tcp: Deal with DSACKs, and adjust rescue hole on success.
When a rescue retransmission is successful, rather than
inserting new holes to the left of it, adjust the old
rescue entry to cover the missed sequence space.
Also, as snd_fack may be stale by that point, pull it forward
in order to never create a hole left of snd_una/th_ack.
Finally, with DSACKs, tcp_sack_doack() may be called
with new full ACKs but a DSACK block. Account for this
eventuality properly to keep sacked_bytes >= 0.
MFC after: 3 days
Reviewed By: kbowling, tuexen, #transport
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D29835
Alexander Motin [Fri, 16 Apr 2021 19:39:01 +0000 (15:39 -0400)]
pms(4): Limit maximum I/O size to 256KB instead of 1MB.
There is a weird limit of AGTIAPI_MAX_DMA_SEGS (128) S/G segments per
I/O since the initial driver import. I don't know why it was added,
can only guess some hardware limitation, but in worst case it means
maximum I/O size of 508KB. Respect it to be safe, rounding to 256KB.
Maintain code similarity between RACK and base stack
for ECN. This may not strictly be necessary, depending
when a state transition to FIN_WAIT_1 is done in RACK
after a shutdown() or close() syscall.
MFC after: 3 days
Reviewed By: tuexen, #transport
Sponsored by: NetApp, Inc.
Differential Revision: https://reviews.freebsd.org/D29658
tcp: Prepare PRR to work with NewReno LossRecovery
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).
Notable upstream pull request merges:
#11742 When specifying raidz vdev name, parity count should match
#11744 Use a helper function to clarify gang block size
#11771 Support running FreeBSD buildworld on Arm-based macOS hosts
This is the last update that will be MFCed into stable/13.
From now on, the tracking of OpenZFS branches will be different:
- main continues tracking openzfs/zfs/master
- stable/13 is going to track openzfs/zfs/zfs-2.1-release
Radix MMU code was missing TLB invalidations when some Level 3 PDEs were
modified. This caused TLB multi-hit machine check interrupts when
superpages were enabled.
Reviewed by: jhibbits
Sponsored by: Eldorado Research Institute (eldorado.org.br)
Differential Revision: https://reviews.freebsd.org/D29511
Leandro Lupori [Thu, 25 Mar 2021 16:30:56 +0000 (13:30 -0300)]
powerpc64: clear low-order bits of ARPN
PowerISA 2.07B says that the low-order p-12 bits of the real page number
contained in ARPN and LP fields of a PTE must be 0s and are ignored
by the hardware (Book III-S, 5.7.7.1), where 2^p is the actual page size
in bytes, but we were clearing only the LP field.
This worked on bare metal and QEMU with KVM, that ignore these bits,
but caused a kernel panic on QEMU with TCG, that expects them to be
cleared.
This fixes running FreeBSD with HPT superpages enabled on QEMU
with TCG.
Sponsored by: Eldorado Research Institute (eldorado.org.br)
Rick Macklem [Thu, 8 Apr 2021 21:04:22 +0000 (14:04 -0700)]
nfsd: fix replies from session cache for retried RPCs
Recent testing of network partitioning a FreeBSD NFSv4.1
server from a Linux NFSv4.1 client identified problems
with both the FreeBSD server and Linux client.
The FreeBSD server failec to reply using the cached
reply in the session slot when an RPC was retried on
the session slot, as indicated by same slot sequence#.
This patch fixes this. It should also fix a similar
failure for NFSv4.0 mounts, when the sequence# in
the open/lock_owner requires a reply be done from
an entry locked into the DRC.
This fix affects the fairly rare case where a NFSv4
client retries a non-idempotent RPC, such as a lock
operation. Note that retries only occur after the
client has needed to create a new TCP connection.
Alex Richardson [Mon, 19 Apr 2021 23:22:45 +0000 (00:22 +0100)]
Allow lib/msun/logarithm_test to pass on ld128 platforms
For some reason the ld128 log1pl() implementation is less accurate than
logl(), but does at least guarantee precision >= the ld80 implementation.
Mark log1p_accuracy_tests as XFAIL for ld128 and increase the log1p tolerance
to the ld80 equivalent in accuracy_tests to avoid losing test coverage for
the other functions.
PR: 253984
Reviewed By: ngie, dim
Differential Revision: https://reviews.freebsd.org/D29039
Alex Richardson [Thu, 15 Apr 2021 15:27:52 +0000 (16:27 +0100)]
lib/msun: Exclude ignored-pragmas from -Werror
This avoids build failures due to the clang 12 warning:
'#pragma FENV_ACCESS' is not supported on this target - ignored
Clang 12 currently emits this warning for all non-x86 architectures.
While this can result in incorrect code generation (e.g. on AArch64 some
exceptions are not raised as expected), this is a pre-existing issue and
we should not fail the build due to this warning.
Alex Richardson [Mon, 22 Mar 2021 16:54:02 +0000 (16:54 +0000)]
Fix lib/msun/tests/csqrt_test on platforms with 128-bit long double
If long double has more than 64 mantissa bits, using uint64_t to hold the
mantissa bits will truncate the value and result in test failures. To fix
this problem use __uint128_t since all platforms that have
__LDBL_MANT_DIG__ > 64 also have compiler support for 128-bit integers.
Alex Richardson [Mon, 22 Mar 2021 13:05:28 +0000 (13:05 +0000)]
Fix unused functions in invtrig_test.c
I only tested the WARNS=6 change on AArch64 and AMD64, but this file has
unused functions for architectures with LDBL_PREC == 53.
While touching this file change the LDBL_PREC == 53 checks to i386 checks.
The long double tests should only be disabled for i386 (due to the rather
odd rounding mode that it uses) not all architectures where long double
is the same as double.
PR: 205449
Fixes: 87d65c747a43 ("lib/msun: Allow building tests with WARNS=6")
Reported by: Jenkins
Alex Richardson [Mon, 22 Mar 2021 11:55:45 +0000 (11:55 +0000)]
Don't add -Winline for WARNS=6
This warning is very rarely useful (inline is a hint and not mandatory).
This flag results in many warnings being printed when compiling C++
code that uses the standard library with GCC.
This flag was originally added in back in r94332 but the flag is a no-op
in Clang ("This diagnostic flag exists for GCC compatibility, and has no
effect in Clang"). Removing it should make the GCC build output slightly
more readable.
Alex Richardson [Mon, 22 Mar 2021 11:53:40 +0000 (11:53 +0000)]
lib/msun/tests: Skip fenv_test:masking if exceptions can't be trapped
Some CPUs (e.g. AArch64 QEMU) cannot trap on floating point exceptions and
therefore ignore the writes to the floating point control register inside
feenableexcept(). If no exceptions are enabled after
feenableexcept(FE_ALL_EXCEPT), we can assume that the CPU does not
support exceptions and we can then skip the test.
Reviewed By: dim
Differential Revision: https://reviews.freebsd.org/D29095
Alex Richardson [Mon, 22 Mar 2021 11:15:13 +0000 (11:15 +0000)]
Convert the msun tests to ATF
This provides better error messages that just an assertion failure and
also makes it easier to mark individual tests as XFAIL.
It was also helpful when coming up with D28786 and D28787.
Apparently GCC only supports arithmetic expressions that use static
const variables in initializers starting with GCC8. To keep older
versions happy use a macro instead.
Fixes: 221622ec0c ("lib/msun: Avoid FE_INEXACT for x86 log2l/log10l")
Reported by: Jenkins
Reviewed By: imp
Differential Revision: https://reviews.freebsd.org/D29233
Alex Richardson [Mon, 8 Mar 2021 09:39:29 +0000 (09:39 +0000)]
lib/msun: Avoid FE_INEXACT for x86 log2l/log10l
This fixes tests/lib/msun/logarithm_test after compiling the test with
-fno-builtin (D28577). Adding invln10_lo + invln10_10 results in
FE_INEXACT (for all inputs) and the same for the log2l invln2_lo + invln2_hi.
This patch avoids FE_INEXACT (for exact results such as 0) by defining a
constant and using that.
Reviewed By: dim
Differential Revision: https://reviews.freebsd.org/D28786
Alex Richardson [Mon, 1 Mar 2021 19:50:38 +0000 (19:50 +0000)]
Allow ssp_test:read to pass more reliably
It appears that the stackframe layout can be slightly different depending on
compiler and target architecture. For example, when using CHERI LLVM for RISC-V
we can actually overflow the buffer by up to 8 bytes without SSP detecting it.
Fix this by increasing the overflow to 15 bytes.
Alex Richardson [Wed, 24 Feb 2021 16:56:53 +0000 (16:56 +0000)]
Update scalbn* functions to the musl versions
The only diff compared to musl is a minor change to scalbnl() to replace
musl's union ldshape with union IEEEl2bits.
This fixes the scalbn tests on non-x86 (since x86 has an assembly version
that is used instead).
in nearest rounding mode scalbn could introduce double rounding error
when an intermediate value and the final result were both in the
subnormal range e.g.
scalbn(0x1.7ffffffffffffp-1, -1073)
returned 0x1p-1073 instead of 0x1p-1074, because the intermediate
computation got rounded to 0x1.8p-1023.
with the fix an intermediate value can only be in the subnormal range
if the final result is 0 which is correct even after double rounding.
(there still can be two roundings so signals may be raised twice, but
that's only observable with trapping exceptions which is not supported.)
old code was correct only if the result was stored (without the
excess precision) or musl was compiled with -ffloat-store.
now we use STRICT_ASSIGN to work around the issue.
(see note 160 in c11 section 6.8.6.4)
math: fix scalbn and scalbnf on overflow/underflow
old code was correct only if the result was stored (without the
excess precision) or musl was compiled with -ffloat-store.
(see note 160 in n1570.pdf section 6.8.6.4)
Alex Richardson [Mon, 22 Feb 2021 17:19:06 +0000 (17:19 +0000)]
Update libm tests from NetBSD
I did this without a full vendor update since that would cause too many
conflicts. Since these files now almost match the NetBSD sources the
next git subtree merge should work just fine.
Alex Richardson [Wed, 3 Feb 2021 09:31:32 +0000 (09:31 +0000)]
sbin/bectl: Skip tests if sparse files are not supported
The tests create a 1GB test file and this causes the tests to fail in the
CheriBSD CI setup where we run tests with a tmpfs mount on /tmp. Tmpfs
does not support sparse files and it appears that tmpfs default to creating
a 1GB mount, so there is not enough space to run these tests.
Instead of checking for at least 1GB of free space, this commit skips the
tests on file systems that do not support sparse files.
Alex Richardson [Wed, 3 Feb 2021 09:29:08 +0000 (09:29 +0000)]
Fix build with read-only source dir after 83c20b8a2da0
I changed the Makefile to use SRCS instead of LDADD, but since there is
still and absolute path to the source the .o file was created inside the
source directory instead of the build directory.
It would be nice if this was an error/warning by default, but for now just
fix this issue by using .PATH and the base name of the file.
Alex Richardson [Fri, 29 Jan 2021 09:29:26 +0000 (09:29 +0000)]
Avoid double output in fenv_test
This tests fork()s, so if there is still data in the stdout buffer on fork
it will print it again in the child process. This was happening in the
CheriBSD CI and caused the test to complain about malformed TAP output.
Improve size readability.
Preserve more space for swap devise names.
Prevent line overflow with long devise name.
Don't draw a bar when swap is not used at all.
Simplify and optimize code.
Change the label to end at end of 100%.
PR: 251655
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D27496
Mitchell Horne [Mon, 8 Mar 2021 19:04:51 +0000 (15:04 -0400)]
gdb: report specific stop reason for watchpoints
The remote protocol allows for implementations to report more specific
reasons for the break in execution back to the client [1]. This is
entirely optional, so it is only implemented for amd64, arm64, and i386
at the moment.
Mitchell Horne [Thu, 4 Mar 2021 00:14:42 +0000 (20:14 -0400)]
arm: implement kdb watchpoint functions
Implement wrappers around the existing debug_monitor interface, to be
consumed by MI kernel debugger code.
For now, the various db_printf() calls in this code remain. In the
future, they could be converted to printf() or removed altogether, to
properly decouple the DDB and GDB options.
Reviewed by: jhb, kib, markj
Sponsored by: NetApp, Inc.
Sponsored by: Klara, Inc.
Mitchell Horne [Thu, 4 Mar 2021 21:53:21 +0000 (17:53 -0400)]
arm64: implement kdb watchpoint functions
Add wrappers around the debug_monitor interface, to be consumed by MI
kernel debugger code. Update dbg_setup_watchpoint() and
dbg_remove_watchpoint() to return specific error codes, not just -1.
Reviewed by: jhb, kib, markj
Sponsored by: NetApp, Inc.
Sponsored by: Klara, Inc.
Mitchell Horne [Fri, 19 Feb 2021 22:36:08 +0000 (18:36 -0400)]
x86: implement kdb watchpoint functions
Add wrappers around the dbreg interface that can be consumed by MI
kernel debugger code. The dbreg functions themselves are updated to
return error codes, not just -1. dbreg_set_watchpoint() is extended to
accept access bits as an argument.
Reviewed by: jhb, kib, markj
Sponsored by: NetApp, Inc.
Sponsored by: Klara, Inc.
Mitchell Horne [Mon, 8 Mar 2021 15:23:40 +0000 (11:23 -0400)]
Introduce kdb-level watchpoint functions
This basically mirrors what already exists in ddb, but provides a
slightly improved interface. It allows the caller to specify the
watchpoint access type, and returns more specific error codes to
differentiate failure cases.
This will be used to support hardware watchpoints in gdb(4).
Stubs are provided for architectures lacking hardware watchpoint logic
(mips, powerpc, riscv), while other architectures are added individually
in follow-up commits.
Reviewed by: jhb, kib, markj
Sponsored by: NetApp, Inc.
Sponsored by: Klara, Inc.
Mitchell Horne [Fri, 19 Mar 2021 19:39:12 +0000 (16:39 -0300)]
x86: consolidate hw watchpoint logic into new file
This is a prerequisite to using these functions outside of ddb, but also
provides some cleanup and minor refactoring. This code is almost
entirely duplicated between the two implementations, the only
significant difference being the lack of dbreg synchronization on i386.
Cleanups are:
- demote some internal functions to static
- use the constant NDBREGS instead of a '4' literal
- remove K&R definitions
- some added comments
Reviewed by: kib, jhb
Sponsored by: NetApp, Inc.
Sponsored by: Klara, Inc.
Justin Hibbits [Wed, 31 Mar 2021 16:54:27 +0000 (11:54 -0500)]
powerpc/aim: Update timebase directly on resume instead of through platform
This only works on single-CPU G4 systems, and more work is needed for
dual-CPU systems. That said, platform sleep does not work, and this is
currently only used for PMU-based CPU speed change.
The elimination of the platform_smp_timebase_sync() call is so that the
timebase sync rendezvous can be enhanced to perform better
synchronization, which requires a full rendezvous. This would be
impossible to do on this single-threaded run.
Justin Hibbits [Wed, 31 Mar 2021 16:54:11 +0000 (11:54 -0500)]
powerpc/powermac: Constrain 'cpu_sleep()' for AIM to mpc745x
Rename cpu_sleep() to mpc745x_sleep() to denote what it's actually
intended for. This function is very G4-specific, and will not work on
any other CPU. This will afterward eliminate a
platform_smp_timebase_sync() call by directly updating the timebase
instead.
When we find a state for packets that was created by a reply-to rule we
still need to process the packet. The state may require us to modify the
packet (e.g. in rdr or nat cases), which we won't do with the shortcut.
pf: change pf_route so pf only runs when packets enter and leave the stack.
before this change pf_route operated on the semantic that pf runs
when packets go over an interface, so when pf_route changed which
interface the packet was on it would run pf_test again. this change
changes (restores) the semantic that pf is only supposed to run
when packets go in or out of the network stack, even if route-to
is responsibly for short circuiting past the network stack.
just to be clear, for normal packets (ie, those not touched by
route-to/reply-to/dup-to), there isn't a difference between running
pf when packets enter or leave the stack, or having pf run when a
packet goes over an interface.
the main reason for this change is that running the same packet
through pf multiple times creates confusion for the state table.
by default, pf states are floating, meaning that packets are matched
to states regardless of which interface they're going over. if a
packet leaving on em0 is rerouted out em1, both traversals will end
up using the same state, which at best will make the accounting
look weird, or at worst fail some checks in the state and get
dropped.
another reason for this commit is is to make handling of the changes
that route-to makes consistent with other changes that are made to
packet. eg, when nat is applied to a packet, we don't run pf_test
again with the new addresses.
the main caveat with this diff is you can't have one rule that
pushes a packet out a different interface, and then have a rule on
that second interface that NATs the packet. i'm not convinced this
ever worked reliably or was used much anyway, so we don't think
it's a big concern.
discussed with many, with special thanks to bluhm@, sashan@ and
sthen@ for weathering most of that pain.
ok claudio@ sashan@ jmatthew@
Slighly relax the gateway validation rules imposed by the 2fe5a79425c7, by requiring only first 8 bytes (everyhing
before sdl_data to be present in the AF_LINK gateway.
Mark Johnston [Mon, 12 Apr 2021 13:32:30 +0000 (09:32 -0400)]
Rename struct device to struct _device
types.h defines device_t as a typedef of struct device *. struct device
is defined in subr_bus.c and almost all of the kernel uses device_t.
The LinuxKPI also defines a struct device, so type confusion can occur.
This causes bugs and ambiguity for debugging tools. Rename the FreeBSD
struct device to struct _device.
Reviewed by: gbe (man pages)
Reviewed by: rpokala, imp, jhb
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D29676
Mark Johnston [Mon, 12 Apr 2021 13:32:08 +0000 (09:32 -0400)]
qlnxr: Properly initialize the Linux device structure
The driver needs to provide a LinuxKPI device structure to register
itself with the IB subsystem. It was erroneously using a copy of its
FreeBSD device structure for this purpose.
Use linux_pci_attach_device() instead, following the example of the
Chelsio iwarp driver. Also ensure that we don't leak the faked device
during detach.
Reviewed by: hselasky
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D29595
Alex Richardson [Mon, 19 Apr 2021 08:36:47 +0000 (09:36 +0100)]
Revert "lib/libc/net/nsdispatch.c: Fix missing unlock and add locking annotations"
This commit should not have introduced any functional changes, but
apparently it did. This appears to have broken LDAP setups.
Reverting for now. Will reland once I have fixed the breakage.
Rick Macklem [Mon, 5 Apr 2021 01:15:54 +0000 (18:15 -0700)]
nfsd: make the server repeat CB_RECALL every couple of seconds
Commit 01ae8969a9ee stopped the NFSv4.1/4.2 server from implicitly
binding the back channel to a new TCP connection so that it
conforms to RFC5661, for NFSv4.1/4.2. An effect of this
for the Linux NFS client is that it will do a
BindConnectionToSession when it sees NFSV4SEQ_CBPATHDOWN
set in a sequence reply. This will fix the back channel, but the
first attempt at a callback like CB_RECALL will already have
failed. Without this patch, a CB_RECALL will not be retried
and that can result in a 5 minute delay until the delegation
times out.
This patch modifies the code so that it will retry the
CB_RECALL every couple of seconds, often avoiding the
5 minute delay.
This is not critical for correct behaviour, but avoids
the 5 minute delay for the case where the Linux client
re-binds the back channel via BindConnectionToSession.
Rick Macklem [Sun, 4 Apr 2021 22:05:39 +0000 (15:05 -0700)]
nfsd: fix BindConnectionToSession so that it clears "cb path down"
Commit 01ae8969a9ee stopped the NFSv4.1/4.2 server from implicitly
binding the back channel to a new TCP connection so that it
conforms to RFC5661, for NFSv4.1/4.2. An effect of this
for the Linux NFS client is that it will do a
BindConnectionToSession when it sees NFSV4SEQ_CBPATHDOWN
set in a sequence reply. It will do this for every RPC
reply until it no longer sees the flag.
Without that patch, this will happen until the client does
an Open, which will clear LCL_CBDOWN.
This patch clears LCL_CBDOWN right away, so that
NFSV4SEQ_CBPATHDOWN will no longer be sent to the client
in Sequence replies and the Linux client will not repeat
the BindConnectionToSession RPCs.
This is not critical for correct behaviour, but reduces
RPC overheads for cases where the Open will not be done
for a while.
tcp: Add hash histogram output and validate bucket length accounting
Provide a histogram output to check, if the hashsize or
bucketlimit could be optimized. Also add some basic sanity
checks around the accounting of the hash utilization.
tcp: For hostcache performance, use atomics instead of counters
As accessing the tcp hostcache happens frequently on some
classes of servers, it was recommended to use atomic_add/subtract
rather than (per-CPU distributed) counters, which have to be
summed up at high cost to cache efficiency in a hot codepath.
tcp: Make hostcache.cache_count MPSAFE by using a counter_u64_t
Addressing the underlying root cause for cache_count to
show unexpectedly high values, by protecting all arithmetic on
that global variable by using counter(9).