| Message ID | 20220623043449.1217288-1-edumazet@google.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | ipmr: get rid of rwlocks | expand |
Hello: This series was applied to netdev/net-next.git (master) by David S. Miller <davem@davemloft.net>: On Thu, 23 Jun 2022 04:34:30 +0000 you wrote: > We need to get rid of rwlocks in networking stacks, > if read_lock() is (ab)used from softirq context. > > As discussed recently [1], rwlock are unfair by design in this case, > and writers can starve and trigger soft lockups. > > This series convert ipmr code (both IPv4 and IPv6 families) > to RCU and spinlocks. > > [...] Here is the summary with links: - [v2,net-next,01/19] ip6mr: do not get a device reference in pim6_rcv() https://git.kernel.org/netdev/net-next/c/0a24c43f54b2 - [v2,net-next,02/19] ipmr: add rcu protection over (struct vif_device)->dev https://git.kernel.org/netdev/net-next/c/ebc3197963fc - [v2,net-next,03/19] ipmr: change igmpmsg_netlink_event() prototype https://git.kernel.org/netdev/net-next/c/0b490b51d226 - [v2,net-next,04/19] ipmr: ipmr_cache_report() changes https://git.kernel.org/netdev/net-next/c/646679881a02 - [v2,net-next,05/19] ipmr: do not acquire mrt_lock in __pim_rcv() https://git.kernel.org/netdev/net-next/c/121fefc669bf - [v2,net-next,06/19] ipmr: do not acquire mrt_lock in ioctl(SIOCGETVIFCNT) https://git.kernel.org/netdev/net-next/c/559260fd9d9a - [v2,net-next,07/19] ipmr: do not acquire mrt_lock before calling ipmr_cache_unresolved() https://git.kernel.org/netdev/net-next/c/9094db4b8004 - [v2,net-next,08/19] ipmr: do not acquire mrt_lock while calling ip_mr_forward() https://git.kernel.org/netdev/net-next/c/4eadb88244d1 - [v2,net-next,09/19] ipmr: do not acquire mrt_lock in ipmr_get_route() https://git.kernel.org/netdev/net-next/c/e4cd9868e8ec - [v2,net-next,10/19] ip6mr: ip6mr_cache_report() changes https://git.kernel.org/netdev/net-next/c/3493a5b730e5 - [v2,net-next,11/19] ip6mr: do not acquire mrt_lock in pim6_rcv() https://git.kernel.org/netdev/net-next/c/6d08658736fc - [v2,net-next,12/19] ip6mr: do not acquire mrt_lock in ioctl(SIOCGETMIFCNT_IN6) https://git.kernel.org/netdev/net-next/c/638cf4a24a09 - [v2,net-next,13/19] ip6mr: do not acquire mrt_lock before calling ip6mr_cache_unresolved https://git.kernel.org/netdev/net-next/c/db9eb7c8ae34 - [v2,net-next,14/19] ip6mr: do not acquire mrt_lock while calling ip6_mr_forward() https://git.kernel.org/netdev/net-next/c/9b1c21d898fd - [v2,net-next,15/19] ip6mr: switch ip6mr_get_route() to rcu_read_lock() https://git.kernel.org/netdev/net-next/c/6fa40a290219 - [v2,net-next,16/19] ipmr: adopt rcu_read_lock() in mr_dump() https://git.kernel.org/netdev/net-next/c/194366b28b83 - [v2,net-next,17/19] ipmr: convert /proc handlers to rcu_read_lock() https://git.kernel.org/netdev/net-next/c/b96ef16d2f83 - [v2,net-next,18/19] ipmr: convert mrt_lock to a spinlock https://git.kernel.org/netdev/net-next/c/3f55211ecf6a - [v2,net-next,19/19] ip6mr: convert mrt_lock to a spinlock https://git.kernel.org/netdev/net-next/c/a96f7a6a60b3 You are awesome, thank you!
We need to get rid of rwlocks in networking stacks, if read_lock() is (ab)used from softirq context. As discussed recently [1], rwlock are unfair by design in this case, and writers can starve and trigger soft lockups. This series convert ipmr code (both IPv4 and IPv6 families) to RCU and spinlocks. [1] https://lkml.org/lkml/2022/6/17/272 v2: fixed two typos, and resent because patch 19/19 did not make it to patchwork. Eric Dumazet (19): ip6mr: do not get a device reference in pim6_rcv() ipmr: add rcu protection over (struct vif_device)->dev ipmr: change igmpmsg_netlink_event() prototype ipmr: ipmr_cache_report() changes ipmr: do not acquire mrt_lock in __pim_rcv() ipmr: do not acquire mrt_lock in ioctl(SIOCGETVIFCNT) ipmr: do not acquire mrt_lock before calling ipmr_cache_unresolved() ipmr: do not acquire mrt_lock while calling ip_mr_forward() ipmr: do not acquire mrt_lock in ipmr_get_route() ip6mr: ip6mr_cache_report() changes ip6mr: do not acquire mrt_lock in pim6_rcv() ip6mr: do not acquire mrt_lock in ioctl(SIOCGETMIFCNT_IN6) ip6mr: do not acquire mrt_lock before calling ip6mr_cache_unresolved ip6mr: do not acquire mrt_lock while calling ip6_mr_forward() ip6mr: switch ip6mr_get_route() to rcu_read_lock() ipmr: adopt rcu_read_lock() in mr_dump() ipmr: convert /proc handlers to rcu_read_lock() ipmr: convert mrt_lock to a spinlock ip6mr: convert mrt_lock to a spinlock include/linux/mroute_base.h | 15 ++- net/ipv4/ipmr.c | 215 +++++++++++++++++++----------------- net/ipv4/ipmr_base.c | 53 +++++---- net/ipv6/ip6mr.c | 202 ++++++++++++++++----------------- 4 files changed, 255 insertions(+), 230 deletions(-)