Up to [cvs.NetBSD.org] / src / sys / netinet
Request diff between arbitrary revisions
Keyword substitution: kv
Default branch: MAIN
Mostly merge changes from HEAD upto 20200411
Sync with head.
Use percpu_foreach_xcall() to gather volatile per-cpu counters. These must be serialized against the interrupts / soft-interrupts in which they're manipulated, as well as protected from non-atomic 64-bit memory loads on 32-bit platforms.
Switch sys/net to percpu_create.
Pull up following revision(s) (requested by knakahara in ticket #1385):
sys/net/if.c 1.461
sys/net/if.h 1.277
sys/net/if_gif.c 1.149
sys/net/if_gif.h 1.33
sys/net/if_ipsec.c 1.19,1.20,1.24
sys/net/if_ipsec.h 1.5
sys/net/if_l2tp.c 1.33,1.36-1.39
sys/net/if_l2tp.h 1.7,1.8
sys/net/route.c 1.220,1.221
sys/net/route.h 1.125
sys/netinet/in_gif.c 1.95
sys/netinet/in_l2tp.c 1.17
sys/netinet/ip_input.c 1.391,1.392
sys/netinet/wqinput.c 1.6
sys/netinet6/in6_gif.c 1.94
sys/netinet6/in6_l2tp.c 1.18
sys/netinet6/ip6_forward.c 1.97
sys/netinet6/ip6_input.c 1.210,1.211
sys/netipsec/ipsec_output.c 1.82,1.83 (patched)
sys/netipsec/ipsecif.c 1.12,1.13,1.15,1.17 (patched)
sys/netipsec/key.c 1.259,1.260
ipsecif(4) support input drop packet counter.
ipsecif(4) should not increment drop counter by errors not related to if_snd. Pointed out by ozaki-r@n.o, thanks.
Remove unnecessary addresses in PF_KEY message.
MOBIKE Extensions for PF_KEY draft-schilcher-mobike-pfkey-extension-01.txt says
====================
5. SPD Update
// snip
SADB_X_SPDADD:
// snip
sadb_x_ipsecrequest_reqid:
An ID for that SA can be passed to the kernel in the
sadb_x_ipsecrequest_reqid field.
If tunnel mode is specified, the sadb_x_ipsecrequest structure is
followed by two sockaddr structures that define the tunnel
endpoint addresses. In the case that transport mode is used, no
additional addresses are specified.
====================
see: <a rel="nofollow" href="https://tools.ietf.org/html/draft-schilcher-mobike-pfkey-extension-01">https://tools.ietf.org/html/draft-schilcher-mobike-pfkey-extension-01</a>
ipsecif(4) uses transport mode, so it should not add addresses.
ipsecif(4) supports multiple peers in the same NAPT.
E.g. ipsec0 connects between NetBSD_A and NetBSD_B, ipsec1 connects
NetBSD_A and NetBSD_C at the following figure.
+----------+
+----| NetBSD_B |
+----------+ +------+ | +----------+
| NetBSD_A |--- ... ---| NAPT |---+
+----------+ +------+ | +----------+
+----| NetBSD_C |
+----------+
Add ATF later.
l2tp(4): fix output bytes counter. Pointed by k-goda@IIJ, thanks.
remove a variable which is no longer used.
l2tp: initialize mowner variables for MBUFTRACE
Avoid having a rtcache directly in a percpu storage
percpu(9) has a certain memory storage for each CPU and provides it by the piece
to users. If the storages went short, percpu(9) enlarges them by allocating new
larger memory areas, replacing old ones with them and destroying the old ones.
A percpu storage referenced by a pointer gotten via percpu_getref can be
destroyed by the mechanism after a running thread sleeps even if percpu_putref
has not been called.
Using rtcache, i.e., packet processing, typically involves sleepable operations
such as rwlock so we must avoid dereferencing a rtcache that is directly stored
in a percpu storage during packet processing. Address this situation by having
just a pointer to a rtcache in a percpu storage instead.
Reviewed by knakahara@ and yamaguchi@
wqinput: avoid having struct wqinput_worklist directly in a percpu storage
percpu(9) has a certain memory storage for each CPU and provides it by the piece
to users. If the storages went short, percpu(9) enlarges them by allocating new
larger memory areas, replacing old ones with them and destroying the old ones.
A percpu storage referenced by a pointer gotten via percpu_getref can be
destroyed by the mechanism after a running thread sleeps even if percpu_putref
has not been called.
Input handlers of wqinput normally involves sleepable operations so we must
avoid dereferencing a percpu data (struct wqinput_worklist) after executing
an input handler. Address this situation by having just a pointer to the data
in a percpu storage instead.
Reviewed by knakahara@ and yamaguchi@
Add missing #include <sys/kmem.h>
Divide Tx context of l2tp(4) to improve performance.
It seems l2tp(4) call path is too long for instruction cache. So, dividing
l2tp(4) Tx context improves CPU use efficiency.
After this commit, l2tp(4) throughput gains 10% on my machine(Atom C3000).
Apply some missing changes lost on the previous commit
Avoid having a rtcache directly in a percpu storage for tunnel protocols.
percpu(9) has a certain memory storage for each CPU and provides it by the piece
to users. If the storages went short, percpu(9) enlarges them by allocating new
larger memory areas, replacing old ones with them and destroying the old ones.
A percpu storage referenced by a pointer gotten via percpu_getref can be
destroyed by the mechanism after a running thread sleeps even if percpu_putref
has not been called.
Using rtcache, i.e., packet processing, typically involves sleepable operations
such as rwlock so we must avoid dereferencing a rtcache that is directly stored
in a percpu storage during packet processing. Address this situation by having
just a pointer to a rtcache in a percpu storage instead.
Reviewed by ozaki-r@ and yamaguchi@
l2tp(4): avoid having struct ifqueue directly in a percpu storage.
percpu(9) has a certain memory storage for each CPU and provides it by the piece
to users. If the storages went short, percpu(9) enlarges them by allocating new
larger memory areas, replacing old ones with them and destroying the old ones.
A percpu storage referenced by a pointer gotten via percpu_getref can be
destroyed by the mechanism after a running thread sleeps even if percpu_putref
has not been called.
Tx processing of l2tp(4) uses normally involves sleepable operations so we
must avoid dereferencing a percpu data (struct ifqueue) after executing Tx
processing. Address this situation by having just a pointer to the data in
a percpu storage instead.
Reviewed by ozaki-r@ and yamaguchi@
Pull up following revision(s) (requested by ozaki-r in ticket #238): sys/netipsec/ipsec_output.c: revision 1.83 sys/net/route.h: revision 1.125 sys/netinet6/ip6_input.c: revision 1.210 sys/netinet6/ip6_input.c: revision 1.211 sys/net/if.c: revision 1.461 sys/net/if_gif.h: revision 1.33 sys/net/route.c: revision 1.220 sys/net/route.c: revision 1.221 sys/net/if.h: revision 1.277 sys/netinet6/ip6_forward.c: revision 1.97 sys/netinet/wqinput.c: revision 1.6 sys/net/if_ipsec.h: revision 1.5 sys/netinet6/in6_l2tp.c: revision 1.18 sys/netinet6/in6_gif.c: revision 1.94 sys/net/if_l2tp.h: revision 1.7 sys/net/if_gif.c: revision 1.149 sys/net/if_l2tp.h: revision 1.8 sys/netinet/in_gif.c: revision 1.95 sys/netinet/in_l2tp.c: revision 1.17 sys/netipsec/ipsecif.c: revision 1.17 sys/net/if_ipsec.c: revision 1.24 sys/net/if_l2tp.c: revision 1.37 sys/netinet/ip_input.c: revision 1.391 sys/net/if_l2tp.c: revision 1.38 sys/netinet/ip_input.c: revision 1.392 sys/net/if_l2tp.c: revision 1.39 Avoid having a rtcache directly in a percpu storage percpu(9) has a certain memory storage for each CPU and provides it by the piece to users. If the storages went short, percpu(9) enlarges them by allocating new larger memory areas, replacing old ones with them and destroying the old ones. A percpu storage referenced by a pointer gotten via percpu_getref can be destroyed by the mechanism after a running thread sleeps even if percpu_putref has not been called. Using rtcache, i.e., packet processing, typically involves sleepable operations such as rwlock so we must avoid dereferencing a rtcache that is directly stored in a percpu storage during packet processing. Address this situation by having just a pointer to a rtcache in a percpu storage instead. Reviewed by knakahara@ and yamaguchi@ - wqinput: avoid having struct wqinput_worklist directly in a percpu storage percpu(9) has a certain memory storage for each CPU and provides it by the piece to users. If the storages went short, percpu(9) enlarges them by allocating new larger memory areas, replacing old ones with them and destroying the old ones. A percpu storage referenced by a pointer gotten via percpu_getref can be destroyed by the mechanism after a running thread sleeps even if percpu_putref has not been called. Input handlers of wqinput normally involves sleepable operations so we must avoid dereferencing a percpu data (struct wqinput_worklist) after executing an input handler. Address this situation by having just a pointer to the data in a percpu storage instead. Reviewed by knakahara@ and yamaguchi@ - Add missing #include <sys/kmem.h> - Divide Tx context of l2tp(4) to improve performance. It seems l2tp(4) call path is too long for instruction cache. So, dividing l2tp(4) Tx context improves CPU use efficiency. After this commit, l2tp(4) throughput gains 10% on my machine(Atom C3000). - Apply some missing changes lost on the previous commit - Avoid having a rtcache directly in a percpu storage for tunnel protocols. percpu(9) has a certain memory storage for each CPU and provides it by the piece to users. If the storages went short, percpu(9) enlarges them by allocating new larger memory areas, replacing old ones with them and destroying the old ones. A percpu storage referenced by a pointer gotten via percpu_getref can be destroyed by the mechanism after a running thread sleeps even if percpu_putref has not been called. Using rtcache, i.e., packet processing, typically involves sleepable operations such as rwlock so we must avoid dereferencing a rtcache that is directly stored in a percpu storage during packet processing. Address this situation by having just a pointer to a rtcache in a percpu storage instead. Reviewed by ozaki-r@ and yamaguchi@ - l2tp(4): avoid having struct ifqueue directly in a percpu storage. percpu(9) has a certain memory storage for each CPU and provides it by the piece to users. If the storages went short, percpu(9) enlarges them by allocating new larger memory areas, replacing old ones with them and destroying the old ones. A percpu storage referenced by a pointer gotten via percpu_getref can be destroyed by the mechanism after a running thread sleeps even if percpu_putref has not been called. Tx processing of l2tp(4) uses normally involves sleepable operations so we must avoid dereferencing a percpu data (struct ifqueue) after executing Tx processing. Address this situation by having just a pointer to the data in a percpu storage instead. Reviewed by ozaki-r@ and yamaguchi@
wqinput: avoid having struct wqinput_worklist directly in a percpu storage percpu(9) has a certain memory storage for each CPU and provides it by the piece to users. If the storages went short, percpu(9) enlarges them by allocating new larger memory areas, replacing old ones with them and destroying the old ones. A percpu storage referenced by a pointer gotten via percpu_getref can be destroyed by the mechanism after a running thread sleeps even if percpu_putref has not been called. Input handlers of wqinput normally involves sleepable operations so we must avoid dereferencing a percpu data (struct wqinput_worklist) after executing an input handler. Address this situation by having just a pointer to the data in a percpu storage instead. Reviewed by knakahara@ and yamaguchi@
Sync with HEAD
Sync with HEAD Resolve a couple of conflicts (result of the uimin/uimax changes)
Change the type of wqinput's drop counter to uint64_t. OK'd by ozaki-r@.
Pull up following revision(s) (requested by ozaki-r in ticket #588): sys/netinet6/in6.c: revision 1.260 sys/netinet/in.c: revision 1.219 sys/netinet/wqinput.c: revision 1.4 sys/rump/net/lib/libnetinet/netinet_component.c: revision 1.11 sys/netinet/ip_input.c: revision 1.376 sys/netinet6/ip6_input.c: revision 1.193 Avoid a deadlock between softnet_lock and IFNET_LOCK A deadlock occurs because there is a violation of the rule of lock ordering; softnet_lock is held with hodling IFNET_LOCK, which violates the rule. To avoid the deadlock, replace softnet_lock in in_control and in6_control with KERNEL_LOCK. We also need to add some KERNEL_LOCKs to protect the network stack surely. This is required, for example, for PR kern/51356. Fix PR kern/53043
Avoid a deadlock between softnet_lock and IFNET_LOCK A deadlock occurs because there is a violation of the rule of lock ordering; softnet_lock is held with hodling IFNET_LOCK, which violates the rule. To avoid the deadlock, replace softnet_lock in in_control and in6_control with KERNEL_LOCK. We also need to add some KERNEL_LOCKs to protect the network stack surely. This is required, for example, for PR kern/51356. Fix PR kern/53043
update from HEAD
Sync with HEAD
file wqinput.c was added on branch tls-maxphys on 2017-12-03 11:39:04 +0000
pool_init does not copy its name argument therefore don't pass in a stack allocated buffer vmstat -mv shows pool(s) with broken name(s) use the name argument passed into wqinput_create directly which is a static string in all 4 callee cases (workqueue_create/workqueue_init copies the name argument)
Add missing NULL check for pool_get call with PR_NOWAIT This should fix a kernel panic reported by wiz@ on current-users ML: http://mail-index.netbsd.org/current-users/2017/05/03/msg031646.html
Sync with HEAD
Sync with HEAD
Sync with HEAD
file wqinput.c was added on branch bouyer-socketcan on 2017-04-21 16:54:06 +0000
file wqinput.c was added on branch pgoyette-localcount on 2017-03-20 06:57:51 +0000
file wqinput.c was added on branch nick-nhusb on 2017-02-05 13:40:59 +0000
Defer some pr_input to workqueue pr_input is currently called in softint. Some pr_input such as ICMP, ICMPv6 and CARP can add/delete/update IP addresses and routing table entries. For example, icmp6_redirect_input updates an a routing table entry and nd6_ra_input may delete an IP address. Basically such operations shouldn't be done in softint. That aside, we have a reason to avoid the situation; psz/psref waits cannot be used in softint, however they are required to work in such pr_input in the MP-safe world. The change implements the workqueue pr_input framework called wqinput which provides a means to defer pr_input of a protocol to workqueue easily. Currently icmp_input, icmp6_input, carp_proto_input and carp6_proto_input are deferred to workqueue by the framework. Proposed and discussed on tech-kern and tech-net