Delete becomes very, very fast - almost free, but after setsockopt() syscall returns, the key is still alive until next RCU grace period. Which is fine for listen sockets as userspace needs to be aware of setsockopt(TCP_AO) and accept() race and resolve it with verification by getsockopt() after TCP connection was accepted. The benchmark results (on non-loaded box, worse with more RCU work pending): > ok 33 Worst case delete 16384 keys: min=5ms max=10ms mean=6.93904ms stddev=0.263421 > ok 34 Add a new key 16384 keys: min=1ms max=4ms mean=2.17751ms stddev=0.147564 > ok 35 Remove random-search 16384 keys: min=5ms max=10ms mean=6.50243ms stddev=0.254999 > ok 36 Remove async 16384 keys: min=0ms max=0ms mean=0.0296107ms stddev=0.0172078
Unified · Split
| @@ -396,7 +396,8 @@ struct tcp_ao_del { /* setsockopt(TCP_AO_DEL_KEY) */ · 395 unmodified lines | |||
| 396 | __s32 ifindex; /* L3 dev index for VRF */ | 396 | __s32 ifindex; /* L3 dev index for VRF */ |
| 397 | __u32 set_current :1, /* corresponding ::current_key */ | 397 | __u32 set_current :1, /* corresponding ::current_key */ |
| 398 | set_rnext :1, /* corresponding ::rnext */ | 398 | set_rnext :1, /* corresponding ::rnext */ |
| 399 | reserved :30; /* must be 0 */ | 399 | del_async :1, /* only valid for listen sockets */ |
| 400 | reserved :29; /* must be 0 */ | ||
| 400 | __u16 reserved2; /* padding, must be 0 */ | 401 | __u16 reserved2; /* padding, must be 0 */ |
| 401 | __u8 prefix; /* peer's address prefix */ | 402 | __u8 prefix; /* peer's address prefix */ |
| 402 | __u8 sndid; /* SendID for outgoing segments */ | 403 | __u8 sndid; /* SendID for outgoing segments */ |
| @@ -1628,7 +1628,7 @@ err_free_ao: · 1627 unmodified lines | |||
| 1628 | } | 1628 | } |
| 1629 | 1629 | ||
| 1630 | static int tcp_ao_delete_key(struct sock *sk, struct tcp_ao_info *ao_info, | 1630 | static int tcp_ao_delete_key(struct sock *sk, struct tcp_ao_info *ao_info, |
| 1631 | struct tcp_ao_key *key, | 1631 | bool del_async, struct tcp_ao_key *key, |
| 1632 | struct tcp_ao_key *new_current, | 1632 | struct tcp_ao_key *new_current, |
| 1633 | struct tcp_ao_key *new_rnext) | 1633 | struct tcp_ao_key *new_rnext) |
| 1634 | { | 1634 | { |
| @@ -1636,11 +1636,24 @@ static int tcp_ao_delete_key(struct sock *sk, struct tcp_ao_info *ao_info, · 1 unmodified line | |||
| 1636 | 1636 | ||
| 1637 | hlist_del_rcu(&key->node); | 1637 | hlist_del_rcu(&key->node); |
| 1638 | 1638 | ||
| 1639 | /* Support for async delete on listening sockets: as they don't | ||
| 1640 | * need current_key/rnext_key maintaining, we don't need to check | ||
| 1641 | * them and we can just free all resources in RCU fashion. | ||
| 1642 | */ | ||
| 1643 | if (del_async) { | ||
| 1644 | atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc); | ||
| 1645 | call_rcu(&key->rcu, tcp_ao_key_free_rcu); | ||
| 1646 | return 0; | ||
| 1647 | } | ||
| 1648 | |||
| 1639 | /* At this moment another CPU could have looked this key up | 1649 | /* At this moment another CPU could have looked this key up |
| 1640 | * while it was unlinked from the list. Wait for RCU grace period, | 1650 | * while it was unlinked from the list. Wait for RCU grace period, |
| 1641 | * after which the key is off-list and can't be looked up again; | 1651 | * after which the key is off-list and can't be looked up again; |
| 1642 | * the rx path [just before RCU came] might have used it and set it | 1652 | * the rx path [just before RCU came] might have used it and set it |
| 1643 | * as current_key (very unlikely). | 1653 | * as current_key (very unlikely). |
| 1654 | * Free the key with next RCU grace period (in case it was | ||
| 1655 | * current_key before tcp_ao_current_rnext() might have | ||
| 1656 | * changed it in forced-delete). | ||
| 1644 | */ | 1657 | */ |
| 1645 | synchronize_rcu(); | 1658 | synchronize_rcu(); |
| 1646 | if (new_current) | 1659 | if (new_current) |
| @@ -1711,6 +1724,8 @@ static int tcp_ao_del_cmd(struct sock *sk, unsigned short int family, · 64 unmodified lines | |||
| 1711 | if (!new_rnext) | 1724 | if (!new_rnext) |
| 1712 | return -ENOENT; | 1725 | return -ENOENT; |
| 1713 | } | 1726 | } |
| 1727 | if (cmd.del_async && sk->sk_state != TCP_LISTEN) | ||
| 1728 | return -EINVAL; | ||
| 1714 | 1729 | ||
| 1715 | if (family == AF_INET) { | 1730 | if (family == AF_INET) { |
| 1716 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.addr; | 1731 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.addr; |
| @@ -1758,8 +1773,8 @@ static int tcp_ao_del_cmd(struct sock *sk, unsigned short int family, · 41 unmodified lines | |||
| 1758 | if (key == new_current || key == new_rnext) | 1773 | if (key == new_current || key == new_rnext) |
| 1759 | continue; | 1774 | continue; |
| 1760 | 1775 | ||
| 1761 | return tcp_ao_delete_key(sk, ao_info, key, | 1776 | return tcp_ao_delete_key(sk, ao_info, cmd.del_async, key, |
| 1762 | new_current, new_rnext); | 1777 | new_current, new_rnext); |
| 1763 | } | 1778 | } |
| 1764 | return -ENOENT; | 1779 | return -ENOENT; |
| 1765 | } | 1780 | } |