Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9179

The -fno-strict-aliasing flag was added to fix compilation warnings when
building Node.js with GCC <= 4.4

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9179

process.send() should be synchronous, it should block until the message
has been sent in full, but it wasn't after the second-to-last libuv
upgrade because of commit libuv/libuv@393c1c5 ("unix: set non-block
mode in uv_{pipe,tcp,udp}_open"), which made its way into io.js in
commit 07bd05b ("deps: update libuv to 1.2.1").

Commit libuv/libuv@b36d4ff ("unix: implement uv_stream_set_blocking()")
as landed in io.js in commit 9681fca ("deps: update libuv to 1.4.0")
makes it possible to restore the synchronous behavior again and that's
precisely what this commit does.

The same line of reasoning applies to `net.Socket({ fd: 1 })`: creating
a socket object from a stdio file descriptor, like the `process.stdout`
getter does, should put the file descriptor in blocking mode for
compatibility reasons.

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9179

Float patch to fix pipe on Windows. Original commit message:

  win: fix pipe blocking writes

  In the code path for pipe blocking writes, WriteFile is already
  posting a completion packet to the I/O completion port.
  POST_COMPLETION_FOR_REQ was causing the same request to get
  returned twice by GetCompletionStatusEx.
  Also on the same code path, we were waiting on the wrong event.

  We need to update queued_bytes and write_queue_size when a
  blocking write request completes asynchronously.

Ref: libuv/libuv#238

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9179

Float patch to fix setsockopt for multicast on Solaris and derivatives.
Original commit message:

  solaris: fix setsockopt for multicast options

  On Solaris and derivatives such as SmartOS, the length of socket options
  for multicast and ttl options is not always sizeof(char).

  This fixes the udp_options and udp_options6 tests.

Ref: libuv/libuv#243

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9179

Original commit message:

  darwin: fix size calculation in select() fallback

  Apple's `fd_set` stores its bits in an array of 32-bit integers, which
  means `FD_ISSET()` may read out of bounds if we allocate storage at
  byte granularity. There's also a chance that the `select()` call could
  corrupt the heap, although I didn't investigate that.

  This issue was discovered by LLVM's AddressSanitizer which caught
  `FD_ISSET()` trying to read out of bounds.

Ref: libuv/libuv#241

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9179

Because we are floating several patches on top of libuv, make that
apparent in the version number.

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9179

Extract the tarball into deps/openssl/openssl

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

change all openssl/include/openssl/*.h to include resolved symbolic
links and openssl/crypto/opensslconf.h to refer config/opensslconf.h

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

sha256-x86_64.pl does not exist in the origin openssl distribution. It
was copied from sha512-x86_64.pl and both sha256/sha512 scripts were
modified so as to generates only one asm file specified as its key
hash length.

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

`x86masm.pl` was mistakenly using .486 instruction set, why `cpuid` (and
perhaps others) are requiring .686 .

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

It is no longer used.

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

deps/openssl/asm/x64-win32-masm/x86_64cpuid.asm.orig was accidentally
committed before.

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

reapply b910613

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

* openssl: upgrade to 1.0.1m (Addressing multiple CVEs)

Switch from running the loop with UV_RUN_ONCE to UV_RUN_DEFAULT, because
it's possible that the poll returns earlier than expected and thus the
timer is not run on a single interation.

The loop is not stopped either from the timer callback or from the async
handle's.

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9410

Extract the tarball into deps/openssl/openssl

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

Conflicts:
	deps/openssl/openssl/apps/s_client.c

change all openssl/include/openssl/*.h to include resolved symbolic
links and openssl/crypto/opensslconf.h to refer config/opensslconf.h

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

sha256-x86_64.pl does not exist in the origin openssl distribution. It
was copied from sha512-x86_64.pl and both sha256/sha512 scripts were
modified so as to generates only one asm file specified as its key
hash length.

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

`x86masm.pl` was mistakenly using .486 instruction set, why `cpuid` (and
perhaps others) are requiring .686 .

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

It is no longer used.

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

deps/openssl/asm/x64-win32-masm/x86_64cpuid.asm.orig was accidentally
committed before.

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

reapply b910613

PR: #9451
PR-URL: #9451
Reviewed-By: Julien Gilli <julien.gilli@joyent.com>

* openssl: upgrade to 1.0.1m (Addressing multiple CVES)

Conflicts:
	ChangeLog

Conflicts:
	ChangeLog
	src/node_version.h

Improve performance by:
+ Not leaking the `arguments` object!
+ Getting the last character of a string by index, instead of
  with `.substr()` or `.slice()`

Improve code consistency by:
+ Using `[]` instead of `.charAt()` where possible
+ Using a function declaration instead of a var declaration
+ Using `.slice()` with clearer arguments
+ Checking if `dir` is truthy in `win32.format`
  (added tests for this)

Improve both by:
+ Making the reusable `trimArray()` function
+ Standardizing getting certain path statistics with
  the new `win32StatPath()` function

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9289

At the uv layer pipes are connected with uv_pipe_connect.
The current spec for this method indicates that the maximum
length is limited to the size of length of
sizeof(sockaddr_un.sun_path), typically between 92 and
108 bytes. Anything longer than that just gets truncated.

The simple testsuite currently creates pipes in directories
under the directory where node was built.  In our jenkins
jobs this sometimes ends up being a deep enough path that
the path for the pipes is getting truncated.  The result
is that tests using pipes fail with errors that don't
make it obvious what the problem is.

Even if the errors were helpful, we still need a way
to avoid the truncation.

This patch adds the environment variable NODE_PIPE_DIR.
If set the tests create pipes in this directory instead of
the current defaults.  In addition the test harness is
updated to remove/delete this directory before/after
each test is run.

	modified:   test/common.js
	modified:   test/simple/test-net-pipe-connect-errors.js
	modified:   test/testpy/__init__.py
	modified:   test/simple/test-cluster-eaccess.js

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #9381

simple tests test-http-request-end.js, test-http-default-encoding.js
hangs in AIX. The root cause for both the failures is related to the
timing with which packets are sent between the client and server.
On the client side, one factor that affects the timing is Nagle's
algorithm. With Nagle enabled there may be a delay between two packets
as the stack may wait until either:
  a. An acknowledgement for the first packet is received, or
  b. 200 ms elapses.
before sending the second packet.

Similarly at the server side 2 sequential packages can be delivered to
the application either together or separatly.

On AIX we see that they are delivered separately to the server, while on
Linux delivered together. If we change the timing, for example disabling
Nagle on AIX we see the 2 packets delivered together and the tests pass.

In the test case simple/test-http-request-end.js, the client request
handler of the server receives and stores the data in a data callback,
closes the server in a request end callback, and writes to the client
and ends the response, in-line with the request receipt. An HTTP parser
module parses the incoming message, and invokes callback routines which
are registered for HTTP events (such as header, body, end etc.)

Because the termination sequence arrive in a separate packet, there is a
delay in parsing that message and identify that the client request ended
(and thereby invoke the request end call backhandler). Due to this delay,
the response close happens first, which in-turn destroys the server
socket leading to the fd and watcher removal from the uv loop abandoning
further events on this connection, and end call back never being called,
causing the reported hang.  simple/test-http-default-encoding.js suffers
from the same problem.

Also, remove the timer logic from the test case. Test harness anyways
contain a timer which controls the individual tests so remove such
controls from the test case, as suggested by @tjfontaine

Reviewed-By: Michael Dawson <michael_dawson@ca.ibm.com>
PR-URL: #9432

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #14180

This change is a backport of 1a3ca82
from io.js.

Original commit message:

  Read all pending data out of the socket on `error` event and ensure that
  no `data`/`end` handlers will be invoked on `socket.destroy()`.
  Otherwise following assertion happens:

      AssertionError: null == true
          at TLSSocket.socketOnData (_http_client.js:308:3)
          at TLSSocket.emit (events.js:107:17)
          at TLSSocket.Readable.read (_stream_readable.js:373:10)
          at TLSSocket.socketCloseListener (_http_client.js:229:10)
          at TLSSocket.emit (events.js:129:20)
          at TCP.close (net.js:476:12)

  Fix: #9348
  PR-URL: nodejs/node#1103
  Reviewed-By: Rod Vagg <rod@vagg.org>
  Reviewed-By: Nicu Micleușanu <micnic90@gmail.com>

Fixes #9348.

Reviewed-By: Julien Gilli <julien.gilli@joyent.com>
PR-URL: #14087

This is a backport of b64983d.

Original commit message:

  src: reset signal handler to SIG_DFL on FreeBSD

  FreeBSD has a nasty bug with SA_RESETHAND reseting the SA_SIGINFO,
  that is in turn set for a libthr wrapper. This leads to a crash.
  Work around the issue by manually setting SIG_DFL in the signal
  handler.

  Fix: #9326
  PR-URL: nodejs/node#1218
  Reviewed-By: Johan Bergström <bugs@bergstroem.nu>
  Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

Fixes #9326.

Reviewed-By: Trevor Norris <trev.norris@gmail.com>
PR-URL: #14184

simple tests test-cluster-master-error.js, test-cluster-master-kill.js
fails in AIX with assertion failure indicating that the workers were
alive even after the master terminated. A 200ms leeway is provided for
the workers to actually terminate, but the isAlive check returns
true in both the cases.

In AIX, the workers were actually terminating, but they took more time
- as much as 800ms (normal) to 1000ms (in rare cases).

Based on a C test we ran, it is found that the exit routines in AIX
is a bit more longer than that in Linux. There are a number of cleanup
activities performed in exit() system call, and depending on when the
signal handlers are shutdown in that sequence, the process will be
deemed as dead or alive, from another process's perspective.

process.kill(pid) is used in the test case to check the liveliness of
the worker, and when the kill() call is issued, even if the target
process is in it's exit sequences, if the signal handlers are not shut
down, it will respond to external signals, causing those calls to pass.

This fix extends the additional timeout for all platforms

Reviewed-By: Michael Dawson <michael_dawson@ca.ibm.com>
PR-URL: #9431

* uv: Upgrade to 1.4.2

* npm: Upgrade to 2.7.4

* V8: do not add extra newline in log file (Julien Gilli)

* V8: Fix --max_old_space_size=4096 integer overflow (Andrei Sedoi)

* asyncwrap: fix constructor condition for early ret (Trevor Norris)

* buffer: align chunks on 8-byte boundary (Fedor Indutny)

* buffer: fix pool offset adjustment (Trevor Norris)

* build: fix use of strict aliasing (Trevor Norris)

* console: allow Object.prototype fields as labels (Colin Ihrig)

* fs: make F_OK/R_OK/W_OK/X_OK not writable (Jackson Tian)

* fs: properly handle fd passed to truncate() (Bruno Jouhier)

* http: fix assert on data/end after socket error (Fedor Indutny)

* lib: fix max size check in Buffer constructor (Ben Noordhuis)

* lib: fix stdio/ipc sync i/o regression (Ben Noordhuis)

* module: replace NativeModule.require (Herbert Vojčík)

* net: allow port 0 in connect() (cjihrig)

* net: unref timer in parent sockets (Fedor Indutny)

* path: refactor for performance and consistency (Nathan Woltman)

* smalloc: extend user API (Trevor Norris)

* src: fix for SIGINT crash on FreeBSD (Fedor Indutny)

* src: fix builtin modules failing with --use-strict (Julien Gilli)

* watchdog: fix timeout for early polling return (Saúl Ibarra Corretgé)

In Linux, simple/test-child-process-fork-net2.js fails intermittently.

In SuSE Linux system, under network high load situations, this failure is
consistently reproducible.

The test case tests whether the TCP connections which were established between
the processes terminate in a timely and clean manner. After some iterations of
data transfer on established connections, the server is closed. The server does
not get closed immediately, instead waits for all the active connections to
terminate. A timed (200ms) callback closes the connections, which eventually
closes the server.

The start is the time when the server close is invoked.
The end is the time when the server is actually closed(onClose call back invoked).

Given that there is a minimum delay of 200ms before the connections are
terminated, expecting the elapsed time above 190 is reasonable and fair,
but looks like the leeway of 800ms for the upper bounds seem to be too
stringent, and breaking some scenarios of network load.

Reviewed-By: Michael Dawson <michael_dawson@ca.ibm.com>
PR-URL: #14129

This is a backport of ea37ac0

Original commit message:

  On AIX, OS X and the BSDs, calling shutdown() on one end of a pipe
  when the other end has closed the connection fails with ENOTCONN.

  The sequential/test-child-process-execsync test failed sporadically
  because of a race between the parent and the child where one closed
  its end of the pipe before the other got around to calling shutdown()
  on its end of the pipe.

  Libuv is not the right place to handle that because it can't tell if
  the ENOTCONN error is genuine but io.js can.

  Refs: libuv/libuv#268
  PR-URL: iojs#1214
  Reviewed-By: Bert Belder <bertbelder@gmail.com>

Fixes: #9444.