tun_select_queue >>>>>>> can initialize. >>>>>> >>>>>> The problem is that tun_select_queue() is not always called. >>>>>> netdev_core_pick_tx() ensures dev->real_num_tx_queues != 1 before >>>>>> calling it, but this variable may change later and result in a race >>>>>> condition. Another case is that XDP with predefined queue. >>>>>> >>>>>>> >>>>>>> I would still use skb->hash to encode the hash. That hash type of that >>>>>>> field is not strictly defined. It can be siphash from ___skb_get_hash >>>>>>> or a device hash, which most likely also uses Toeplitz. Then you also >>>>>>> don't run into the problem of growing the struct size. >>>>>> >>>>>> I'm concerned exactly because it's not strictly defined. Someone may >>>>>> decide to overwrite it later if we are not cautious enough. qdisc_skb_cb >>>>>> also has sufficient space to contain both of the hash value and type. >>>>> >>>>> How about using skb extensions? >>>> >>>> I think it will work. I'll try it in the next version. >>> >>> Btw, I still think using eBPF for hash might be better. >>> >>> Though the hashing rule is defined in the spec, it may be extended in >>> the future. For example, several extensions has been proposed: >>> >>> 1) RSS context >>> 2) encapsulated packet hashing >> >> Looking at the proposals, I'm now more inclined to extend the BPF >> steering program. > > Just to make sure we are at the same page. > > If the eBPF program needs to access skb extensions, it would not be a > steering program anymore (not a filter). > > Or do you mean it is a dedicated eBPF program that calculates the hash? I think the BPF program should be a steering program but extended for hash reporting. Since we need a hash reporting feature that is not present in a socket filter, the BPF program should have a dedicated bpf_prog_type (not BPF_PROG_TYPE_SOCKET_FILTER). However, as its functionality is the superset of the conventional steering program, I'm planning to use the existing TUNSETSTEERINGEBPF ioctl to set it.
diff --git a/drivers/net/tun.c b/drivers/net/tun.c index 89ab9efe522c..561a573cd008 100644 --- a/drivers/net/tun.c +++ b/drivers/net/tun.c @@ -171,6 +171,9 @@ struct tun_prog { struct bpf_prog *prog; }; +#define TUN_VNET_HASH_MAX_KEY_SIZE 40 +#define TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH 128 + /* Since the socket were moved to tun_file, to preserve the behavior of persist * device, socket filter, sndbuf and vnet header size were restore when the * file were attached to a persist device. @@ -209,6 +212,9 @@ struct tun_struct { struct tun_prog __rcu *steering_prog; struct tun_prog __rcu *filter_prog; struct ethtool_link_ksettings link_ksettings; + struct tun_vnet_hash vnet_hash; + u16 vnet_hash_indirection_table[TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH]; + u32 vnet_hash_key[TUN_VNET_HASH_MAX_KEY_SIZE / 4]; /* init args */ struct file *file; struct ifreq *ifr; @@ -219,6 +225,13 @@ struct veth { __be16 h_vlan_TCI; }; +static const struct tun_vnet_hash_cap tun_vnet_hash_cap = { + .max_indirection_table_length = + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH, + + .types = VIRTIO_NET_SUPPORTED_HASH_TYPES +}; + static void tun_flow_init(struct tun_struct *tun); static void tun_flow_uninit(struct tun_struct *tun); @@ -320,10 +333,16 @@ static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp) if (get_user(be, argp)) return -EFAULT; - if (be) + if (be) { + if (!(tun->flags & TUN_VNET_LE) && + (tun->vnet_hash.flags & TUN_VNET_HASH_REPORT)) { + return -EINVAL; + } + tun->flags |= TUN_VNET_BE; - else + } else { tun->flags &= ~TUN_VNET_BE; + } return 0; } @@ -558,15 +577,47 @@ static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb) return ret % numqueues; } +static u16 tun_vnet_select_queue(struct tun_struct *tun, struct sk_buff *skb) +{ + u32 value = qdisc_skb_cb(skb)->tun_vnet_hash_value; + u32 numqueues; + u32 index; + u16 queue; + + numqueues = READ_ONCE(tun->numqueues); + if (!numqueues) + return 0; + + index = value & READ_ONCE(tun->vnet_hash.indirection_table_mask); + queue = READ_ONCE(tun->vnet_hash_indirection_table[index]); + if (!queue) + queue = READ_ONCE(tun->vnet_hash.unclassified_queue); + + return queue % numqueues; +} + static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb, struct net_device *sb_dev) { struct tun_struct *tun = netdev_priv(dev); + u8 vnet_hash_flags = READ_ONCE(tun->vnet_hash.flags); + struct virtio_net_hash hash; u16 ret; + if (vnet_hash_flags & (TUN_VNET_HASH_RSS | TUN_VNET_HASH_REPORT)) { + virtio_net_hash(skb, READ_ONCE(tun->vnet_hash.types), + tun->vnet_hash_key, &hash); + + skb->tun_vnet_hash = true; + qdisc_skb_cb(skb)->tun_vnet_hash_value = hash.value; + qdisc_skb_cb(skb)->tun_vnet_hash_report = hash.report; + } + rcu_read_lock(); if (rcu_dereference(tun->steering_prog)) ret = tun_ebpf_select_queue(tun, skb); + else if (vnet_hash_flags & TUN_VNET_HASH_RSS) + ret = tun_vnet_select_queue(tun, skb); else ret = tun_automq_select_queue(tun, skb); rcu_read_unlock(); @@ -2088,10 +2139,15 @@ static ssize_t tun_put_user(struct tun_struct *tun, struct iov_iter *iter) { struct tun_pi pi = { 0, skb->protocol }; + struct virtio_net_hash vnet_hash = { + .value = qdisc_skb_cb(skb)->tun_vnet_hash_value, + .report = qdisc_skb_cb(skb)->tun_vnet_hash_report + }; ssize_t total; int vlan_offset = 0; int vlan_hlen = 0; int vnet_hdr_sz = 0; + size_t vnet_hdr_content_sz; if (skb_vlan_tag_present(skb)) vlan_hlen = VLAN_HLEN; @@ -2116,31 +2172,49 @@ static ssize_t tun_put_user(struct tun_struct *tun, } if (vnet_hdr_sz) { - struct virtio_net_hdr gso; + union { + struct virtio_net_hdr hdr; + struct virtio_net_hdr_v1_hash v1_hash_hdr; + } hdr; + int ret; if (iov_iter_count(iter) < vnet_hdr_sz) return -EINVAL; - if (virtio_net_hdr_from_skb(skb, &gso, - tun_is_little_endian(tun), true, - vlan_hlen)) { + if ((READ_ONCE(tun->vnet_hash.flags) & TUN_VNET_HASH_REPORT) && + vnet_hdr_sz >= sizeof(hdr.v1_hash_hdr) && + skb->tun_vnet_hash) { + vnet_hdr_content_sz = sizeof(hdr.v1_hash_hdr); + ret = virtio_net_hdr_v1_hash_from_skb(skb, + &hdr.v1_hash_hdr, + true, + vlan_hlen, + &vnet_hash); + } else { + vnet_hdr_content_sz = sizeof(hdr.hdr); + ret = virtio_net_hdr_from_skb(skb, &hdr.hdr, + tun_is_little_endian(tun), + true, vlan_hlen); + } + + if (ret) { struct skb_shared_info *sinfo = skb_shinfo(skb); pr_err("unexpected GSO type: " "0x%x, gso_size %d, hdr_len %d\n", - sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size), - tun16_to_cpu(tun, gso.hdr_len)); + sinfo->gso_type, tun16_to_cpu(tun, hdr.hdr.gso_size), + tun16_to_cpu(tun, hdr.hdr.hdr_len)); print_hex_dump(KERN_ERR, "tun: ", DUMP_PREFIX_NONE, 16, 1, skb->head, - min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true); + min((int)tun16_to_cpu(tun, hdr.hdr.hdr_len), 64), true); WARN_ON_ONCE(1); return -EINVAL; } - if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso)) + if (copy_to_iter(&hdr, vnet_hdr_content_sz, iter) != vnet_hdr_content_sz) return -EFAULT; - iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso)); + iov_iter_advance(iter, vnet_hdr_sz - vnet_hdr_content_sz); } if (vlan_hlen) { @@ -3007,24 +3081,27 @@ static int tun_set_queue(struct file *file, struct ifreq *ifr) return ret; } -static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p, - void __user *data) +static struct bpf_prog *tun_set_ebpf(struct tun_struct *tun, + struct tun_prog __rcu **prog_p, + void __user *data) { struct bpf_prog *prog; int fd; + int ret; if (copy_from_user(&fd, data, sizeof(fd))) - return -EFAULT; + return ERR_PTR(-EFAULT); if (fd == -1) { prog = NULL; } else { prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); if (IS_ERR(prog)) - return PTR_ERR(prog); + return prog; } - return __tun_set_ebpf(tun, prog_p, prog); + ret = __tun_set_ebpf(tun, prog_p, prog); + return ret ? ERR_PTR(ret) : prog; } /* Return correct value for tun->dev->addr_len based on tun->dev->type. */ @@ -3082,6 +3159,11 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, int le; int ret; bool do_notify = false; + struct bpf_prog *bpf_ret; + struct tun_vnet_hash vnet_hash; + u16 vnet_hash_indirection_table[TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH]; + u8 vnet_hash_key[TUN_VNET_HASH_MAX_KEY_SIZE]; + size_t len; if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) { @@ -3295,7 +3377,10 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, ret = -EFAULT; break; } - if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) { + if (vnet_hdr_sz < + (int)((tun->vnet_hash.flags & TUN_VNET_HASH_REPORT) ? + sizeof(struct virtio_net_hdr_v1_hash) : + sizeof(struct virtio_net_hdr))) { ret = -EINVAL; break; } @@ -3314,10 +3399,16 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, ret = -EFAULT; break; } - if (le) + if (le) { tun->flags |= TUN_VNET_LE; - else + } else { + if (!tun_legacy_is_little_endian(tun)) { + ret = -EINVAL; + break; + } + tun->flags &= ~TUN_VNET_LE; + } break; case TUNGETVNETBE: @@ -3360,11 +3451,17 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, break; case TUNSETSTEERINGEBPF: - ret = tun_set_ebpf(tun, &tun->steering_prog, argp); + bpf_ret = tun_set_ebpf(tun, &tun->steering_prog, argp); + if (IS_ERR(bpf_ret)) + ret = PTR_ERR(bpf_ret); + else if (bpf_ret) + tun->vnet_hash.flags &= ~TUN_VNET_HASH_RSS; break; case TUNSETFILTEREBPF: - ret = tun_set_ebpf(tun, &tun->filter_prog, argp); + bpf_ret = tun_set_ebpf(tun, &tun->filter_prog, argp); + if (IS_ERR(bpf_ret)) + ret = PTR_ERR(bpf_ret); break; case TUNSETCARRIER: @@ -3382,6 +3479,54 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd, ret = open_related_ns(&net->ns, get_net_ns); break; + case TUNGETVNETHASHCAP: + if (copy_to_user(argp, &tun_vnet_hash_cap, + sizeof(tun_vnet_hash_cap))) + ret = -EFAULT; + break; + + case TUNSETVNETHASH: + len = sizeof(vnet_hash); + if (copy_from_user(&vnet_hash, argp, len)) { + ret = -EFAULT; + break; + } + + if (((vnet_hash.flags & TUN_VNET_HASH_REPORT) && + (tun->vnet_hdr_sz < sizeof(struct virtio_net_hdr_v1_hash) || + !tun_is_little_endian(tun))) || + vnet_hash.indirection_table_mask >= + TUN_VNET_HASH_MAX_INDIRECTION_TABLE_LENGTH) { + ret = -EINVAL; + break; + } + + argp = (u8 __user *)argp + len; + len = (vnet_hash.indirection_table_mask + 1) * 2; + if (copy_from_user(vnet_hash_indirection_table, argp, len)) { + ret = -EFAULT; + break; + } + + argp = (u8 __user *)argp + len; + len = virtio_net_hash_key_length(vnet_hash.types); + + if (copy_from_user(vnet_hash_key, argp, len)) { + ret = -EFAULT; + break; + } + + tun->vnet_hash = vnet_hash; + memcpy(tun->vnet_hash_indirection_table, + vnet_hash_indirection_table, + (vnet_hash.indirection_table_mask + 1) * 2); + memcpy(tun->vnet_hash_key, vnet_hash_key, len); + + if (vnet_hash.flags & TUN_VNET_HASH_RSS) + __tun_set_ebpf(tun, &tun->steering_prog, NULL); + + break; + default: ret = -EINVAL; break; diff --git a/include/uapi/linux/if_tun.h b/include/uapi/linux/if_tun.h index 287cdc81c939..dc591cd897c8 100644 --- a/include/uapi/linux/if_tun.h +++ b/include/uapi/linux/if_tun.h @@ -61,6 +61,8 @@ #define TUNSETFILTEREBPF _IOR('T', 225, int) #define TUNSETCARRIER _IOW('T', 226, int) #define TUNGETDEVNETNS _IO('T', 227) +#define TUNGETVNETHASHCAP _IO('T', 228) +#define TUNSETVNETHASH _IOW('T', 229, unsigned int) /* TUNSETIFF ifr flags */ #define IFF_TUN 0x0001 @@ -115,4 +117,18 @@ struct tun_filter { __u8 addr[][ETH_ALEN]; }; +struct tun_vnet_hash_cap { + __u16 max_indirection_table_length; + __u32 types; +}; + +#define TUN_VNET_HASH_RSS 0x01 +#define TUN_VNET_HASH_REPORT 0x02 +struct tun_vnet_hash { + __u8 flags; + __u32 types; + __u16 indirection_table_mask; + __u16 unclassified_queue; +}; + #endif /* _UAPI__IF_TUN_H */
virtio-net have two usage of hashes: one is RSS and another is hash reporting. Conventionally the hash calculation was done by the VMM. However, computing the hash after the queue was chosen defeats the purpose of RSS. Another approach is to use eBPF steering program. This approach has another downside: it cannot report the calculated hash due to the restrictive nature of eBPF. Introduce the code to compute hashes to the kernel in order to overcome thse challenges. An alternative solution is to extend the eBPF steering program so that it will be able to report to the userspace, but it makes little sense to allow to implement different hashing algorithms with eBPF since the hash value reported by virtio-net is strictly defined by the specification. The hash value already stored in sk_buff is not used and computed independently since it may have been computed in a way not conformant with the specification. Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com> --- drivers/net/tun.c | 187 ++++++++++++++++++++++++++++++++---- include/uapi/linux/if_tun.h | 16 +++ 2 files changed, 182 insertions(+), 21 deletions(-)