libc/getaddrinfo(2): return EAI_AGAIN on nameserver timeout
A nameserver timeout is a soft failure, future attempts may succeed.
Returning EAI_AGAIN is crucial for API users to tell a soft name
resolution failure from negative resolution result.
Before the change we would return EAI_ADDRFAMILY, which I believe, is a
regression from 144361386696, and before that revision we used to return
EAI_NONAME in most of the cases.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D49411
libc/tests: enable getaddrinfo(1) tests from NetBSD
Enable running of contrib/netbsd-tests/lib/libc/net/getaddrinfo without
modifications to the test program itself. However, create our own version
of expected results files. They are produced from the NetBSD files, with
the following list of changes:
o AF_INET6 value is different, hence: s/family 24/family 28/.
o SCTP result (socktype 5 protocol 132) needs to be added to many queries
results.
o NetBSD libc sorts multiple results first by protocol, then by address
family. FreeBSD first sorts on address family, then on protocol. Some
results need to be reordered.
o Error strings need to be fixed:
s/hostname nor servname provided, or not known/Name does not resolve/
s/ai_family not supported/Address family not recognized/
s/servname not supported for ai_socktype/\
Service was not recognized for socket type/.
[2 lines not shown]
libc/tests: add getaddrinfo test
A test suite for getaddrinfo(3) written in C. Unlike NetBSD test, this
one will be mostly focused on what the API should return when something
isn't good with your DNS. Test emulates bad DNS servers in resolv.conf
intercepting fopen(2) and emulates downed network intercepting send(2).
Initial version covers three main scenarios: all good, server(s) timed
out, network down. For each scenario we test hostname with trailing dot
and without, since libc resolver uses quite different code paths,
sometimes even yielding with different error codes. All current error
codes in the test are what our libc returns right now, meaning the test
documents what we have, not what there should be.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D49410
mount_nfs: make temporary DNS failure non-fatal with background mode
Typical problem with network mounts is remote equipment not being
available when our host boots up after a power failure. Even if you
properly configure boot order of all local services and wait for link
coming up on your NIC, you still may boot faster than some intermediate
switch on the network or the DNS server itself. Let's refer to this as a
"server room boot race". For NFS mounts with hostname in hosts(5) the
race is addressed by a retry loop on NFS mount timeout. However, a DNS
resolution timeout is treated differently to NFS mount timeout. We fail
on the former and keep retrying on the latter.
With feedback received on current@, I see that the problem is so old, that
people got used to it and see it as a desired behavior rather than a
problem. And for those who is affected by the problem, they suggest
hosts(5) as a solution. Note that using hosts(5) isn't scalable, and
using bare IP addresses is neither scalable, nor compatible with
Kerberized mounts.
[7 lines not shown]
athpoke: Use a format string with err()
Don't do err(1, ptr); that's a very old security risk turned warning.
Differential Revision: https://reviews.freebsd.org/D49546
Reviewed by: thj
ath: fix athtools common code to correctly init state
Check req->s instead of s to know if the state needs to be
cleaned up first.
This error is from a refactor I did years ago. Oops!
Differential Revision: https://reviews.freebsd.org/D49545
Reviewed by: thj
net80211: document the 802.11 specification for ieee80211_getcapinfo()
Add a reference to the 802.11 specification for the management frame
capability info field.
Differential Revision: https://reviews.freebsd.org/D49513
Reviewed by: thj
mtw: fix compilation with IEEE80211_SUPPORT_SUPERG enabled
It looks like this was missed during the driver development and
porting.
Differential Revision: https://reviews.freebsd.org/D49512
Reviewed by: thj
release/azure: Update instruction of publishing to Azure Marketplace
Using Compute Gallery as the source is the preferred method now.
Reviewed by: brooks, emaste
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D49528
hier.7: Storage partitions are not filesystems
I recommended the incorrect text to the submitter.
MFC after: 3 days
Fixes: f47cbb29e1c2 (Add /dev/gpt)
Reported by: Mark Millard <marklmi at yahoo.com>
Reviewed by: imp, mhorne
Approved by: mhorne (mentor)
Differential Revision: https://reviews.freebsd.org/D49523
top: Sync usage and synopsis
Switching between io and cpu sorting uses a great example in the usage.
[-m io | cpu]. Use that everywhere.
MFC after: 3 days
Reviewed by: mhorne
Approved by: mhorne (mentor)
Differential Revision: https://reviews.freebsd.org/D49515
arm64: Fix off-by-one in its_init_cpu_lpi
The low bits of GICR_PROPBASER are defined as
The number of bits of LPI INTID supported, minus one, by the LPI
Configuration table starting at Physical_Address.
but flsl(1 << n) returns n + 1; use ilog2_long instead.
PR: 285677
Reported by: Julien Grall
Reviewed by: andrew
Sponsored by: Amazon
Differential Revision: https://reviews.freebsd.org/D49542
kern.opts.mk: Remove EFI from riscv BROKEN_OPTIONS
This discrepancy previously caused `make makeman` to warn
`riscv/riscv64: ignoring duplicate option EFI`. Make the kernel EFI
option consistent with EFI's per-arch defaults in src.opts.mk to remove
the warning. Note that we do build EFI boot programs for RISC-V.
This is a NFC because the only component gated by the kernel's EFI
option is sys/modules/efirt, which is also inside of a MACHINE_CPUARCH
block that excludes riscv64 anyway.
Reviewed by: mhorne
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D49549
zfs: remove inclusion of machine/pcpu.h
It was necessary in the beginning for a definition of curthread,
but in the later versions of the patch turned out to be not needed.
Fixes: 5b02365ac656e1cccf293ec1c57a8eb6c5cd51e2
Reported by: mav
zfs: enable FPU on powerpc*
Summary:
MFC after: 1 week
Test Plan: make kernel and boot with ZFS on /
Reviewers: #powerpc
Subscribers: imp, delphij
Differential Revision: https://reviews.freebsd.org/D49538
pfctl: Split pool parsing into separate functions
The pf pools are used in NAT, route-to and af-to rules. Some parts of
code are duplicated between them. Create functions apply_redirspec(),
apply_nat_ports() and apply_rdr_ports() to handle the common tasks.
Simplify data structures used for pool parsing. Move the contents of
struct redirection to struct redirspec. Map all ways of parsing pools
directly onto struct redirspec. Name various forms of struct redirspect
to hint where they are used.
Remove struct redirspec *rroute from struct filter_opts, because
filter_opts is bzero()'ed after the route part of rule is parsed, and
thus can't be used.
Add tests to ensure that parsing and error messages behave as expected.
The tests have been written and tested with pfctl from before this
patch.
[6 lines not shown]
Bump .Dd for recently changed aio_*(2) manpages
The ones that were effectively unchanged from
d97e44784bb5a^..e24279e0f9e did not have `.Dd` bumped. Only
the ones that had a net content change between those
revisions.
MFC after: 2 weeks
MFC with: d97e44784bb5ae24279e0f9e
Remove mentions of ENOSYS added in d97e44784bb5aio(4) is a hard requirement in the kernel as of f3215338ef82. The
scenario that the patch was submitted for is no longer possible.
This isn't a straight up revert since the previous change also addressed
some minor issues.
PR: 190942
Reported by: asomers
MFC after: 2 weeks
MFC with: d97e44784bb5
Fixes: d97e44784bb5 ("aio_*(2): mention ENOSYS under ERRORS")
Differential Revision: https://reviews.freebsd.org/D49541
powernv: Explicitly enable the facilities we support
Thus far we've relied on the firmware setting HFSCR correctly, enabling
all facilities. However, it appears some recent firmwares do not.
Linux explicitly enables all facilities it supports (including
transactional memory, which new firmwares apparently do not enable), so
do the same here, explicitly enable all the facilities FreeBSD supports.
MFC after: 1 week
(cherry picked from commit fe73eb441f1c98fe350f4822c7d2abfeef046bf8)
powerpc: Add Hypervisor Facility Unavailable handling
Treat Hypervisor Facility Unavailable as an illegal instruction in user
space. This way programs that use facilities like transactional memory,
which will never be supported as it has been removed from the ISA, will
not crash the kernel.
Tested by: pkubaj
MFC after: 1 week
(cherry picked from commit 0b663e0cee89dedf5b070ff4e839d5ad95061535)
release/vm: fix setting NO_ROOT and WITHOUT_QEMU
To support common shell patterns (test -n and -z) we want NO_ROOT and
WITHOUT_QEMU to be set and not empty in mk-vmimage.sh if they are set at
all in the Makefile.
Actually do this rather than making them always set (which propogates to
submakes run by mk-vmimage.sh and unconditionally enables NO_ROOT with
undesirable side effects such as installing a /METALOG in vm images).
Reported by: cperciva
Sponsored by: DARPA, AFRL
Differential Revision: https://reviews.freebsd.org/D49544
pci: Use a single variable for the offset of the power management registers
This is the more typical pattern for other capability register sets,
and two of these variables weren't used.
Differential Revision: https://reviews.freebsd.org/D49267
