explicitly select write locking for all use of the inpcb mutex.
Update some pcbinfo lock assertions to assert locked rather than
write-locked, although in practice almost all uses of the pcbinfo
rwlock main exclusive, and all instances of inpcb lock acquisition
are exclusive.
This change should introduce (ideally) little functional change.
However, it lays the groundwork for significantly increased
parallelism in the TCP/IP code.
MFC after: 3 months
Tested by: kris (superset of committered patch)
while in principle a good idea, opened us up to a race inherrent to
the syncache's direct insertion of incoming TCP connections into the
"completed connection" listen queue, as it transpires that the socket
is inserted before the inpcb is fully filled in by syncache_expand().
The bug manifested with the occasional returning of 0.0.0.0:0 in the
address returned by the accept() system call, which occurred if accept
managed to execute tcp_usr_accept() before syncache_expand() had copied
the endpoint addresses into inpcb connection state.
Re-add tcbinfo locking around the address copyout, which has the effect
of delaying the copy until syncache_expand() has finished running, as
it is run while the tcbinfo lock is held. This is undesirable in that
it increases contention on tcbinfo further, but a more significant
change will be required to how the syncache inserts new sockets in
order to fix this and keep more granular locking here. In particular,
either more state needs to be passed into sonewconn() so that
pru_attach() can fill in the fields *before* the socket is inserted, or
the socket needs to be inserted in the incomplete connection queue
until it is actually ready to be used.
Reported by: glebius (and kris)
Tested by: glebius
drop the lock and then re-acquire it, revalidating TCP connection state
assumptions when we do so. This avoids a potential lock order reversal
(and potential deadlock, although none have been reported) due to the
inpcb lock being held over a page fault.
MFC after: 1 week
PR: 102752
Reviewed by: bz
Reported by: Václav Haisman <v dot haisman at sh dot cvut dot cz>
- Rename output routines tcp_gen_* -> tcp_output_*.
- Rename notification routines that turn in to no-ops in the absence of TOE
from tcp_gen_* -> tcp_offload_*.
- Fix some minor comment nits.
- Add a /* FALLTHROUGH */
Reviewed by: Sam Leffler, Robert Watson, and Mike Silbersack
us to scale up to sb_max, aka kern.ipc.maxsockbuf.
We do this because there are broken firewalls that will corrupt the window
scale option, leading to the other endpoint believing that our advertised
window is unscaled. At scale factors larger than 5 the unscaled window will
drop below 1500 bytes, leading to serious problems when traversing these
broken firewalls.
With the default maxsockbuf of 256K, a scale factor of 3 will be chosen by
this algorithm. Those who choose a larger maxsockbuf should watch out
for the compatiblity problems mentioned above.
Reviewed by: andre
- Reintegrate the ANSI C function declaration change
from tcp_timer.c rev 1.92
- Reorganize the tcpcb structure so that it has a single
pointer to the "tcp_timer" structure which contains all
of the tcp timer callouts. This change means that when
the single tcp timer change is reintegrated, tcpcb will
not change in size, and therefore the ABI between
netstat and the kernel will not change.
Neither of these changes should have any functional
impact.
Reviewed by: bmah, rrs
Approved by: re (bmah)
TCP timers as a single timer, but retain the API changes necessary to
reintroduce this change. This will back out the source of at least two
reported problems: lock leaks in certain timer edge cases, and TCP timers
continuing to fire after a connection has closed (a bug previously fixed and
then reintroduced with the timer rewrite).
In a follow-up commit, some minor restylings and comment changes performed
after the TCP timer rewrite will be reapplied, and a further change to allow
the TCP timer rewrite to be added back without disturbing the ABI. The new
design is believed to be a good thing, but the outstanding issues are
leading to significant stability/correctness problems that are holding
up 7.0.
This patch was generated by silby, but is being committed by proxy due to
poor network connectivity for silby this week.
Approved by: re (kensmith)
Submitted by: silby
Tested by: rwatson, kris
Problems reported by: peter, kris, others
<netinet/tcp_fsm.h> is included into any compilation unit that needs
tcpstates[]. Also remove incorrect extern declarations and TCPDEBUG
conditionals. This allows kernels both with and without TCPDEBUG to
build, and unbreaks the tinderbox.
Approved by: re (rwatson)
protocol entry points using functions named proto_getsockaddr and
proto_getpeeraddr rather than proto_setsockaddr and proto_setpeeraddr.
While it's true that sockaddrs are allocated and set, the net effect is
to retrieve (get) the socket address or peer address from a socket, not
set it, so align names to that intent.
directly to a merged model where only one callout, the next to fire,
is registered.
Instead of callout_reset(9) and callout_stop(9) the new function
tcp_timer_activate() is used which then internally manages the callout.
The single new callout is a mutex callout on inpcb simplifying the
locking a bit.
tcp_timer() is the called function which handles all race conditions
in one place and then dispatches the individual timer functions.
Reviewed by: rwatson (earlier version)
potential issues where the peer does not close, potentially leaving
thousands of connections in FIN_WAIT_2. This is controlled by a new sysctl
fast_finwait2_recycle, which is disabled by default.
Reviewed by: gnn, silby.
socket option TCP_INFO.
Note that the units used in the original Linux API are in microseconds,
so use a 64-bit mantissa to convert FreeBSD's internal measurements
from struct tcpcb from ticks.
Normally the socket buffers are static (either derived from global
defaults or set with setsockopt) and do not adapt to real network
conditions. Two things happen: a) your socket buffers are too small
and you can't reach the full potential of the network between both
hosts; b) your socket buffers are too big and you waste a lot of
kernel memory for data just sitting around.
With automatic TCP send and receive socket buffers we can start with a
small buffer and quickly grow it in parallel with the TCP congestion
window to match real network conditions.
FreeBSD has a default 32K send socket buffer. This supports a maximal
transfer rate of only slightly more than 2Mbit/s on a 100ms RTT
trans-continental link. Or at 200ms just above 1Mbit/s. With TCP send
buffer auto scaling and the default values below it supports 20Mbit/s
at 100ms and 10Mbit/s at 200ms. That's an improvement of factor 10, or
1000%. For the receive side it looks slightly better with a default of
64K buffer size.
New sysctls are:
net.inet.tcp.sendbuf_auto=1 (enabled)
net.inet.tcp.sendbuf_inc=8192 (8K, step size)
net.inet.tcp.sendbuf_max=262144 (256K, growth limit)
net.inet.tcp.recvbuf_auto=1 (enabled)
net.inet.tcp.recvbuf_inc=16384 (16K, step size)
net.inet.tcp.recvbuf_max=262144 (256K, growth limit)
Tested by: many (on HEAD and RELENG_6)
Approved by: re
MFC after: 1 month
upper-bounding it to the size of the initial socket buffer lower-bound it
to the smallest MSS we accept. Ideally we'd use the actual MSS information
here but it is not available yet.
For socket buffer auto sizing to be effective we need room to grow the
receive window. The window scale shift is determined at connection setup
and can't be changed afterwards. The previous, original, method effectively
just did a power of two roundup of the socket buffer size at connection
setup severely limiting the headroom for larger socket buffers.
Tested by: many (as part of the socket buffer auto sizing patch)
MFC after: 1 month
marked INP_DROPPED or INP_TIMEWAIT:
o return ECONNRESET instead of EINVAL for close, disconnect, shutdown,
rcvd, rcvoob, and send operations
o return ECONNABORTED instead of EINVAL for accept
These changes should reduce confusion in applications since EINVAL is
normally interpreted to mean an invalid file descriptor. This change
does not conflict with POSIX or other standards I checked. The return
of EINVAL has always been possible but rare; it's become more common
with recent changes to the socket/inpcb handling and with finer-grained
locking and preemption.
Note: there are other instances of EINVAL for this state that were
left unchanged; they should be reviewed.
Reviewed by: rwatson, andre, ru
MFC after: 1 month
function, pru_close, to notify protocols that the file descriptor or
other consumer of a socket is closing the socket. pru_abort is now a
notification of close also, and no longer detaches. pru_detach is no
longer used to notify of close, and will be called during socket
tear-down by sofree() when all references to a socket evaporate after
an earlier call to abort or close the socket. This means detach is now
an unconditional teardown of a socket, whereas previously sockets could
persist after detach of the protocol retained a reference.
This faciliates sharing mutexes between layers of the network stack as
the mutex is required during the checking and removal of references at
the head of sofree(). With this change, pru_detach can now assume that
the mutex will no longer be required by the socket layer after
completion, whereas before this was not necessarily true.
Reviewed by: gnn
( and where appropriate the destruction) of the pcb mutex to the init/finit
functions of the pcb zones.
This allows locking of the pcb entries and race condition free comparison
of the generation count.
Rearrange locking a bit to avoid extra locking operation to update the generation
count in in_pcballoc(). (in_pcballoc now returns the pcb locked)
I am planning to convert pcb list handling from a type safe to a reference count
model soon. ( As this allows really freeing the PCBs)
Reviewed by: rwatson@, mohans@
MFC after: 1 week
common pcb tear-down logic into tcp_detach(), which is called from
either. Invoke tcp_drop() from the tcp_usr_abort() path rather than
tcp_disconnect(), as we want to drop it immediately not perform a
FIN sequence. This is one reason why some people were experiencing
panics in sodealloc(), as the netisr and aborting thread were
simultaneously trying to tear down the socket. This bug could often
be reproduced using repeated runs of the listenclose regression test.
MFC after: 3 months
PR: 96090
Reported by: Peter Kostouros <kpeter at melbpc dot org dot au>, kris
Tested by: Peter Kostouros <kpeter at melbpc dot org dot au>, kris
immediately rather than jumping to the normal output handling, which
assumes we've pulled out the inpcb, which hasn't happened at this
point (and isn't necessary).
Return ECONNABORTED instead of EINVAL when the inpcb has entered
INP_TIMEWAIT or INP_DROPPED, as this is the documented error value.
This may correct the panic seen by Ganbold.
MFC after: 1 month
Reported by: Ganbold <ganbold at micom dot mng dot net>
disconnect for fully connected sockets was dropped, meaning that if
the socket was closed while the connection was alive, it would be
leaked. Structure tcp_usr_detach() so that there are two clear
parts: initiating disconnect, and reclaiming state, and reintroduce
the tcp_disconnect() call in the first part.
MFC after: 3 months
socket can have a tcp connection that has entered time wait
attached to it, in the event that shutdown() is called on the
socket and the FINs properly exchange before close(). In this
case we don't detach or free the inpcb, just leave the tcptw
detached and freed, but we must release the inpcb lock (which we
didn't previously).
MFC after: 3 months
pru_abort(), pru_detach(), and in_pcbdetach():
- Universally support and enforce the invariant that so_pcb is
never NULL, converting dozens of unnecessary NULL checks into
assertions, and eliminating dozens of unnecessary error handling
cases in protocol code.
- In some cases, eliminate unnecessary pcbinfo locking, as it is no
longer required to ensure so_pcb != NULL. For example, the receive
code no longer requires the pcbinfo lock, and the send code only
requires it if building a new connection on an otherwise unconnected
socket triggered via sendto() with an address. This should
significnatly reduce tcbinfo lock contention in the receive and send
cases.
- In order to support the invariant that so_pcb != NULL, it is now
necessary for the TCP code to not discard the tcpcb any time a
connection is dropped, but instead leave the tcpcb until the socket
is shutdown. This case is handled by setting INP_DROPPED, to
substitute for using a NULL so_pcb to indicate that the connection
has been dropped. This requires the inpcb lock, but not the pcbinfo
lock.
- Unlike all other protocols in the tree, TCP may need to retain access
to the socket after the file descriptor has been closed. Set
SS_PROTOREF in tcp_detach() in order to prevent the socket from being
freed, and add a flag, INP_SOCKREF, so that the TCP code knows whether
or not it needs to free the socket when the connection finally does
close. The typical case where this occurs is if close() is called on
a TCP socket before all sent data in the send socket buffer has been
transmitted or acknowledged. If INP_SOCKREF is found when the
connection is dropped, we release the inpcb, tcpcb, and socket instead
of flagging INP_DROPPED.
- Abort and detach protocol switch methods no longer return failures,
nor attempt to free sockets, as the socket layer does this.
- Annotate the existence of a long-standing race in the TCP timer code,
in which timers are stopped but not drained when the socket is freed,
as waiting for drain may lead to deadlocks, or have to occur in a
context where waiting is not permitted. This race has been handled
by testing to see if the tcpcb pointer in the inpcb is NULL (and vice
versa), which is not normally permitted, but may be true of a inpcb
and tcpcb have been freed. Add a counter to test how often this race
has actually occurred, and a large comment for each instance where
we compare potentially freed memory with NULL. This will have to be
fixed in the near future, but requires is to further address how to
handle the timer shutdown shutdown issue.
- Several TCP calls no longer potentially free the passed inpcb/tcpcb,
so no longer need to return a pointer to indicate whether the argument
passed in is still valid.
- Un-macroize debugging and locking setup for various protocol switch
methods for TCP, as it lead to more obscurity, and as locking becomes
more customized to the methods, offers less benefit.
- Assert copyright on tcp_usrreq.c due to significant modifications that
have been made as part of this work.
These changes significantly modify the memory management and connection
logic of our TCP implementation, and are (as such) High Risk Changes,
and likely to contain serious bugs. Please report problems to the
current@ mailing list ASAP, ideally with simple test cases, and
optionally, packet traces.
MFC after: 3 months
rather than an error. Detaches do not "fail", they other occur or
the protocol flags SS_PROTOREF to take ownership of the socket.
soclose() no longer looks at so_pcb to see if it's NULL, relying
entirely on the protocol to decide whether it's time to free the
socket or not using SS_PROTOREF. so_pcb is now entirely owned and
managed by the protocol code. Likewise, no longer test so_pcb in
other socket functions, such as soreceive(), which have no business
digging into protocol internals.
Protocol detach routines no longer try to free the socket on detach,
this is performed in the socket code if the protocol permits it.
In rts_detach(), no longer test for rp != NULL in detach, and
likewise in other protocols that don't permit a NULL so_pcb, reduce
the incidence of testing for it during detach.
netinet and netinet6 are not fully updated to this change, which
will be in an upcoming commit. In their current state they may leak
memory or panic.
MFC after: 3 months
than an int, as an error here is not meaningful. Modify soabort() to
unconditionally free the socket on the return of pru_abort(), and
modify most protocols to no longer conditionally free the socket,
since the caller will do this.
This commit likely leaves parts of netinet and netinet6 in a situation
where they may panic or leak memory, as they have not are not fully
updated by this commit. This will be corrected shortly in followup
commits to these components.
MFC after: 3 months
following the protocol pru_listen() call to solisten_proto(), so
that it occurs under the socket lock acquisition that also sets
SO_ACCEPTCONN. This requires passing the new backlog parameter
to the protocol, which also allows the protocol to be aware of
changes in queue limit should it wish to do something about the
new queue limit. This continues a move towards the socket layer
acting as a library for the protocol.
Bump __FreeBSD_version due to a change in the in-kernel protocol
interface. This change has been tested with IPv4 and UNIX domain
sockets, but not other protocols.
- most of the kernel code will not care about the actual encoding of
scope zone IDs and won't touch "s6_addr16[1]" directly.
- similarly, most of the kernel code will not care about link-local
scoped addresses as a special case.
- scope boundary check will be stricter. For example, the current
*BSD code allows a packet with src=::1 and dst=(some global IPv6
address) to be sent outside of the node, if the application do:
s = socket(AF_INET6);
bind(s, "::1");
sendto(s, some_global_IPv6_addr);
This is clearly wrong, since ::1 is only meaningful within a single
node, but the current implementation of the *BSD kernel cannot
reject this attempt.
Submitted by: JINMEI Tatuya <jinmei__at__isl.rdc.toshiba.co.jp>
Obtained from: KAME
