Improve the TCP blackhole detection. The principle is to reduce the
MSS in two steps and try each candidate two times. However, if two
candidates are the same (which is the case in TCP/IPv6), this candidate
was tested four times. This patch ensures that each candidate actually
reduced the MSS and is only tested 2 times. This reduces the time window
of missclassifying a temporary outage as an MTU issue.
manu [Tue, 14 Apr 2020 16:34:13 +0000 (16:34 +0000)]
arm: allwinner: aw_mmc: Use the mmc_fdt_helper
The fdt properties are now parsed via the help of mmc_fdt_helper functions.
This also adds card detection.
Note that on some boards (like the Pine64) card detection is broken due to
a missing resistor on the cd pin.
manu [Tue, 14 Apr 2020 16:30:54 +0000 (16:30 +0000)]
Those functions are here to help fdt mmc controller drivers to parse
the dts to find the supported speeds and the regulators.
Not all DTS have every settings properly defined so host controller
will still have to add some caps themselves.
It also add a mmc_fdt_gpio_setup function which will read the cd-gpios
property and register it as the CD pin.
If the pin support interrupts one will be registered and the cd_helper
function will be called.
If the pin doesn't support interrupts the internal taskqueue will poll
for change and call the same cd_helper function.
mmc_fdt_gpio_setup will also parse the wp-gpio property and MMC drivers
can know the write-protect pin value by calling the
mmc_fdt_gpio_get_readonly function.
Make sonewconn() overflow messages have per-socket rate-limits and values.
sonewconn() emits debug-level messages when a listen socket's queue
overflows. Currently, sonewconn() tracks overflows on a global basis. It
will only log one message every 60 seconds, regardless of how many sockets
experience overflows. And, when it next logs at the end of the 60 seconds,
it records a single message referencing a single PCB with the total number
of overflows across all sockets.
This commit changes to per-socket overflow tracking. The code will now
log one message every 60 seconds per socket. And, the code will provide
per-socket queue length and overflow counts. It also provides a way to
change the period between log messages using a sysctl.
Print more detail as part of the sonewconn() overflow message.
When a socket's listen queue overflows, sonewconn() emits a debug-level
log message. These messages are sometimes useful to systems administrators
in highlighting a process which is not keeping up with its listen queue.
This commit attempts to enhance the usefulness of this message by printing
more details about the socket's address. If all else fails, it will at
least print the domain name of the socket.
KTLS: Re-work unmapped mbufs to carry ext_pgs in the mbuf itself.
While the original implementation of unmapped mbufs was a large
step forward in terms of reducing cache misses by enabling mbufs
to carry more than a single page for sendfile, they are rather
cache unfriendly when accessing the ext_pgs metadata and
data. This is because the ext_pgs part of the mbuf is allocated
separately, and almost guaranteed to be cold in cache.
This change takes advantage of the fact that unmapped mbufs
are never used at the same time as pkthdr mbufs. Given this
fact, we can overlap the ext_pgs metadata with the mbuf
pkthdr, and carry the ext_pgs meta directly in the mbuf itself.
Similarly, we can carry the ext_pgs data (TLS hdr/trailer/array
of pages) directly after the existing m_ext.
In order to be able to carry 5 pages (which is the minimum
required for a 16K TLS record which is not perfectly aligned) on
LP64, I've had to steal ext_arg2. The only user of this in the
xmit path is sendfile, and I've adjusted it to use arg1 when
using unmapped mbufs.
This change is almost entirely mechanical, except that we
change mb_alloc_ext_pgs() to no longer allow allocating
pkthdrs, the change to avoid ext_arg2 as mentioned above,
and the removal of the ext_pgs zone,
This change saves roughly 2% "raw" CPU (~59% -> 57%), or over
3% "scaled" CPU on a Netflix 100% software kTLS workload at
90+ Gb/s on Broadwell Xeons.
In a follow-on commit, I plan to remove some hacks to avoid
access ext_pgs fields of mbufs, since they will now be in
cache.
Many thanks to glebius for helping to make this better in
the Netflix tree.
Similar to mmap'ing vnodes, posixshm should count any mapping where maxprot
contains VM_PROT_WRITE (i.e. fd opened r/w with no write-seal applied) as
writable and thus blocking of any write-seal.
The memfd tests have been amended to reflect the fixes here, which notably
includes:
1. Fix for error return bug; EPERM is not a documented failure mode for mmap
2. Fix rejection of write-seal with active mappings that can be upgraded via
mprotect(2).
Re-organize the NFS file handle affinity code for the NFS server.
The file handle affinity code was configured to be used by both the
old and new NFS servers. This no longer makes sense, since there is
only one NFS server.
This patch copies a majority of the code in sys/nfs/nfs_fha.c and
sys/nfs/nfs_fha.h into sys/fs/nfsserver/nfs_fha_new.c and
sys/fs/nfsserver/nfs_fha_new.h, so that the files in sys/nfs can be
deleted. The code is simplified by deleting the function callback pointers
used to call functions in either the old or new NFS server and they were
replaced by calls to the functions.
As well as a cleanup, this re-organization simplifies the changes
required for handling of external page mbufs, which is required for KERN_TLS.
This patch should not result in a semantic change to file handle affinity.
Now that armv4/v5 is gone, remove the bits that implemented atomic operations
by disabling interrupts.
Those were specific to FreeBSD and never reached upstream.
lagg: stop double-counting output errors and counting drops as errors
Before this change, lagg double-counted errors from lagg members, and counted
every drop by a lagg member as an error. Eg, if lagg sent a packet, and the
underlying hardware driver dropped it, a counter would be incremented by both
lagg and the underlying driver.
This change attempts to fix that by incrementing lagg's counters only for
errors that do not come from underlying drivers.
Checks here against useracc are not useful and are racy.
copyin/copyout are sufficient to guard against bad addresses. They will return
EFAULT if the user is up to no good (by choice or ignorance). There's no point
in checking, since it doesn't even improve the error messages.
uart(4) backends currently detect RX FIFO overrun errors and report
them to the uart(4) core layer. They are then reported to the generic
TTY layer which promptly ignores them. As a result, there is
currently no good way to determine if a uart is experiencing RX FIFO
overruns. One could add a generic per-tty counter, but there did not
appear to be a good way to export those. Instead, add a sysctl under
the uart(4) sysctl tree to export the count of overruns.
Shifting right by 1 is not the same as dividing by 2 for signed
values. In particular, dividing a signed value by 2 gives the integer
ceiling of the (e.g. -5 / 2 == -2) whereas shifting right by 1 always
gives the floor (-5 >> 1 == -3).
An embedded board with a 25 Mhz base clock results in an error of
-30.5% when used with a baud rate of 115200. Using division, this
truncates to -30% and is permitted. Using the shift, this fails and
is rejected causing TIOCSETA requests to fail with EINVAL and breaking
getty(8).
Using division gives the same error range for both over and under baud
rates and also makes the code match the behavior documented in the
existing comment about supporting boards with 25 Mhz clocks.
cem [Mon, 13 Apr 2020 20:25:01 +0000 (20:25 +0000)]
Disable QUEUE_MACRO_DEBUG_TRACE in LINT kernels
It changes the size of TAILQ_ENTRY, which obviously impacts ABI in a variety of
ways. Some of these things are _Static_asserted. For now, mask the option
from LINT.
Fix sendto() on unconnected SOCK_STREAM/SEQPACKET unix sockets.
Previously the unpcb pointer of the newly connected remote socket was
not initialized correctly, so attempting to lock it would result in a
null pointer dereference.
Reported by: syzkaller
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Relax restrictions on private mappings of POSIX shm objects.
When creating a private mapping of a POSIX shared memory object,
VM_PROT_WRITE should always be included in maxprot regardless of
permissions on the underlying FD. Otherwise it is possible to open a
shm object read-only, map it with MAP_PRIVATE and PROT_WRITE, and
violate the invariant in vm_map_insert() that (prot & maxprot) == prot.
Reported by: syzkaller
Reviewed by: kevans, kib
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D24398
close_range/closefrom: fix regression from close_range introduction
close_range will clamp the range between [0, fdp->fd_lastfile], but failed
to take into account that fdp->fd_lastfile can become -1 if all fds are
closed. =-( In this scenario, just return because there's nothing further we
can do at the moment.
Add a test case for this, fork() and simply closefrom(0) twice in the child;
on the second invocation, fdp->fd_lastfile == -1 and will trigger a panic
before this change.
This had been introduced to ease any pain for using slightly older kernels
with a newer libc, e.g., for bisecting a kernel across the introduction of
shm_open2(2). 6 months has passed, retire the fallback and let shm_open()
unconditionally call shm_open2().
arc4random.c: In the incredibly unbelievable circumstance where
_rs_init() fails to allocate pages, don't call abort() because of
corefile data leakage concerns, but simply _exit(). The reasoning
is _rs_init() will only fail if someone finds a way to apply
specific pressure against this failure point, for the purpose of
leaking information into a core which they can read. We don't
need a corefile in this instance to debug that. So take this
"lever" away from whoever in the future wants to do that.
arc4random.3: reference random(4)
arc4random_uniform.c: include stdint.h over sys/types.h
Delete the mbuf macros that were used for the Mac OS/X port.
When the code was ported to Mac OS/X, mbuf handling functions were
converted to using the Mac OS/X accessor functions. For FreeBSD, they
are a simple set of macros in sys/fs/nfs/nfskpiport.h.
Since r359757, r359780, r359785, r359810, r359811 have removed all uses
of these macros, this patch deleted the macros from the .h files.
My eventual goal is deleting nfskpiport.h, but that will take some more
editting to replace uses of the remaining macros.
close_range(min, max, flags) allows for a range of descriptors to be
closed. The Python folk have indicated that they would much prefer this
interface to closefrom(2), as the case may be that they/someone have special
fds dup'd to higher in the range and they can't necessarily closefrom(min)
because they don't want to hit the upper range, but relocating them to lower
isn't necessarily feasible.
sys_closefrom has been rewritten to use kern_close_range() using ~0U to
indicate closing to the end of the range. This was chosen rather than
requiring callers of kern_close_range() to hold FILEDESC_SLOCK across the
call to kern_close_range for simplicity.
The flags argument of close_range(2) is currently unused, so any flags set
is currently EINVAL. It was added to the interface in Linux so that future
flags could be added for, e.g., "halt on first error" and things of this
nature.
This patch is based on a syscall of the same design that is expected to be
merged into Linux.
cem [Sun, 12 Apr 2020 18:04:20 +0000 (18:04 +0000)]
Add queue(2) debug macros as build options
Add QUEUE_MACRO_DEBUG_TRACE and QUEUE_MACRO_DEBUG_TRASH as proper kernel
options. While here, alpha-sort the debug section of sys/conf/options.
Enable QUEUE_MACRO_DEBUG_TRASH in amd64 GENERIC (but not GENERIC-NODEBUG)
kernels. It is similar in nature and cost to other use-after-free pointer
trashing we do in GENERIC. It is probably reasonable to enable in any arch
GENERIC kernel that defines INVARIANTS.
dim [Sun, 12 Apr 2020 16:06:59 +0000 (16:06 +0000)]
Merge commit 30588a739 from llvm git (by Erich Keane):
Make target features check work with ctor and dtor-
The problem was reported in PR45468, applying target features to an
always_inline constructor/destructor runs afoul of GlobalDecl
construction assert when checking for target-feature compatibility.
The core problem is fixed by using the version of the check that
takes a FunctionDecl rather than the GlobalDecl. However, while
writing the test, I discovered that source locations weren't properly
set for this check on ctors/dtors. This patch also fixes constructors
and CALLED destructors.
Unfortunately, it doesn't seem too possible to get a meaningful
source location for a 'cleanup' destructor, so those are still
'frontend' level errors unfortunately. A fixme was added to the test
to cover that situation.
This should fix 'Assertion failed: (!isa<CXXConstructorDecl>(D) && "Use
other ctor with ctor decls!"), function Init, file
/usr/src/contrib/llvm-project/clang/include/clang/AST/GlobalDecl.h, line
45' when compiling the security/botan2 port.
This is the foundational change for the routing subsytem rearchitecture.
More details and goals are available in https://reviews.freebsd.org/D24141 .
This patch introduces concept of nexthop objects and new nexthop-based
routing KPI.
Nexthops are objects, containing all necessary information for performing
the packet output decision. Output interface, mtu, flags, gw address goes
there. For most of the cases, these objects will serve the same role as
the struct rtentry is currently serving.
Typically there will be low tens of such objects for the router even with
multiple BGP full-views, as these objects will be shared between routing
entries. This allows to store more information in the nexthop.
These 2 function are intended to replace all all flavours of
<in_|in6_>rtalloc[1]<_ign><_fib>, mpath functions and the previous
fib[46]-generation functions.
Upon successful lookup, they return nexthop object which is guaranteed to
exist within current NET_EPOCH. If longer lifetime is desired, one can
specify NHR_REF as a flag and get a referenced version of the nexthop.
Reference semantic closely resembles rtentry one, allowing sed-style conversion.
Additionally, another 2 functions are introduced to support uRPF functionality
inside variety of our firewalls. Their primary goal is to hide the multipath
implementation details inside the routing subsystem, greatly simplifying
firewalls implementation:
All functions have a separate scopeid argument, paving way to eliminating IPv6 scope
embedding and allowing to support IPv4 link-locals in the future.
Structure changes:
* rtentry gets new 'rt_nhop' pointer, slightly growing the overall size.
* rib_head gets new 'rnh_preadd' callback pointer, slightly growing overall sz.
Old KPI:
During the transition state old and new KPI will coexists. As there are another 4-5
decent-sized conversion patches, it will probably take a couple of weeks.
To support both KPIs, fields not required by the new KPI (most of rtentry) has to be
kept, resulting in the temporary size increase.
Once conversion is finished, rtentry will notably shrink.
More details:
* architectural overview: https://reviews.freebsd.org/D24141
* list of the next changes: https://reviews.freebsd.org/D24232
The intended change was
sp->next.tqe_next = NULL;
sp->next.tqe_prev = NULL;
which doesn't fix the issue I'm seeing and the committed fix is
not the intended fix due to copy-and-paste.
Thanks a lot to Conrad Meyer for making me aware of the problem.
Replace mbuf macros with the code they would generate in the NFS code.
When the code was ported to Mac OS/X, mbuf handling functions were
converted to using the Mac OS/X accessor functions. For FreeBSD, they
are a simple set of macros in sys/fs/nfs/nfskpiport.h.
Since porting to Mac OS/X is no longer a consideration, replacement of
these macros with the code generated by them makes the code more
readable.
When support for external page mbufs is added as needed by the KERN_TLS,
the patch becomes simpler if done without the macros.
This patch should not result in any semantic change.
This is the final patch of this series and the macros should now be
able to be deleted from the .h files in a future commit.
Replace mbuf macros with the code they would generate in the NFS code.
When the code was ported to Mac OS/X, mbuf handling functions were
converted to using the Mac OS/X accessor functions. For FreeBSD, they
are a simple set of macros in sys/fs/nfs/nfskpiport.h.
Since porting to Mac OS/X is no longer a consideration, replacement of
these macros with the code generated by them makes the code more
readable.
When support for external page mbufs is added as needed by the KERN_TLS,
the patch becomes simpler if done without the macros.
This patch should not result in any semantic change.
This commits add option 'M' which attempts to forcibly unmount the
dataset. Thanks to this we can enforce receiving snapshots in a
single step.
Note that this functionality is not supported on Linux because the
VFS will prevent active mounted filesystems from being unmounted,
even with the force option. This is the intended VFS behavior.
Discussed-with: Pawel Jakub Dawidek <pjd@FreeBSD.org> Reviewed-by: Ryan Moeller <ryan@iXsystems.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Allan Jude <allanjude@freebsd.org>
Differential Revision: https://reviews.freebsd.org/D22306
Tested on stable r359316 @ Sleep mode on custom hw, Power off on BBB and PB
OFF bit [1] in status register control the pmic behaviour when PWR_EN pin
is pulled low.
On most AM335x hardware [beaglebone *] the desired behaviour are in fact
power off due to some hardware designs - read more in the comments around
pmic in sys/gnu/dts/arm/am335x-bone-common.dtsi
This patch let the device-tree decide with ti,pmic-shutdown-controller[2]
the state of off bit in status register.
manu [Sat, 11 Apr 2020 15:25:40 +0000 (15:25 +0000)]
gpioctl: Print interrupts capabilities
GPIO drivers who supports interrupts report them in the caps
(obtain via the getcaps method) but gpioctl doesn't know
how to interpret this and print "UNKNOWN" for each one of them.
Even if we don't have userland gpio interrupts support for now
let gpioctl print the supported caps.
Split their functionality by moving random seed allocation
to SYSINIT and calling (new) generic multipath function from
standard IPv4/IPv5 RIB init handlers.
powerpc/booke: Use power-of-two mappings in 64-bit pmap_mapdev
Summary:
This reduces the precious TLB1 entry consumption (64 possible in
existing 64-bit cores), by adjusting the size and alignment of a device
mapping to a power of 2, to encompass the full mapping and its
surroundings.
One caveat with this: If a mapping really is smaller than a power of 2,
it's possible to get a machine check or hang if the 'missing' physical
space is accessed. In practice this should not be an issue for users,
as devices overwhelmingly have physical spaces on power-of-two sizes and
alignments, and any design that includes devices which don't follow this
can be addressed by undefining the POW2_MAPPINGS guard.
powerpc/booke: Add pte_find_next() to find the next in-use PTE
Summary:
Iterating over VM_MIN_ADDRESS->VM_MAXUSER_ADDRESS can take a very long
time iterating one page at a time (2**(log_2(SIZE)-12) operations),
yielding possibly several days or even weeks on 64-bit Book-E, even for
a largely empty, which can happen when swapping out a process by
vmdaemon. Speed this up by instead finding the next PTE at or equal to
the given VA.
powerpc/booke: Change Book-E 64-bit pmap to 4-level table
Summary:
The existing page table is fraught with errors, since it creates a hole
in the address space bits. Fix this by taking a cue from the POWER9
radix pmap, and make the page table 4 levels, 52 bits.
Inode check-hash errors were being reported after system crashes.
Trace the cause down to journalled soft updates recovery code in
fsck failing to recompute the check-hash after updating an inode.
As inode check-hash was first introduced to UFS in FreeBSD 13,
there is no need to MFC this commit.
Replace mbuf macros with the code they would generate in the NFS code.
When the code was ported to Mac OS/X, mbuf handling functions were
converted to using the Mac OS/X accessor functions. For FreeBSD, they
are a simple set of macros in sys/fs/nfs/nfskpiport.h.
Since porting to Mac OS/X is no longer a consideration, replacement of
these macros with the code generated by them makes the code more
readable.
When support for external page mbufs is added as needed by the KERN_TLS,
the patch becomes simpler if done without the macros.
This patch should not result in any semantic change.
This conversion will be committed one file at a time.
A T6 adapter contains two crypto engines on separate channels. This
commit distributes sessions between the two engines. Previously, only
the first engine was used.
Replace mbuf macros with the code they would generate in the NFS code.
When the code was ported to Mac OS/X, mbuf handling functions were
converted to using the Mac OS/X accessor functions. For FreeBSD, they
are a simple set of macros in sys/fs/nfs/nfskpiport.h.
Since porting to Mac OS/X is no longer a consideration, replacement of
these macros with the code generated by them makes the code more
readable.
When support for external page mbufs is added as needed by the KERN_TLS,
the patch becomes simpler if done without the macros.
This patch should not result in any semantic change.
This conversion will be committed one file at a time.
sbappendcontrol() needs to avoid clearing M_NOTREADY on data mbufs.
If LOCAL_CREDS is set on a unix socket and sendfile() is called,
sendfile will call uipc_send(PRUS_NOTREADY), prepending a control
message to the M_NOTREADY mbufs. uipc_send() then calls
sbappendcontrol() instead of sbappend(), and sbappendcontrol() would
erroneously clear M_NOTREADY.
Pass send flags to sbappendcontrol(), like we do for sbappend(), to
preserve M_READY when necessary.
Reported by: syzkaller
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D24333
Properly handle disconnected sockets in uipc_ready().
When transmitting over a unix socket, data is placed directly into the
receiving socket's receive buffer, instead of the transmitting socket's
send buffer. This means that when pru_ready is called during
sendfile(), the passed socket does not contain M_NOTREADY mbufs in its
buffers; uipc_ready() must locate the linked socket.
Currently uipc_ready() frees the mbufs if the socket is disconnected,
but this is wrong since the mbufs may still be present in the receiving
socket's buffer after a disconnect. This can result in a use-after-free
and potentially a double free if the receive buffer is flushed after
uipc_ready() frees the mbufs.
Fix the problem by trying harder to locate the correct socket buffer and
calling sbready(): use the global list of SOCK_STREAM unix sockets to
search for a sockbuf containing the now-ready mbufs. Only free the
mbufs if we fail this search.
Reviewed by: jah, kib
Reported and tested by: pho
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D24332
There are several reports of "hdac0: Command timeout on address 2"
messages emitted during playback on a variety of contemporary machines.
Show the command that timed out in case it might provide a clue in
finding the cause.
PR: 229190
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
The comment previously stated the delay must be at least 250us but that
was insufficient and so should be doubled, but the delay was actually
1000. The HDA spec actually says the delay must be 521 us (25 frames)
so update the comment to match.
Split rtrequest1_fib() into smaller manageable chunks.
No functional changes.
* Move route addition / route deletion code from rtrequest1_fib()
to add_route() and del_route() respectively.
* Rename rtrequest1_fib_change() to change_route() for consistency.
* Shrink the scope of ugly info #defines.
userland build: replace -fno-common with ${CFCOMMONFLAG}
This change allows any downstream or otherwise consumer to easily override
the new -fno-common default on a temporary basis without having to hack into
src.sys.mk, and also makes it a bit easier to search for these specific
cases where -fno-common must be overridden with -fcommon or else the build
will fail.
The gdb build, the only program requiring -fcommon on head/, is switched
over as an example usage. It will need it on all branches, so this does not
harm future mergability.
Forced rw unmounts and remounts from rw to ro already suspend
filesystem, which closes races with writers instantiating new vnodes
while unmount flushes the queue. Original intent of not including
non-forced unmounts into this regime was to allow such unmounts to
fail if writer was active, but this did not worked well.
Similar change, but causing all unmount, even involving only ro
filesystem, were proposed in D24088, but I believe that suspending ro
is undesirable, and definitely spends CPU time.
Reported by: markj
Discussed with: chs, mckusick
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Fixing the soft update macros in -r359612 triggered a previously
hidden bug in the file truncation code. Until that bug is tracked
down and fixed, revert to the old behavior.
Reported by: Peter Holm
Reviewed by: kib, Chuck Silvers
When running with a kernel compiled with DEBUG_LOCKS, before
panic'ing for recusing on a non-recursive lock, print out the
kernel stack where the lock was originally acquired.
This is an application of the kernel overflow fix from r357948 to
userspace, based on the algorithm developed by Bruce Evans. To keep
the ABI of the vds_timekeep stable, instead of adding the large_delta
member, MSB of both multipliers are added to quickly estimate the overflow.
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Replace mbuf macros with the code they would generate in the NFS code.
When the code was ported to Mac OS/X, mbuf handling functions were
converted to using the Mac OS/X accessor functions. For FreeBSD, they
are a simple set of macros in sys/fs/nfs/nfskpiport.h.
Since porting to Mac OS/X is no longer a consideration, replacement of
these macros with the code generated by them makes the code more
readable.
When support for external page mbufs is added as needed by the KERN_TLS,
the patch becomes simpler if done without the macros.
This patch should not result in any semantic change.
This conversion will be committed one file at a time.
Remove the old NFS lock device driver that uses Giant.
This NFS lock device driver was replaced by the kernel NLM around FreeBSD7 and
has not normally been used since then.
To use it, the kernel had to be built without "options NFSLOCKD" and
the nfslockd.ko had to be deleted as well.
Since it uses Giant and is no longer used, this patch removes it.
With this device driver removed, there is now a lot of unused code
in the userland rpc.lockd. That will be removed on a future commit.