Version 3.5 brings:
- Atomic commits of ruleset changes (reduce the chance of ending up in an
inconsistent state).
- A 30% reduction in the size of state table entries.
- Source-tracking (limit number of clients and states per client).
- Sticky-address (the flexibility of round-robin with the benefits of
source-hash).
- Significant improvements to interface handling.
- and many more ...
- Remove pflog and pfsync modules. Things will change in such a fashion
that there will be one module with pf+pflog that can be loaded into
GENERIC without problems (which is what most people want). pfsync is no
longer possible as a module.
- Add multicast address for in-kernel multicast pfsync protocol. Protocol
glue will follow once the import is done.
- Add one more mbuf tag
do not pick up the first local ip address for the source
ip address, return ENETUNREACH instead.
Submitted by: Gleb Smirnoff
Reviewed by: -current (silence)
mode tunnel, take the per-route MTU into account, *if* and *only if* it
is non-zero (as found in struct rt_metrics/rt_metrics_lite).
PR: kern/42727
Obtained from: NetBSD (ip_input.c rev 1.151)
fixes the problem of UDP sockets getting wedged in a connected state (and
bound to their destination) under heavy load.
Temporary bind/connect should probably be deleted in future
as an optimization, as described in "A Faster UDP" [Partridge/Pink 1993].
Notes:
- INP_LOCK() is already held in udp_output(). The connection is in effect
happening at a layer lower than the socket layer, therefore in theory
socket locking should not be needed.
- Inlining the in_pcbdisconnect() operation buys us nothing (in the case
of the current state of the code), as laddr is not part of the
inpcb hash or the udbinfo hash. Therefore there should be no need
to rehash after restoring laddr in the error case (this was a
concern of the original author of the patch).
PR: kern/41765
Requested by: gnn
Submitted by: Jinmei Tatuya (with cleanups)
Tested by: spray(8)
flags relating to several aspects of socket functionality. This change
breaks out several bits relating to send and receive operation into a
new per-socket buffer field, sb_state, in order to facilitate locking.
This is required because, in order to provide more granular locking of
sockets, different state fields have different locking properties. The
following fields are moved to sb_state:
SS_CANTRCVMORE (so_state)
SS_CANTSENDMORE (so_state)
SS_RCVATMARK (so_state)
Rename respectively to:
SBS_CANTRCVMORE (so_rcv.sb_state)
SBS_CANTSENDMORE (so_snd.sb_state)
SBS_RCVATMARK (so_rcv.sb_state)
This facilitates locking by isolating fields to be located with other
identically locked fields, and permits greater granularity in socket
locking by avoiding storing fields with different locking semantics in
the same short (avoiding locking conflicts). In the future, we may
wish to coallesce sb_state and sb_flags; for the time being I leave
them separate and there is no additional memory overhead due to the
packing/alignment of shorts in the socket buffer structure.
your (network) modules as well as any userland that might make sense of
sizeof(struct ifnet).
This does not change the queueing yet. These changes will follow in a
seperate commit. Same with the driver changes, which need case by case
evaluation.
__FreeBSD_version bump will follow.
Tested-by: (i386)LINT
conform to the rfc2734 and rfc3146 standard for IP over firewire and
should eventually supercede the fwe driver. Right now the broadcast
channel number is hardwired and we don't support MCAP for multicast
channel allocation - more infrastructure is required in the firewire
code itself to fix these problems.
SOCK_LOCK(so):
- Hold socket lock over calls to MAC entry points reading or
manipulating socket labels.
- Assert socket lock in MAC entry point implementations.
- When externalizing the socket label, first make a thread-local
copy while holding the socket lock, then release the socket lock
to externalize to userspace.
reference count:
- Assert SOCK_LOCK(so) macros that directly manipulate so_count:
soref(), sorele().
- Assert SOCK_LOCK(so) in macros/functions that rely on the state of
so_count: sofree(), sotryfree().
- Acquire SOCK_LOCK(so) before calling these functions or macros in
various contexts in the stack, both at the socket and protocol
layers.
- In some cases, perform soisdisconnected() before sotryfree(), as
this could result in frobbing of a non-present socket if
sotryfree() actually frees the socket.
- Note that sofree()/sotryfree() will release the socket lock even if
they don't free the socket.
Submitted by: sam
Sponsored by: FreeBSD Foundation
Obtained from: BSD/OS
ruleset, the pcb is looked up once per ipfw_chk() activation.
This is done by extracting the required information out of the PCB
and caching it to the ipfw_chk() stack. This should greatly reduce
PCB looking contention and speed up the processing of UID/GID based
firewall rules (especially with large UID/GID rulesets).
Some very basic benchmarks were taken which compares the number
of in_pcblookup_hash(9) activations to the number of firewall
rules containing UID/GID based contraints before and after this patch.
The results can be viewed here:
o http://people.freebsd.org/~csjp/ip_fw_pcb.png
Reviewed by: andre, luigi, rwatson
Approved by: bmilekic (mentor)
versions of various routers seen:
- Introduce igmp_mtx.
- Protect global variable 'router_info_head' and list fields
in struct router_info with this mutex, as well as
igmp_timers_are_running.
- find_rti() asserts that the caller acquires igmp_mtx.
- Annotate a failure to check the return value of
MALLOC(..., M_NOWAIT).
that m_prepend() is not called with possibility to wait while the
pcb lock is held. What still needs revisiting is whether the
ripcbinfo lock is really required here.
Discussed with: rwatson
process is a non-prison root. The security.jail.allow_raw_sockets
sysctl variable is disabled by default, however if the user enables
raw sockets in prisons, prison-root should not be able to interact
with firewall rule sets.
Approved by: rwatson, bmilekic (mentor)
unless it's in the closed or listening state (remote address
== INADDR_ANY).
If a TCP inpcb is in any other state, it's impossible to steal
its local port or use it for port theft. And if there are
both closed/listening and connected TCP inpcbs on the same
localIP:port couple, the call to in_pcblookup_local() will
find the former due to the design of that function.
No objections raised in: -net, -arch
MFC after: 1 month
of IP options.
net.inet.ip.process_options=0 Ignore IP options and pass packets unmodified.
net.inet.ip.process_options=1 Process all IP options (default).
net.inet.ip.process_options=2 Reject all packets with IP options with ICMP
filter prohibited message.
This sysctl affects packets destined for the local host as well as those
only transiting through the host (routing).
IP options do not have any legitimate purpose anymore and are only used
to circumvent firewalls or to exploit certain behaviours or bugs in TCP/IP
stacks.
Reviewed by: sam (mentor)
labeling new mbufs created from sockets/inpcbs in IPv4. This helps avoid
the need for socket layer locking in the lower level network paths
where inpcb locks are already frequently held where needed. In
particular:
- Use the inpcb for label instead of socket in raw_append().
- Use the inpcb for label instead of socket in tcp_output().
- Use the inpcb for label instead of socket in tcp_respond().
- Use the inpcb for label instead of socket in tcp_twrespond().
- Use the inpcb for label instead of socket in syncache_respond().
While here, modify tcp_respond() to avoid assigning NULL to a stack
variable and centralize assertions about the inpcb when inp is
assigned.
Obtained from: TrustedBSD Project
Sponsored by: DARPA, McAfee Research
o New function ip_findroute() to reduce code duplication for the
route lookup cases. (luigi)
o Store ip_len in host byte order on the stack instead of using
it via indirection from the mbuf. This allows to defer the host
byte conversion to a later point and makes a quicker fallback to
normal ip_input() processing. (luigi)
o Check if route is dampned with RTF_REJECT flag and drop packet
already here when ARP is unable to resolve destination address.
An ICMP unreachable is sent to inform the sender.
o Check if interface output queue is full and drop packet already
here. No ICMP notification is sent because signalling source quench
is depreciated.
o Check if media_state is down (used for ethernet type interfaces)
and drop the packet already here. An ICMP unreachable is sent to
inform the sender.
o Do not account sent packets to the interface address counters. They
are only for packets with that 'ia' as source address.
o Update and clarify some comments.
Submitted by: luigi (most of it)
uncommitted):
Rename ip_claim_next_hop() to m_claim_next_hop(), give it an extra arg
(the type of tag to claim) and push it out of ip_var.h into mbuf.h
alongside all of the other macros that work ok mbuf's and tag's.
jail, which is less restrictive but allows for more flexible
jail usage (for those who are willing to make the sacrifice).
The default is off, but allowing raw sockets within jails can
now be accomplished by tuning security.jail.allow_raw_sockets
to 1.
Turning this on will allow you to use things like ping(8)
or traceroute(8) from within a jail.
The patch being committed is not identical to the patch
in the PR. The committed version is more friendly to
APIs which pjd is working on, so it should integrate
into his work quite nicely. This change has also been
presented and addressed on the freebsd-hackers mailing
list.
Submitted by: Christian S.J. Peron <maneo@bsdpro.com>
PR: kern/65800
possible while maintaining compatibility with the widest range of TCP stacks.
The algorithm is as follows:
---
For connections in the ESTABLISHED state, only resets with
sequence numbers exactly matching last_ack_sent will cause a reset,
all other segments will be silently dropped.
For connections in all other states, a reset anywhere in the window
will cause the connection to be reset. All other segments will be
silently dropped.
---
The necessity of accepting all in-window resets was discovered
by jayanth and jlemon, both of whom have seen TCP stacks that
will respond to FIN-ACK packets with resets not meeting the
strict last_ack_sent check.
Idea by: Darren Reed
Reviewed by: truckman, jlemon, others(?)
1. rt_check() cleanup:
rt_check() is only necessary for some address families to gain access
to the corresponding arp entry, so call it only in/near the *resolve()
routines where it is actually used -- at the moment this is
arpresolve(), nd6_storelladdr() (the call is embedded here),
and atmresolve() (the call is just before atmresolve to reduce
the number of changes).
This change will make it a lot easier to decouple the arp table
from the routing table.
There is an extra call to rt_check() in if_iso88025subr.c to
determine the routing info length. I have left it alone for
the time being.
The interface of arpresolve() and nd6_storelladdr() now changes slightly:
+ the 'rtentry' parameter (really a hint from the upper level layer)
is now passed unchanged from *_output(), so it becomes the route
to the final destination and not to the gateway.
+ the routines will return 0 if resolution is possible, non-zero
otherwise.
+ arpresolve() returns EWOULDBLOCK in case the mbuf is being held
waiting for an arp reply -- in this case the error code is masked
in the caller so the upper layer protocol will not see a failure.
2. arpcom untangling
Where possible, use 'struct ifnet' instead of 'struct arpcom' variables,
and use the IFP2AC macro to access arpcom fields.
This mostly affects the netatalk code.
=== Detailed changes: ===
net/if_arcsubr.c
rt_check() cleanup, remove a useless variable
net/if_atmsubr.c
rt_check() cleanup
net/if_ethersubr.c
rt_check() cleanup, arpcom untangling
net/if_fddisubr.c
rt_check() cleanup, arpcom untangling
net/if_iso88025subr.c
rt_check() cleanup
netatalk/aarp.c
arpcom untangling, remove a block of duplicated code
netatalk/at_extern.h
arpcom untangling
netinet/if_ether.c
rt_check() cleanup (change arpresolve)
netinet6/nd6.c
rt_check() cleanup (change nd6_storelladdr)
from tcp_hostcache would have overridden a (now) lower MTU of
an interface or route that changed since first PMTU discovery.
The bug would have caused TCP to redo the PMTU discovery when
not strictly necessary.
Make a comment about already pre-initialized default values
more clear.
Reviewed by: sam
source address of a packet exists in the routing table. The
default route is ignored because it would match everything and
render the check pointless.
This option is very useful for routers with a complete view of
the Internet (BGP) in the routing table to reject packets with
spoofed or unrouteable source addresses.
Example:
ipfw add 1000 deny ip from any to any not versrcreach
also known in Cisco-speak as:
ip verify unicast source reachable-via any
Reviewed by: luigi
implementation taken directly from OpenBSD.
I've resisted committing this for quite some time because of concern over
TIME_WAIT recycling breakage (sequential allocation ensures that there is a
long time before ports are recycled), but recent testing has shown me that
my fears were unwarranted.
TIME_WAIT recycling cases I was able to generate with http testing tools.
In short, as the old algorithm relied on ticks to create the time offset
component of an ISN, two connections with the exact same host, port pair
that were generated between timer ticks would have the exact same sequence
number. As a result, the second connection would fail to pass the TIME_WAIT
check on the server side, and the SYN would never be acknowledged.
I've "fixed" this by adding random positive increments to the time component
between clock ticks so that ISNs will *always* be increasing, no matter how
quickly the port is recycled.
Except in such contrived benchmarking situations, this problem should never
come up in normal usage... until networks get faster.
No MFC planned, 4.x is missing other optimizations that are needed to even
create the situation in which such quick port recycling will occur.
in favour of rtalloc_ign(), which is what would end up being called
anyways.
There are 25 more instances of rtalloc() in net*/ and
about 10 instances of rtalloc_ign()
we convert ip_len into a network byte order; in_delayed_cksum() still
expects it in host byte order.
The symtom was the ``in_cksum_skip: out of data by %d'' complaints
from the kernel.
To add to the previous commit log. These fixes make tcpdump(1) happy
by not complaining about UDP/TCP checksum being bad for looped back
IP multicast when multicast router is deactivated.
Reported by: Vsevolod Lobko
to implement this mistake.
Fixed some nearby style bugs (initialization in declaration, misformatting
of this initialization, missing blank line after the declaration, and
comparision of the non-boolean result of the initialization with 0 using
"!". In KNF, "!" is not even used to compare booleans with 0).
It was fixed by moving problemetic checks, as well as checks that
doesn't need locking before locks are acquired.
Submitted by: Ryan Sommers <ryans@gamersimpact.com>
In co-operation with: cperciva, maxim, mlaier, sam
Tested by: submitter (previous patch), me (current patch)
Reviewed by: cperciva, mlaier (previous patch), sam (current patch)
Approved by: sam
Dedicated to: enough!
+ struct ifnet: remove unused fields, move ipv6-related field close
to each other, add a pointer to l3<->l2 translation tables (arp,nd6,
etc.) for future use.
+ struct route: remove an unused field, move close to each
other some fields that might likely go away in the future
Previously, Giant would be grabbed at entry to the IP local delivery code
when debug.mpsafenet was set to true, as that implied Giant wouldn't be
grabbed in the driver path. Now, we will use this primitive to
conditionally grab Giant in the event the entire network stack isn't
running MPSAFE (debug.mpsafenet == 0).
Compute the payload checksum for a locally originated IP multicast where
God intended, in ip_mloopback(), rather than doing it in ip_output() and
only when multicast router is active. This is more correct as we do not
fool ip_input() that the packet has the correct payload checksum when in
fact it does not (when multicast router is inactive). This is also more
efficient if we don't join the multicast group we send to, thus allowing
the hardware to checksum the payload.
- Add gre_mtx to protect global softc list.
- Hold gre_mtx over various list operations (insert, delete).
- Centralize if_gre interface teardown in gre_destroy(), and call this
from modevent unload and gre_clone_destroy().
- Export gre_mtx to ip_gre.c, which walks the gre list to look up gre
interfaces during encapsulation. Add a wonking comment on how we need
some sort of drain/reference count mechanism to keep gre references
alive while in use and simultaneous destroy.
This commit does not lockdown softc data, which follows in a future
commit.
- Add encapmtx to protect ip_encap.c global variables (encapsulation
list).
- Unifdef #ifdef 0 pieces of encap_init() which was (and now really
is) basically a no-op.
- Lock encapmtx when walking encaptab, modifying it, comparing
entries, etc.
- Remove spl's.
Note that currently there's no facilite to make sure outstanding
use of encapsulation methods on a table entry have drained bfore
we allow a table entry to be removed. As such, it's currently the
caller's responsibility to make sure that draining takes place.
Reviewed by: mlaier
ifp is now passed explicitly to ether_demux; no need to look it up again.
Make mtag a global var in ip_input.
Noticed by: rwatson
Approved by: bms(mentor)
to NET_UNLOCK_GIANT(). While they are used in similar ways, the
semantics are quite different -- NET_LOCK_GIANT() and NET_UNLOCK_GIANT()
directly wrap mutex lock and unlock operations, whereas drop/pickup
special case the handling of Giant recursion. Add a comment saying
as much.
Add NET_ASSERT_GIANT(), which conditionally asserts Giant based
on the value of debug_mpsafenet.
This enables pf to track dynamic address changes on interfaces (dailup) with
the "on (<ifname>)"-syntax. This also brings hooks in anticipation of
tracking cloned interfaces, which will be in future versions of pf.
Approved by: bms(mentor)
pf/pflog/pfsync as modules. Do not list them in NOTES or modules/Makefile
(i.e. do not connect it to any (automatic) builds - yet).
Approved by: bms(mentor)
the syscall arguments and does the suser() permission check, and
kern_mlock(), which does the resource limit checking and calls
vm_map_wire(). Split munlock() in a similar way.
Enable the RLIMIT_MEMLOCK checking code in kern_mlock().
Replace calls to vslock() and vsunlock() in the sysctl code with
calls to kern_mlock() and kern_munlock() so that the sysctl code
will obey the wired memory limits.
Nuke the vslock() and vsunlock() implementations, which are no
longer used.
Add a member to struct sysctl_req to track the amount of memory
that is wired to handle the request.
Modify sysctl_wire_old_buffer() to return an error if its call to
kern_mlock() fails. Only wire the minimum of the length specified
in the sysctl request and the length specified in its argument list.
It is recommended that sysctl handlers that use sysctl_wire_old_buffer()
should specify reasonable estimates for the amount of data they
want to return so that only the minimum amount of memory is wired
no matter what length has been specified by the request.
Modify the callers of sysctl_wire_old_buffer() to look for the
error return.
Modify sysctl_old_user to obey the wired buffer length and clean up
its implementation.
Reviewed by: bms
increased <netinet/tcp_var>'s already large set of prerequisites, and
this was handled badly. Just don't declare the complete syncache struct
unless <netinet/pcb.h> is included before <netinet/tcp_var.h>.
Approved by: jlemon (years ago, for a more invasive fix)
amount of segments it will hold.
The following tuneables and sysctls control the behaviour of the tcp
segment reassembly queue:
net.inet.tcp.reass.maxsegments (loader tuneable)
specifies the maximum number of segments all tcp reassemly queues can
hold (defaults to 1/16 of nmbclusters).
net.inet.tcp.reass.maxqlen
specifies the maximum number of segments any individual tcp session queue
can hold (defaults to 48).
net.inet.tcp.reass.cursegments (readonly)
counts the number of segments currently in all reassembly queues.
net.inet.tcp.reass.overflows (readonly)
counts how often either the global or local queue limit has been reached.
Tested by: bms, silby
Reviewed by: bms, silby
