openbsd
23 months agostyle(9) fix. No functional change.
mvs [Tue, 15 Nov 2022 22:47:15 +0000 (22:47 +0000)]
style(9) fix. No functional change.

23 months agoNot assigning a link local address by default breaks various things.
claudio [Tue, 15 Nov 2022 18:42:46 +0000 (18:42 +0000)]
Not assigning a link local address by default breaks various things.
A lot of code depends on a eui64 address to be present. So revert this
bit of the last commit.
OK florian@

23 months agodocument BN_mod_sqrt(3)
schwarze [Tue, 15 Nov 2022 17:55:00 +0000 (17:55 +0000)]
document BN_mod_sqrt(3)

23 months agoTake `vmobjlock' around uao_grow() and uao_shrink() calls to fix tmpfs
mvs [Tue, 15 Nov 2022 17:16:44 +0000 (17:16 +0000)]
Take `vmobjlock' around uao_grow() and uao_shrink() calls to fix tmpfs
panics [1].

1. https://marc.info/?l=openbsd-bugs&m=165012301707403&w=2

from Leo Larnack <leo at pseven.xyz>

ok kn@ mpi@

23 months agosync
sthen [Tue, 15 Nov 2022 10:44:06 +0000 (10:44 +0000)]
sync

23 months agoUnlock SIOCGIFG{MEMB,ATTR,LIST}
kn [Mon, 14 Nov 2022 22:45:02 +0000 (22:45 +0000)]
Unlock SIOCGIFG{MEMB,ATTR,LIST}

The global interface group list is also protected by the net lock and all
access to it (all within if.c) take it accordingly.

Getting all
- members    of a group (SIOCGIFGMEMB),
- attributes of a group (SIOCGIFGATTR),
- groups                (SIOCGIFGLIST)
are each read-only operations on the global interface group `ifg_head'.

The global interface list `ifnetlist' or its per-interface group lists are
not used in these ioctls.

OK mvs

23 months agoUnlock SIOCGIFCONF
kn [Mon, 14 Nov 2022 22:07:30 +0000 (22:07 +0000)]
Unlock SIOCGIFCONF

As netintro(4) explains, this copies a bunch of data from the global
interface list as well as its per-interface address lists.

All of this is never written to by ifconf(), protected by the net lock
and documented as such in the struct comments already.

OK mvs

23 months agoDocument global interface group list locking
kn [Mon, 14 Nov 2022 22:06:25 +0000 (22:06 +0000)]
Document global interface group list locking

The per-interface group list is protected by the net lock and already
documented as such.

The global interface group list `ifg_head' is also protected by the net
lock and all access to it (all within if.c) take it accordingly.

Feedback OK mvs

23 months agomerge conflicts
sthen [Mon, 14 Nov 2022 21:09:32 +0000 (21:09 +0000)]
merge conflicts

23 months agoimport NSD 4.6.1, ok florian@, thanks aoyama@ for checking autoconf C99 bits
sthen [Mon, 14 Nov 2022 21:07:49 +0000 (21:07 +0000)]
import NSD 4.6.1, ok florian@, thanks aoyama@ for checking autoconf C99 bits

23 months agodocument BN_kronecker(3)
schwarze [Mon, 14 Nov 2022 19:51:35 +0000 (19:51 +0000)]
document BN_kronecker(3)

23 months agodocument BN_reciprocal(3)
schwarze [Mon, 14 Nov 2022 18:28:29 +0000 (18:28 +0000)]
document BN_reciprocal(3)

23 months agoHide public symbols in libcrypto/x509 .c files
beck [Mon, 14 Nov 2022 17:48:49 +0000 (17:48 +0000)]
Hide public symbols in libcrypto/x509 .c files

ok tb@

23 months agoFix tag type and buffer limit in DT_TEXTREL lookup.
visa [Mon, 14 Nov 2022 17:25:00 +0000 (17:25 +0000)]
Fix tag type and buffer limit in DT_TEXTREL lookup.

OK deraadt@ kettenis@

23 months agoWith the commit to in6_ifattach mpe(4) now also works with IPv6
claudio [Mon, 14 Nov 2022 17:23:43 +0000 (17:23 +0000)]
With the commit to in6_ifattach mpe(4) now also works with IPv6

23 months agoAdd missing clock trigger to loongson_isa_splx().
visa [Mon, 14 Nov 2022 17:15:41 +0000 (17:15 +0000)]
Add missing clock trigger to loongson_isa_splx().

Fixes unexpected delays that have occurred with mips64 clock(4).

23 months agoRelax the list of interfaces which support IPv6 a bit.
claudio [Mon, 14 Nov 2022 17:12:55 +0000 (17:12 +0000)]
Relax the list of interfaces which support IPv6 a bit.

No longer require IFF_MULTICAST for all interfaces. It is save to skip
this for interfaces that don't require a ND cache. Also do not assign a
link-local address in such cases. This affects point-to-point interfaces
and the NBMA / point-to-multipoint interfaces like mpe(4), mgre(4) and wg(4).
The NBMA interfaces need some alternative way to figure out the address
mapping. In the end this allows non-multicast interfaces to work with IPv6.
OK dlg@ kn@

23 months agofix swapped user/group introduced in sync-with-mount_tmpfs commit;
sthen [Mon, 14 Nov 2022 14:35:39 +0000 (14:35 +0000)]
fix swapped user/group introduced in sync-with-mount_tmpfs commit;
from Josiah Frentsos

23 months agomake the EFI partition quite large, like on arm64 x13s, because newer x86
deraadt [Mon, 14 Nov 2022 14:04:25 +0000 (14:04 +0000)]
make the EFI partition quite large, like on arm64 x13s, because newer x86
((Lenovo in particular) firmware updating methods do the same -- placing
a payload in there, so a small space won't do.
ok kettenis

23 months agoPreserve original order of nameservers
kn [Mon, 14 Nov 2022 13:57:46 +0000 (13:57 +0000)]
Preserve original order of nameservers

RFC 2132 "DHCP Options and BOOTP Vendor Extensions"
3.8. Domain Name Server Option says
Servers SHOULD be listed in order of preference.

tcpdump(8), route(8) monitor and dhcpleasectl(8) -l athn0 show servers from
the DHCP OFFER in their original order, as expected.

resolvd(8) however sorts proposals by priority and IP address before writing
them to resolv.conf(5).

But as the system resolver tries this file's `nameserver' options in the
order appearance, sorting by IP breaks DHCP's intended order and thus may
result in the wrong nameserver being queried.

Sorting by IP is done to later remove duplicates from the file.

Sort by priority alone and ensure uniqueness by iterating over the list of
of proposals and zeroeing duplicates instead to preserve any proposal's
original order.

Spotted on a public wifi OFFERing two local IPs plus 8.8.8.8 in this order
which ended up with 8.8.8.8 being the first entry in my /etc/resolv.conf.

In other words, `route nameserver lo0 2.2.2.2 1.1.1.1 1.1.1.1' now yields
nameserver 2.2.2.2 # resolvd: lo0
nameserver 1.1.1.1 # resolvd: lo0
rather than
nameserver 1.1.1.1 # resolvd: lo0
nameserver 2.2.2.2 # resolvd: lo0

Feedback OK deraadt

23 months agoRecognize PT_OPENBSD_MUTABLE with LLVM's readobj / objdump.
deraadt [Mon, 14 Nov 2022 13:42:19 +0000 (13:42 +0000)]
Recognize PT_OPENBSD_MUTABLE with LLVM's readobj / objdump.
from brad

23 months agoCheck for offline chunks first, then check for keydisks
kn [Mon, 14 Nov 2022 13:39:37 +0000 (13:39 +0000)]
Check for offline chunks first, then check for keydisks

Like keydisks, offline data chunks also report a size of zero, so skipping
zero sized chunks before doing the offline check means not printing the
"softraid chunk N not online - skipping..." warning.

Restore order to make this warning appear on degraded volumes again.

23 months agoSince the introduction of automatic immutable from the kernel, the munmap()
deraadt [Mon, 14 Nov 2022 13:21:50 +0000 (13:21 +0000)]
Since the introduction of automatic immutable from the kernel, the munmap()
of ld.so boot.text region is now (silently) failing because the region is
contained within the text LOAD, which is immutable.  So create a new btext
LOAD with flags PF_X|PF_R|PF_OPENBSD_MUTABLE, and place all boot.text objects
in there.  This LOAD must also be page-aligned so it doesn't skip unmapping
some of the object region, previously it was hilariously unaligned.
ok kettenis and guenther seemed to like it also
This one is for powerpc64 and a modified version of the diff deraadt@ mailed
out to make sure the LOADs are in increasing address order.
this is the alpha version

23 months agoSince the introduction of automatic immutable from the kernel, the munmap()
deraadt [Mon, 14 Nov 2022 13:20:23 +0000 (13:20 +0000)]
Since the introduction of automatic immutable from the kernel, the munmap()
of ld.so boot.text region is now (silently) failing because the region is
contained within the text LOAD, which is immutable.  So create a new btext
LOAD with flags PF_X|PF_R|PF_OPENBSD_MUTABLE, and place all boot.text objects
in there.  This LOAD must also be page-aligned so it doesn't skip unmapping
some of the object region, previously it was hilariously unaligned.
ok kettenis and guenther seemed to like it also
This one is for powerpc64 and a modified version of the diff deraadt@ mailed
out to make sure the LOADs are in increasing address order.
this is the i386 version

23 months agoSince the introduction of automatic immutable from the kernel, the munmap()
deraadt [Mon, 14 Nov 2022 13:13:19 +0000 (13:13 +0000)]
Since the introduction of automatic immutable from the kernel, the munmap()
of ld.so boot.text region is now (silently) failing because the region is
contained within the text LOAD, which is immutable.  So create a new btext
LOAD with flags PF_X|PF_R|PF_OPENBSD_MUTABLE, and place all boot.text objects
in there.  This LOAD must also be page-aligned so it doesn't skip unmapping
some of the object region, previously it was hilariously unaligned.
ok kettenis and guenther seemed to like it also
This one is for powerpc64 and a modified version of the diff deraadt@ mailed
out to make sure the LOADs are in increasing address order.
this is the version for sparc64, tested by pascal also

23 months agoSince the introduction of automatic immutable from the kernel, the munmap()
deraadt [Mon, 14 Nov 2022 13:12:59 +0000 (13:12 +0000)]
Since the introduction of automatic immutable from the kernel, the munmap()
of ld.so boot.text region is now (silently) failing because the region is
contained within the text LOAD, which is immutable.  So create a new btext
LOAD with flags PF_X|PF_R|PF_OPENBSD_MUTABLE, and place all boot.text objects
in there.  This LOAD must also be page-aligned so it doesn't skip unmapping
some of the object region, previously it was hilariously unaligned.
ok kettenis and guenther seemed to like it also
This one is for powerpc64 and a modified version of the diff deraadt@ mailed
out to make sure the LOADs are in increasing address order.
this is the version for powerpc, tested by pascal

23 months agoTurns out the power button on the Mac mini and the touchid button on the
kettenis [Mon, 14 Nov 2022 11:11:17 +0000 (11:11 +0000)]
Turns out the power button on the Mac mini and the touchid button on the
various Macbook models generate different events.  So rename the existing
constant for a "short press" of the touchid button and add support for the
power button on the Mac mini such that is can be resumed.

ok tobhe@, patrick@

23 months ago- show -R with -s in options list
jmc [Mon, 14 Nov 2022 09:56:09 +0000 (09:56 +0000)]
- show -R with -s in options list
- show -t with -T in options list
- sort the -T commands
- small text/formatting tweaks

ok sthen
ok kn on an earlier version

23 months agoDocument arm64 as supporting RAID 1C boot
kn [Mon, 14 Nov 2022 09:34:51 +0000 (09:34 +0000)]
Document arm64 as supporting RAID 1C boot

Now that installboot(8) -p is softraid(4) aware, fresh arm64 installations
actually boot out of the box without manual intervention, as expected.

23 months agoWhen the fdt does not provide a list of brightness states, pwmbl(4)
miod [Mon, 14 Nov 2022 07:22:44 +0000 (07:22 +0000)]
When the fdt does not provide a list of brightness states, pwmbl(4)
builds a 256 state ramp (i.e. state[i] = i with 0 <= i < 256).

Change the logic to not need to malloc an array of values in this case,
since they are trivially known.

test&ok patrick@

23 months agoinclude sys/device.h, on some architectures this gets pulled in, but others
deraadt [Mon, 14 Nov 2022 00:16:44 +0000 (00:16 +0000)]
include sys/device.h, on some architectures this gets pulled in, but others
(like sparc64) require it explicitly

23 months agoFix comment style
tb [Sun, 13 Nov 2022 23:38:42 +0000 (23:38 +0000)]
Fix comment style

23 months agoVarious improvements; joint work with beck@:
schwarze [Sun, 13 Nov 2022 22:11:44 +0000 (22:11 +0000)]
Various improvements; joint work with beck@:
1. Explain up front what "ASN1_TIME" is (suggested by beck@, wording by me).
2. For opaque structs, use the generic term "object",
like we already do it in many other LibreSSL manual pages.
3. Drop some redundant phrases.
4. Improve the EXAMPLES section (by beck@, with fixes by me).
6. Add a STANDARDS section.
...and some other minor polishing.
OK beck@

23 months agoMake sure csa->csa_bundled is NULL after freeing to prevent a
mbuhl [Sun, 13 Nov 2022 21:19:40 +0000 (21:19 +0000)]
Make sure csa->csa_bundled is NULL after freeing to prevent a
possible use after free.
ok tobhe@

23 months agoSync config.h, we have ed25519 support now.
florian [Sun, 13 Nov 2022 18:40:45 +0000 (18:40 +0000)]
Sync config.h, we have ed25519 support now.

23 months agoCheck certificate extensions in trusted certificates.
beck [Sun, 13 Nov 2022 18:37:32 +0000 (18:37 +0000)]
Check certificate extensions in trusted certificates.

Historically the standards let the implementation decide to
either check or ignore the certificate properties of trust anchors.
You could either use them simply as a source of a public key which
was trusted for everything, or you were also permitted to check the
certificate properties and fully enforce them. Hooray for freedumb.

OpenSSL changed to checking these with :
commit 0daccd4dc1f1ac62181738a91714f35472e50f3c
Author: Viktor Dukhovni <openssl-users@dukhovni.org>
Date:   Thu Jan 28 03:01:45 2016 -0500

BoringSSL currently does not check them, as it also inherited
the previous OpenSSL behaviour. It will change to check them in
the future.
(https://bugs.chromium.org/p/boringssl/issues/detail?id=533)

23 months agoSync data and instruction cache before entering an executable page
jca [Sun, 13 Nov 2022 16:14:06 +0000 (16:14 +0000)]
Sync data and instruction cache before entering an executable page

Adapted from kettenis' fix for arm64/pmap.c.  Since we haven't been able
to reliably reproduce the clang crashes that affect base and ports, it
not yet known whether riscv64 was badly affected by this misordering.
Time will tell.  ok kettenis@

23 months agoSplit out handlers for SOCK_DGRAM unix(4) sockets from SOCK_STREAM and
mvs [Sun, 13 Nov 2022 16:01:32 +0000 (16:01 +0000)]
Split out handlers for SOCK_DGRAM unix(4) sockets from SOCK_STREAM and
SOCK_SEQPACKET. Introduce `uipc_dgram_usrreqs' to store pointers for
dgram specific handlers.

The dgram pru_shutdown and pru_send handlers were splitted to
uipc_dgram_shutdown() and uipc_dgram_send(). The pru_accept, pru_rcvd
and pru_abort handlers are not required for dgram sockets.

The unp_disconnect() remains shared between all unix(4) sockets because
it called from common paths too.

Proposed by and ok guenther@

23 months agosync
tb [Sun, 13 Nov 2022 14:09:44 +0000 (14:09 +0000)]
sync

23 months agoAvoid symbol collision with libcrypto
tb [Sun, 13 Nov 2022 14:09:17 +0000 (14:09 +0000)]
Avoid symbol collision with libcrypto

libcrypto now provides Ed25519 and the raw public key API, so neuter the
compat implementations. Also link libfido against libcrypto.

The #ifdefs were upstreamed in https://github.com/Yubico/libfido2/pull/645

no objection djm
ok jsing

23 months agoBump libtls minor to match libcrypto and libssl
tb [Sun, 13 Nov 2022 14:07:15 +0000 (14:07 +0000)]
Bump libtls minor to match libcrypto and libssl

23 months agoBump libssl minor to match libcrypto
tb [Sun, 13 Nov 2022 14:06:58 +0000 (14:06 +0000)]
Bump libssl minor to match libcrypto

23 months agoBump minor after symbol addition
tb [Sun, 13 Nov 2022 14:06:35 +0000 (14:06 +0000)]
Bump minor after symbol addition

23 months agoUpdate Symbols.list
tb [Sun, 13 Nov 2022 14:06:15 +0000 (14:06 +0000)]
Update Symbols.list

23 months agoExpose direct access API for Ed25519.
tb [Sun, 13 Nov 2022 14:05:04 +0000 (14:05 +0000)]
Expose direct access API for Ed25519.

23 months agoExpose various EVP hooks for Ed25519 and X25519
tb [Sun, 13 Nov 2022 14:04:13 +0000 (14:04 +0000)]
Expose various EVP hooks for Ed25519 and X25519

This adds the EVP_PKEY_ED25519 and EVP_PKEY_X25519 aliases for the NIDs
and exposes the raw public key API.

The ED25519_KEYLEN and X25519_KEYLEN defines are still kept internal for
now to match what OpenSSL have. We may want to expose those later.

23 months agoAdd ED25519 aliases for NID, SN and OBJ
tb [Sun, 13 Nov 2022 14:03:13 +0000 (14:03 +0000)]
Add ED25519 aliases for NID, SN and OBJ

The Ed25519 versions already existed, but OpenSSL chose to uppercase the D.

23 months agoExpose ASN1_buf_print() in asn1.h
tb [Sun, 13 Nov 2022 13:59:46 +0000 (13:59 +0000)]
Expose ASN1_buf_print() in asn1.h

This is needed to print the 32-byte Ed25519 keys which aren't handled
as BNs.

23 months agoRemove unused fields from struct in6_ifextra
kn [Sat, 12 Nov 2022 16:36:07 +0000 (16:36 +0000)]
Remove unused fields from struct in6_ifextra

This is all under _KERNEL:
- rs_lhcookie was added in 2014 110585f259f4974284e531f0a1e121b001a580dc
  "Move sending of router solicitations to the kernel; [...]"
  but never used
- nprefixes and ndefrouters became obsolete with 2017
  4a2f474d14c160dc7829cce0149ead09d473ece9
  "Remove sending of router solicitations and processing of router
   advertisements from the kernel. [...]"

OK mpi

23 months agoAdd support for string comparison in filters.
mpi [Sat, 12 Nov 2022 14:19:08 +0000 (14:19 +0000)]
Add support for string comparison in filters.

It is now possible to filter by process name, like:

syscall:mmap:entry
/comm == "ld"/
{
...
}

Currently the parser treats C-string like any other expression member even
if arithmetic operations do no apply to strings.

23 months agoHide symbols in libcrypto/ui
beck [Sat, 12 Nov 2022 13:16:10 +0000 (13:16 +0000)]
Hide symbols in libcrypto/ui

ok jsing@

23 months agoHide symbols in libcrypto/pkcs12
beck [Sat, 12 Nov 2022 13:03:28 +0000 (13:03 +0000)]
Hide symbols in libcrypto/pkcs12

ok jsing@

23 months agoFix a long-standing pmap bug, where we would enter an executable mapping
kettenis [Sat, 12 Nov 2022 12:58:34 +0000 (12:58 +0000)]
Fix a long-standing pmap bug, where we would enter an executable mapping
for a page before synchronizing the data and instruction cache.  This means
that another thread that is executing code on this page may not fault, but
see stale contennts until the data cache flushes and/or instruction cache
invalidation propagates.  The bug surfaced when testing a change that would
recycle code pages quickly instead of keeping them around.

Fix the issue by synchronizing the caches before entering an executable
mapping for a page.  Also make sure we mark the page as "clean" after
synchronization instead of before.

ok patrick@, jca@ (and mpi@ and dlg@ for an earlier version of this diff)

23 months agoHide symbols in libcrypto/pkcs7
beck [Sat, 12 Nov 2022 12:11:14 +0000 (12:11 +0000)]
Hide symbols in libcrypto/pkcs7

This applies the guentherizer 9000(tm) to pkcs7, after moving
several pkcs7 funcitions back to pkcs7 that were in x509/x_all.c
for reasons known only to the miasma.

ok jsing@

23 months agoRemove unused RS_LHCOOKIE macro
kn [Sat, 12 Nov 2022 02:53:17 +0000 (02:53 +0000)]
Remove unused RS_LHCOOKIE macro

Added in 2014 110585f259f4974284e531f0a1e121b001a580dc
    Move sending of router solicitations to the kernel; [...]
but never used.

23 months agoZap comment about dead linkmtu
kn [Sat, 12 Nov 2022 02:50:59 +0000 (02:50 +0000)]
Zap comment about dead linkmtu

Removed in 2015 a923c8476c6771493ee2bc4f277a429c9ae3eaff
    Remove linkmtu and maxmtu from struct nd_ifinfo.

23 months agoZap dead ip6_auto_linklocal
kn [Sat, 12 Nov 2022 02:49:34 +0000 (02:49 +0000)]
Zap dead ip6_auto_linklocal

Removed in 2015 15f0a5b39a2bb226b44457ed81a5d989cfee0393
    Ensure that link-local addresses are correctly configured on loopback
    interfaces.

23 months agoPut pf_state_import() under NPFSYNC>0 to fix build without pfsync
kn [Sat, 12 Nov 2022 02:48:14 +0000 (02:48 +0000)]
Put pf_state_import() under NPFSYNC>0 to fix build without pfsync

23 months agoAdd support for storing builtins arguments into local variables.
mpi [Fri, 11 Nov 2022 22:43:09 +0000 (22:43 +0000)]
Add support for storing builtins arguments into local variables.

23 months agoAdd support for multiple statements in if () blocks.
mpi [Fri, 11 Nov 2022 22:40:41 +0000 (22:40 +0000)]
Add support for multiple statements in if () blocks.

23 months agofix ounctuation and SEE ALSO;
jmc [Fri, 11 Nov 2022 21:41:45 +0000 (21:41 +0000)]
fix ounctuation and SEE ALSO;

23 months agoHide symbols in libcrypto/stack
beck [Fri, 11 Nov 2022 19:18:55 +0000 (19:18 +0000)]
Hide symbols in libcrypto/stack

Automated change from the first attempts at the semi automated
Guentherizer 2000.

ok jsing@ tb@ joshua@

23 months agoRevert last until we can guarantee gh_guid can't escape to the
krw [Fri, 11 Nov 2022 18:53:29 +0000 (18:53 +0000)]
Revert last until we can guarantee gh_guid can't escape to the
d_uid written to disk.

23 months agoClean up openssl(1) command execution.
joshua [Fri, 11 Nov 2022 18:24:32 +0000 (18:24 +0000)]
Clean up openssl(1) command execution.

This cleans up the code that handles command execution for openssl(1),
displays the help message when 'openssl help' is executed, and exits
with code 1 when an invalid command is executed, matching the behaviour
of OpenSSL version 1.1+ and above.

ok tb@

23 months agotimeout(9): remove timeout_set_kclock(), TIMEOUT_INITIALIZER_KCLOCK()
cheloha [Fri, 11 Nov 2022 18:09:58 +0000 (18:09 +0000)]
timeout(9): remove timeout_set_kclock(), TIMEOUT_INITIALIZER_KCLOCK()

We have too many timeout(9) initialization functions and macros.
Let's slim it down and combine some interfaces.

- Remove timeout_set_kclock(), TIMEOUT_INITIALIZER_KCLOCK().
- Expand timeout_set_flags(), TIMEOUT_INITIALIZER_FLAGS() to accept
  an additional "kclock" parameter.
- Reimplement timeout_set(), timeout_set_proc() with timeout_set_flags().
- Reimplement TIMEOUT_INITIALIZER() with TIMEOUT_INITIALIZER_FLAGS().
- Update the sole timeout_set_flags() user to pass a kclock parameter.
- Update the sole timeout_set_kclock() user to call timeout_set_flags().
- Update the sole TIMEOUT_INITIALIZER_FLAGS() user to provide a kclock
  parameter.

The timeout(9) code is now a bit out of sync with the manpage.  This
will be corrected in a subsequent commit.

ok kn@

23 months agoMake sure csa->csa_bundled is NULL after freeing to prevent a
mbuhl [Fri, 11 Nov 2022 17:58:14 +0000 (17:58 +0000)]
Make sure csa->csa_bundled is NULL after freeing to prevent a
possible use after free.

ok tobhe@

23 months agoDebug was accidentally left enabled in previous commit.
matthieu [Fri, 11 Nov 2022 17:23:09 +0000 (17:23 +0000)]
Debug was accidentally left enabled in previous commit.

23 months agoConvert the legacy TLS stack to tls_content.
jsing [Fri, 11 Nov 2022 17:15:26 +0000 (17:15 +0000)]
Convert the legacy TLS stack to tls_content.

This converts the legacy TLS stack to tls_content - records are now
opened into a tls_content structure, rather than being written back into
the same buffer that the sealed record was read into.

This will allow for further clean up of the legacy record layer.

ok tb@

23 months agome and my text editor are not getting along today
dlg [Fri, 11 Nov 2022 17:12:30 +0000 (17:12 +0000)]
me and my text editor are not getting along today

23 months agoRemove the legacy interactive mode from openssl(1).
joshua [Fri, 11 Nov 2022 17:07:38 +0000 (17:07 +0000)]
Remove the legacy interactive mode from openssl(1).

This removes the legacy interactive mode from openssl(1) since it is
rarely used, complicates the code, and has also been removed from
OpenSSL in version 3.x.x.

ok tb@ jsing@

23 months agoDue to checksum offloading one more element could be written to the ring
mbuhl [Fri, 11 Nov 2022 16:41:44 +0000 (16:41 +0000)]
Due to checksum offloading one more element could be written to the ring
than there is space available.
ok dlg

23 months agoIn case of an invalid SA resp is passed to ikev2_msg_cleanup without
mbuhl [Fri, 11 Nov 2022 16:17:16 +0000 (16:17 +0000)]
In case of an invalid SA resp is passed to ikev2_msg_cleanup without
initialization where the msg_parent field is accessed.
ok tobhe

23 months agotry pf.c r1.1143 again: move pf_purge out from under the kernel lock
dlg [Fri, 11 Nov 2022 16:12:08 +0000 (16:12 +0000)]
try pf.c r1.1143 again: move pf_purge out from under the kernel lock

this also avoids holding NET_LOCK too long.

the main change is done by running the purge tasks in systqmp instead
of systq. the pf state list was recently reworked so iteration over
the state can be done without blocking insertions.

however, scanning a lot of states can still take a lot of time, so
this also makes the state list scanner yield if it has spent too
much time running.

the other purge tasks for source nodes, rules, and fragments have
been moved to their own timeout/task pair to simplify the time
accounting.

in my environment, before this change pf purges often took 10 to
50ms. the softclock thread runs next to it often took a similar
amount of time, presumably because they ended up spinning waiting
for each other. after this change the pf_purges are more like 6 to
12ms, and dont block softclock. most of the variability in the runs
now seems to come from contention on the net lock.

tested by me sthen@ chris@
ok sashan@ kn@ claudio@

the diff was backed out because it made things a bit more racey,
but sashan@ has squashed those races this week. let's try it again.

23 months agosync
matthieu [Fri, 11 Nov 2022 16:06:13 +0000 (16:06 +0000)]
sync

23 months agoEnable icc(4). ok anton@ patrick@
matthieu [Fri, 11 Nov 2022 15:27:39 +0000 (15:27 +0000)]
Enable icc(4). ok anton@ patrick@

23 months agoAdd icc(4) to handle Customer Control keyboards attached to i2c bus.
matthieu [Fri, 11 Nov 2022 15:25:13 +0000 (15:25 +0000)]
Add icc(4) to handle Customer Control keyboards attached to i2c bus.
ok anton@ patrick@

23 months agoadd a mutex to struct pf_state and init it.
dlg [Fri, 11 Nov 2022 15:02:31 +0000 (15:02 +0000)]
add a mutex to struct pf_state and init it.

nothing is protected by it yet but it will allow us to provide
consistent updates to individual states without relying on a global
lock. getting that right between the packet processing in pf itself,
pfsync, the pf purge code, the ioctl paths, etc is not worth the
required contortions.

while pf_state does grow, it doesn't use more cachelines on machines
where we will want to run in parallel with a lot of states.

stolen from and ok sashan@

23 months agoihidev can pass a buffer larger than the size of the claimed input
anton [Fri, 11 Nov 2022 13:59:40 +0000 (13:59 +0000)]
ihidev can pass a buffer larger than the size of the claimed input
report since it unconditionally uses the size of the longest input
report. This is quite harmless and the defensive check in hidcc can
therefore be relaxed and instead only examining the number of expected
bytes.

ok matthieu@ patrick@

23 months agocvs(1) populates the author keyword using getlogin(2), therefore favor
anton [Fri, 11 Nov 2022 13:59:02 +0000 (13:59 +0000)]
cvs(1) populates the author keyword using getlogin(2), therefore favor
logname(1) when normalizing the output.

23 months agoSymbols.list: Drop comments and sort.
tb [Fri, 11 Nov 2022 13:08:29 +0000 (13:08 +0000)]
Symbols.list: Drop comments and sort.

While grouping the API by its purpose is nice, it doesn't help much if >90%
is "general API".

ok jsing

23 months agokn points out that brackets are not parentheses
dlg [Fri, 11 Nov 2022 12:50:45 +0000 (12:50 +0000)]
kn points out that brackets are not parentheses

23 months agofix a misuse of vi.
dlg [Fri, 11 Nov 2022 12:36:05 +0000 (12:36 +0000)]
fix a misuse of vi.

23 months agokn@ points out that lock annotations are usually wrapped in ()
dlg [Fri, 11 Nov 2022 12:29:32 +0000 (12:29 +0000)]
kn@ points out that lock annotations are usually wrapped in ()

23 months agoUse named initialisers.
jsing [Fri, 11 Nov 2022 12:23:49 +0000 (12:23 +0000)]
Use named initialisers.

Requested by tb@

23 months agoMerge bf_pi.h into bf_skey.c.
jsing [Fri, 11 Nov 2022 12:22:21 +0000 (12:22 +0000)]
Merge bf_pi.h into bf_skey.c.

There's not much point having a static table in a header file that is only
included in one source file.

Discussed with tb@

23 months agoTidy includes, fix comment style and mop up some blank lines.
jsing [Fri, 11 Nov 2022 12:18:25 +0000 (12:18 +0000)]
Tidy includes, fix comment style and mop up some blank lines.

23 months agoUse the first 8 bytes of GPT gh_guid as the default d_uid. gh_guid is
krw [Fri, 11 Nov 2022 12:14:01 +0000 (12:14 +0000)]
Use the first 8 bytes of GPT gh_guid as the default d_uid. gh_guid is
initialized with arc4random().

23 months agoWhack blowfish with a style(9) bat.
jsing [Fri, 11 Nov 2022 12:08:29 +0000 (12:08 +0000)]
Whack blowfish with a style(9) bat.

23 months agosteal a change by sashan@ to say which bits of pf_state need which locks.
dlg [Fri, 11 Nov 2022 12:06:17 +0000 (12:06 +0000)]
steal a change by sashan@ to say which bits of pf_state need which locks.

not all members are annotated yet, but that's because there's no clear
protection for them yet.

ok sashan@

23 months agoNuke trailing space in a comment.
krw [Fri, 11 Nov 2022 12:05:32 +0000 (12:05 +0000)]
Nuke trailing space in a comment.

23 months agoStart CBS-ifying the name constraints code.
beck [Fri, 11 Nov 2022 12:02:34 +0000 (12:02 +0000)]
Start CBS-ifying the name constraints code.

ok jsing@ tb@

23 months agoStop pretending that obj_mac.h is optional.
jsing [Fri, 11 Nov 2022 11:53:24 +0000 (11:53 +0000)]
Stop pretending that obj_mac.h is optional.

This is effectively:

  unifdef -m -DUSE_OBJ_MAC objects/objects.h

ok beck@, with extreme prejudice.

23 months agorename pfsync_up() to pfsync_is_up()
dlg [Fri, 11 Nov 2022 11:47:12 +0000 (11:47 +0000)]
rename pfsync_up() to pfsync_is_up()

foo_up() where foo is a network driver is usually a function that
configures and brings an interface up into a running state. this
small tweak just makes the code a bit easier for me to read.

23 months agoUnmap and free memory when we shutdown rtkit. This fixes an issue where
kettenis [Fri, 11 Nov 2022 11:45:10 +0000 (11:45 +0000)]
Unmap and free memory when we shutdown rtkit.  This fixes an issue where
we run out of SART entries after a few suspend/resume cycles.

ok patrick@

23 months agoresync with ISO web site; ok kmos@
daniel [Fri, 11 Nov 2022 11:27:05 +0000 (11:27 +0000)]
resync with ISO web site; ok kmos@

23 months agoAdd support for symbol hiding disabled by default.
beck [Fri, 11 Nov 2022 11:25:18 +0000 (11:25 +0000)]
Add support for symbol hiding disabled by default.

Fully explained in libcrypto/README. TL;DR make sure libcrypto
and libssl's function calls internally and to each other are via
symbol names that won't get overridden by linking other libraries.

Mostly work by guenther@, which will currently be gated behind a
build setting NAMESPACE=yes. once we convert all the symbols to
this method we will do a major bump and pick up the changes.

ok tb@ jsing@

23 months agoTurn KASSERT() into if() to prevent state being inserted to
sashan [Fri, 11 Nov 2022 11:22:48 +0000 (11:22 +0000)]
Turn KASSERT() into if() to prevent state being inserted to
pfsync snapshot multiple times.

OK dlg@

23 months agorewrite the pf_state_peer_ntoh and pf_state_peer_hton macros as functions.
dlg [Fri, 11 Nov 2022 11:02:35 +0000 (11:02 +0000)]
rewrite the pf_state_peer_ntoh and pf_state_peer_hton macros as functions.

i can read this code as functions, but it takes too much effort as macros.

23 months agomove struct pf_state from pfvar.h to pfvar_priv.h.
dlg [Fri, 11 Nov 2022 10:55:48 +0000 (10:55 +0000)]
move struct pf_state from pfvar.h to pfvar_priv.h.

we (sashan) are going to add a mutex to the pf_state struct, but a
mutex is a kernel data structure that changes shape depending on
things like whether MULTIPROCESSOR is enabled, and should therefore
not be visible to userland. when we added a mutex to pf_state,
compiling pfctl failed because it doesn't know what a mutex is and
it can't know which version of it the current kernel is running
with.

moving struct pf_state to pfvar_priv.h makes it clear it is a private
kernel only data structure, and avoids this leak into userland.

tested by me and make build
ok sashan@

23 months agoRevert removal of toggling interrupt line in vmd vcpu run loop.
dv [Fri, 11 Nov 2022 10:52:44 +0000 (10:52 +0000)]
Revert removal of toggling interrupt line in vmd vcpu run loop.

phessler reports a performance regression. Needs more testing.