In the transport call on 12/3 Gleb asked to move the CC modules towards
using reference counting to prevent folks from unloading a module in use.
It was also agreed that Michael would do a user space utility like tcp_drop
that could be used to move all connections that are using a specific CC
to some other CC.
This is the half I committed to doing, making it so that we maintain a refcount
on a cc module every time a pcb refers to it and decrementing that every
time a pcb no longer uses a cc module. This also helps us simplify the
whole unloading process by getting rid of tcp_ccunload() which munged
through all the tcb's. Instead we mark a module as being removed and
prevent further references to it. We also make sure that if a module is
marked as being removed it cannot be made as the default and also
the opposite of that, if its a default it fails and does not mark it as being
removed.
Reviewed by: Michael Tuexen, Gleb Smirnoff
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D33249
The padding in CURRENT shall not shrink. It is
designed that in CURRENT at always stays
the same, and then when a new stable is branched, it
inherits 6 pointer placeholders that can be used
withing this stable/X lifetime to extend the structure.
Reviewed By: tuexen, #transport
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D34269
Reserve couters in the tcps struct in preparation
for AccECN, extend the debugging output for TF2
flags, optimize the syncache flags from individual
bits to a codepoint for the specifc ECN handshake.
This is in preparation of AccECN.
No functional chance except for extended debug
output capabilities.
Reviewed By: #transport, rrs
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D34161
tcp_ctloutput_set() will be used via the sysctl interface in a
upcoming command line tool tcpsso.
Reviewed by: glebius, rscheff
Sponsored by: Netflix, Inc.
Differential Revision: https://reviews.freebsd.org/D34164
In order to consistently provide access to all
(including reserved) TCP header flag bits,
use an accessor function tcp_get_flags and
tcp_set_flags. Also expand any flag variable from
uint8_t / char to uint16_t.
Reviewed By: hselasky, tuexen, glebius, #transport
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D34130
Consistently only pass the inp and the sopt around. Don't pass the
so around, since in a upcoming commit tcp_ctloutput_set() will be
called from a context different from setsockopt(). Also expect
the inp to be locked when calling tcp_ctloutput_[gs]et(), this is
also required for the upcoming use by tcpsso, a command line tool
to set socket options.
Reviewed by: glebius, rscheff
Sponsored by: Netflix, Inc.
Differential Revision: https://reviews.freebsd.org/D34151
Under rare circumstances, a spurious retranmission is
incorrectly detected and rewound, messing up various tcpcb values,
which can lead to a panic when SACK is in use.
Reviewed By: tuexen, chengc_netapp.com, #transport
MFC after: 3 days
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D33979
The historical BSD network stack loop that rolls over domains and
over protocols has no advantages over more modern SYSINIT(9).
While doing the sweep, split global and per-VNET initializers.
Getting rid of pr_init allows to achieve several things:
o Get rid of ifdef's that protect against double foo_init() when
both INET and INET6 are compiled in.
o Isolate initializers statically to the module they init.
o Makes code easier to understand and maintain.
Reviewed by: melifaro
Differential revision: https://reviews.freebsd.org/D33537
The advanced TCP stacks (bbr, rack) may decide to drop a TCP connection
when they do output on it. The default stack never does this, thus
existing framework expects tcp_output() always to return locked and
valid tcpcb.
Provide KPI extension to satisfy demands of advanced stacks. If the
output method returns negative error code, it means that caller must
call tcp_drop().
In tcp_var() provide three inline methods to call tcp_output():
- tcp_output() is a drop-in replacement for the default stack, so that
default stack can continue using it internally without modifications.
For advanced stacks it would perform tcp_drop() and unlock and report
that with negative error code.
- tcp_output_unlock() handles the negative code and always converts
it to positive and always unlocks.
- tcp_output_nodrop() just calls the method and leaves the responsibility
to drop on the caller.
Sweep over the advanced stacks and use new KPI instead of using HPTS
delayed drop queue for that.
Reviewed by: rrs, tuexen
Differential revision: https://reviews.freebsd.org/D33370
When a connection is established to use TCP-MD5, tcp_twrespond() doesn't
respond with a signed segment. This results in the host performing the
active close to remain in a TIME_WAIT state and the other host in the
LAST_ACK state. Fix this by sending a signed segment when the connection
is established to use TCP-MD5.
Reviewed by: glebius
Differential Revision: https://reviews.freebsd.org/D33490
The structure goes away anyway, but it would be interesting to know
how much memory we used to save with it. So for the record, structure
size with this revision is 64 bytes.
This reverts commit 266f97b5e9a7958e365e78288616a459b40d924a, reversing
changes made to a10253cffea84c0c980a36ba6776b00ed96c3e3b.
A mismerge of a merge to catch up to main resulted in files being
committed which should not have been.
With introduction of epoch(9) synchronization to network stack the
inpcb database became protected by the network epoch together with
static network data (interfaces, addresses, etc). However, inpcb
aren't static in nature, they are created and destroyed all the
time, which creates some traffic on the epoch(9) garbage collector.
Fairly new feature of uma(9) - Safe Memory Reclamation allows to
safely free memory in page-sized batches, with virtually zero
overhead compared to uma_zfree(). However, unlike epoch(9), it
puts stricter requirement on the access to the protected memory,
needing the critical(9) section to access it. Details:
- The database is already build on CK lists, thanks to epoch(9).
- For write access nothing is changed.
- For a lookup in the database SMR section is now required.
Once the desired inpcb is found we need to transition from SMR
section to r/w lock on the inpcb itself, with a check that inpcb
isn't yet freed. This requires some compexity, since SMR section
itself is a critical(9) section. The complexity is hidden from
KPI users in inp_smr_lock().
- For a inpcb list traversal (a pcblist sysctl, or broadcast
notification) also a new KPI is provided, that hides internals of
the database - inp_next(struct inp_iterator *).
Reviewed by: rrs
Differential revision: https://reviews.freebsd.org/D33022
Until this change there were two places where we would free tcpcb -
tcp_discardcb() in case if all timers are drained and tcp_timer_discard()
otherwise. They were pretty much copy-n-paste, except that in the
default case we would run tcp_hc_update(). Merge this into single
function tcp_freecb() and move new short version of tcp_timer_discard()
to tcp_timer.c and make it static.
Reviewed by: rrs, hselasky
Differential revision: https://reviews.freebsd.org/D32965
Pass control for IP/IP6 level options from generic tcp_ctloutput_set()
down to per-stack ctloutput.
Call tcp6_use_min_mtu() from tcp stack tcp_default_ctloutput().
Reviewed by: rrs
Differential Revision: https://reviews.freebsd.org/D32655
TCP stack sysctl nodes are currently inserted using the stack
name alias. Allow the user to get the current stack's alias to
allow for programatic sysctl access.
Obtained from: Netflix
DSACK accounting has been for quite some time under a NETFLIX_STATS ifdef. Statistics
on DSACKs however are very useful in figuring out how much bad retransmissions you
are doing. This is further complicated, however, by stacks that do TLP. A TLP
when discovering a lost ack in the reverse path will cause the generation
of a DSACK. For this situation we introduce a new dsack-tlp-bytes as well
as the more traditional dsack-bytes and dsack-packets. These will now
all display in netstat -p tcp -s. This also updates all stacks that
are currently built to keep track of these stats.
Reviewed by: tuexen
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D32158
If the "len" variable is non-zero, we can assume that the sum of
"tp->t_snd_rxt_bytes + tp->t_sndbytes" is also non-zero.
It is also assumed that the 64-bit byte counters will never wrap around.
Differential Revision: https://reviews.freebsd.org/D31959
Reviewed by: gallatin, rrs and tuexen
Found by: "I told you so", also called hselasky
MFC after: 1 week
Sponsored by: NVIDIA Networking
In reviewing tcp_lro.c we have a possibility that some drives may send a mbuf into
LRO without making sure that the checksum passes. Some drivers actually are
aware of this and do not call lro when the csum failed, others do not do this and
thus could end up sending data up that we think has a checksum passing when
it does not.
This change will fix that situation by properly verifying that the mbuf
has the correct markings (CSUM VALID bits as well as csum in mbuf header
is set to 0xffff).
Reviewed by: tuexen, hselasky, gallatin
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D31155
Ifnet (inline) hw kTLS NICs typically keep state within
a TLS record, so that when transmitting in-order,
they can continue encryption on each segment sent without
DMA'ing extra state from the host.
This breaks down when transmits are out of order (eg,
TCP retransmits). In this case, the NIC must re-DMA
the entire TLS record up to and including the segment
being retransmitted. This means that when re-transmitting
the last 1448 byte segment of a TLS record, the NIC will
have to re-DMA the entire 16KB TLS record. This can lead
to the NIC running out of PCIe bus bandwidth well before
it saturates the network link if a lot of TCP connections have
a high retransmoit rate.
This change introduces a new sysctl (kern.ipc.tls.ifnet_max_rexmit_pct),
where TCP connections with higher retransmit rate will be
switched to SW kTLS so as to conserve PCIe bandwidth.
Reviewed by: hselasky, markj, rrs
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D30908
Hardware TLS is now supported in some interface cards and it works well. Except that
when we have connections that retransmit a lot we get into trouble with all the retransmits.
This prep step makes way for change that Drew will be making so that we can "kick out" a
session from hardware TLS.
Reviewed by: mtuexen, gallatin
Sponsored by: Netflix Inc
Differential Revision: https://reviews.freebsd.org/D30895
When running at NF the current Rack and BBR changes with the recent
commits from Richard that cause the socket buffer lock to be held over
the ip_output() call and then finally culminating in a call to tcp_handle_wakeup()
we get a lot of leaked mbufs. I don't think that this leak is actually caused
by holding the lock or what Richard has done, but is exposing some other
bug that has probably been lying dormant for a long time. I will continue to
look (using his changes) at what is going on to try to root cause out the issue.
In the meantime I can't leave the leaks out for everyone else. So this commit
will revert all of Richards changes and move both Rack and BBR back to just
doing the old sorwakeup_locked() calls after messing with the so_rcv buffer.
We may want to look at adding back in Richards changes after I have pinpointed
the root cause of the mbuf leak and fixed it.
Reviewed by: mtuexen,rscheff
Sponsored by: Netflix Inc
Differential Revision: https://reviews.freebsd.org/D30704
r367492 would unlock the socket buffer before eventually calling the upcall.
This leads to problematic interaction with NFS kernel server/client components
(MP threads) accessing the socket buffer with potentially not correctly updated
state.
Reported by: rmacklem
Reviewed By: tuexen, #transport
Tested by: rmacklem, otis
MFC after: 2 weeks
Sponsored By: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D29690
Recover from excessive losses without reverting to a
retransmission timeout (RTO). Disabled by default, enable
with sysctl net.inet.tcp.do_lrd=1
Reviewed By: #transport, rrs, tuexen, #manpages
Sponsored by: Netapp, Inc.
Differential Revision: https://reviews.freebsd.org/D28931
This fixes several breakages (panics) since the tcp_lro code was
committed that have been reported. Quite a few new features are
now in rack (prefecting of DGP -- Dynamic Goodput Pacing among the
largest). There is also support for ack-war prevention. Documents
comming soon on rack..
Sponsored by: Netflix
Reviewed by: rscheff, mtuexen
Differential Revision: https://reviews.freebsd.org/D30036
This change makes the TCP LRO code more generic and flexible with regards
to supporting multiple different TCP encapsulation protocols and in general
lays the ground for broader TCP LRO support. The main job of the TCP LRO code is
to merge TCP packets for the same flow, to reduce the number of calls to upper
layers. This reduces CPU and increases performance, due to being able to send
larger TSO offloaded data chunks at a time. Basically the TCP LRO makes it
possible to avoid per-packet interaction by the host CPU.
Because the current TCP LRO code was tightly bound and optimized for TCP/IP
over ethernet only, several larger changes were needed. Also a minor bug was
fixed in the flushing mechanism for inactive entries, where the expire time,
"le->mtime" was not always properly set.
To avoid having to re-run time consuming regression tests for every change,
it was chosen to squash the following list of changes into a single commit:
- Refactor parsing of all address information into the "lro_parser" structure.
This easily allows to reuse parsing code for inner headers.
- Speedup header data comparison. Don't compare field by field, but
instead use an unsigned long array, where the fields get packed.
- Refactor the IPv4/TCP/UDP checksum computations, so that they may be computed
recursivly, only applying deltas as the result of updating payload data.
- Make smaller inline functions doing one operation at a time instead of
big functions having repeated code.
- Refactor the TCP ACK compression code to only execute once
per TCP LRO flush. This gives a minor performance improvement and
keeps the code simple.
- Use sbintime() for all time-keeping. This change also fixes flushing
of inactive entries.
- Try to shrink the size of the LRO entry, because it is frequently zeroed.
- Removed unused TCP LRO macros.
- Cleanup unused TCP LRO statistics counters while at it.
- Try to use __predict_true() and predict_false() to optimise CPU branch
predictions.
Bump the __FreeBSD_version due to changing the "lro_ctrl" structure.
Tested by: Netflix
Reviewed by: rrs (transport)
Differential Revision: https://reviews.freebsd.org/D29564
MFC after: 2 week
Sponsored by: Mellanox Technologies // NVIDIA Networking
Adding support for TCP over UDP allows communication with
TCP stacks which can be implemented in userspace without
requiring special priviledges or specific support by the OS.
This is joint work with rrs.
Reviewed by: rrs
Sponsored by: Netflix, Inc.
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D29469
As full support of RFC6675 is in place, deprecating
net.inet.tcp.rfc6675_pipe and enabling by default
net.inet.tcp.sack.revised.
Reviewed By: #transport, kbowling, rrs
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D28702
Add proper PRR vnet declarations for consistency.
Also add pointer to tcpopt struct to tcp_do_prr_ack, in preparation
for it to deal with non-SACK window reduction (after loss).
No functional change.
MFC after: 2 weeks
Reviewed By: tuexen, #transport
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D29440
a further CPU enhancements for compressed acks. These
are acks that are compressed into an mbuf. The transport
has to be aware of how to process these, and an upcoming
update to rack will do so. You need the rack changes
to actually test and validate these since if the transport
does not support mbuf compression, then the old code paths
stay in place. We do in this commit take out the concept
of logging if you don't have a lock (which was quite
dangerous and was only for some early debugging but has
been left in the code).
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D28374
When timestamp support has been negotiated, TCP segements received
without a timestamp should be discarded. However, there are broken
TCP implementations (for example, stacks used by Omniswitch 63xx and
64xx models), which send TCP segments without timestamps although
they negotiated timestamp support.
This patch adds a sysctl variable which tolerates such TCP segments
and allows to interoperate with broken stacks.
Reviewed by: jtl@, rscheff@
Differential Revision: https://reviews.freebsd.org/D28142
Sponsored by: Netflix, Inc.
PR: 252449
MFC after: 1 week
PRR improves loss recovery and avoids RTOs in a wide range
of scenarios (ACK thinning) over regular SACK loss recovery.
PRR is disabled by default, enable by net.inet.tcp.do_prr = 1.
Performance may be impeded by token bucket rate policers at
the bottleneck, where net.inet.tcp.do_prr_conservate = 1
should be enabled in addition.
Submitted by: Aris Angelogiannopoulos
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D18892
Under specific conditions, a window update can be sent with
outdated SACK information. Some clients react to this by
subsequently delaying loss recovery, making TCP perform very
poorly.
Reported by: chengc_netapp.com
Reviewed by: rrs, jtl
MFC after: 2 weeks
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D24237
Extend netstat to display TCP stack and detailed congestion state
Adding the "-c" option used to show detailed per-connection
congestion control state for TCP sessions.
This is one summary patch, which adds the relevant variables into
xtcpcb. As previous "spare" space is used, these changes are ABI
compatible.
Reviewed by: tuexen
MFC after: 2 weeks
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D26518
Adding proper accounting of sacked_bytes and (per-ACK)
delivered data to the SACK scoreboard. This will
allow more aspects of RFC6675 to be implemented as well
as Proportional Rate Reduction (RFC6937).
Prior to this change, the pipe calculation controlled with
net.inet.tcp.rfc6675_pipe was also susceptible to incorrect
results when more than 3 (or 4) holes in the sequence space
were present, which can no longer all fit into a single
ACK's SACK option.
Reviewed by: kbowling, rgrimes (mentor)
Approved by: rgrimes (mentor, blanket)
MFC after: 3 weeks
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D18624
The use of t_rcvtime as proxy for the last transmission
fails for transactional IO, where the client requests
data before the server can respond with a bulk transfer.
Set aside a dedicated variable to actually track the last
locally sent segment going forward.
Reported by: rrs
Reviewed by: rrs, tuexen (mentor)
Approved by: tuexen (mentor), rgrimes (mentor)
MFC after: 2 weeks
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D25016
is to know the time to first byte in and time to first byte out. Currently we
have no way to know these all we have is t_starttime. That (t_starttime) tells us
what time the 3 way handshake completed. We don't know when the first
request came in or how quickly we responded. Nor from a client perspective
do we know how long from when we sent out the first byte before the
server responded.
This small change adds the ability to track the TTFB's. This will show up in
BB logging which then can be pulled for later analysis. Note that currently
the tracking is via the ticks variable of all three variables. This provides
a very rough estimate (hz=1000 its 1ms). A follow-on set of work will be
to change all three of these values into something with a much finer resolution
(either microseconds or nanoseconds), though we may want to make the resolution
configurable so that on lower powered machines we could still use the much
cheaper ticks variable.
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D24902
if it can support the PRUS option (OOB). And then have
the new function call that to validate and give the
correct error response if needed to the user (rack
and bbr do not support obsoleted OOB data).
Sponsoered by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D24574
latest rack and bbr in from the NF repo. When those come
in the OOB data handling will be fixed where Skyzaller crashes.
Differential Revision: https://reviews.freebsd.org/D24575
MSS in two steps and try each candidate two times. However, if two
candidates are the same (which is the case in TCP/IPv6), this candidate
was tested four times. This patch ensures that each candidate actually
reduced the MSS and is only tested 2 times. This reduces the time window
of missclassifying a temporary outage as an MTU issue.
Reviewed by: jtl
MFC after: 1 week
Sponsored by: Netflix, Inc.
Differential Revision: https://reviews.freebsd.org/D24308
these are kernel modules. Also add a KMOD_TCPSTAT_ADD and use that
instead of TCPSTAT_ADD.
Reviewed by: jtl@, rrs@
MFC after: 1 week
Sponsored by: Netflix, Inc.
Differential Revision: https://reviews.freebsd.org/D23904
