Remove tcp_input lock statistics; these are intended for debugging only
and are not intended to ship in 8.0 as they dirty additional cache
lines in a performance-critical per-packet path.
Approved by: re (kib, bz)
In tcp_input(), we acquire a global write lock at first only if a
segment is likely to trigger a TCP state change (i.e., FIN/RST/SYN).
If we later have to upgrade the lock, we acquire an inpcb reference
and drop both global/inpcb locks before reacquiring in-order. In
that gap, the connection may transition into TIMEWAIT, so we need
to loop back and reevaluate the inpcb after relocking.
Reported by: Kamigishi Rei <spambox at haruhiism.net>
Reviewed by: bz
Approved by: re (kib)
Remove a log message from production code. This log message can be
triggered by a misconfigured host that is sending out gratuious ARPs.
This log message can also be triggered during a network renumbering
event when multiple prefixes co-exist on a single network segment.
Approved by: re
Previously, if an address alias is configured on an interface, and
this address alias has a prefix matching that of another address
configured on the same interface, then the ARP entry for the alias
is not deleted from the ARP table when that address alias is removed.
This patch fixes the aforementioned issue.
PR: kern/139113
Reviewed by: bz
Approved by: re
Fixes to mcast userland API.
--
Fix an API issue in leave processing for IPv4 multicast groups.
* Do not assume that the group lookup performed by imo_match_group()
is valid when ifp is NULL in this case.
* Instead, return EADDRNOTAVAIL if the ifp cannot be resolved for the
membership we are being asked to leave.
Caveat user:
* The way IPv4 multicast memberships are implemented in the inpcb layer
at the moment, has the side-effect that struct ip_moptions will
still hold the membership, under the old ifp, until ip_freemoptions()
is called for the parent inpcb.
* The underlying issue is: the inpcb layer does not get notification
of ifp being detached going away in a thread-safe manner.
This is non-trivial to fix.
--
Fix an obvious logic error in the IPv4 multicast leave processing,
where the filter mode vector was not updated correctly after the leave.
--
Tighten input checking in inp_join_group():
* Don't try to use the source address, when its family is unspecified.
* If we get a join without a source, on an existing inclusive
mode group, this is an error, as it would change the filter mode.
Fix a problem with the handling of in_mfilter for new memberships:
* Do not rely on imf being NULL; it is explicitly initialized to a
non-NULL pointer when constructing a membership.
* Explicitly initialize *imf to EX mode when the source address
is unspecified.
This fixes a problem with in_mfilter slot recycling in the join path.
--
Don't allow joins w/o source on an existing group.
This is almost always pilot error.
We don't need to check for group filter UNDEFINED state at t1,
because we only ever allocate filters with their groups, so we
unconditionally reject such calls with EINVAL.
Trying to change the active filter mode w/o going through IP_MSFILTER
is also disallowed.
Deals with the case described in PR 137164 upfront, cumulative
with the fix in svn rev 197132 which only calls imo_match_source()
if the source address family was not unspecified.
--
Revision 197136 has a text conflict, however it is a comment only change.
PR: 137164, 138689, 138690, 138691
Submitted by: Stef Walter (with fixups)
Approved by: re (kib)
Fix a bug reported by Daniel Mentz:
When authenticating DATA chunks some DATA chunks
might get stuck when the MTU gets decreased via
an ICMP message.
Approved by: re, rrs (mentor)
1) A lock issue, if we ever had to try again
we would double lock the INP lock.
2) We were allowing (at wrap) associd 0... which really
we cannot allow since 0 normally means in most socket
API calls that we are wishing to effect something on
the INP not TCB.
Approved by: re, rrs (mentor)
Self pointing routes are installed for configured interface addresses
and address aliases. After an interface is brought down and brought
back up again, those self pointing routes disappeared. This patch
ensures after an interface is brought back up, the loopback routes
are reinstalled properly.
Reviewed by: bz
Approved by: re
The bootp code installs an interface address and the nfs client
module tries to install the same address again. This extra code
is removed, which was discovered by the removal of a call to
in_ifscrub() in r196714. This call to in_ifscrub is put back here
because the SIOCAIFADDR command can be used to change the prefix
length of an existing alias.
r197235 reverts file nfs_vfsops.c
Reviewed by: kmacy
Approved by: re
This patch fixes the following issues:
- Routing messages are not generated when adding and removing
interface address aliases.
- Loopback route installed for an interface address alias is
not deleted from the routing table when that address alias
is removed from the associated interface.
- Function in_ifscrub() is called extraneously.
Reviewed by: gnn, kmacy, sam
Approved by: re
Previously local end of point-to-point interface is not reachable
within the system that owns the interface. Packets destined to
the local end point leak to the wire towards the default gateway
if one exists. This behavior is changed as part of the L2/L3
rewrite efforts. The local end point is now reachable within the
system. The inpcb code needs to consider this fact during the
address selection process.
Reviewed by: bz
Approved by: re
When joining a multicast group, the inp_lookup_mcast_ifp call
does a KASSERT that the group address is multicast, so the
check if this is indeed true and eventually return a EINVAL if not,
should be done before calling inp_lookup_mcast_ifp. This fixes a kernel
crash when calling setsockopt (sock, IPPROTO_IP, IP_ADD_MEMBERSHIP,...)
with invalid group address.
Reviewed by: bms
Approved by: re (kib)
In case an upper layer protocol tries to send a packet but the
L2 code does not have the ethernet address for the destination
within the broadcast domain in the table, we remember the
original mbuf in `la_hold' in arpresolve() and send out a
different packet with an arp request.
In case there will be more upper layer packets to send we will
free an earlier one held in `la_hold' and queue the new one.
Once we get a packet in, with which we can perfect our arp table
entry we send out the original 'on hold' packet, should there
be any.
Rather than continuing to process the packet that we received,
we returned without freeing the packet that came in, which
basically means that we leaked an mbuf for every arp request
we sent.
Rather than freeing the received packet and returning, continue
to process the incoming arp packet as well.
This should (a) improve some setups, also proxy-arp, in case it was an
incoming arp request and (b) resembles the behaviour FreeBSD had
from day 1, which alignes with RFC826 "Packet reception" (merge case).
Rename 'm0' to 'hold' to make the code more understandable as
well as diffable to earlier versions more easily.
Handle the link-layer entry 'la' lock comepletely in the block
where needed and release it as early as possible, rather than
holding it longer, down to the end of the function.
Found by: pointyhat, ns1
Bug hunting session with: erwin, simon, rwatson
Tested by: simon on cluster machines
Reviewed by: ratson, kmacy, julian
Approved by: re (kib)
Do not try to free the rt_lle entry of the cached route in
ip_output() if the cached route was not initialized from the
flow-table. The rt_lle entry is invalid unless it has been
initialized through the flow-table.
Reviewed by: kmacy, rwatson
Approved by: re
Use locks specific to the lltable code, rather than borrow the ifnet
list/index locks, to protect link layer address tables. This avoids
lock order issues during interface teardown, but maintains the bug that
sysctl copy routines may be called while a non-sleepable lock is held.
Reviewed by: bz, kmacy, qingli
Approved by: re (kib)
Rework global locks for interface list and index management, correcting
several critical bugs, including race conditions and lock order issues:
Replace the single rwlock, ifnet_lock, with two locks, an rwlock and an
sxlock. Either can be held to stablize the lists and indexes, but both
are required to write. This allows the list to be held stable in both
network interrupt contexts and sleepable user threads across sleeping
memory allocations or device driver interactions. As before, writes to
the interface list must occur from sleepable contexts.
Reviewed by: bz, julian
Approved by: re (kib)
Introduce a div_destroy() function which takes over per-vnet cleanup tasks
from the existing modevent / MOD_UNLOAD handler, and register div_destroy()
in protosw as per-vnet .pr_destroy() handler for options VIMAGE builds. In
nooptions VIMAGE builds, div_destroy() will be invoked from the modevent
handler, resulting in effectively identical operation as it was prior this
change. div_destroy() also tears down hashtables used by ipdivert, which
were previously left behind on ipdivert kldunloads.
For options VIMAGE builds only, temporarily disable kldunloading of ipdivert,
because without introducing additional locking logic it is impossible to
atomically check whether all ipdivert instances in all vnets are idle, and
proceed with cleanup without opening a race window for a vnet to open an
ipdivert socket while ipdivert tear-down is in progress.
While here, staticize div_init(), because it is not used outside of
ip_divert.c.
In cooperation with: julian
Approved by: re (rwatson), julian (mentor)
Approved by: re (rwatson)
Fix ipfw's initialization functions to get the correct order of evaluation
to allow vnet and non vnet operation. Move some functions from ip_fw_pfil.c
to ip_fw2.c and mode to mostly using the SYSINIT and VNET_SYSINIT handlers
instead of the modevent handler. Correct some spelling errors in comments
in the affected code. Note this bug fixes a crash in NON VIMAGE kernels when
ipfw is unloaded.
This patch is a minimal patch for 8.0
I have a much larger patch that actually fixes the underlying problems
that will be applied after 8.0
Reviewed by: zec@, rwatson@, bz@(earlier version)
Approved by: re (rwatson)
Fix CARP memory leaks on carp_if's malloc'd using M_CARP. This occurs when
CARP tries to free them using M_IFADDR after the last address for a virtual
host is removed and when detaching from the parent interface.
Approved by: re (kib), ken (mentor)
- change the interface to flowtable_lookup so that we don't rely on
the mbuf for obtaining the fib index
- check that a cached flow corresponds to the same fib index as the
packet for which we are doing the lookup
- at interface detach time flush any flows referencing stale rtentrys
associated with the interface that is going away (fixes reported
panics)
- reduce the time between cleans in case the cleaner is running at
the time the eventhandler is called and the wakeup is missed less
time will elapse before the eventhandler returns
- separate per-vnet initialization from global initialization
(pointed out by jeli@)
Reviewed by: sam@
Approved by: re@
* Fix a bug where PR-SCTP settings are ignore when using implicit
association setup.
* Fix a bug where message with illegal stream ids are not deleted.
* Fix a crash when reporting back unsent messages from the send_queue.
* Fix a bug related to INIT retransmission when the socket is already
closed.
* Fix a bug where associations were stalled when partial delivery API
was enabled.
* Fix a bug where the receive buffer size was smaller than the
partial_delivery_point.
Approved by: re, rrs (mentor)
In function ip_output(), the cached route is flushed when there is a
mismatch between the cached entry and the intended destination. The
cached rtentry{} is flushed but the associated llentry{} is not. This
causes the wrong destination MAC address being used in the output
packets. The fix is to flush the llentry{} when rtentry{} is cleared.
Reviewed by: kmacy, rwatson
Approved by: re
SCTP is not yet compatible with options VIMAGE kernels although it compiles
with VIMAGE defined, so explicitly disallow building such kernels.
Reviewed by: rrs
Approved by: re (rwatson), julian (mentor)
Approved by: re (rwatson)
URL: http://svn.freebsd.org/changeset/base/196201
Fix ipfw crash on uid or gid check.
Receiving any ip packet for which there is no existing socket will
crash if ipfw has a uid or gid test rule, as the uid/gid
of the non existent owner of said non existent socket is tested.
Brooks introduced this error as part of his >16 gids patch.
It appears to be a cut-n-paste error from similar code a few lines
before. The old code used the 'pcb' variable here, but in the
new code that switched the 'inp' variable, which is often NULL
and what is tested in the code further up. The rest of the multi-gid
patch for ipfw seems solid (and cleaner than previous code).
p.s. What's up with all the properties changing? It is a fresh checkout.
Reviewed by: brooks
Approved by: re (rwatson)
all pertinent statatistics for the subsystem. These structures are
sometimes "borrowed" by kernel modules that require a place to store
statistics for similar events.
Add KPI accessor functions for statistics structures referenced by kernel
modules so that they no longer encode certain specifics of how the data
structures are named and stored. This change is intended to make it
easier to move to per-CPU network stats following 8.0-RELEASE.
The following modules are affected by this change:
if_bridge
if_cxgb
if_gif
ip_mroute
ipdivert
pf
In practice, most of these statistics consumers should, in fact, maintain
their own statistics data structures rather than borrowing structures
from the base network stack. However, that change is too agressive for
this point in the release cycle.
Reviewed by: bz
Approved by: re (kib)
vnet.h, we now use jails (rather than vimages) as the abstraction
for virtualization management, and what remained was specific to
virtual network stacks. Minor cleanups are done in the process,
and comments updated to reflect these changes.
Reviewed by: bz
Approved by: re (vimage blanket)
processing code holds the read lock (when processing a
FWD-TSN for pr-sctp). If it finds stranded data that
can be given to the application, it calls sctp_add_to_readq().
The readq function also grabs this lock. So if INVAR is on
we get a double recurse on a non-recursive lock and panic.
This fix will change it so that readq() function gets a
flag to tell if the lock is held, if so then it does not
get the lock.
Approved by: re@freebsd.org (Kostik Belousov)
MFC after: 1 week
- Allow loopback route to be installed for address assigned to
interface of IFF_POINTOPOINT type.
- Install loopback route for an IPv4 interface addreess when the
"useloopback" sysctl variable is enabled. Similarly, install
loopback route for an IPv6 interface address when the sysctl variable
"nd6_useloopback" is enabled. Deleting loopback routes for interface
addresses is unconditional in case these sysctl variables were
disabled after an interface address has been assigned.
Reviewed by: bz
Approved by: re
network stacks, VNET_SYSINIT:
- Add VNET_SYSINIT and VNET_SYSUNINIT macros to declare events that will
occur each time a network stack is instantiated and destroyed. In the
!VIMAGE case, these are simply mapped into regular SYSINIT/SYSUNINIT.
For the VIMAGE case, we instead use SYSINIT's to track their order and
properties on registration, using them for each vnet when created/
destroyed, or immediately on module load for already-started vnets.
- Remove vnet_modinfo mechanism that existed to serve this purpose
previously, as well as its dependency scheme: we now just use the
SYSINIT ordering scheme.
- Implement VNET_DOMAIN_SET() to allow protocol domains to declare that
they want init functions to be called for each virtual network stack
rather than just once at boot, compiling down to DOMAIN_SET() in the
non-VIMAGE case.
- Walk all virtualized kernel subsystems and make use of these instead
of modinfo or DOMAIN_SET() for init/uninit events. In some cases,
convert modular components from using modevent to using sysinit (where
appropriate). In some cases, do minor rejuggling of SYSINIT ordering
to make room for or better manage events.
Portions submitted by: jhb (VNET_SYSINIT), bz (cleanup)
Discussed with: jhb, bz, julian, zec
Reviewed by: bz
Approved by: re (VIMAGE blanket)
non-vrtiualized sysctls so we cannot used one common function.
Add a macro to convert the arg1 in the virtualized case to
vnet.h to not expose the maths to all over the code.
Add a wrapper for the single virtualized call, properly handling
arg1 and call the default implementation from there.
Convert the two over places to use the new macro.
Reviewed by: rwatson
Approved by: re (kib)
back to the bottom of ip_init() as found in 7.x. I missed the fact that
the bottom half of the init routine only runs in the !VNET case.
Submitted by: zec
Approved by: re (vimage blanket)
nor destructors, as there's no actual work to do.
In most cases, the constructors weren't needed because of the existing
protocol initialization functions run by net_init_domain() as part of
VNET_MOD_NET, or they were eliminated when support for static
initialization of virtualized globals was added.
Garbage collect dependency references to modules without constructors or
destructors, notably VNET_MOD_INET and VNET_MOD_INET6.
Reviewed by: bz
Approved by: re (vimage blanket)
