New sysctl node "net.inet6.ip6.gifpmtu" means
- 0 (default)
Fragment by IPV6_MMTU. All packets reach the destination certainly,
however the long packet performance is poor.
This is same behavior as before.
- 1
Fragment by outer interface's MTU. The long packet performance would
be good, however the packets may be dropped in some network paths
whose path MTU less than the interface's MTU.
- others
undefined yet
New sysctl node "net.interfaces.gif*.pmtu" means
- -1 (default)
Use system default value (net.inet6.ip6.gifpmtu).
- 0
Fragment by IPV6_MMTU for this gif(4) tunnel.
- 1
Fragment by outer interface's MTU for this gif(4) tunnel.
- others
undefined yet
See RFC4459 for more information and other solutions.
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@
The tunnel interfaces are gif(4), l2tp(4), and ipsecif(4). They use mutex
itself in percpu area. When percpu_cpu_enlarge() run, the address of the
mutex in percpu area becomes different from the address which lockdebug
saved. That can cause "already initialized" false detection.
if_gif.c:r1.133 introduces gif_update_variant() which ensure ifp->if_flags
is set IFF_RUNNING when gif_softc->gif_var->gv_{psrc,pdst} are not null.
So, in6_gif_ctlinput() is not required IFF_RUNNING checking. In contrast,
it is required gv_{psrc,pdst} NULL checking.
See the following descriptions for details.
Proposed on tech-kern and tech-net
Overview
--------
We protect the routing table with a rwock and protect
rtcaches with another rwlock. Each rtentry is protected
from being freed or updated via reference counting and psref.
Global rwlocks
--------------
There are two rwlocks; one for the routing table (rt_lock) and
the other for rtcaches (rtcache_lock). rtcache_lock covers
all existing rtcaches; there may have room for optimizations
(future work).
The locking order is rtcache_lock first and rt_lock is next.
rtentry references
------------------
References to an rtentry is managed with reference counting
and psref. Either of the two mechanisms is used depending on
where a rtentry is obtained. Reference counting is used when
we obtain a rtentry from the routing table directly via
rtalloc1 and rtrequest{,1} while psref is used when we obtain
a rtentry from a rtcache via rtcache_* APIs. In both cases,
a caller can sleep/block with holding an obtained rtentry.
The reasons why we use two different mechanisms are (i) only
using reference counting hurts the performance due to atomic
instructions (rtcache case) (ii) ease of implementation;
applying psref to APIs such rtaloc1 and rtrequest{,1} requires
additional works (adding a local variable and an argument).
We will finally migrate to use only psref but we can do it
when we have a lockless routing table alternative.
Reference counting for rtentry
------------------------------
rt_refcnt now doesn't count permanent references such as for
rt_timers and rtcaches, instead it is used only for temporal
references when obtaining a rtentry via rtalloc1 and rtrequest{,1}.
We can do so because destroying a rtentry always involves
removing references of rt_timers and rtcaches to the rtentry
and we don't need to track such references. This also makes
it easy to wait for readers to release references on deleting
or updating a rtentry, i.e., we can simply wait until the
reference counter is 0 or 1. (If there are permanent references
the counter can be arbitrary.)
rt_ref increments a reference counter of a rtentry and rt_unref
decrements it. rt_ref is called inside APIs (rtalloc1 and
rtrequest{,1} so users don't need to care about it while
users must call rt_unref to an obtained rtentry after using it.
rtfree is removed and we use rt_unref and rt_free instead.
rt_unref now just decrements the counter of a given rtentry
and rt_free just tries to destroy a given rtentry.
See the next section for destructions of rtentries by rt_free.
Destructions of rtentries
-------------------------
We destroy a rtentry only when we call rtrequst{,1}(RTM_DELETE);
the original implementation can destroy in any rtfree where it's
the last reference. If we use reference counting or psref, it's
easy to understand if the place that a rtentry is destroyed is
fixed.
rt_free waits for references to a given rtentry to be released
before actually destroying the rtentry. rt_free uses a condition
variable (cv_wait) (and psref_target_destroy for psref) to wait.
Unfortunately rtrequst{,1}(RTM_DELETE) can be called in softint
that we cannot use cv_wait. In that case, we have to defer the
destruction to a workqueue.
rtentry#rt_cv, rtentry#rt_psref and global variables
(see rt_free_global) are added to conduct the procedure.
Updates of rtentries
--------------------
One difficulty to use refcnt/psref instead of rwlock for rtentry
is updates of rtentries. We need an additional mechanism to
prevent readers from seeing inconsistency of a rtentry being
updated.
We introduce RTF_UPDATING flag to rtentries that are updating.
While the flag is set to a rtentry, users cannot acquire the
rtentry. By doing so, we avoid users to see inconsistent
rtentries.
There are two options when a user tries to acquire a rtentry
with the RTF_UPDATING flag; if a user runs in softint context
the user fails to acquire a rtentry (NULL is returned).
Otherwise a user waits until the update completes by waiting
on cv.
The procedure of a updater is simpler to destruction of
a rtentry. Wait on cv (and psref) and after all readers left,
proceed with the update.
Global variables (see rt_update_global) are added to conduct
the procedure.
Currently we apply the mechanism to only RTM_CHANGE in
rtsock.c. We would have to apply other codes. See
"Known issues" section.
psref for rtentry
-----------------
When we obtain a rtentry from a rtcache via rtcache_* APIs,
psref is used to reference to the rtentry.
rtcache_ref acquires a reference to a rtentry with psref
and rtcache_unref releases the reference after using it.
rtcache_ref is called inside rtcache_* APIs and users don't
need to take care of it while users must call rtcache_unref
to release the reference.
struct psref and int bound that is needed for psref is
embedded into struct route. By doing so we don't need to
add local variables and additional argument to APIs.
However this adds another constraint to psref other than
reference counting one's; holding a reference of an rtentry
via a rtcache is allowed by just one caller at the same time.
So we must not acquire a rtentry via a rtcache twice and
avoid a recursive use of a rtcache. And also a rtcache must
be arranged to be used by a LWP/softint at the same time
somehow. For IP forwarding case, we have per-CPU rtcaches
used in softint so the constraint is guaranteed. For a h
rtcache of a PCB case, the constraint is guaranteed by the
solock of each PCB. Any other cases (pf, ipf, stf and ipsec)
are currently guaranteed by only the existence of the global
locks (softnet_lock and/or KERNEL_LOCK). If we've found the
cases that we cannot guarantee the constraint, we would need
to introduce other rtcache APIs that use simple reference
counting.
psref of rtcache is created with IPL_SOFTNET and so rtcache
shouldn't used at an IPL higher than IPL_SOFTNET.
Note that rtcache_free is used to invalidate a given rtcache.
We don't need another care by my change; just keep them as
they are.
Performance impact
------------------
When NET_MPSAFE is disabled the performance drop is 3% while
when it's enabled the drop is increased to 11%. The difference
comes from that currently we don't take any global locks and
don't use psref if NET_MPSAFE is disabled.
We can optimize the performance of the case of NET_MPSAFE
on by reducing lookups of rtcache that uses psref;
currently we do two lookups but we should be able to trim
one of two. This is a future work.
Known issues
------------
There are two known issues to be solved; one is that
a caller of rtrequest(RTM_ADD) may change rtentry (see rtinit).
We need to prevent new references during the update. Or
we may be able to remove the code (perhaps, need more
investigations).
The other is rtredirect that updates a rtentry. We need
to apply our update mechanism, however it's not easy because
rtredirect is called in softint and we cannot apply our
mechanism simply. One solution is to defer rtredirect to
a workqueue but it requires some code restructuring.
In the MP-safe world, a rtentry stemming from a rtcache can be freed at any
points. So we need to protect rtentries somehow say by reference couting or
passive references. Regardless of the method, we need to call some release
function of a rtentry after using it.
The change adds a new function rtcache_unref to release a rtentry. At this
point, this function does nothing because for now we don't add a reference
to a rtentry when we get one from a rtcache. We will add something useful
in a further commit.
This change is a part of changes for MP-safe routing table. It is separated
to avoid one big change that makes difficult to debug by bisecting.
A panic cause in rn_match() called by encap[46]_lookup(). The reason is that
gif(4) does not suspend receive packet processing in spite of suspending
transmit packet processing while anyone is doing gif(4) ioctl.
To prevent calling softint_schedule() after called softint_disestablish(),
the following modifications are added
+ ioctl (writing configuration) side
- off IFF_RUNNING flag before changing configuration
- wait softint handler completion before changing configuration
+ packet processing (reading configuraiotn) side
- if IFF_RUNNING flag is on, do nothing
+ in whole
- add gif_list_lock_{enter,exit} to prevent the same configuration is
set to other gif(4) interfaces
Having a pointer of an interface in a mbuf isn't safe if we remove big
kernel locks; an interface object (ifnet) can be destroyed anytime in any
packet processing and accessing such object via a pointer is racy. Instead
we have to get an object from the interface collection (ifindex2ifnet) via
an interface index (if_index) that is stored to a mbuf instead of an
pointer.
The change provides two APIs: m_{get,put}_rcvif_psref that use psref(9)
for sleep-able critical sections and m_{get,put}_rcvif that use
pserialize(9) for other critical sections. The change also adds another
API called m_get_rcvif_NOMPSAFE, that is NOT MP-safe and for transition
moratorium, i.e., it is intended to be used for places where are not
planned to be MP-ified soon.
The change adds some overhead due to psref to performance sensitive paths,
however the overhead is not serious, 2% down at worst.
Proposed on tech-kern and tech-net.
This change was intended, but Nakahara-san had already made a better
one locally! So I'll let him commit that one, and I'll try not to
step on anyone's toes again.
Mostly mechanical change to replace it, culling some now-needless
boilerplate around all the users.
This does not substantively change the ip_encap API or eliminate
abuse of sketchy pointer casts -- that will come later, and will be
easier now that it is not tangled up with struct protosw.
You can't use this unless you know what it is a priori: the formal
prototype is variadic, and the different instances (e.g., ip_output,
route_output) have different real prototypes.
Convert the only user of it, raw_send in net/raw_cb.c, to take an
explicit callback argument. Convert the only instances of it,
route_output and key_output, to such explicit callbacks for raw_send.
Use assertions to make sure the conversion to explicit callbacks is
warranted.
Discussed on tech-net with no objections:
https://mail-index.netbsd.org/tech-net/2016/01/16/msg005484.html
dismantling of pr_usrreq in the protocols; no functional change intended.
PRU_ATTACH/PRU_DETACH changes will follow soon.
Bump for struct protosw. Welcome to 6.99.62!
When a link-layer address changes (e.g., ifconfig ex0 link
02🇩🇪ad:be:ef:02 active), send a gratuitous ARP and/or a Neighbor
Advertisement to update the network-/link-layer address bindings
on our LAN peers.
Refuse a change of ethernet address to the address 00:00:00:00:00:00
or to any multicast/broadcast address. (Thanks matt@.)
Reorder ifnet ioctl operations so that driver ioctls may inherit
the functions of their "class"---ether_ioctl(), fddi_ioctl(), et
cetera---and the class ioctls may inherit from the generic ioctl,
ifioctl_common(), but both driver- and class-ioctls may override
the generic behavior. Make network drivers share more code.
Distinguish a "factory" link-layer address from others for the
purposes of both protecting that address from deletion and computing
EUI64.
Return consistent, appropriate error codes from network drivers.
Improve readability. KNF.
*** Details ***
In if_attach(), always initialize the interface ioctl routine,
ifnet->if_ioctl, if the driver has not already initialized it.
Delete if_ioctl == NULL tests everywhere else, because it cannot
happen.
In the ioctl routines of network interfaces, inherit common ioctl
behaviors by calling either ifioctl_common() or whichever ioctl
routine is appropriate for the class of interface---e.g., ether_ioctl()
for ethernets.
Stop (ab)using SIOCSIFADDR and start to use SIOCINITIFADDR. In
the user->kernel interface, SIOCSIFADDR's argument was an ifreq,
but on the protocol->ifnet interface, SIOCSIFADDR's argument was
an ifaddr. That was confusing, and it would work against me as I
make it possible for a network interface to overload most ioctls.
On the protocol->ifnet interface, replace SIOCSIFADDR with
SIOCINITIFADDR. In ifioctl(), return EPERM if userland tries to
invoke SIOCINITIFADDR.
In ifioctl(), give the interface the first shot at handling most
interface ioctls, and give the protocol the second shot, instead
of the other way around. Finally, let compatibility code (COMPAT_OSOCK)
take a shot.
Pull device initialization out of switch statements under
SIOCINITIFADDR. For example, pull ..._init() out of any switch
statement that looks like this:
switch (...->sa_family) {
case ...:
..._init();
...
break;
...
default:
..._init();
...
break;
}
Rewrite many if-else clauses that handle all permutations of IFF_UP
and IFF_RUNNING to use a switch statement,
switch (x & (IFF_UP|IFF_RUNNING)) {
case 0:
...
break;
case IFF_RUNNING:
...
break;
case IFF_UP:
...
break;
case IFF_UP|IFF_RUNNING:
...
break;
}
unifdef lots of code containing #ifdef FreeBSD, #ifdef NetBSD, and
#ifdef SIOCSIFMTU, especially in fwip(4) and in ndis(4).
In ipw(4), remove an if_set_sadl() call that is out of place.
In nfe(4), reuse the jumbo MTU logic in ether_ioctl().
Let ethernets register a callback for setting h/w state such as
promiscuous mode and the multicast filter in accord with a change
in the if_flags: ether_set_ifflags_cb() registers a callback that
returns ENETRESET if the caller should reset the ethernet by calling
if_init(), 0 on success, != 0 on failure. Pull common code from
ex(4), gem(4), nfe(4), sip(4), tlp(4), vge(4) into ether_ioctl(),
and register if_flags callbacks for those drivers.
Return ENOTTY instead of EINVAL for inappropriate ioctls. In
zyd(4), use ENXIO instead of ENOTTY to indicate that the device is
not any longer attached.
Add to if_set_sadl() a boolean 'factory' argument that indicates
whether a link-layer address was assigned by the factory or some
other source. In a comment, recommend using the factory address
for generating an EUI64, and update in6_get_hw_ifid() to prefer a
factory address to any other link-layer address.
Add a routing message, RTM_LLINFO_UPD, that tells protocols to
update the binding of network-layer addresses to link-layer addresses.
Implement this message in IPv4 and IPv6 by sending a gratuitous
ARP or a neighbor advertisement, respectively. Generate RTM_LLINFO_UPD
messages on a change of an interface's link-layer address.
In ether_ioctl(), do not let SIOCALIFADDR set a link-layer address
that is broadcast/multicast or equal to 00:00:00:00:00:00.
Make ether_ioctl() call ifioctl_common() to handle ioctls that it
does not understand.
In gif(4), initialize if_softc and use it, instead of assuming that
the gif_softc and ifp overlap.
Let ifioctl_common() handle SIOCGIFADDR.
Sprinkle rtcache_invariants(), which checks on DIAGNOSTIC kernels
that certain invariants on a struct route are satisfied.
In agr(4), rewrite agr_ioctl_filter() to be a bit more explicit
about the ioctls that we do not allow on an agr(4) member interface.
bzero -> memset. Delete unnecessary casts to void *. Use
sockaddr_in_init() and sockaddr_in6_init(). Compare pointers with
NULL instead of "testing truth". Replace some instances of (type
*)0 with NULL. Change some K&R prototypes to ANSI C, and join
lines.
- Socket layer becomes MP safe.
- Unix protocols become MP safe.
- Allows protocol processing interrupts to safely block on locks.
- Fixes a number of race conditions.
With much feedback from matt@ and plunky@.
to _ro_rt. Use rtcache_getrt() to access a route cache's struct
rtentry *.
Introduce struct ifnet->if_dl that always points at the interface
identifier/link-layer address. Make code that treated the first
ifaddr on struct ifnet->if_addrlist as the interface address use
if_dl, instead.
Remove stale debugging code from net/route.c. Move the rtflush()
code into rtcache_clear() and delete rtflush(). Delete rtalloc(),
because nothing uses it any more.
Make ND6_HINT an inline, lowercase subroutine, nd6_hint.
I've done my best to convert IP Filter, the ISO stack, and the
AppleTalk stack to rtcache_getrt(). They compile, but I have not
tested them. I have given the changes to PF, GRE, IPv4 and IPv6
stacks a lot of exercise.
route_in6, struct route_iso), replacing all caches with a struct
route.
The principle benefit of this change is that all of the protocol
families can benefit from route cache-invalidation, which is
necessary for correct routing. Route-cache invalidation fixes an
ancient PR, kern/3508, at long last; it fixes various other PRs,
also.
Discussions with and ideas from Joerg Sonnenberger influenced this
work tremendously. Of course, all design oversights and bugs are
mine.
DETAILS
1 I added to each address family a pool of sockaddrs. I have
introduced routines for allocating, copying, and duplicating,
and freeing sockaddrs:
struct sockaddr *sockaddr_alloc(sa_family_t af, int flags);
struct sockaddr *sockaddr_copy(struct sockaddr *dst,
const struct sockaddr *src);
struct sockaddr *sockaddr_dup(const struct sockaddr *src, int flags);
void sockaddr_free(struct sockaddr *sa);
sockaddr_alloc() returns either a sockaddr from the pool belonging
to the specified family, or NULL if the pool is exhausted. The
returned sockaddr has the right size for that family; sa_family
and sa_len fields are initialized to the family and sockaddr
length---e.g., sa_family = AF_INET and sa_len = sizeof(struct
sockaddr_in). sockaddr_free() puts the given sockaddr back into
its family's pool.
sockaddr_dup() and sockaddr_copy() work analogously to strdup()
and strcpy(), respectively. sockaddr_copy() KASSERTs that the
family of the destination and source sockaddrs are alike.
The 'flags' argumet for sockaddr_alloc() and sockaddr_dup() is
passed directly to pool_get(9).
2 I added routines for initializing sockaddrs in each address
family, sockaddr_in_init(), sockaddr_in6_init(), sockaddr_iso_init(),
etc. They are fairly self-explanatory.
3 structs route_in6 and route_iso are no more. All protocol families
use struct route. I have changed the route cache, 'struct route',
so that it does not contain storage space for a sockaddr. Instead,
struct route points to a sockaddr coming from the pool the sockaddr
belongs to. I added a new method to struct route, rtcache_setdst(),
for setting the cache destination:
int rtcache_setdst(struct route *, const struct sockaddr *);
rtcache_setdst() returns 0 on success, or ENOMEM if no memory is
available to create the sockaddr storage.
It is now possible for rtcache_getdst() to return NULL if, say,
rtcache_setdst() failed. I check the return value for NULL
everywhere in the kernel.
4 Each routing domain (struct domain) has a list of live route
caches, dom_rtcache. rtflushall(sa_family_t af) looks up the
domain indicated by 'af', walks the domain's list of route caches
and invalidates each one.
parentheses in return statements.
Cosmetic: don't open-code TAILQ_FOREACH().
Cosmetic: change types of variables to avoid oodles of casts: in
in6_src.c, avoid casts by changing several route_in6 pointers
to struct route pointers. Remove unnecessary casts to caddr_t
elsewhere.
Pave the way for eliminating address family-specific route caches:
soon, struct route will not embed a sockaddr, but it will hold
a reference to an external sockaddr, instead. We will set the
destination sockaddr using rtcache_setdst(). (I created a stub
for it, but it isn't used anywhere, yet.) rtcache_free() will
free the sockaddr. I have extracted from rtcache_free() a helper
subroutine, rtcache_clear(). rtcache_clear() will "forget" a
cached route, but it will not forget the destination by releasing
the sockaddr. I use rtcache_clear() instead of rtcache_free()
in rtcache_update(), because rtcache_update() is not supposed
to forget the destination.
Constify:
1 Introduce const accessor for route->ro_dst, rtcache_getdst().
2 Constify the 'dst' argument to ifnet->if_output(). This
led me to constify a lot of code called by output routines.
3 Constify the sockaddr argument to protosw->pr_ctlinput. This
led me to constify a lot of code called by ctlinput routines.
4 Introduce const macros for converting from a generic sockaddr
to family-specific sockaddrs, e.g., sockaddr_in: satocsin6,
satocsin, et cetera.
rtcache_init and rtcache_init_noclone lookup ro_dst and store
the result in ro_rt, taking care of the reference counting and
calling the domain specific route cache.
rtcache_free checks if a route was cashed and frees the reference.
rtcache_copy copies ro_dst of the given struct route, checking that
enough space is available and incrementing the reference count of the
cached rtentry if necessary.
rtcache_check validates that the cached route is still up. If it isn't,
it tries to look it up again. Afterwards ro_rt is either a valid again
or NULL.
rtcache_copy is used internally.
Adjust to callers of rtalloc/rtflush in the tree to check the sanity of
ro_dst first (if necessary). If it doesn't fit the expectations, free
the cache, otherwise check if the cached route is still valid. After
that combination, a single check for ro_rt == NULL is enough to decide
whether a new lookup needs to be done with a different ro_dst.
Make the route checking in gre stricter by repeating the loop check
after revalidation.
Remove some unused RADIX_MPATH code in in6_src.c. The logic is slightly
changed here to first validate the route and check RTF_GATEWAY
afterwards. This is sementically equivalent though.
etherip doesn't need sc_route_expire similiar to the gif changes from
dyoung@ earlier.
Based on the earlier patch from dyoung@, reviewed and discussed with
him.
knakahara