Revisit the fix from commit b705b53. The problem with using sigset_t
and _NSIG is that the size of sigset_t and the value of _NSIG depend
on what headers libuv picks up first, <signal.h> or <asm/signal.h>.
With the former, sizeof(sigset_t) = 128; with the latter, it's 8.
Simply sidestep the issue by calculating the signal mask as a 64 bits
integer, without using sigset_t or _NSIG.
This is a partial cherry-pick of commit 751ac48 from the v1.x branch.
Original PR-URL: https://github.com/libuv/libuv/pull/83
PR-URL: https://github.com/libuv/libuv/pull/84
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
Add a per-event loop flag for blocking SIGPROF signals when polling for
events.
The motivation for this addition is to reduce the number of wakeups and
subsequent clock_gettime() system calls when using a sampling profiler.
On Linux, this switches from epoll_wait() to epoll_pwait() when enabled.
Other platforms bracket the poll syscall with pthread_sigmask() calls.
Refs strongloop/strong-agent#3 and strongloop-internal/scrum-cs#37.
PR-URL: https://github.com/libuv/libuv/pull/15
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
linux-syscalls.h assumes that on all Linux platforms the value
of O_NONBLOCK is the same as SOCK_NONBLOCK.
This commit fixes it, as it is at least not true for hppa, and
resolves#1442.
Signed-off-by: Luca Bruno <lucab@debian.org>
sizeof(sigset_t) = 128 whereas the kernel expects 8, the size of a long.
It made the system call fail with EINVAL when a non-NULL sigset was
passed in. Fortunately, it's academic because there is just one call
site and it passes in NULL.
Fixeslibuv/libuv#4.
It seems that epoll_wait is implemented in glibc in terms of epoll_pwait and
new architectures (like arm64) do not implement the epoll_wait syscall at all.
So if epoll_wait errors with ENOSYS, just call epoll_pwait.
This is a backport of 861de3d71d for v0.10
branch.
`fd_set`s are way too small for `select()` hack when stream's fd is
bigger than 1023. Make `fd_set`s a part of `uv__stream_select_t`
structure.
fix#1461
Conflicts:
src/unix/stream.c
According to @aktau, the call to `mach_timebase_info` costs 90% of the
total execution time of `uv_hrtime()`. The result of the call is static
on all existing platforms, so there is no need in invoking it multiple
times.
Signed-off-by: Fedor Indutny <fedor@indutny.com>
Invalidate file descriptor when closing `uv_fs_event_t` handle. Note
that `uv__io_stop` is just removing `fd` from `loop->watchers` and not
actually invalidating all consequent events in a `kevent()` results.
fixjoyent/node#1101
If the same file description is open in two different processes, then
closing the file descriptor is not sufficient to deregister it from the
epoll instance (as described in epoll(7)), resulting in spurious events
that cause the event loop to spin repeatedly. So always explicitly
deregister it.
Fixes#1099.
Conflicts:
test/test-spawn.c
Changed the order of the member assignments since the thread
may start before the parameters are assigned. This especially
happens when the osx scheduler is very busy.
Thus allow passing the same file descriptor as the source of multiple
stdios.
Committed with the help and code parts from:
* Sam Roberts <sam@strongloop.com>
fix#1074
CPU entries in /proc/stat are not guaranteed to be monotonically
increasing, asserting on this assumption can break in cases such
as the UltraSparc II machine shown in #1080.
Signed-off-by: Luca Bruno <lucab@debian.org>
In our build infrastructure, I've seen a lot of segfaults recently that
were all only happening on OSX. Upon inspecting the coredumps, it
appearded that all segfaults happened at the same instruction, and upon
translating the assembly back to the source, I found that an array could
be indexed with a -1 index before the index was checked to be not -1.
As concrete evidence, here is the situation that I found caused the
segfault. The instruction in question along with the relevant register
values was:
mov (%r8,%r15,8),%r12
r8 = 0x7fb0ba800000
r15 = 0xffffffffffffffff
r8 + r15 * 8 == 0x7fb0ba7ffff8
It appears that the base of loop->watchers was page aligned, and by
going back one word I guess that the page wasn't mapped, causing our
segfaults.
Some scandir implementations allocate the dirent struct even if the
directory is empty, so if `scandir` returns 0 there may still be memory
that needs to get deallocated. I have altered uv__fs_readdir to go to
the "deallocation exit area" when 0 files are found in the directory
and continue to return early on a return value of -1.
I went to add a test for this functionality, but it appears that one
already exists (reading an empty directory), so I imagine that the
valgrind builds must only be happening on linux instead of OSX as well?
I have confirmed manually that a program without this fix will
infinitely leak memory, and with this fix the memory usage stays
constant.
Commit 3d2c820 back-ports a patch from the master branch that disables
the use of SO_REUSEPORT on Linux for reasons mentioned in the commit
log.
Unfortunately, the back-port was incomplete; another setsockopt() call
site in src/unix/udp.c was overlooked. This commit rectifies that.
Hat tip to Luca Bruno for helping troubleshoot the issue.
When fd is closed and new one (with the same number) is opened inside
kqueue/epoll/port loop's callback - stale events might invoke callbacks
on wrong watchers.
Check if watcher was changed after invocation and invalidate all events
with the same fd.
fix#826
Work around an epoll quirk where it sometimes reports just the EPOLLERR
or EPOLLHUP event. In order to force the event loop to move forward,
we merge in the read/write events that the watcher is interested in;
uv__read() and uv__write() will then deal with the error or hangup in
the usual fashion.
Fixes#982.
This is a back-port of commit 24bfef2 from the master branch.
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.
This is a back-port of commit 9d60f1e from the master branch.
Fixesjoyent/node#6454.
Watchers could be stopped between two `kevent()`/`epoll_wait()` calls
(which may happen in the same loop in `uv__io_poll()`), in such cases
`watcher->events` could be stale and won't be updated to
`watcher->pevents`.
Try to use and rely on `watcher->pevents` instead of blindly expecting
`watcher->events` to be always correct.
This commit reverts the following commits:
983fa68 darwin: fix 10.6 build error in fsevents.c
684e212 fsevents: use shared FSEventStream
ea4cb77 fsevents: FSEvents is most likely not thread-safe
9bae606 darwin: create fsevents thread on demand
Several people have reported stability issues on OS X 10.8 and bus
errors on the 10.9 developer preview.
See also joyent/node#6296 and joyent/node#6251.
The cleanup-after-error code path in uv_spawn() was closing file
descriptors indiscriminately. Only close file descriptors that we
created ourselves, not the ones that are passed in by the user.
Fixesjoyent/node#6297.
It turns out that node.js relies on the blocking behavior of pipes in
some cases, notably when forking worker processes. Reopens#941.
This reverts commit 8fe4ca686b.
Don't rely on the caller to set the O_NONBLOCK flag on the file
descriptor.
Prevents sporadic stalls when the file descriptor is in blocking mode
and exactly as many bytes are read as there are available; in that case,
libuv will try to read again and block. Node.js was guilty of this.
Fixes#941.
Work around an 'initializer element is not constant' build error in
src/unix/fsevents.c by turning the const int flags into #defines.
Only an issue on OS X 10.6 due to the old compiler it uses.
Fixes#908.
This is a back-port of commit 82f2472 from the master branch.
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
Conflicts:
include/uv-private/uv-darwin.h
src/unix/fsevents.c
* 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.
Conflicts:
src/unix/darwin.c
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.
Before this commit, uv_fs_chown() and uv_fs_fchown() took the uid and
gid as signed integers which is wrong because uid_t and gid_t are
unsigned on most all platforms and IDs that don't fit in a signed
integer do exist.
This is not an ABI change because the size of the uid and gid arguments
do not change, only their sign.
On Windows, uv_uid_t and uv_gid_t are typedef'd as unsigned char for
reasons that are unclear. It doesn't matter: they get cast to ints when
used as function arguments. The arguments themselves are unused.
Partial fix for joyent/node#5890.
