Don't set SO_RCVBUF/SO_SNDBUF to fixed values, thus disabling autotuning.
Patch modeled after a patch suggestion from Daniel Havey <dhavey@gmail.com>
in https://cygwin.com/ml/cygwin-patches/2017-q1/msg00010.html:
At Windows we love what you are doing with Cygwin. However, we have
been getting reports from our hardware vendors that iperf is slow on
Windows. Iperf is of course compiled against the cygwin1.dll and we
believe we have traced the problem down to the function fdsock in
net.cc. SO_RCVBUF and SO_SNDBUF are being manually set. The comments
indicate that the idea was to increase the buffer size, but, this code
must have been written long ago because Windows has used autotuning
for a very long time now. Please do not manually set SO_RCVBUF or
SO_SNDBUF as this will limit your internet speed.
I am providing a patch, an STC and my cygcheck -svr output. Hope we
can fix this. Please let me know if I can help further.
Simple Test Case:
I have a script that pings 4 times and then iperfs for 10 seconds to
debit.k-net.fr
With patch
$ bash buffer_test.sh 178.250.209.22
usage: bash buffer_test.sh <iperf server name>
Pinging 178.250.209.22 with 32 bytes of data:
Reply from 178.250.209.22: bytes=32 time=167ms TTL=34
Reply from 178.250.209.22: bytes=32 time=173ms TTL=34
Reply from 178.250.209.22: bytes=32 time=173ms TTL=34
Reply from 178.250.209.22: bytes=32 time=169ms TTL=34
Ping statistics for 178.250.209.22:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 167ms, Maximum = 173ms, Average = 170ms
------------------------------------------------------------
Client connecting to 178.250.209.22, TCP port 5001
TCP window size: 64.0 KByte (default)
------------------------------------------------------------
[ 3] local 10.137.196.108 port 58512 connected with 178.250.209.22 port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0- 1.0 sec 768 KBytes 6.29 Mbits/sec
[ 3] 1.0- 2.0 sec 9.25 MBytes 77.6 Mbits/sec
[ 3] 2.0- 3.0 sec 18.0 MBytes 151 Mbits/sec
[ 3] 3.0- 4.0 sec 18.0 MBytes 151 Mbits/sec
[ 3] 4.0- 5.0 sec 18.0 MBytes 151 Mbits/sec
[ 3] 5.0- 6.0 sec 18.0 MBytes 151 Mbits/sec
[ 3] 6.0- 7.0 sec 18.0 MBytes 151 Mbits/sec
[ 3] 7.0- 8.0 sec 18.0 MBytes 151 Mbits/sec
[ 3] 8.0- 9.0 sec 18.0 MBytes 151 Mbits/sec
[ 3] 9.0-10.0 sec 18.0 MBytes 151 Mbits/sec
[ 3] 0.0-10.0 sec 154 MBytes 129 Mbits/sec
Without patch:
dahavey@DMH-DESKTOP ~
$ bash buffer_test.sh 178.250.209.22
Pinging 178.250.209.22 with 32 bytes of data:
Reply from 178.250.209.22: bytes=32 time=168ms TTL=34
Reply from 178.250.209.22: bytes=32 time=167ms TTL=34
Reply from 178.250.209.22: bytes=32 time=170ms TTL=34
Reply from 178.250.209.22: bytes=32 time=169ms TTL=34
Ping statistics for 178.250.209.22:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 167ms, Maximum = 170ms, Average = 168ms
------------------------------------------------------------
Client connecting to 178.250.209.22, TCP port 5001
TCP window size: 208 KByte (default)
------------------------------------------------------------
[ 3] local 10.137.196.108 port 58443 connected with 178.250.209.22 port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0- 1.0 sec 512 KBytes 4.19 Mbits/sec
[ 3] 1.0- 2.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 2.0- 3.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 3.0- 4.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 4.0- 5.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 5.0- 6.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 6.0- 7.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 7.0- 8.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 8.0- 9.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 9.0-10.0 sec 1.50 MBytes 12.6 Mbits/sec
[ 3] 0.0-10.1 sec 14.1 MBytes 11.7 Mbits/sec
The output shows that the RTT from my machine to the iperf server is
similar in both cases (about 170ms) however with the patch the
throughput averages 129 Mbps while without the patch the throughput
only averages 11.7 Mbps. If we calculate the maximum throughput using
Bandwidth = Queue/RTT we get (212992 * 8)/0.170 = 10.0231 Mbps. This
is just about what iperf is showing us without the patch since the
buffer size is set to 212992 I believe that the buffer size is
limiting the throughput. With the patch we have no buffer limitation
(autotuning) and can develop the full potential bandwidth on the link.
If you want to duplicate the STC you will have to find an iperf server
(I found an extreme case) that has a large enough RTT distance from
you and try a few times. I get varying results depending on Internet
traffic but without the patch never exceed the limit caused by the
buffering.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
GDB since commit 24cdb46e [1] can report and use these names.
Add utility function SetThreadName(), which sends a thread name to the
debugger.
Use that:
- to set the default thread name for main thread and newly created pthreads.
- in pthread_setname_np() for user thread names.
- for helper thread names in cygthread::create()
- for helper threads which are created directly with CreateThread.
Note that there can still be anonymous threads, created by system or
injected DLLs.
[1] https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=24cdb46e9f0a694b4fbc11085e094857f08c0419
Revamp device parsing code. Introducing support for more partitions
into the shilka-generated parser has the unfortunate side-effect of
raising the size of the DLL by almost 2 Megs. Therefore we split out
the handling for /dev/sdXY devices into a tiny bit of hand-written
code.
While at it, remove some unused cruft from devices.* and generally
clean up the device class to provide access methods instead of direct
access to members.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Bump GPLv2+ to GPLv3+ for some files, clarify BSD 2-clause.
Everything else stays under GPLv3+.
New Linking Exception exempts resulting executables from LGPLv3 section 4.
Add CONTRIBUTORS file to keep track of licensing.
Remove 'Copyright Red Hat Inc' comments.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Don't use u_char, u_short, u_int or u_long in Cygwin, unless it refers
to the Winsock types. Use u_intN_t in BSD-based sources, unsigned char
where strings are concerned, uintN_t otherwise. Also:
* net.cc: Fix comment, we're not using u_long anymore.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Add experimental code to workaround the issue described in the thread
starting at
https://cygwin.com/ml/cygwin/2015-07/msg00350.html
There's a hint in https://communities.vmware.com/message/2577858#2577858
that this problem is related to using the AI_ALL flag.
This patch checks if GetAddrInfoW returned with WSANO_RECOVERY and if
the AI_ALL flag was set, it retries GetAddrInfo without the AI_ALL flag.
* net.cc (cygwin_getaddrinfo): Add experimental code to retry
GetAddrInfoW without AI_ALL flag if it returned with WSANO_RECOVERY.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
According to https://cygwin.com/ml/cygwin/2016-03/msg00124.html it's a
problem to collect friendlyname info using AF_INET6 sockets. Fix problem
by exposing additional hardware info for all collected interfaces via the
pointer in the ifaddrs::ifa_data member.
* include/ifaddrs.h (struct ifaddrs_hwdata): Define as struct of
not yet exposed members of struct ifall, defined in net.cc.
* net.cc (struct ifall): Replace hardware dta members with struct
ifaddrs_hwdata. Accommodate throughout.
(get_ifs): Let ifaddrs ifa_data member point to ifall::ifa_hwdata
member.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
* strfuncs.cc (sys_cp_wcstombs): Always return number of multibytes
without trailing NUL as the documentation implies. Throughout Cygwin,
fix usage to align to this pattern.
* fhandler_process.cc (format_process_winexename): Drop trailing NUL
and LF from output.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
initial handshake on AF_LOCAL sockets.
* fhandler.h (class fhandler_socket): Add no_getpeereid status flag.
(fhandler_socket::af_local_set_no_getpeereid): New prototype.
* fhandler_socket.cc (fhandler_socket::af_local_connect): Skip handshake
if no_getpeereid is set. Add debug output.
(fhandler_socket::af_local_accept): Likewise.
(fhandler_socket::af_local_set_no_getpeereid): New function.
(fhandler_socket::af_local_copy): Copy no_getpeereid.
(fhandler_socket::getpeereid): Fail if no_getpeereid is set.
* net.cc (cygwin_setsockopt): Add SO_PEERCRED for AF_LOCAL/SOCK_STREAM
sockets. Add comment to explain why we need it.
* include/cygwin/version.h (CYGWIN_VERSION_API_MINOR): Bump.
handler.
* cygtls.cc (_cygtls::remove): Accommodate the fact that pathbufs
has been moved from _local_storage to _cygtls.
* cygtls.h (class tls_pathbuf): Add comment to hint to gendef usage
of counters. Change type of counters to uint32_t for clarity.
Remove _cygtls as friend class.
(struct _local_storage): Move pathbufs from here...
(struct _cygtls): ...to here, allowing to access it from _sigbe.
(class san): Only define on 32 bit. Remove errno, _c_cnt and _w_cnt
members.
(san::setup): Drop parameter. Don't initialize removed members.
(san::leave): Don't set removed members.
(class myfault): Only define on 32 bit.
(myfault::faulted): Only keep implementation not taking any parameter.
Drop argument in call to sebastian.setup.
(__try/__leave/__except/__endtry): Implement to support real SEH. For
now stick to SJLJ on 32 bit.
* dcrt0.cc (dll_crt0_0): Drop 64 bit call to
exception::install_myfault_handler.
* exception.h (exception_handler): Define with EXCEPTION_DISPOSITION
as return type.
(PDISPATCHER_CONTEXT): Define as void * on 32 bit. Define as pointer
to _DISPATCHER_CONTEXT on 64 bit.
(class exception): Define separately for 32 and 64 bit.
(exception::myfault): Add handler for myfault SEH handling on 64 bit.
(exception::exception): Fix mangled method name to account for change
in type of last parameter.
(exception::install_myfault_handler): Remove.
* exceptions.cc (exception::myfault_handle): Remove.
(exception::myfault): New SEH handler for 64 bit.
* gendef (_sigbe): Set tls_pathbuf counters to 0 explicitely when
returning to the caller.
* ntdll.h: Move a comment to a better place.
(struct _SCOPE_TABLE): Define on 64 bit.
* thread.cc (verifyable_object_isvalid): Remove gcc 4.7 workaround.
* tls_pbuf.cc (tls_pbuf): Fix to accommodate new place of pathbufs.
(tls_pathbuf::destroy): Change type of loop variables to uint32_t.
* tls_pbuf.h (class tmp_pathbuf): Change type of buffer counters to
uint32_t. Accommodate new place of pathbufs.
* tlsoffsets.h: Regenerate.
* tlsoffsets64.h: Regenerate.
from GetAdaptersAddresses (CID 60218).
(get_ifs): Add missing braces in AF_INET6 case which broke netmask
computation. Break out of loop if prefix gets <= 0 (CID 59939).
including the windows headers. Explain why.
(get_inet_addr): Convert ANY address to LOOPBACK address. Explain why.
(fhandler_socket::evaluate_events): Forcibly set SO_ERROR socket option
in case a connection attempt failed. Explain why.
(fhandler_socket::ioctl): Drop x86_64 re-definition of u_long here.
* fhandler_procnet.cc: On x86_64, define u_long as __ms_u_long before
including the windows headers. Explain why.
* net.cc: Ditto.
comment.
* mount.cc (mount_info::from_fstab): Use tmp_pathbuf rather than
stack for big local buffer.
* net.cc (cygwin_gethostname): Call GetComputerNameExA rather than
GetComputerNameA if gethostname failed.
* shared.cc (user_info::initialize): Fix formatting.
* include/sys/file.h: Define flock and accompanying macros if not
already defined in sys/_default_fcntl.h.
included by default.
* winlean.h: Add long comment to explain why we have to define certain
symbols.
(_NORMALIZE_): Define.
(_WINNLS_): Drop definition and subsequent undef.
(_WINNETWK_): Ditto.
(_WINSVC_): Ditto.
2013-11-23 Eric Blake <eblake@redhat.com>
hints is NULL. Add a comment and set whints.ai_family explicitely to
AF_UNSPEC to follow glibc man page closely. Check flags against whints
for locality.
(IdnToUnicode): Define.
(FreeAddrInfoW): Define.
(GetAddrInfoW): Define.
(GetNameInfoW): Define.
* net.cc: Drop W. Richard STEVENS libgai implementation.
(cygwin_freeaddrinfo): Move code from ipv4_freeaddrinfo here. Move
definition up in file to avoid forward declaration in ga_duplist.
(ga_dup): Take addrinfoW as input. Add parameters to get IDN flags
and to set error value. Handle IDN flags and especially AI_CANONIDN.
Convert input wchar_t ai_canonname string to multibyte for output.
(ga_duplist): Add parameters to get IDN flags and to set error value
and propagate to ga_dup. Call cygwin_freeaddrinfo.
(gai_errmap): Add comments. Align error strings to GLibc. Add
EAI_IDN_ENCODE entry.
(get_ipv6_funcs): Remove.
(load_ipv6_guard): Remove.
(ipv6_inited): Remove.
(load_ipv6): Remove.
(load_ipv6_funcs): Remove.
(cygwin_getaddrinfo): Drop calling load_ipv6. Handle AI_IDN* flags.
Convert input strings to wchar_t and call GetAddrInfoW/FreeAddrInfoW.
In case hints is NULL, set default ai_flags explicitely to
AI_V4MAPPED | AI_ADDRCONFIG, as documented for Glibc. Only add AI_ALL
if AI_ADDRCONFIG is not given. Unconditionally add Windows-specific
AI_DISABLE_IDN_ENCODING to ai_flags to make IDN behaviour compatible
to Glibc even on Windows 8 and later.
(cygwin_getnameinfo): Drop calling load_ipv6. Handle NI_IDN* flags.
Call GetNameInfoW and convert returned strings from wchar_t to
multibyte.
* include/netdb.h: Add comments to describe flags and error values.
(AI_*): Define all flags using hex values for clearness.
(AI_IDN): Define.
(AI_CANONIDN): Define.
(AI_IDN_ALLOW_UNASSIGNED): Define.
(AI_IDN_USE_STD3_ASCII_RULES): Define.
(NI_*): Define all flags using hex values for clearness.
(NI_IDN): Define.
(NI_IDN_ALLOW_UNASSIGNED): Define.
(NI_IDN_USE_STD3_ASCII_RULES): Define.
(EAI_IDN_ENCODE): Define.
* devices.cc: Regenerate.
* dtable.h (struct dtable): Make fhandler_base friend, rather
than fhandler_disk_file.
* fhandler.cc (fhandler_base::open_with_arch): Create unique id.
(fhandler_base::cleanup): Call del_my_locks.
(fhandler_base::fcntl): Handle F_GETLK, F_SETLK and F_SETLKW.
* fhandler.h (fhandler_base::get_dev): Return real device number.
(fhandler_base::set_unique_id): New inline method.
(fhandler_disk_file::lock): Drop declaration.
(fhandler_disk_file::get_dev): New method, return pc.fs_serial_number.
(fhandler_dev_zero::open): Drop declaration.
* fhandler_disk_file.cc (fhandler_disk_file::close): Move
del_my_locks call to fhandler_base::open_with_arch.
(fhandler_disk_file::fcntl): Move handling of locking commands to
fhandler_base::fcntl.
(fhandler_base::open_fs): Drop call to NtAllocateLocallyUniqueId.
* fhandler_zero.cc (fhandler_dev_zero::open): Remove so that default
fhandler_base::open is used to open \Device\Null.
* flock.cc (fixup_lockf_after_exec): Finding a single fhandler is
enough here.
(fhandler_base::lock): Replace fhandler_disk_file::lock. Refuse to lock
nohandle devices. Handle read/write test using POSIX flags. Explain
why. Never fail on SEEK_CUR or SEEK_END, rather assume position 0,
just as Linux.
* net.cc (fdsock): Create unique id.
why -fomit-frame-pointer doesn't work right for passwd.o and path.o. Add
-static to link command line for cygwin0.dll.
* fhandler_disk_file.cc (fhandler_disk_file::facl): Reorganize slightly to
silence compiler warning when compiling with -fstack-check.
* net.cc (inet_ntop6): Initialize structure members to silence compiler warning
when compiling with -fstack-check.
* pseudo-reloc.cc (_pei386_runtime_relocator): Make this a C function. Detect
NULL u.
* winsup.h (_pei386_runtime_relocator): Declare this as extern "C".
* lib/_cygwin_crt0_common.cc (_pei386_runtime_relocator): Call with NULL
argument.
* signal.cc (sigaction_worker): Eliminate last argument. Let callers report
their own strace info. Regparmize.
(sigaction): Reflect sigaction_worker changes.
(siginterrupt): Ditto.
* exceptions.cc: Update copyright.
systems. Change comment acordingly.
* wincap.h (wincaps::has_gaa_largeaddress_bug): New element.
* wincap.cc: Implement above element throughout.
(wincap_8): New globale wincaps to support Windows 8.
(wincapc::init): Take Windows 8 into account. Set new
has_gaa_largeaddress_bug member to false on 32 bit systems.
Corinna Vinschen
Erik M. Bray
Jon Turney
Takashi Yano
Renato Silva
Alexey Pavlov
Christopher Faylor