This commit fixes the following valgrind warning:
==26443== Conditional jump or move depends on uninitialised value(s)
==26443== at 0x44AAFF: uv__read (stream.c:872)
==26443== by 0x44ADD4: uv__stream_io (stream.c:1051)
==26443== by 0x4377B8: uv__io_rw (core.c:539)
==26443== by 0x43C761: ev_invoke_pending (ev.c:2145)
==26443== by 0x43724D: uv__poll (core.c:260)
==26443== by 0x437297: uv__run (core.c:269)
==26443== by 0x4372F6: uv_run (core.c:277)
==26443== by 0x42094B: run_test_pipe_ref4 (test-ref.c:334)
==26443== by 0x406551: run_test_part (runner.c:302)
==26443== by 0x405384: main (run-tests.c:57)
Don't return req->result after posting the work request. It's possible (if
unlikely) for a worker thread to process the request before the main thread
reaches the return statement.
Some platforms don't support sendfile() (netbsd, openbsd), others don't support
arbitrary file descriptors (freebsd, darwin) and yet others have quirks in their
implementation (sunos).
Work around the quirks. If all else fails, fall back to sendfile() emulation.
uv_fs_stat and uv_fs_lstat removed the trailing backslash (if any) from the path
in order to please a test case that was written for Windows. Remove the
compatibility hack and fix the test.
Free the memory that is used to store the new argv and envp. It's not strictly
necessary (the OS is going to reclaim the memory anyway) but it makes the output
from valgrind a lot less noisy.
Benchmarks demonstrated that the idle timer handle approach didn't balance the
load quite fair enough, the majority of new connections still ended up in one
or two processes.
The new approach voluntarily gives up a scheduler timeslice by calling
nanosleep() with a one nanosecond timeout.
Why not sched_yield()? Because on Linux (and this is probably true for other
Unices as well), sched_yield() only yields if there are other processes running
on the same CPU.
nanosleep() on the other hand always forces the process to sleep, which gives
other processes a chance to accept our pending connections.
Make uv_cpu_info() understand the ARM and MIPS versions of /proc/cpuinfo,
it only knew how to deal with the x86 version
This commit also fixes a buglet where uv_cpu_info() reported the maximum CPU
frequency instead of the actual CPU frequency. That is, before this commit
`out/Debug/run-tests platform_output | grep speed | sort | uniq -c` on my
system always reported:
8 speed: 3400
Now it reports (for example):
2 speed: 3400
6 speed: 1600
In other words, two CPUs are running at full speed while the others have been
scaled back because they're mostly idle.
This is a back-port of commit 54bfb66 from the master branch.
Fixes#526.
The OS X version was being checked with the __MAC_OS_X_VERSION_MIN_REQUIRED__
macro, but this value doesn't match the actual SDK version when it is
overridden with the -mmacosx-version-min command line switch.
Adds initial libuv build/platform support for AIX. Builds work using gcc or the
IBM XL C compiler using its gxlc wrapper. Platform support is added for
uv_hrtime, uv_exepath, uv_get_free_memory, uv_get_total_memory, uv_loadavg,
uv_uptime, uv_cpu_info, uv_interface_addresses.
This will make uv_queue_work consistent with other functions, where it's
possible to omit the callback. I'm pretty sure that this is already
implemented for unix systems, although I didn't test the unix part of
uv_queue_work.
Implement a best effort approach to mitigating accept() EMFILE errors.
We have a spare file descriptor stashed away that we close to get below
the EMFILE limit. Next, we accept all pending connections and close them
immediately to signal the clients that we're overloaded - and we are, but
we still keep on trucking.
There is one caveat: it's not reliable in a multi-threaded environment.
The file descriptor limit is per process. Our party trick fails if another
thread opens a file or creates a socket in the time window between us
calling close() and accept().
Fixes#315.
Don't spin in epoll_wait() / kevent() / port_getn() / etc. when we can't
accept() a new connection due to having reached the file descriptor limit.
Pass the error to the connection_cb and let the libuv user deal with it.
Don't spin in epoll_wait() / kevent() / port_getn() / etc. when we can't
accept() a new connection due to having reached the file descriptor limit.
Pass the error to the connection_cb and let the libuv user deal with it.
Some memory was leaked when the uv_udp_t handle was closed when there were
in-flight send requests with a heap allocated buffer list.
That doesn't happen much in practice. In the common case (writing < 5 buffers),
the buffer list is stored inside the uv_udp_send_t structure, not allocated on
the heap.
uv_async_send would always return 1 when non-gcc compilers were used.
When uv_async_send returns 1 no attempt is made to make port_getn
return, so in this situation uv_async_send didn't wake up the event
loop.
Set the non-blocking and close-on-exec flags with ioctl() instead of fcntl(),
it's about 10-25% faster.
Stick with fcntl() on Solaris. ioctl(FIONBIO) works but is twice as slow as
fcntl(O_NONBLOCK). ioctl(FIOCLEX) doesn't raise an error but doesn't actually
work either.
uv_update_time does not overwrite the high 32 bits of uv_loop_t.time.
It merely increments it by one when the low 32 bits have wrapped. That
means that `time` needs to be initialized to zero before
uv_update_time() is called for the first time.
Don't try the pipe2() or socketpair(O_CLOEXEC) syscalls when a previous call to
uv__make_pipe() or uv__make_socketpair() call established that the kernel
doesn't support it. Speeds up pipe and socketpair creation on older kernels.
Variables tagged with __read_mostly are put into a separate ELF section to
improve the cache locality of data that is read often but seldom written to.
kqueue(2) on osx doesn't work (emits EINVAL error) with specific fds
(i.e. /dev/tty, /dev/null, etc). When given such descriptors - start
select(2) watcher thread that will emit io events.
uv_fs_poll_t has an embedded uv_timer_t handle that got closed at a time when
the memory of the encapsulating handle might already have been deallocated.
Solve that by moving the poller's state into a structure that is allocated on
the heap and can be freed independently.
Readable tty handles need to be able to update the virtual window,
so if uv_tty_t is initialized with a console input fd, additionally
open the console output.
Make uv_cpu_info() understand the ARM and MIPS versions of /proc/cpuinfo,
it only knew how to deal with the x86 version
This commit also fixes a buglet where uv_cpu_info() reported the maximum CPU
frequency instead of the actual CPU frequency. That is, before this commit
`out/Debug/run-tests platform_output | grep speed | sort | uniq -c` on my
system always reported:
8 speed: 3400
Now it reports (for example):
2 speed: 3400
6 speed: 1600
In other words, two CPUs are running at full speed while the others have been
scaled back because they're mostly idle.
Fixes#526.
Formerly spawn errors would be reported as a message printed to the
process' stderr, to match unix behaviour. Unix has now been fixed to
be more sensible, so this hack can now be removed.
This also fixes a race condition that could occur when the user closes
a process handle before the exit callback has been made.
Start the SIGCHLD signal watcher before calling fork(). There was a very subtle
race where a child process could quit (and generate a SIGCHILD) before the
signal handler was installed.
To reproduce, call uv_spawn() repeatedly with the below x86_64 program:
// compile with `gcc -O2 -nostdlib`
void _start(void)
{
// syscall(SYS_exit, 0)
__asm__ __volatile__ (
"xorq %rdi, %rdi;"
"xorq %rax, %rax;"
"mov $0x3c, %al;"
"syscall;"
);
for (;;);
}
poll() on newer versions of OS X sets POLLHUP|POLLIN whereas older versions
(and other Unices) only set POLLHUP. It was tripping up a check that expected
to read data when POLLIN was set.
While easy to work around, I switched it to a blocking read instead:
it's less code and avoids surprises like the one above altogether.
Fixes#522.
OS X has no public API for fdatasync. And as pointed out in `man fsync(2)`:
For applications that require tighter guarantees about the integrity of
their data, Mac OS X provides the F_FULLFSYNC fcntl. The F_FULLFSYNC
fcntl asks the drive to flush all buffered data to permanent storage.
Applications, such as databases, that require a strict ordering of writes
should use F_FULLFSYNC to ensure that their data is written in the order
they expect. Please see fcntl(2) for more detail.
uv_spawn() saves and restores the environ in case the child clobbers it -
which is impossible because the child process runs in a separate address space.
Problem: registering two uv_fs_event_t watchers for the same path, then closing
them, caused a segmentation fault. While active, the watchers didn't work right
either, only one would receive events.
Cause: each watcher has a wd (watch descriptor) that's used as its key in a
binary tree. When you call inotify_watch_add() twice with the same path, the
second call doesn't return a new wd - it returns the existing one. That in turn
resulted in the first handle getting ousted from the binary tree, leaving
dangling pointers.
This commit addresses that by storing the watchers in a queue and storing the
queue in the binary tree instead of storing the watchers directly in the tree.
Fixesjoyent/node#3789.
uv_set_process_title() was susceptible to a format string vulnerability:
$ node -e 'process.title = Array(42).join("%s")'
Segmentation fault: 11 (core dumped)
The fix is trivial - call setproctitle("%s", s) instead of setproctitle(s) -
but valgrind complains loudly about reads from and writes to uninitialized
memory in libc. It's not a libuv bug because the test case below triggers the
same warnings:
#include <sys/types.h>
#include <unistd.h>
int main(void)
{
setproctitle("%s", "test");
return 0;
}
That's why this commit replaces setproctitle() with sysctl(KERN_PROC_ARGS).
This commit reapplies commit a9f6f06, which got reverted in 69a6afe. The revert
turned out to be unnecessary.
It's making node.js crash when run as root. Backtrace:
(gdb) bt
#0 0x00007fff856e3ff9 in __findenv ()
#1 0x00007fff856e404c in getenv ()
#2 0x000000010004c850 in loop_init (loop=0x10045a792, flags=8) at ev.c:1707
#3 0x000000010004cb3b in ev_backend [inlined] () at /Users/tjfontaine/Development/node/deps/uv/src/unix/ev/ev.c:2090
#4 0x000000010004cb3b in ev_default_loop (flags=1606417108) at ev.c:2092
#5 0x000000010004e5c6 in uv__loop_init (loop=0x10066e330, default_loop=1) at loop.c:52
#6 0x0000000100044367 in uv_default_loop () at core.c:196
#7 0x0000000100004625 in node::Init (argc=1606417456, argv=0x100b0f490) at node.cc:2761
#8 0x000000010000797d in node::Start (argc=1606417600, argv=0x0) at node.cc:2888
#9 0x0000000100000ca4 in start ()
This reverts commits:
b49d6f7 unix: fix uv_set_process_title()
a9f6f06 unix: fix format string vulnerability in freebsd.c
a87abc7 unix: avoid buffer overflow in proctitle.c
dc97d44 unix: move uv_set_process_title() to proctitle.c
Use strncpy() to set the process title, it pads the remainder with nul bytes.
Avoids garbage output on systems where `ps aux` prints the entire proctitle
buffer, not just the characters up to the first '\0'.
Fixesjoyent/node#3726.
uv_set_process_title() was susceptible to a format string vulnerability:
$ node -e 'process.title = Array(42).join("%s")'
Segmentation fault: 11 (core dumped)
The fix is trivial - call setproctitle("%s", s) instead of setproctitle(s) -
but valgrind complains loudly about reads from and writes to uninitialized
memory in libc. It's not a libuv bug because the test case below triggers the
same warnings:
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <unistd.h>
int main(void)
{
setproctitle("%s", "test");
return 0;
}
That's why this commit replaces setproctitle() with sysctl(KERN_PROC_ARGS).
Use hijacking argv array to change process' title. It seems to be working fine
on almost every platform (at least it should not break anything as it's used in
nginx in a similar way).
This is a back-port of commit e1320757 from the master branch.
Newer versions of gyp do not support files with the same basenames (example:
core.c and linux/core.c).
The nominal reason is consistency across build systems. Apparently, msbuild
doesn't support it either.
Somewhere, someplace, baby Jesus cries sad little tears...
Fixes#464.
Use atomic compare-and-swap to detect if we've been preempted by another thread
and therefore can avoid making the expensive write() syscall.
Speeds up the heavily contended case by about 1-2% and has little if any impact
on the non-contended case. I wasn't able to measure the difference at any rate.
Don't start reading immediately after connecting. If the server sends a message
and the client hasn't called uv_read_start() yet, the event loop will busy loop
because the pending message keeps waking it up.
Don't make the event loop spin when connect() returns EINPROGRESS.
Test case:
#include "uv.h"
static void connect_cb(uv_connect_t* req, int status) {
// ...
}
int main() {
uv_tcp_t handle;
uv_connect_t req;
struct sockaddr_in addr;
addr = uv_ip4_addr("8.8.8.8", 1234); // unreachable
uv_tcp_init(uv_default_loop(), &handle);
uv_tcp_connect(&req, (uv_stream_t*)&handle, addr, connect_cb);
uv_run(uv_default_loop()); // busy loops until connection times out
return 0;
}
After EINPROGRESS, there are zero active handles and one active request. That
in turn makes uv__poll_timeout() believe that it should merely poll, not block,
in epoll() / kqueue() / port_getn().
Sidestep that by artificially starting the handle on connect() and stopping it
again once the TCP handshake completes / is rejected / times out.
It's a slightly hacky approach because I don't want to change the ABI of the
stable branch. I'll address it properly in the master branch.
Incidentally fixes a rather obscure bug where uv_tcp_connect() reconnected
and leaked a file descriptor when the handle was already busy connecting,
handle->fd was zero (unlikely) and uv_tcp_connect() got called again.
accept4() piggybacks on the socketcall() on i386. socketcall() has the flaw
that it returns EINVAL instead of ENOSYS when the operation is not supported.
The problem is that accept4() also returns EINVAL when its flag argument is
invalid.
Try to discern between the two failure cases to the best of our abilities.
This fixes the issue that a mount point would be treated as a
symlink, but readlink would subsequently return an NT namespaced
path that is unusable for many purposes.
This also pre-emptively fixes the problems that would arise when
an user has a reparse point whose tag is neither
IO_REPARSE_TAG_MOUNT_POINT nor IO_REPARSE_TAG_SYMLINK.
Finally uv_lstat() will now return the correct length in st_size.
Previously the length was computed incorrectly for relative
symlinks, and those that had non-ascii characters in their target.
Or rather, don't assume that dlopen() does *not* clobber dlerror().
Joe Ferner reports that loading libjava on OS X sets dlerror() even when
dlopen() succeeds, which makes uv_dlopen() raise an error when it shouldn't.
I haven't been able to reproduce it but it's possible that libjava clobbers
dlerror() by trying (and failing) to load other libraries.
At any rate, Joe confirmed that this patch addresses the issue.
Fixes#462.
Fixes a lifecycle issue where the req got created, added to the list of active
reqs but wasn't properly activated because uv__connect() returned early.
produces better error message from test-dgram-multicast-multi-process
when run w/o network.
before:
=== release test-dgram-multicast-multi-process ===
Path: simple/test-dgram-multicast-multi-process
dgram.js:287
throw new errnoException(errno, 'addMembership');
^
Error: addMembership Unknown system errno 19
at new errnoException (dgram.js:356:11)
at Socket.addMembership (dgram.js:287:11)
at Object.<anonymous> (/home/roman/wc/node/test/simple/test-dgram-multicast-multi-process.js:224:16)
at Module._compile (module.js:449:26)
at Object.Module._extensions..js (module.js:467:10)
at Module.load (module.js:356:32)
at Function.Module._load (module.js:312:12)
at Module.runMain (module.js:487:10)
at process.startup.processNextTick.process._tickCallback (node.js:244:9)
[PARENT] Worker 9223 died. 1 dead of 3
dgram.js:287
throw new errnoException(errno, 'addMembership');
^
Error: addMembership Unknown system errno 19
at new errnoException (dgram.js:356:11)
at Socket.addMembership (dgram.js:287:11)
at Object.<anonymous> (/home/roman/wc/node/test/simple/test-dgram-multicast-multi-process.js:224:16)
at Module._compile (module.js:449:26)
at Object.Module._extensions..js (module.js:467:10)
at Module.load (module.js:356:32)
at Function.Module._load (module.js:312:12)
at Module.runMain (module.js:487:10)
at process.startup.processNextTick.process._tickCallback (node.js:244:9)
[PARENT] sent 'First message to send' to 224.0.0.114:12346
dgram.js:287
[PARENT] sent 'Second message to send' to 224.0.0.114:12346
throw new errnoException(errno, 'addMembership');
[PARENT] sent 'Third message to send' to 224.0.0.114:12346
^
[PARENT] sendSocket closed
[PARENT] Worker 9224 died. 2 dead of 3
Error: addMembership Unknown system errno 19
at new errnoException (dgram.js:356:11)
at Socket.addMembership (dgram.js:287:11)
at Object.<anonymous> (/home/roman/wc/node/test/simple/test-dgram-multicast-multi-process.js:224:16)
at Module._compile (module.js:449:26)
at Object.Module._extensions..js (module.js:467:10)
at Module.load (module.js:356:32)
at Function.Module._load (module.js:312:12)
at Module.runMain (module.js:487:10)
at process.startup.processNextTick.process._tickCallback (node.js:244:9)
[PARENT] Worker 9225 died. 3 dead of 3
[PARENT] All workers have died.
[PARENT] Fail
Command: out/Release/node /home/roman/wc/node/test/simple/test-dgram-multicast-multi-process.js
after:
=== release test-dgram-multicast-multi-process ===
Path: simple/test-dgram-multicast-multi-process
dgram.js:287
throw new errnoException(errno, 'addMembership');
^
Error: addMembership ENODEV
at new errnoException (dgram.js:356:11)
at Socket.addMembership (dgram.js:287:11)
at Object.<anonymous> (/home/roman/wc/node/test/simple/test-dgram-multicast-multi-process.js:224:16)
at Module._compile (module.js:449:26)
at Object.Module._extensions..js (module.js:467:10)
at Module.load (module.js:356:32)
at Function.Module._load (module.js:312:12)
at Module.runMain (module.js:487:10)
at process.startup.processNextTick.process._tickCallback (node.js:244:9)
[PARENT] Worker 13141 died. 1 dead of 3
dgram.js:287
throw new errnoException(errno, 'addMembership');
^
[PARENT] sent 'First message to send' to 224.0.0.114:12346
[PARENT] sent 'Second message to send' to 224.0.0.114:12346
[PARENT] sent 'Third message to send' to 224.0.0.114:12346
[PARENT] sent 'Fourth message to send' to 224.0.0.114:12346
[PARENT] sendSocket closed
dgram.js:287
throw new errnoException(errno, 'addMembership');
^
Error: addMembership ENODEV
at new errnoException (dgram.js:356:11)
at Socket.addMembership (dgram.js:287:11)
at Object.<anonymous> (/home/roman/wc/node/test/simple/test-dgram-multicast-multi-process.js:224:16)
at Module._compile (module.js:449:26)
at Object.Module._extensions..js (module.js:467:10)
at Module.load (module.js:356:32)
at Function.Module._load (module.js:312:12)
at Module.runMain (module.js:487:10)
at process.startup.processNextTick.process._tickCallback (node.js:244:9)
[PARENT] Worker 13142 died. 2 dead of 3
Error: addMembership ENODEV
at new errnoException (dgram.js:356:11)
at Socket.addMembership (dgram.js:287:11)
at Object.<anonymous> (/home/roman/wc/node/test/simple/test-dgram-multicast-multi-process.js:224:16)
at Module._compile (module.js:449:26)
at Object.Module._extensions..js (module.js:467:10)
at Module.load (module.js:356:32)
at Function.Module._load (module.js:312:12)
at Module.runMain (module.js:487:10)
at process.startup.processNextTick.process._tickCallback (node.js:244:9)
[PARENT] Worker 13143 died. 3 dead of 3
[PARENT] All workers have died.
[PARENT] Fail
Command: out/Release/node /home/roman/wc/node/test/simple/test-dgram-multicast-multi-process.js
Newer versions of gyp do not support files with the same basenames (example:
core.c and linux/core.c).
The nominal reason is consistency across build systems. Apparently, msbuild
doesn't support it either.
Somewhere, someplace, baby Jesus cries sad little tears...
Fixes#464.
Just closing the listening socket handle does not cancel AcceptEx
operations when another handle is keeping the underlying socket
open. Thus the AcceptEx operations have to be explicitly canceled
in uv_tcp_close.
* the callback gets called only once on error, not repeatedly...
* ...unless the error reason changes from e.g. UV_ENOENT to UV_EACCES
* the callback receives pointers to uv_statbuf_t objects so it can inspect what
changed
