Currently when the peer collapses its rwnd, we mark packets to be retransmitted
and use the must_retran flags like we do when a PMTU collapses to retransmit the
collapsed packets. However this causes a problem with some middle boxes that
play with the rwnd to control flow. As soon as the rwnd increases we start resending
which may be not even a rtt.. and in fact the peer may have gotten the packets. Which
means we gratuitously retransmit packets we should not.
The fix here is to make sure that a rack time has passed before retransmitting the packets.
This makes sure that the rwnd collapse was real and the packets do need retransmission.
Reviewed by: tuexen
Sponsored by: Netflix Inc
Differential Revision: https://reviews.freebsd.org/D35166
A while back Hans optimized the LRO code. This is great but one
optimization he did degrades the timestamp precision so that
all flushed LRO entries end up with the same LRO timestamp
if there is not a hardware timestamp. The intent of the LRO timestamp
is to get as close to the time that the packet arrived as possible. Without
the LRO queuing this works out fine since a binuptime is taken and then
the rx_common code is called. But when you go through the queue path
you end up *not* updating the M_LRO_TSTMP fields.
Another issue in the LRO code is several places that cause packet reordering. In
general TCP can handle reordering but it can cause extra un-needed retransmission
as well as other oddities. We will fix all of the reordering problems.
Lets fix this so that we restore the precision to the timestamp.
Reviewed by: tuexen, gallatin
Sponsored by: Netflix Inc
Differential Revision: https://reviews.freebsd.org/D36043
Get rid of struct pfsync_pkt. It was used to store data on the stack to
pass to all the submessage handlers, but only the flags part of it was
ever used. Just pass the flags directly instead.
Reviewed by: kp
Obtained from: OpenBSD
Sponsored by: InnoGames GmbH
Differential Revision: https://reviews.freebsd.org/D36294
During a discussion with someone working on NFS-over-TLS
for a non-FreeBSD platform, we agreed that a single server
daemon for TLS handshakes could become a bottleneck when
an NFS server first boots, if many concurrent NFS-over-TLS
connections are attempted.
This patch modifies the kernel RPC code so that it can
handle multiple rpc.tlsservd daemons. A separate commit
currently under review as D35886 for the rpc.tlsservd
daemon.
Fix the off-by-one in the KASSERT in arm_gic_reserve_msi_range()
allowing the MSIs being allocated at the end of the IRQ range.
Reported by: bz
Submitted by: andrew
Reviewed by: andrew
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D36291
Classify VLAN devices as ethernet in rdma_copy_addr().
This fixes the following error message:
rdma_bind_addr: No such file or directory
Submitted by: bartosz.sobczak@intel.com (Bartosz Sobczak)
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D36120
Sponsored by: NVIDIA Networking
When specifying a custom band-pass filter for the sound(4) equalizer,
use the same format as used by the resulting C-code define to avoid
confusion.
PR: 195760
Sponsored by: NVIDIA Networking
NXP defines unique name for LS1046A UART - "NXP0018".
It is ns8250 compatible, adding a new uart compat data entry is enough
to make it work.
Tested on LS1046ARDB.
Approved by: mw(mentor)
Reviewed by: mw, kd
Obtained from: Semihalf
Differential Revision: https://reviews.freebsd.org/D36260
For now it is read-only, but eventually the cycle that goes over
all fragments should be refactored and this MIB should also become
read/write.
This MIB will allow SNMP daemons to implement MIB-II ipReasmTimeout MIB
straightfoward. Right now net-snmp compilation is broken by 1922eb3e9c.
The base system bsnmpd is not broken just because it ignored PR_SLOWTIMO,
and thus always returned incorrectly doubled value for ipReasmTimeout.
We cannot see a thread with the flag set in unsuspend, after we stopped
doing SINGLE_ALLPROC from user processes.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D36207
It does not serve any purpose after we stopped doing
thread_single(SINGLE_ALLPROC) from stoppable user processes.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D36207
There is a problem still left after the fixes to REAP_KILL_PROC. The
handling of the stopping signals by sig_suspend_threads() can occur
outside the stopping process context by tdsendsignal(), and it uses
mostly the same mechanism of aborting sleeps as suspension. In other
words, it badly interacts with thread_single(SINGLE_ALLPROC).
But unlike single threading from the process context, we cannot wait by
sleep for other single threading requests to pass, because we own
spinlock(s).
Fix this by moving both the thread_single(p2, SINGLE_ALLPROC), and the
signalling, to the threaded taskqueue which cannot be single-threaded
itself.
Reported and tested by: pho
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D36207
The function result is already used as bool.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D36207
We do not check single-threading conditions in trap, or when sleeping
uninterruptible.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential revision: https://reviews.freebsd.org/D36207
It is sbt now. Also, explain what flags are.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential revision: https://reviews.freebsd.org/D36207
The relock order is important not only for a signal delivery, but also
for the suspension requests.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential revision: https://reviews.freebsd.org/D36207
Use memset instead of bzero, memcpy instead of bcopy, and add some
optional {}s.
Reviewed by: kp
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D36267
It is only ever xlocked in drain_dev_clone_events and the only consumer of
that routine does not need it -- eventhandler code already makes sure the
relevant callback is no longer running.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D36268
RB_ROTATE_LEFT (and it symmetric twin) modify the rb-tree, adjusting
pointers so that what started as a proper tree ends up a proper
tree. When two consecutive rotations move the same node up the tree,
some of the pointers changed in the first rotation are immediately
changed again in the second - namely, the pointer from the rising node
to its new parent, and the pointer from that parent back to the rising
node. This change removes from RB_ROTATE macros the responsibility for
managing those two pointers, and leaves it to the code that calls for
rotations to fix up those pointers afterward. That drops a comparison
and a pair of assignments from every INSERT_COLOR or REMOVE_COLOR call
that ends in a double rotation.
A side-effect of this change is that the SWAP_CHILD macro must take as
a parameter a pointer to the node that is changing children, where it
is now computed from the old child. Since this macro is called in a
couple of places besides the RB_ROTATE macros, those calls are also
affected.
Reviewed by: alc
MFC after: 3 weeks
Differential Revision: https://reviews.freebsd.org/D36266
Reject attempts to map host physical address ranges that are not
subsets of a passthrough device's BAR into a guest.
Reviewed by: markj, emaste
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D36238
During attach we release the lock to call e6000sw_attach_miibus(), which
calls mii_attach(). The mii_attach() function calls miibus_readreg() /
miibus_writereg(). However, these are set to be
e6000sw_readphy_locked().
That is, the read/write phy functions do not acquire the lock, but
expect to be called while locked. Simply do not unlock and relock while
calling mii_attach().
Reviewed by: Hubert Mazur <hum@semihalf.com>
Sponsored by: Rubicon Communications, LLC ("Netgate")
Differential Revision: https://reviews.freebsd.org/D36117
It supports gpio type checking. Depending on gpio type some
register addresses are different.
Reviewed by: manu
Differential Revision: https://reviews.freebsd.org/D36262
Ethernet rule cleanup is postponed to an epoch callback. Ensure it's
been called before we remove the entire vnet, or we risk the rules still
getting hit after we've freed the uma zone, i.e. a use-after-free.
Sponsored by: Rubicon Communications, LLC ("Netgate")
Linux is using 7 bit addressing while FreeBSD uses 8 bit addresses
internally, but i2c(8) uses 7 bit address.
This confused me when originally doing the code and I thought that
0x50 was the 8bit EDID address while it's the 7bit address and since
I did all my testing using this I didn't noticed the problem.
Reported by: avg
PR: 265920 (somewhat)
Rather than defining the same values in two places and having to do
conflict resulution on the name in LKPI, change the defines to an
enum in net80211. In addition to de-duplication this also gives us
value checks in certain cases.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Reviewed by: adrian
Differential Revision: https://reviews.freebsd.org/D36250
Replace a malloc() by IEEE80211_MALLOC().
For malloc flags even in the local ieee80211_freebsd.c there was a mix
of both versions M_ and IEEE80211_M_.
Consistently use the IEEE80211_M_ malloc options everywhere.
If the field is changed for malloc, it'll also be changed for the
other accessor functions taking a "how" field to avoid any confusion.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Reviewed by: adrian
Differential Revision: https://reviews.freebsd.org/D36249
Make sure the given IE length fits into the total length left when
parsing through the information elements. In theory I would say
discard everything if there is an error but that proves hard with
the current code.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Reviewed by: adrian
Differential Revision: https://reviews.freebsd.org/D36245
The IFF_DRV_RUNNING flag is used to see if the interface needs
to be temporarily brought down during MTU change sequence.
The problem here is that this flag is cleared in mvneta_stop_locked,
resulting in the reinitialization logic never being executed after
MTU has been changed.
Fix that by saving the flag value before the interface is brought down.
Reported by: Jérôme Tomczyk <jerome.tomczyk@stormshield.eu>
Approved by: mw(mentor)
Obtained from: Semihalf
Sponsored by: Stormshield
MFC after: 2 weeks
Both are needed by drm-kmod
Obtained from: drm-kmod
Sponsored by: Beckhoff Automation GmbH & Co. KG
Differential Revision: //reviews.freebsd.org/D36212
Those are needed and also included in linux (via polution).
Reviewed by: bz
Sponsored by: Beckhoff Automation GmbH & Co. KG
Differential Revision: https://reviews.freebsd.org/D36111
All those are needed for drm-kmod.
Add them to base in another directory that will be append in the CFLAGS.
Reviewed by: bz
Sponsored by: Beckhoff Automation GmbH & Co. KG
Differential Revision: https://reviews.freebsd.org/D36110
Add needed includes so we can use it.
Reviewed by: bz
Fixes: c3f4f28c63 ("linuxkpi: Add some basic swap functions")
Sponsored by: Beckhoff Automation GmbH & Co. KG
Differential Revision: https://reviews.freebsd.org/D36109
