freebsd-dev/lib/libc/sys/sigfastblock.2
Konstantin Belousov 146fc63fce Add a way to manage thread signal mask using shared word, instead of syscall.
A new syscall sigfastblock(2) is added which registers a uint32_t
variable as containing the count of blocks for signal delivery.  Its
content is read by kernel on each syscall entry and on AST processing,
non-zero count of blocks is interpreted same as the signal mask
blocking all signals.

The biggest downside of the feature that I see is that memory
corruption that affects the registered fast sigblock location, would
cause quite strange application misbehavior. For instance, the process
would be immune to ^C (but killable by SIGKILL).

With consumers (rtld and libthr added), benchmarks do not show a
slow-down of the syscalls in micro-measurements, and macro benchmarks
like buildworld do not demonstrate a difference. Part of the reason is
that buildworld time is dominated by compiler, and clang already links
to libthr. On the other hand, small utilities typically used by shell
scripts have the total number of syscalls cut by half.

The syscall is not exported from the stable libc version namespace on
purpose.  It is intended to be used only by our C runtime
implementation internals.

Tested by:	pho
Disscussed with:	cem, emaste, jilles
Sponsored by:	The FreeBSD Foundation
Differential revision:	https://reviews.freebsd.org/D12773
2020-02-09 11:53:12 +00:00

167 lines
5.5 KiB
Groff

.\" Copyright (c) 2016 The FreeBSD Foundation, Inc.
.\"
.\" This documentation was written by
.\" Konstantin Belousov <kib@FreeBSD.org> under sponsorship
.\" from the FreeBSD Foundation.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\" notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\" notice, this list of conditions and the following disclaimer in the
.\" documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.\" $FreeBSD$
.\"
.Dd December 13, 2019
.Dt SIGFASTBLOCK 2
.Os
.Sh NAME
.Nm sigfastblock
.Nd controls signals blocking with a simple memory write
.Sh LIBRARY
.Lb libc
.Sh SYNOPSIS
.In sys/signalvar.h
.Ft int
.Fn sigfastblock "int cmd" "void *ptr"
.Sh DESCRIPTION
.Bf -symbolic
This function is not intended for a direct usage by applications.
The functionality is provided for implementing some optimizations in
.Xr ld-elf.so.1 8
and
.Lb libthr .
.Ef
.Pp
The function configures the kernel facility that allows a thread to
block asynchronous signals delivery with a single write to userspace
memory, avoiding overhead of system calls like
.Xr sigprocmask 2
for establishing critical sections.
The C runtime uses it to optimize implementation of async-signal-safe
functionality.
.Pp
A thread might register a
.Dv sigblock
variable of type
.Vt int
as a location which is consulted by kernel when calculating the
blocked signal mask for delivery of asynchronous signals.
If the variable indicates that blocking is requested, then the kernel
effectively operates as if the mask containing all blockable signals was
supplied to
.Xr sigprocmask 2 .
.Pp
The variable is supposed to be modified only from the owning thread,
there is no way to guarantee visibility of update from other thread
to kernel when signals are delivered.
.Pp
Lower bits of the sigblock variable are reserved as flags,
which might be set or cleared by kernel at arbitrary moments.
Userspace code should use
.Xr atomic 9
operations of incrementing and decrementing by
.Dv SIGFASTBLOCK_INC
quantity to recursively block or unblock signals delivery.
.Pp
If a signal would be delivered when unmasked, kernel might set the
.Dv SIGFASTBLOCK_PEND
.Dq pending signal
flag in the sigblock variable.
Userspace should perform
.Dv SIGFASTBLOCK_UNBLOCK
operation when clearing the variable if it notes the pending signal
bit is set, which would deliver the pending signals immediately.
Otherwise, signals delivery might be postponed.
.Pp
The
.Fa cmd
argument specifies one of the following operations:
.Bl -tag -width SIGFASTBLOCK_UNSETPTR
.It Dv SIGFASTBLOCK_SETPTR
Register the variable of type
.Vt int
at location pointed to by the
.Fa ptr
argument as sigblock variable for the calling thread.
.It Dv SIGFASTBLOCK_UNSETPTR
Unregister the currently registered sigblock location.
Kernel stops inferring the blocked mask from non-zero value of its
blocked count.
New location can be registered after previous one is deregistered.
.It Dv SIGFASTBLOCK_UNBLOCK
If there are pending signals which should be delivered to the calling
thread, they are delivered before returning from the call.
The sigblock variable should have zero blocking count, and indicate
that the pending signal exists.
Effectively this means that the variable should have the value
.Dv SIGFASTBLOCK_PEND .
.El
.Sh RETURN VALUES
.Rv -std
.Sh ERRORS
The operation may fail with the following errors:
.Bl -tag -width Er
.It Bq Er EBUSY
The
.Dv SIGFASTBLOCK_SETPTR
attempted while the sigblock address was already registered.
The
.Dv SIGFASTBLOCK_UNBLOCK
was called while sigblock variable value is not equal to
.Dv SIGFASTBLOCK_PEND .
.It Bq Er EINVAL
The variable address passed to
.Dv SIGFASTBLOCK_SETPTR
is not aligned naturally.
The
.Dv SIGFASTBLOCK_UNSETPTR
operation was attempted without prior successfull call to
.Dv SIGFASTBLOCK_SETPTR .
.It Bq Er EFAULT
Attempt to read or write to the sigblock variable failed.
Note that kernel generates the
.Dv SIGSEGV
signal if an attempt to read from the sigblock variable faulted
during implicit accesses from syscall entry.
.El
.Sh SEE ALSO
.Xr kill 2 ,
.Xr signal 2 ,
.Xr sigprocmask 2 ,
.Xr libthr 3 ,
.Xr ld-elf.so.1 8
.Sh STANDARDS
The
.Nm
function is non-standard, although a similar functionality is a common
optimization provided by several other systems.
.Sh HISTORY
The
.Nm
function was introduced in
.Fx 13.0 .
.Sh BUGS
The
.Nm
symbol is currently not exported by libc, on purpose.
Consumers should either use the
.Dv __sys_fast_sigblock
symbol from the private libc namespace, or utilize
.Xr syscall 2 .