Since Linux 4.5, sendmsg(2) fails with ETOOMANYREFS if the number of
"in-flight" IOs, which has been sent by sendmsg(2) but has not yet
accepted by recvmsg(2), exceeds the RLIMIT_NOFILE resource limit.
https://rubyci.s3.amazonaws.com/arm64-neoverse-n1/ruby-master/log/20231025T090004Z.fail.html.gz
```
1) Error:
TestSocket_UNIXSocket#test_fd_passing_race_condition:
Errno::ETOOMANYREFS: Too many references: cannot splice - sendmsg(2)
```
This change reduces the number of times of IO passing under 1024,
which is a default limit in many environments.
OpenBSD and Solaris behave differently here.
Linux does deliver the empty packet, which is questionable
as it's undistinguishable from a closed connection.
It seems that OpenBSD and Solaris simply drop it.
We could test the platform before doing the assertion, but
it would likely be fragile, and the entire web recommend
to not ever send an empty packet, so the value of this
assertion is low.
[Bug #19012]
man recvmsg(2) states:
> Return Value
> These calls return the number of bytes received, or -1 if an error occurred.
> The return value will be 0 when the peer has performed an orderly shutdown.
Not too sure how one is supposed to make the difference between a packet of
size 0 and a closed connection.
* Windows: Fix warning about undefined if_indextoname()
* Windows: Fix UNIXSocket on MINGW and make .pair more reliable
* Windows: Use nonblock=true for read tests with scheduler
* Windows: Move socket detection from File.socket? to File.stat
Add S_IFSOCK to Windows and interpret reparse points accordingly.
Enable tests that work now.
* Windows: Use wide-char functions to UNIXSocket
This fixes behaviour with non-ASCII characters.
It also fixes deletion of temporary UNIXSocket.pair files.
* Windows: Add UNIXSocket tests for specifics of Windows impl.
* Windows: fix VC build due to missing _snwprintf
Avoid usage of _snwprintf, since it fails linking ruby.dll like so:
linking shared-library x64-vcruntime140-ruby320.dll
x64-vcruntime140-ruby320.def : error LNK2001: unresolved external symbol snwprintf
x64-vcruntime140-ruby320.def : error LNK2001: unresolved external symbol vsnwprintf_l
whereas linking miniruby.exe succeeds.
This patch uses snprintf on the UTF-8 string instead.
Also remove branch GetWindowsDirectoryW, since it doesn't work.
* Windows: Fix dangling symlink test failures
Co-authored-by: Lars Kanis <kanis@comcard.de>
Receiving UnixSocket works fine if you don't provide a mode, and
I think it is reasonable to expect that you should not provide
a mode if klass.for_fd would not accept a mode.
Fixes [Bug #11778]
* ext/socket/unixsocket.c (unix_send_io): document args
(unix_recv_io): ditto
* test/socket/test_unix.rb (test_fd_passing_class_mode): added
I was working on these when I encountered the problem in
with BasicSocket.for_fd not handling mode args:
https://bugs.ruby-lang.org/issues/11778
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@52922 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Not sure what drugs I was on, but blindly sleeping instead of
using IO#wait or IO.select to wait for data on a socket is
completely wrong.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@50951 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This reduces GC overhead and makes the API more consistent
with IO#read and IO#read_nonblock.
* ext/socket/basicsocket.c (bsock_recv): document outbuf
* ext/socket/unixsocket.c (unix_recvfrom): ditto
* ext/socket/init.c (rsock_strbuf, recvfrom_locktmp): new functions
(rsock_s_recvfrom): support destination buffer as 3rd arg
(rsock_s_recvfrom_nonblock): ditto
* string.c (rb_str_locktmp_ensure): export for internal ext
* test/socket/test_nonblock.rb: test recv_nonblock
* test/socket/test_unix.rb: test recv
[ruby-core:69543] [Feature #11242]
Benchmark results:
user system total real
alloc 0.130000 0.280000 0.410000 ( 0.420656)
extbuf 0.100000 0.220000 0.320000 ( 0.318708)
-------------------8<--------------------
require 'socket'
require 'benchmark'
nr = 100000
msg = ' ' * 16384
size = msg.bytesize
buf = ' ' * size
UNIXSocket.pair(:DGRAM) do |a, b|
Benchmark.bmbm do |x|
x.report('alloc') do
nr.times do
b.send(msg, 0)
a.recv(size, 0)
end
end
x.report('extbuf') do
nr.times do
b.send(msg, 0)
a.recv(size, 0, buf)
end
end
end
end
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@50912 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This is analogous to functionality found in IO#read_nonblock and
IO#wait_nonblock. Raising exceptions for common failures on
non-blocking servers is expensive and makes $DEBUG too noisy.
Benchmark results:
user system total real
default 2.790000 0.870000 3.660000 ( 3.671597)
exception: false 1.120000 0.800000 1.920000 ( 1.922032)
exception: false (cached arg) 0.820000 0.770000 1.590000 ( 1.589267)
--------------------- benchmark script ------------------------
require 'socket'
require 'benchmark'
require 'tmpdir'
nr = 1000000
Dir.mktmpdir('nb_bench') do |path|
sock_path = "#{path}/test.sock"
s = UNIXServer.new(sock_path)
Benchmark.bmbm do |x|
x.report("default") do
nr.times do
begin
s.accept_nonblock
rescue IO::WaitReadable
end
end
end
x.report("exception: false") do
nr.times do
begin
s.accept_nonblock(exception: false)
rescue IO::WaitReadable
abort "should not raise"
end
end
end
x.report("exception: false (cached arg)") do
arg = { exception: false }
nr.times do
begin
s.accept_nonblock(arg)
rescue IO::WaitReadable
abort "should not raise"
end
end
end
end
end
* ext/socket/init.c (rsock_s_accept_nonblock):
support exception: false
[ruby-core:66385] [Feature #10532]
* ext/socket/init.c (rsock_init_socket_init): define new symbols
* ext/socket/rubysocket.h: adjust prototype
* ext/socket/socket.c (sock_accept_nonblock): support exception: false
* ext/openssl/ossl_ssl.c (ossl_ssl_accept_nonblock): ditto
* ext/socket/socket.c (Init_socket): adjust accept_nonblock definition
* ext/openssl/ossl_ssl.c (Init_ossl_ssl): ditto
* ext/socket/tcpserver.c (rsock_init_tcpserver): ditto
* ext/socket/unixserver.c (rsock_init_unixserver): ditto
* ext/socket/tcpserver.c (tcp_accept_nonblock): adjust
rsock_s_accept_nonblock call
* ext/socket/unixserver.c (unix_accept_nonblock): ditto
* ext/openssl/ossl_ssl.c (ossl_start_ssl): support no_exception
* ext/openssl/ossl_ssl.c (ossl_ssl_connect): adjust ossl_start_ssl call
* ext/openssl/ossl_ssl.c (ossl_ssl_connect_nonblock): ditto
* ext/openssl/ossl_ssl.c (ossl_ssl_accept): ditto
* test/socket/test_nonblock.rb (test_accept_nonblock): test for
"exception :false"
* test/socket/test_tcp.rb (test_accept_nonblock): new test
* test/socket/test_unix.rb (test_accept_nonblock): ditto
* test/openssl/test_pair.rb (test_accept_nonblock_no_exception): ditto
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@49948 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
after path length check.
This fixes a fd leak by TestSocket_UNIXSocket#test_too_long_path.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@46218 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ext/socket/ancdata.c (bsock_sendmsg_internal): only retry on error
(bsock_recvmsg_internal): ditto
* test/socket/test_unix.rb: test above for infinite loop
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@45066 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
subclasses that include WaitReadable or WaitWritable rather than
extending them into the exception object each time.
* error.c: Capture EGAIN, EWOULDBLOCK, EINPROGRESS exceptions and
export them for use in WaitReadable/Writable exceptions.
* io.c: Create versions of EAGAIN, EWOULDBLOCK, EINPROGRESS that
include WaitReadable and WaitWritable. Add rb_readwrite_sys_fail
for nonblocking failures using those exceptions. Use that
function in io_getpartial and io_write_nonblock instead of
rb_mod_sys_fail
* ext/openssl/ossl_ssl.c: Add new SSLError subclasses that include
WaitReadable and WaitWritable. Use those classes for
write_would_block and read_would_block instead of rb_mod_sys_fail.
* ext/socket/ancdata.c: Use rb_readwrite_sys_fail instead of
rb_mod_sys_fail in bsock_sendmsg_internal and
bsock_recvmsg_internal.
* ext/socket/init.c: Use rb_readwrite_sys_fail instead of
rb_mod_sys_fail in rsock_s_recvfrom_nonblock and
rsock_s_connect_nonblock.
* ext/socket/socket.c: Use rb_readwrite_sys_fail instead of
rb_mod_sys_fail in sock_connect_nonblock.
* include/ruby/ruby.h: Export rb_readwrite_sys_fail for use instead
of rb_mod_sys_fail. Introduce new constants RB_IO_WAIT_READABLE and
RB_IO_WAIT_WRITABLE for first arg to rb_readwrite_sys_fail.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@40195 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
sizeof(sa_familiy_t) if path is empty. see "Autobind Feature" in
unix(7) for details.
* ext/socket/lib/socket.rb (unix_socket_abstract_name?): treat an
empty path as an abstract name.
* test/socket/test_unix.rb: related test.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@38994 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
because rb_sys_fail_str() fails if its argument contains NUL.
* test/socket/test_unix.rb: related test.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@38991 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
correct address length of an abstract socket.
* test/socket/test_unix.rb: related test.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@38963 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
ext/socket/unixsocket.c (unixsock_connect_internal,
rsock_init_unixsock): calculate the correct address length of
an abstract socket. Without this fix, sizeof(struct sockaddr_un)
is specified as the length of an abstract socket for bind(2) or
connect(2), so the address of the socket is filled with extra NUL
characters. See unix(7) for details.
* ext/socket/lib/socket.rb (unix_server_socket): don't access the
file system if the platform is Linux and path starts with NUL,
which means that the socket is an abstract socket.
* test/socket/test_unix.rb: related test.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@38939 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
filename shorter for less possibility of Unix socket path over
107 bytes when TMPDIR has long path.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@35436 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
(unix_recv_io): ditto.
fixed by Eric Wong. [ruby-core:35574]
* test/socket/test_unix.rb: test added for above problem.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@32062 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
(make_io_for_rights): new function to allocate
IOs for FDs in SCM_RIGHTS message.
(bsock_recvmsg_internal): use make_io_for_rights. So the FDs can be
closed by GC.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@22426 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
