On the BSDs, SO_REUSEPORT is pretty much SO_REUSEADDR with some special
casing for IP multicast. When two processes (that don't do multicast)
bind to the same address, only the last one receives traffic. It allows
one to "steal" the bound address from another process. (Both processes
have to enable SO_REUSEPORT though, so it only works in a cooperative
setting.)
On Linux however, it enables port sharing, not stealing - both processes
receive a share of the traffic. This is a desirable trait but pre-3.9
kernels don't support the socket option and a libuv program therefore
behaves differently with older kernels or on another platform.
The difference in behavior (sharing vs. stealing) is, in my opinion,
big enough and confusing enough that it merits a rollback. People
that want this kind of functionality can prepare the socket manually
and hand it off to uv_udp_open().
This commit effectively reverts commit 17452cd.
* windows: Re-implement uv_fs_stat. The st_ctime field now contains the
change time, not the creation time, like on unix systems. st_dev,
st_ino, st_blocks and st_blksize are now also filled out. (Bert
Belder)
* linux: fix setsockopt(SO_REUSEPORT) error handling (Ben Noordhuis)
* windows: report uv_process_t exit code correctly (Bert Belder)
* windows: make uv_fs_chmod() report errors correctly (Bert Belder)
* windows: make some more NT apis available for libuv's internal use
(Bert Belder)
* windows: squelch some compiler warnings (Bert Belder)
This improves the output of uv_fs_stat:
* `st_ctime` now contains the change time, not the creation time.
* `st_ino` is now filled in with an fs-specific unique number.
* `st_dev` is set to the serial number of the containing file system.
* `st_blocks` now gets set.
* `st_blksize` is no longer zero, but set to a reasonable default.
This commit adds bootstrap code, function signatures, struct definitions
etc. to make it possible for libuv to use NtQueryVolumeInformationFile.
Most of this comes from the NT DDK, which libuv mirrors so that people
don't need the DDK to compile it.
Before this patch libuv would attempt to use GetLastError() to retrieve
the cause of NtQueryInformationFile failure, but that's not how it
should be done.
Linux as of 3.9 has a SO_REUSEPORT option that is similar but not
identical to its BSD counterpart.
On the BSDs, it turns on SO_REUSEADDR _and_ makes it possible to share
the address and port across processes.
On Linux, it "merely" enables fair load distribution - port sharing
still requires that you set SO_REUSEADDR.
Fair distribution is a desirable trait but not an essential one.
We don't know in advance whether the kernel actually supports
SO_REUSEPORT so don't treat EINVAL or ENOPROTOOPT as errors.
As an aside, on the BSDs we now omit the setsockopt(SO_REUSEADDR)
system call because it's implied by SO_REUSEPORT.
Fixes#870.
Changes since version 0.11.8:
* fsevents: share FSEventStream between multiple FS watchers, which
removes a limit on the maximum number of file watchers that can be
created on OS X. (Fedor Indutny)
* process: the `exit_status` parameter for a uv_process_t's exit
callback now is an int64_t, and no longer an int. (Bert Belder)
* process: make uv_spawn() return some types of errors immediately on
windows, instead of passing the error code the the exit callback. This
brings it on par with libuv's behavior on unix. (Bert Belder)
This means we no longer have to strip the high bit from the process exit
code on Windows, which is problematic because an unhandled SEH exception
can make a process exit with a status code that has the high bit set.
* It will now report some types of errors synchronously, to bring it on
par with uv-unix. Fixes#865.
* Failure to find the target executable will now set up stdio pipes
correctly, so the user can assume that when uv_spawn() returns 0 these
pipes will never trigger asserts in libuv's guts.
It seems that number of simultaneously opened FSEventStreams is
limited on OSX (i.e. you can have only fixed number of them on
one running system), getting past through this limit will cause
`FSEventStreamCreate` to return false and write following message
to stderr:
(CarbonCore.framework) FSEventStreamStart: register_with_server:
ERROR: f2d_register_rpc() => (null) (-21)
To prevent this, we must use only one shared FSEventStream with a
paths for all uv_fsevent_t handles, and then filter out events for
each handle using this paths again.
See https://github.com/joyent/node/issues/5463
Don't include <stdint.h>, it's not available when compiling with MSVC
2008 and we don't need it in the first place - just cast the argument
to `char *` rather than `uintptr_t`.
Fixes#893.
uv_ip6_addr() copies the address part to a temporary buffer when it
contains a link-local suffix ('<address>%<interface>'). Before this
commit, it didn't zero-terminate the address properly: it put the nul
byte at the end of the temporary buffer rather than at the end of the
address string, meaning the buffer looked like this:
<address> <garbage> '\0'
Fixes the following valgrind warning:
==16170== Conditional jump or move depends on uninitialised value(s)
==16170== at 0x43602C: inet_pton6 (inet.c:228)
==16170== by 0x435CE1: uv_inet_pton (inet.c:163)
==16170== by 0x436FD0: uv_ip6_addr (uv-common.c:175)
==16170== by 0x434717: test_ip6_addr_scope (test-ip6-addr.c:89)
==16170== by 0x43455B: call_iface_info_cb (test-ip6-addr.c:45)
==16170== by 0x43462B: foreach_ip6_interface (test-ip6-addr.c:59)
==16170== by 0x434791: run_test_ip6_addr_link_local (test-ip6-add
==16170== by 0x4061E8: run_test_part (runner.c:396)
==16170== by 0x404F4B: main (run-tests.c:58)
==16170==
==16170== Conditional jump or move depends on uninitialised value(s)
==16170== at 0x4C2AD8A: __GI_strchr (mc_replace_strmem.c:224)
==16170== by 0x435ECB: inet_pton6 (inet.c:231)
==16170== by 0x435CE1: uv_inet_pton (inet.c:163)
==16170== by 0x436FD0: uv_ip6_addr (uv-common.c:175)
==16170== by 0x434717: test_ip6_addr_scope (test-ip6-addr.c:89)
==16170== by 0x43455B: call_iface_info_cb (test-ip6-addr.c:45)
==16170== by 0x43462B: foreach_ip6_interface (test-ip6-addr.c:59)
==16170== by 0x434791: run_test_ip6_addr_link_local (test-ip6-add
==16170== by 0x4061E8: run_test_part (runner.c:396)
==16170== by 0x404F4B: main (run-tests.c:58)
Fixes#890.
Include <errno.h> and <stdlib.h>. They're being pulled in implicitly
on OS X but apparently that's not the case with iOS builds.
Build breakage introduced in 5ff6f85 by yours truly. Mea culpa.
Fixes#885 and #891.
Before this commit, libuv would abort() if waitpid() failed with EINTR.
It's unlikely that anyone actually hit this error condition: the major
UNIX platforms - with the possible exception of Solaris - don't return
EINTR when the WNOHANG flag is specified, as libuv does.
However, POSIX allows for an implementation to do whatever here: unless
explicitly forbidden, it's allowed and POSIX doesn't restrict
implementers in this particular area.
Let's opt for robustness and handle EINTR.
The fsevents watcher thread only needs a minimal amount of stack space
in order to run. Restrict its stack size so there's more address space
left for the libuv user.
* Move CF run loop code to fsevents.c.
* Create the fsevents thread on demand rather than at startup.
* Remove use of ACCESS_ONCE. All accesses to loop->cf_loop are
protected by full memory barriers so no reordering can take place.
Fixes#872.
Fixes the following warning:
src\uv-common.c(476): warning C4715:
'uv__getaddrinfo_translate_error' : not all control paths return
a value
The function never returns - the final statement is a call to abort() -
but it seems MSVC's program flow analyzer is too weak to figure that
out.
mingw32 doesn't have the JOB_OBJECT_XXX defines. Provide them ourselves
when they are not already defined.
This is possibly only an issue with older versions of mingw32 because
mingw-w64 2.0 _does_ export the defines.
Change the uv_fs_write() prototype so the 'buf' argument is now
`const void*` rather than `void*`.
The argument is stored in a non-const field in the uv_fs_t but that's
inconsequential because the memory it points to is not touched.
