openbsd
5 years agobump smtpd version
eric [Thu, 28 Feb 2019 13:23:32 +0000 (13:23 +0000)]
bump smtpd version

on behalf of gilles@

5 years agolist mpls as a depend for mpe and mpw, dont require bridge to build mpw
dlg [Thu, 28 Feb 2019 06:06:28 +0000 (06:06 +0000)]
list mpls as a depend for mpe and mpw, dont require bridge to build mpw

mpe doesnt need ether as a depend while im here.

5 years agoadd mpip(4)
dlg [Thu, 28 Feb 2019 03:36:49 +0000 (03:36 +0000)]
add mpip(4)

ok claudio@

5 years agoAdd mpip(4), an IP tunnel interface for "IP Layer 2" over MPLS pseudowires
dlg [Thu, 28 Feb 2019 03:27:57 +0000 (03:27 +0000)]
Add mpip(4), an IP tunnel interface for "IP Layer 2" over MPLS pseudowires

This is basically mpw(4), but it carries IP directly instead of
Ethernet. On the wire it can look the same as what IP over MPLS
looks like, but because it is a pseudowire you can configure a
control word or the FAT label to improve load balancing.  It can
be used to quickly set up an IP tunnel over an MPLS fabric
without the need to configure bgpd and mpe(4) interfaces.

Because It implements the same pwe3 ioctls that mpw(4) uses ifconfig
already supports configuration of mpip(4) interfaces. ldpd will
grow support for this in the near future.

This is not hooked up to the build yet

discussed with claduio@ at ak219
ok claudio@

5 years agodup stdout/in for proxycommand=-, otherwise stdout might be
markus [Wed, 27 Feb 2019 19:37:01 +0000 (19:37 +0000)]
dup stdout/in for proxycommand=-, otherwise stdout might be
redirected to /dev/null; ok djm@

5 years agoMake iwm(4) use CTS-to-self for HT protection if the AP requests this,
stsp [Wed, 27 Feb 2019 07:47:57 +0000 (07:47 +0000)]
Make iwm(4) use CTS-to-self for HT protection if the AP requests this,
rather than always using RTS for HT protection.

5 years agoupdate RFC references, from tobias_heider at genua.de, ok claudio@
sthen [Wed, 27 Feb 2019 06:33:56 +0000 (06:33 +0000)]
update RFC references, from tobias_heider at genua.de, ok claudio@

5 years agoBack in 2016 we enabled RTS threshold by default because it apparently
stsp [Wed, 27 Feb 2019 06:00:29 +0000 (06:00 +0000)]
Back in 2016 we enabled RTS threshold by default because it apparently
helped us on a loaded hackathon wifi network. In 11n mode we are now
using a heuristic for this. Disable RTS threshold for all other modes
again because it can do more harm than good in many situations.

This change may increase Tx throughput in 11a/g modes, though results depend
on the driver. Some drivers, e.g. urtwn(4), keep using RTS regardless for
unknown reasons and may need to be fixed separately. And we will keep using
RTS if the AP enables 11g protection while 11b clients are around.

5 years agotry to make this check work better and gather more output
benno [Wed, 27 Feb 2019 05:41:32 +0000 (05:41 +0000)]
try to make this check work better and gather more output

5 years agoMake xhci_ring_produce() check the previous TRB to find out if it
patrick [Wed, 27 Feb 2019 05:22:37 +0000 (05:22 +0000)]
Make xhci_ring_produce() check the previous TRB to find out if it
needs the Chain Bit set or not instead of using the last parameter,
which is used to mark the last TRB in a USB transfer, not in a TD.
To make that work we need to setup the recently acquired TRB before
calling xhci_xfer_get_trb() the next time.  Thus setting up the
initial TRB has to happen right away.  To kick the transfer off we
simply flip the toggle bit on the first TRB right at the end.

Fixes regression for jcs@, dhill@ and stsp@
Initially discussed with mpi@
ok stsp@

5 years agoremove unused includes
denis [Wed, 27 Feb 2019 04:52:19 +0000 (04:52 +0000)]
remove unused includes

OK yasuoka@

5 years agoAdjust to the last bgpd change. sa2addr gets an extra agrument and
claudio [Wed, 27 Feb 2019 04:34:21 +0000 (04:34 +0000)]
Adjust to the last bgpd change. sa2addr gets an extra agrument and
local and remote address in struct peer changed to bgpd_addrs.
OK benno@

5 years agoConvert the remote and local addresses in struct peer to be bgpd_addrs
claudio [Wed, 27 Feb 2019 04:31:56 +0000 (04:31 +0000)]
Convert the remote and local addresses in struct peer to be bgpd_addrs
instead of sockaddr_storage. This again helps protability and simplifies
some code. sa2addr now takes an optional pointer to return the port of
the sockaddr.
OK benno@

5 years agoFix export none. none became a keyword some time ago and so this broke.
claudio [Wed, 27 Feb 2019 04:16:02 +0000 (04:16 +0000)]
Fix export none. none became a keyword some time ago and so this broke.
Switch also default-route to a keyword and remove the old 6.3/6.4 announce
compat code.
Reported by florian@
OK benno@

5 years agoTest the export statement for neighbors
claudio [Wed, 27 Feb 2019 04:11:40 +0000 (04:11 +0000)]
Test the export statement for neighbors

5 years agoImplement support for dynamic RTS threshold in MiRA.
stsp [Wed, 27 Feb 2019 04:10:35 +0000 (04:10 +0000)]
Implement support for dynamic RTS threshold in MiRA.
Improves throughput and latency on 11n networks.
Tested by myself, jmc@, jcs@, phessler@, benno@

5 years agoFix iwn(4) scan misreporting the channel of wifi networks in some situations.
stsp [Wed, 27 Feb 2019 01:09:06 +0000 (01:09 +0000)]
Fix iwn(4) scan misreporting the channel of wifi networks in some situations.
Problem noticed by jmc@

5 years agoIn preparation for reusing libLLVM component library Makefiles to build
jsg [Tue, 26 Feb 2019 23:39:51 +0000 (23:39 +0000)]
In preparation for reusing libLLVM component library Makefiles to build
libLLVM from a single directory avoid reused filenames by symlinking
duplicated names with a prefix of the component library name so object
file names will be unique.

symlink approach suggested by deraadt@ ok patrick@

5 years agodefine LLVM_ENABLE_THREADS and LLVM_HAS_ATOMICS
jsg [Tue, 26 Feb 2019 23:16:38 +0000 (23:16 +0000)]
define LLVM_ENABLE_THREADS and LLVM_HAS_ATOMICS
matches the result of building with cmake

ok patrick@

5 years agotweak the mbuf loading in the tx path so it's easier to read.
dlg [Tue, 26 Feb 2019 23:12:58 +0000 (23:12 +0000)]
tweak the mbuf loading in the tx path so it's easier to read.

count mbuf load failures as output errors so i can see if that's
a problem (it's not, but at least i can see it isn't now).

5 years agosync
deraadt [Tue, 26 Feb 2019 22:24:55 +0000 (22:24 +0000)]
sync

5 years agocrank to 6.5-beta
deraadt [Tue, 26 Feb 2019 22:24:41 +0000 (22:24 +0000)]
crank to 6.5-beta

5 years agouse real format strings, resolve compiler warnings; ok millert@
cheloha [Tue, 26 Feb 2019 21:42:45 +0000 (21:42 +0000)]
use real format strings, resolve compiler warnings; ok millert@

5 years agoadd domain-search to the list of options capable of adding to resolv.conf;
jmc [Tue, 26 Feb 2019 20:32:23 +0000 (20:32 +0000)]
add domain-search to the list of options capable of adding to resolv.conf;
confirmed by and ok krw

5 years agoMove release check up front.
ajacoutot [Tue, 26 Feb 2019 18:52:35 +0000 (18:52 +0000)]
Move release check up front.

5 years agoFix sending IKEV2_CFG_INTERNAL_IP6_DNS, IKEV2_CFG_INTERNAL_IP6_NBNS,
patrick [Tue, 26 Feb 2019 18:05:22 +0000 (18:05 +0000)]
Fix sending IKEV2_CFG_INTERNAL_IP6_DNS, IKEV2_CFG_INTERNAL_IP6_NBNS,
IKEV2_CFG_INTERNAL_IP6_DHCP and IKEV2_CFG_INTERNAL_IP6_SERVER by using
the correct member in the iked_addr struct for the address.

From Aram Havarnean

5 years agoCorrect return values for non-fatal alerts.
jsing [Tue, 26 Feb 2019 17:38:39 +0000 (17:38 +0000)]
Correct return values for non-fatal alerts.

In the close notify case we need to signal EOF and in the user cancelled
case we need to return WANT_POLLIN. Returning success results in
tls13_record_layer_read_record() thinking that we have record data when
we do not, which then results in the content type check later failing.

ok tb@

5 years agoCorrectly propagate EOF from BIO in tls13_legacy_wire_read().
jsing [Tue, 26 Feb 2019 17:36:30 +0000 (17:36 +0000)]
Correctly propagate EOF from BIO in tls13_legacy_wire_read().

ok tb@

5 years agoCorrectly handle tls_read() and tls_write() failures.
jsing [Tue, 26 Feb 2019 17:32:47 +0000 (17:32 +0000)]
Correctly handle tls_read() and tls_write() failures.

Otherwise a TLS error (for example the remote end sent a fatal alert) is
silently ignored.

ok bluhm@ tb@

5 years agosync
visa [Tue, 26 Feb 2019 14:30:18 +0000 (14:30 +0000)]
sync

5 years agoAdd manual pages for SMR.
visa [Tue, 26 Feb 2019 14:28:20 +0000 (14:28 +0000)]
Add manual pages for SMR.

OK mpi@ sashan@

5 years agoIntroduce safe memory reclamation, a mechanism for reclaiming shared
visa [Tue, 26 Feb 2019 14:24:21 +0000 (14:24 +0000)]
Introduce safe memory reclamation, a mechanism for reclaiming shared
objects that readers can access without locking. This provides a basis
for read-copy-update operations.

Readers access SMR-protected shared objects inside SMR read-side
critical section where sleeping is not allowed. To reclaim
an SMR-protected object, the writer has to ensure mutual exclusion of
other writers, remove the object's shared reference and wait until
read-side references cannot exist any longer. As an alternative to
waiting, the writer can schedule a callback that gets invoked when
reclamation is safe.

The mechanism relies on CPU quiescent states to determine when an
SMR-protected object is ready for reclamation.

The <sys/smr.h> header additionally provides an implementation of
singly- and doubly-linked lists that can be used together with SMR.
These lists allow lockless read access with a concurrent writer.

Discussed with many
OK mpi@ sashan@

5 years agoikectl's built-in CA command for simple configurations has a fixed certificate
sthen [Tue, 26 Feb 2019 14:21:30 +0000 (14:21 +0000)]
ikectl's built-in CA command for simple configurations has a fixed certificate
validity for the ca certificate. Raise this from 365 days to 4500 as expiry means
installing new CA certificates on all client machines which can cause significant
pain. This doesn't change the default validity for server certificates which
remains at 1 year (controlled by ikeca.cnf) - refreshing key and certificate
on these can be done easily without visiting all machines.   ok deraadt@

5 years agoAdd missing break; in case statement and while there check that the
claudio [Tue, 26 Feb 2019 12:46:08 +0000 (12:46 +0000)]
Add missing break; in case statement and while there check that the
filter_community type is valid. Found while looking into a gcc warning
about uninitialized data.

5 years agoSome betoh64 sneaked in again, change to be64toh to be posix compliant.
claudio [Tue, 26 Feb 2019 12:33:40 +0000 (12:33 +0000)]
Some betoh64 sneaked in again, change to be64toh to be posix compliant.

5 years agoTo simplify the upcoming UTF-8 cleanup in less(1), delete support
schwarze [Tue, 26 Feb 2019 11:01:54 +0000 (11:01 +0000)]
To simplify the upcoming UTF-8 cleanup in less(1), delete support
for ANSI escape sequences introduced by an 8-bit CSI (e.g. "\23343m")
because these are neither compatible with UTF-8 nor strictly
compatible with pure ASCII and for those introduced by an UTF-8 CSI
(e.g. "\302\23343m") because not even xterm(1) supports them at
all, not even with a non-default configuration, because both forms
are very rarely used, if at all, and because the current code trying
to support them doesn't even appear to work according to my tests.
Full support for the ESC-[ CSI (e.g. "\033[43m") remains.

Tweaks and OK millert@, OK nicm@,
and sthen@ agrees with the general direction.

5 years agoExtend ext-community test to include most of the possible wildchar support
claudio [Tue, 26 Feb 2019 10:51:09 +0000 (10:51 +0000)]
Extend ext-community test to include most of the possible wildchar support

5 years agoAdd support for '*', local-as and neighbor-as for ext-community matching
claudio [Tue, 26 Feb 2019 10:49:15 +0000 (10:49 +0000)]
Add support for '*', local-as and neighbor-as for ext-community matching
and setting. This allows rules like:
    ext-community * * # delete any ext-community
    ext-community ovs * # delete any ext-community of specified type
    ext-community rt 1.2.3.4:*
and
    ext-community rt 65001:local-as
    ext-community rt local-as:11111

Note: Sometimes the type of the ext-community is underspecified when using
wildchars or expands.  So 'ext-community rt *' or 'ext-community soo *' will
match for any of the 3 possible types (2-byte AS, 4-byte AS and IP address).
If local-as/neighbor-as is used as an expand of as-number like
    ext-community rt local-as:11111
then bgpd will default to the 4-byte AS type to encode the community.

OK benno@

5 years agoRe-enable interrupts on resume with RF switch disabled
kn [Tue, 26 Feb 2019 08:51:15 +0000 (08:51 +0000)]
Re-enable interrupts on resume with RF switch disabled

When disabling the switch, suspending and eventually resuming again with
wifi still off, re-enabling the switch has no effect due to interrupts
being disabled completely.

To ensure seemless operation, explicitly enable interrupts during hardware
initialization iff the switch is disabled.

While here, initialize the interrupt mask up front to avoid clearing the
previously set mask in any case.

OK mlarkin, Feedback and OK ststp

5 years agosync
tb [Tue, 26 Feb 2019 07:15:36 +0000 (07:15 +0000)]
sync

5 years agog/c unused robots_roll file.
tb [Tue, 26 Feb 2019 07:14:35 +0000 (07:14 +0000)]
g/c unused robots_roll file.

From Randy Hartman

5 years agoi forgot to commit a regen after fixing some bcm entries
dlg [Tue, 26 Feb 2019 06:26:17 +0000 (06:26 +0000)]
i forgot to commit a regen after fixing some bcm entries

5 years agodon't let rxr completely fill the rxr ring.
dlg [Tue, 26 Feb 2019 04:04:30 +0000 (04:04 +0000)]
don't let rxr completely fill the rxr ring.

at the moment there are 256 slots on the right. if rxr raises the
high watermark to 256, we overwrite ring entries.

5 years agoget rid of the mpw goop
dlg [Tue, 26 Feb 2019 03:57:55 +0000 (03:57 +0000)]
get rid of the mpw goop

this reduces the output that mpw interfaces generate from 3 lines
to the one generic mpls and pwe3 line.

mpw can be configured with the individual ioctls, and gains the
ability to do flow aware transport.

5 years agodon't check the pseudowire type in tx and rx paths.
dlg [Tue, 26 Feb 2019 03:44:50 +0000 (03:44 +0000)]
don't check the pseudowire type in tx and rx paths.

whether the mpw interface is advertising "ethernet" or "ethernet-
tagged" is something the ends of the wire agree on (ie, ldpd is
configured a certain way), it is not something that affects ethernet
encap or decap.

the MPW ioctls can still configure it and read it, but it has no
bearing on how the driver operates on packets.

5 years agoimplement support for pwe3 ioctls
dlg [Tue, 26 Feb 2019 03:26:50 +0000 (03:26 +0000)]
implement support for pwe3 ioctls

this allows individual configuration of the use of control words,
flow aware tranport labels, and the local and remote label plus
neighbor individually.

discussed with claudio@ at a2k19
ok mpi@

5 years agouse NET_LOCK to coordinate destroying a cloned interface.
dlg [Tue, 26 Feb 2019 03:23:04 +0000 (03:23 +0000)]
use NET_LOCK to coordinate destroying a cloned interface.

5 years agoadd support for the new pwe3 ioctls.
dlg [Tue, 26 Feb 2019 03:22:36 +0000 (03:22 +0000)]
add support for the new pwe3 ioctls.

the existing mpw ioctl is still available for ldpd to use for a
(short) while.

discussed with claudio@ at a2k19
ok mpi@

5 years agocheck for root on mpls and pwe3 ioctls
dlg [Tue, 26 Feb 2019 03:20:08 +0000 (03:20 +0000)]
check for root on mpls and pwe3 ioctls

part of a larger diff ok mpi@

5 years agoadd some generic ioctls for configuring mpls pseudowire parameters.
dlg [Tue, 26 Feb 2019 03:19:11 +0000 (03:19 +0000)]
add some generic ioctls for configuring mpls pseudowire parameters.

these will replace the monstor mpw ioctl, and will be used by an
upcoming mpls ip tunnel interface driver.

discussed with claudio@ at a2k19
ok mpi@

5 years agoput the rx buffer at the end of the cluster.
dlg [Tue, 26 Feb 2019 03:17:18 +0000 (03:17 +0000)]
put the rx buffer at the end of the cluster.

makes m_prepend later less likely to allocate a new mbuf.

5 years agoavoid a deadlock in ixl_down when calling ifq_barrier.
dlg [Tue, 26 Feb 2019 03:12:34 +0000 (03:12 +0000)]
avoid a deadlock in ixl_down when calling ifq_barrier.

this is particularly noticable on sparc64 when you reboot.

ok jmatthew@

5 years agoput the rx buffer at the end of the mbuf
dlg [Tue, 26 Feb 2019 03:09:50 +0000 (03:09 +0000)]
put the rx buffer at the end of the mbuf

this should give a bit of space if the packet needs an m_pullup.
care is taken to ensure the packet is still aligned for the ip
stack.

ok and tweaks claudio@

5 years agosync
sthen [Mon, 25 Feb 2019 22:22:39 +0000 (22:22 +0000)]
sync

5 years agocreate a RETURN VALUES section and put things that return in there.
dlg [Mon, 25 Feb 2019 22:03:56 +0000 (22:03 +0000)]
create a RETURN VALUES section and put things that return in there.

hopefully it is a bit more obvious that what rw_enter returns can
be passed up to userland.

reads ok jmc@
ok mpi@

5 years agoMacau is MFM
deraadt [Mon, 25 Feb 2019 21:21:45 +0000 (21:21 +0000)]
Macau is MFM

5 years agoFix gcc3 compilation error by using a forward declaration instead of a
tb [Mon, 25 Feb 2019 19:44:04 +0000 (19:44 +0000)]
Fix gcc3 compilation error by using a forward declaration instead of a
repeated typedef. Found the hard way by aoyama who also tested the fix.

ok jsing

5 years agorevert "Move struct ssl_handshake_tls13_st definition", i.e., move
tb [Mon, 25 Feb 2019 19:40:05 +0000 (19:40 +0000)]
revert "Move struct ssl_handshake_tls13_st definition", i.e., move
SSL_HANDSHAKE_TLS13 back to ssl_locl.h.

discussed with jsing and inoguchi

5 years agoCorrectly handle oversize writes.
jsing [Mon, 25 Feb 2019 16:52:34 +0000 (16:52 +0000)]
Correctly handle oversize writes.

If the record layer is asked to write more than fits in a plaintext record,
cap the amount at that limit. This means that we will effectively write out
a single record and return a short-write.

This behaviour matches SSL_write() with SSL_MODE_ENABLE_PARTIAL_WRITE
enabled and the non-SSL_MODE_ENABLE_PARTIAL_WRITE case will be handled
at a higher layer.

ok inoguchi@ tb@

5 years agoAdd a handshake action sent handler and use it for client finished.
jsing [Mon, 25 Feb 2019 16:46:17 +0000 (16:46 +0000)]
Add a handshake action sent handler and use it for client finished.

The write traffic key needs to be changed to the client application traffic
key after the client finished message has been sent. The send handler
generates the client finished message, however we cannot switch keys at
this stage since the client finished message has not yet been protected
by the record layer.

ok tb@

5 years agoCorrect a few error returns.
jsing [Mon, 25 Feb 2019 16:39:14 +0000 (16:39 +0000)]
Correct a few error returns.

5 years agoadd sources matching unresolved symbols seen using Mesa with libLLVM
jsg [Mon, 25 Feb 2019 14:40:16 +0000 (14:40 +0000)]
add sources matching unresolved symbols seen using Mesa with libLLVM
ok patrick@

5 years agoMove struct ssl_handshake_tls13_st definition
inoguchi [Mon, 25 Feb 2019 14:36:25 +0000 (14:36 +0000)]
Move struct ssl_handshake_tls13_st definition

This solves build error on luna88k with gcc3.

ok aoyama@ jca@ jsing@ tb@

5 years agoUse struct bgpd_addr instead of a union of sockaddrs to store addresses
claudio [Mon, 25 Feb 2019 11:51:58 +0000 (11:51 +0000)]
Use struct bgpd_addr instead of a union of sockaddrs to store addresses
in the mrt dump structs. This is more compatible with bgpctl since it uses
struct bgpd_addr in most cases anyway.
OK kn@

5 years agoTypo in comment
jca [Mon, 25 Feb 2019 11:29:30 +0000 (11:29 +0000)]
Typo in comment

spotted by miod

5 years agoFix memory barrier in __mtx_leave(). membar_exit_before_atomic() cannot
visa [Mon, 25 Feb 2019 04:50:25 +0000 (04:50 +0000)]
Fix memory barrier in __mtx_leave(). membar_exit_before_atomic() cannot
be used in the routine because there is no subsequent atomic operation.
membar_exit() has to be used instead.

The mistake has not caused problems because on most platforms
membar_exit_before_atomic() is membar_exit(). Only amd64 and i386 have
a dedicated membar_exit_before_atomic(), and their exit barriers are
no-ops.

OK dlg@

5 years agoget rid of an unused softc member
dlg [Sun, 24 Feb 2019 22:49:31 +0000 (22:49 +0000)]
get rid of an unused softc member

5 years agoAdd Dynastream ANTUSB-2 to the list of supported uscom(4) devices.
patrick [Sun, 24 Feb 2019 17:37:42 +0000 (17:37 +0000)]
Add Dynastream ANTUSB-2 to the list of supported uscom(4) devices.

5 years agoDynastream ANTUSB-2 works well on uscom(4), from Jan Klemkow.
patrick [Sun, 24 Feb 2019 17:36:28 +0000 (17:36 +0000)]
Dynastream ANTUSB-2 works well on uscom(4), from Jan Klemkow.

5 years agoregen
patrick [Sun, 24 Feb 2019 17:35:44 +0000 (17:35 +0000)]
regen

5 years agoDynastream ANTUSB-2, from Jan Klemkow.
patrick [Sun, 24 Feb 2019 17:35:29 +0000 (17:35 +0000)]
Dynastream ANTUSB-2, from Jan Klemkow.

5 years agoRemove -S from install commands
kn [Sun, 24 Feb 2019 12:57:13 +0000 (12:57 +0000)]
Remove -S from install commands

As of usr.bin/xinstall/install.c revision 1.68, -S is a no-op and
install(1) will always create files safely, thus clean the option usage
from the tree.

Diff from Lauri Tirkkonen <lotheac at iki dot fi>, thanks.

5 years agoshort option comes first, followed by the corresponding long option
anton [Sun, 24 Feb 2019 09:57:43 +0000 (09:57 +0000)]
short option comes first, followed by the corresponding long option

ok jmc@

5 years agoRework iwm(4) Tx rate selection.
stsp [Sun, 24 Feb 2019 09:37:18 +0000 (09:37 +0000)]
Rework iwm(4) Tx rate selection.

Ask firmware to retry at a constant Tx rate instead of successively lower
rates. This provides better feedback to Tx rate scaling algorithms in our
kernel and thus increases Tx throughput, especially on clean channels.

Make the driver fall back to CCK rates on 2 GHz channels if 11n MCS
don't work. Previously, we were relying on firmware to do this for us.

Tested on 7260, 7265, 8260, 8265 by me, jcs@, tb@, solene@, Matthias Schmidt.

5 years agoRemove 11n support from AMRR. All 11n drivers switched to MiRA long ago.
stsp [Sun, 24 Feb 2019 09:36:28 +0000 (09:36 +0000)]
Remove 11n support from AMRR. All 11n drivers switched to MiRA long ago.

5 years agoalways include the lock id address in debug output
anton [Sun, 24 Feb 2019 09:24:52 +0000 (09:24 +0000)]
always include the lock id address in debug output

5 years agoinitialize struct lockf fields in a consistent order; ok millert@ mpi@ visa@
anton [Sun, 24 Feb 2019 09:11:09 +0000 (09:11 +0000)]
initialize struct lockf fields in a consistent order; ok millert@ mpi@ visa@

5 years agoTo measure the display width of a wide character in pwidth(), use
schwarze [Sun, 24 Feb 2019 04:54:36 +0000 (04:54 +0000)]
To measure the display width of a wide character in pwidth(), use
the standard function wcwidth(3) instead of several hand-rolled
functions accessing outdated local character tables, making this
part of the code conform to our in-tree Unicode 10.

Of course, with the current hand-rolled (and buggy) UTF-8 parser
contained in less(1), this only works if wchar_t stores UCS-4 values
and is more than 31 bits wide, but both will always be true on
OpenBSD, and ultmately, we shall switch to mbtowc(3) for parsing
anyway, lifting these restrictuons.

The existence of the outdated character tables was originally
called out by Evan Silberman on bugs@.

OK stsp@

5 years agoExplain the ASCII rendering of single quotes because that repeatedly
schwarze [Sat, 23 Feb 2019 18:52:45 +0000 (18:52 +0000)]
Explain the ASCII rendering of single quotes because that repeatedly
caused confusion in the past.  People plainly do not expect that
there are limits to the compatibility between Unicode and ASCII,
but there are.

The information belongs here and not into mandoc_char(7) because
it explains how the specific output device (-T ascii) works and
because it has nothing to do with the question of how characters
are represented on the input side.

5 years agoRemove temporary gotos.
jsing [Sat, 23 Feb 2019 15:02:34 +0000 (15:02 +0000)]
Remove temporary gotos.

In the case of a dummy CCS or post-handshake handshake message, return
TLS13_WANT_POLLIN rather than using a goto internally. This allows the
caller to retry at an appropriate time and reduces the complexity within
the record layer.

ok beck@ tb@

5 years agoSet BIO retry on TLS13_IO_WANT_POLLIN/TLS13_IO_WANT_POLLOUT.
jsing [Sat, 23 Feb 2019 15:00:44 +0000 (15:00 +0000)]
Set BIO retry on TLS13_IO_WANT_POLLIN/TLS13_IO_WANT_POLLOUT.

In most cases a TLS13_IO_WANT_POLLIN or TLS13_IO_WANT_POLLOUT will have
bubbled up from the wire callbacks, in which case the BIO retry flag will
already be set. However, if we return TLS13_IO_WANT_POLLIN or
TLS13_IO_WANT_POLLOUT from a higher layer the BIO retry flag will not be
set and that will cause SSL_get_error() to return SSL_ERROR_SYSCALL rather
than the intended SSL_ERROR_WANT_READ/SSL_ERROR_WANT_WRITE.

ok beck@ tb@

5 years agoCorrect comment on delete_addresses() return value.
krw [Sat, 23 Feb 2019 13:37:34 +0000 (13:37 +0000)]
Correct comment on delete_addresses() return value.

5 years agoRestore pre-a2k19 behaviour of clearing out all existing addresses
krw [Sat, 23 Feb 2019 13:24:19 +0000 (13:24 +0000)]
Restore pre-a2k19 behaviour of clearing out all existing addresses
when setting new address.

That change needs specific consideration rather than inadvertant
inclusion, especially after claiming no functional change was
being made.

Change and related problems noted and reversion tested by Kurt
Mosiejczuk.

5 years agoopenssh-7.9 accidentally reused the server's algorithm lists in the
djm [Sat, 23 Feb 2019 08:20:43 +0000 (08:20 +0000)]
openssh-7.9 accidentally reused the server's algorithm lists in the
client for KEX, ciphers and MACs. The ciphers and MACs were
identical between the client and server, but the error accidentially
disabled the diffie-hellman-group-exchange-sha1 KEX method.

This fixes the client code to use the correct method list, but
because nobody complained, it also disables the
diffie-hellman-group-exchange-sha1 KEX method.

Reported by nuxi AT vault24.org via bz#2697; ok dtucker

5 years agodocument pci_mapreg_map and some related functions.
dlg [Sat, 23 Feb 2019 04:54:25 +0000 (04:54 +0000)]
document pci_mapreg_map and some related functions.

i couldn't figure out which argument becomes the flags for the
wrapped call to bus_space_map(9), so i wrote this page so i wouldnt
have to read the code for this one again.

ok jmc@ schwarze@ mpi@ deraadt@

5 years agorecognise more arm cpus
jsg [Sat, 23 Feb 2019 03:37:50 +0000 (03:37 +0000)]
recognise more arm cpus
ok patrick@

5 years agoUpdate to Unicode 10
afresh1 [Fri, 22 Feb 2019 16:35:16 +0000 (16:35 +0000)]
Update to Unicode 10

Committing the diff in my tree that matched the version sent by
Lauri Tirkkonen <lotheac AT iki.fi>

OK schwarze@

5 years agoImprove the X86FixupGadgets pass:
mortimer [Fri, 22 Feb 2019 15:28:43 +0000 (15:28 +0000)]
Improve the X86FixupGadgets pass:
- Target all four kinds of return bytes (c2, c3, ca, cb)
- Fix up instructions using both ModR/M and SIB bytes
- Force alignment before instructions with return bytes in immediates
- Force alignment before instructions that have return bytes in their encoding
- Add a command line switch to toggle the functionality.

ok deraadt@

5 years agoFor ia32 use a volatile double to force 64 bit rounding. Otherwise
bluhm [Fri, 22 Feb 2019 14:22:51 +0000 (14:22 +0000)]
For ia32 use a volatile double to force 64 bit rounding.  Otherwise
the i387 would use its internal 80 bit stack.  This fixes getround()
on i386.

5 years agoremove __BEGIN_DECLS/__END_DECLS, this .h file wont be
benno [Fri, 22 Feb 2019 09:54:36 +0000 (09:54 +0000)]
remove __BEGIN_DECLS/__END_DECLS, this .h file wont be
read by a C++ compiler, and __* are no lonmger recommended.
https://www.gnu.org/software/libtool/manual/html_node/C-header-files.html
ok deraadt@

5 years agoAllow ssdfb(4) to be mmap(2)-able so that we can use its framebuffer
patrick [Fri, 22 Feb 2019 09:39:33 +0000 (09:39 +0000)]
Allow ssdfb(4) to be mmap(2)-able so that we can use its framebuffer
from the userland.  Also allow changing the brightness levels so that
it can even be turned off to save power and to prevent burn in.

5 years agoat some point the uvm_km_thread learned to free memory, but the comment
tedu [Fri, 22 Feb 2019 07:53:56 +0000 (07:53 +0000)]
at some point the uvm_km_thread learned to free memory, but the comment
was never updated.
from Amit Kulkarni

5 years agoerrror -> error in log messages; from alessandro gallo
jmc [Fri, 22 Feb 2019 07:04:20 +0000 (07:04 +0000)]
errror -> error in log messages; from alessandro gallo

5 years agoperform removal of agent-forwarding directory in forward setup error
djm [Fri, 22 Feb 2019 03:37:11 +0000 (03:37 +0000)]
perform removal of agent-forwarding directory in forward setup error
path with user's privileged. This is a no-op as this code always runs
with user privilege now that we no longer support running sshd with
privilege separation disabled, but as long as the privsep skeleton is
there we should follow the rules.

bz#2969 with patch from Erik Sjölund

5 years agoTest that all supported TLS ciphers actually work. Establish
bluhm [Thu, 21 Feb 2019 23:06:33 +0000 (23:06 +0000)]
Test that all supported TLS ciphers actually work.  Establish
connections between client and server implemented with LibreSSL or
OpenSSL with a fixed cipher on each side.  Check the used cipher
in the session print out.

5 years agoTransfers that span multiple TRBs which wrap around the ring and
patrick [Thu, 21 Feb 2019 22:44:44 +0000 (22:44 +0000)]
Transfers that span multiple TRBs which wrap around the ring and
thus have the Link TRB inbetween must have the Chain Bit set in the
Link TRB.  Otherwise xHCI controllers might think that the transfer
ends at that point.

Fixes an issue that was most prominently seen as Invalid CSW error
when using umass0 on octeon and i.MX8M.

Tested by visa@
ok mpi@

5 years agokristaps f18392f58d1f0bc72917e3ac3a46be74c194688d
benno [Thu, 21 Feb 2019 22:15:09 +0000 (22:15 +0000)]
kristaps f18392f58d1f0bc72917e3ac3a46be74c194688d
Document that the gid/uid list may be empty. Fixes
https://github.com/kristapsdz/openrsync/issues/10

5 years agokristaps cbe83cd64f40e634dbc22d3f2918c41977a6514d
benno [Thu, 21 Feb 2019 22:13:43 +0000 (22:13 +0000)]
kristaps cbe83cd64f40e634dbc22d3f2918c41977a6514d
If we don't get a uid/gid map, such as with an rsync:// address, we
might not be able to map.  So fall back on numeric ids.

5 years agokristaps 7721288a1d170f4d789bf7a7b822f14f91f7bcb9
benno [Thu, 21 Feb 2019 22:12:48 +0000 (22:12 +0000)]
kristaps 7721288a1d170f4d789bf7a7b822f14f91f7bcb9

In order for GPL rsync(1) to work with --delete when openrsync is a
remote sender, we need to set the FLIST_TOP_LEVEL bit for all
top-level directories. This is because the GPL rsync(1) uses this bit
to determine which directories to scan for files for deletion.

5 years agokristaps 156cf3b7d459efe133d603158fc33c243dc07ba0
benno [Thu, 21 Feb 2019 22:11:26 +0000 (22:11 +0000)]
kristaps 156cf3b7d459efe133d603158fc33c243dc07ba0
Use style(9) for bit