GitLab

Port of:
http://hg.openjdk.java.net/jdk8u/jdk8u/jdk/rev/4581ed6af636

This completes the upgrade to jdk8u60 of the classes
DerIndefLenConverter and DerInputStream.

(cherry picked from commit e27611b2c8a2c8cb413ff2f5f0c9cbae288aa835)

Bug: 30703445
Bug: 29631070
Change-Id: I052a1ea9959f9623973bf949a96923b603e17774
(cherry picked from commit 4c307860)

This fixes NPEs introduced by me in sun.security.pkcs.PKCS7 in commit
ddde3e18. In that commit, the standard
PKCS #7 ContentInfo/SignedData parsing works fine, but the "old" and
"Netscape-style" scenarios became broken because I changed them to
use the original encoded form of certificates without requesting to
retain this form during parsing. This commit fixes the issue by
changing the "old" and "Netscape-style" parsing code to retain the
original encoded form of certificates.

Bug: 30148997
Change-Id: I63f5f9148325214e571fdf30382d671b6d3f2a63

This changes sun.security.pkcs.PKCS7 so that the X509Certificate
instances it returns provide the original encoded form of the
certificate via Certificate.getEncoded. Prior to this change,
Certificate.getEncoded of these instances returned the DER form
of the certificate.

Returning the DER form is normally a good idea, but causes trouble
when this sun.security.pkcs.PKCS7 is used for parsing APKs' JAR
signatures. The way Android works is that an APK is permitted to be
updated only if the encoded form of the update's signing certificate
is exactly the same as the one of the already installed version of the
APK. Some APKs use signing certificates which are not DER-encoded,
which will lead to updates of such APKs to be rejected without this
fix.

Bug: 30148997
Change-Id: I44edc63d4923f8fb4f8f7a087611a7496137db52

Adds a test that demonstrates a regression from M in
the HTTP request line when a proxied connection
is made. It only affects URLs with empty paths (i.e. ones
without even a '/' after the host / authority). For
example, on M the request line for a request to
new URL("http://myhost").openConnection() would be:

GET http://myhost HTTP/1.1

but on N, without this change, it would be:

GET http://myhost/ HTTP/1.1

This change reverts to the M behavior.

Bug: 29983827
Change-Id: Ic457230cb9808546e079680b7ace79423ec30544
(cherry picked from commit d1288ce6)

This change fixes reading filesystem-saved
java.util.prefs.Preferences.

Due to a bug in XmlSupport, when reading the
filesystem-saved preferences, the reader was
throwing an exception. The Android
implementation of xml parser adds additional
nodes that were not Elements, this was
causing a cast exception (the old code was expecting
only Elements).

FileSystemPreferences class was swallowing exceptions if
they happen during preferences content reading.
This changes adds additional logging to prevent this.

Bug: 27645233
Change-Id: Ieebd8a6a1002111654005bfcc42009b01c89ea70
(cherry picked from commit 44d29b0a)

For the code:
class Parent {
  some code that will create <clinit>
}
class Child extends Parent {}

java.io.ObjectStreamClass.hasStaticInitializer(Child.class)
was returning false. This is technically valid, but historically
android was returning true in this case. This causes
problems with deserializing classes like Child without
explicit serialVersionUID. To mitigate the problem, this
change reverts to old behavior if sdk target <= 23.

Bug: 29064453
Change-Id: Iaff269ced01543469a30b50673a04d3d8722d261
(cherry picked from commit fd17eabf)

Bug: 27645233
Bug: 27441320
Change-Id: I78422f4314ad082adbd983e04510d4dfabe9792d
(cherry picked from commit e81ecfe3)

In the error log, use a link to a post in the Android Developers
blog instead of the current link to stackoverflow.

Also, add a message when an instance from the Crypto provider
is created for apps targeted at SDK <= 23,  so that when seeing
log messages of applications that failed we can tell quickly
whether the workaround applied or not.

Bug: 27873296

Change-Id: I808ba8d3e5014a754033571d6422a9beecc35e95

This was fixed in Jdk8u60.

changeset:   7887:7d7d553a8c61
user:        igerasim
date:        Tue Aug 13 13:04:21 2013 +0400
summary:     8022584: Memory leak in some NetworkInterface methods

bug: 29247929
Change-Id: I5e53e10f41bda0857871b91541718062898510ce

Libcore was missing few package-info.java/package.html files
for few packages, I've added the relevant files from openJdk 7u40.

In 3 packages (java.nio, java.nio.charser, java.lang.ref) the
package.html file was not being included in the generated documentation
for unknown reasons. Still investigating the cause, but in the meantime
I did found that converting package.html into package-info.java solves the
problem.

nyc-dev specific: Removed not existing classes from java.nio.channels

Bug: 29114138
Change-Id: Id16a645879ec7b34155a8d9d8ae6cd3b0b92b43b
(cherry picked from commit 77bf4f8e)

The collation rules normally returned by RuleBasedCollator.getRules()
are omitted in Android to save space:

https://android.googlesource.com/platform/external/icu/+/9cf98b3

This additional Javadoc comment was previously added:

https://android.googlesource.com/platform/libcore/+/1c13b45

But then it got lost again when the entire source file was replaced:

https://android.googlesource.com/platform/libcore/+/6c0e55c

This commit reinstates a comment about omitted collation rules.

Bug: 27441320
(cherry picked from commit a09dc30505bef70e0e6eb8afa886a59864e25fd2)

Change-Id: I535cee0f0b2c3926ede319b6bb285f68008487cc

The value of the address field should be equals to the first element of
the ByteBufferAsXBuffer class, previously, it was pointing to the first
element of the underlying directbytebuffer object.

Bug: 28964300
(cherry-picked from commit e0f383ff)
Change-Id: I6dfc98ba408c524580198ea168923981b23a0bc2

Bug: 27441320
Change-Id: I2d1230013bd499ff00015890eedb63943f9904ec
(partially cherry picked from commit 1b2f59a5)

The level-1 cache was incorrectly implemented. The check written here :

  if (cache1 != null && charsetName.equals(cache1.getKey()))
      return cache1.getValue();

Is incorrect because a different thread might change the value of cache1
between the null check, the call to getKey and the call to getValue.

The stress test in this change results in a 100% reproducible failure on a
nexus 6P without the included fix.

bug: 26548702
Change-Id: Ib564c5d84282bff23f7b95d3b520d45f740b0ba3

Behaviour before this CL was to avoid selecting a SPI if there
was a previously selected one. That is an incompatibility wrt M.

This CL makes Cipher compatible with the M behaviour

Bug: 29038928
Change-Id: Ie5ddf17e25344d41dc45f0c6df6effbca7dc7021

Addressing flakiness in test_parkFor_3 / test_parkUntil_3.

The wait time permitted was Math.max(expectedMillis * 10, 30)
where expectedMillis = 0. That meant that the test code had
just 30ms between calling Thread.start() and the Parker code being
executed reaching the notifyAll().

Other tests use an expectedMillis of 500 and 300, giving
more time for the Thread to start and complete their wait().
e.g. 4500ms and 2700ms above the expected park time,
respectively.

Bug: 29177727
Change-Id: I92d34a535f98929b767eebb75d4c49487b9b0348

sun.security.provider.SecureRandom is never used on
android, we always use the implementation from
AndroidOpenSSL provider instead.

Bug: 29091543
Change-Id: I20412c0f42c7b397ef5b05e623e61c2742fb935d
(cherry picked from commit 337d8c17)

If ProxySelector#getDefault is not empty, openJdk
SocksSocketImpl (the default SocketImpl) will try to
use it to proxy the socket connection through it, even
if the socket was created without the Proxy instance.

This is a regression from marshmallow release, which
would only tunnel the socket if the Proxy was explicitly
set in constructor.

Bug: 29092095

(cherry picked from commit 4c86b879)

Change-Id: Ie3b505c33e053b9a4240daa61579f7800b74b07e

bug: 29092095

(cherry picked from commit 51dea170)

Change-Id: I74cb36c70d81be0e9333dc71fd1b9580b64d12ca

The dstEnd calculation could previously overflow and a negative
dstBegin wasn't being caught in one specific case.

Bug: 28998511
Change-Id: I144d92d114bab94f5a207c3e3d5f5b55c97e9838