RFR: 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe

[Andrey Turbanov-2]

8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe

-------------

Commit messages:
 - [PATCH] Thread-safe initialization of SunFontManager.platformFontMap
 - [PATCH] Thread-safe initialization of SunFontManager.platformFontMap

Changes: https://git.openjdk.java.net/jdk/pull/2762/files
 Webrev: https://webrevs.openjdk.java.net/?repo=jdk&pr=2762&range=00
  Issue: https://bugs.openjdk.java.net/browse/JDK-8263138
  Stats: 7 lines in 1 file changed: 3 ins; 0 del; 4 mod
  Patch: https://git.openjdk.java.net/jdk/pull/2762.diff
  Fetch: git fetch https://git.openjdk.java.net/jdk pull/2762/head:pull/2762

PR: https://git.openjdk.java.net/jdk/pull/2762

[Sergey Bylokhov-2]

On Sat, 27 Feb 2021 18:33:52 GMT, Andrey Turbanov <[hidden email]> wrote:

> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe

The bug: https://bugs.openjdk.java.net/browse/JDK-8263138

src/java.desktop/share/classes/sun/font/SunFontManager.java line 1452:

> 1450:         if (platformFontMap == null) {
> 1451:             platformFontMap = populateHardcodedFileNameMap();
> 1452:             SunFontManager.platformFontMap = platformFontMap;

I am not sure it is enough for "thread-safe initialization". The "platformFontMap" might not be null, but its content can be broken/incomplete at this point.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Phil Race]

On Sun, 7 Mar 2021 03:32:01 GMT, Sergey Bylokhov <[hidden email]> wrote:

>> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe
>
> The bug: https://bugs.openjdk.java.net/browse/JDK-8263138

I don't know why making this volatile is doing anything.
Let's discuss this further.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[liach]

On Sat, 27 Feb 2021 22:30:18 GMT, Sergey Bylokhov <[hidden email]> wrote:

>> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe
>
> src/java.desktop/share/classes/sun/font/SunFontManager.java line 1452:
>
>> 1450:         if (platformFontMap == null) {
>> 1451:             platformFontMap = populateHardcodedFileNameMap();
>> 1452:             SunFontManager.platformFontMap = platformFontMap;
>
> I am not sure it is enough for "thread-safe initialization". The "platformFontMap" might not be null, but its content can be broken/incomplete at this point.

https://github.com/openjdk/jdk/pull/2691#issuecomment-784367127

In short, previously platformFontMap references were field set/gets and may be inconsistent due to concurrency. It is now moved to a local variable; also the map modifications have been moved to fields to avoid concurrency issues with hash map as far as I can see.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Sergey Bylokhov-2]

On Sun, 28 Feb 2021 03:50:13 GMT, liach <[hidden email]> wrote:

>> src/java.desktop/share/classes/sun/font/SunFontManager.java line 1452:
>>
>>> 1450:         if (platformFontMap == null) {
>>> 1451:             platformFontMap = populateHardcodedFileNameMap();
>>> 1452:             SunFontManager.platformFontMap = platformFontMap;
>>
>> I am not sure it is enough for "thread-safe initialization". The "platformFontMap" might not be null, but its content can be broken/incomplete at this point.
>
> https://github.com/openjdk/jdk/pull/2691#issuecomment-784367127
>
> In short, previously platformFontMap references were field set/gets and may be inconsistent due to concurrency. It is now moved to a local variable; also the map modifications have been moved to fields to avoid concurrency issues with hash map as far as I can see.

One thread may create the local platformFontMap, then set the static SunFontManager.platformFontMap field, and then initialize the platformFontMap or mix these operations. The second thread may see non-null SunFontManager.platformFontMap which is not still initialized.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Andrey Turbanov-2]

On Sun, 28 Feb 2021 04:56:07 GMT, Sergey Bylokhov <[hidden email]> wrote:

>> https://github.com/openjdk/jdk/pull/2691#issuecomment-784367127
>>
>> In short, previously platformFontMap references were field set/gets and may be inconsistent due to concurrency. It is now moved to a local variable; also the map modifications have been moved to fields to avoid concurrency issues with hash map as far as I can see.
>
> One thread may create the local platformFontMap, then set the static SunFontManager.platformFontMap field, and then initialize the platformFontMap or mix these operations. The second thread may see non-null SunFontManager.platformFontMap which is not still initialized.

Made `volatile` to guarantee memory visibility if thread saw non-null value.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Sergey Bylokhov-2]

On Sun, 28 Feb 2021 08:47:22 GMT, Andrey Turbanov <[hidden email]> wrote:

>> One thread may create the local platformFontMap, then set the static SunFontManager.platformFontMap field, and then initialize the platformFontMap or mix these operations. The second thread may see non-null SunFontManager.platformFontMap which is not still initialized.
>
> Made `volatile` to guarantee memory visibility if thread saw non-null value.

I don't have any other questions, will file a bug and run the tests.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Alexey Ivanov-2]

On Sat, 27 Feb 2021 18:33:52 GMT, Andrey Turbanov <[hidden email]> wrote:

> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe

Marked as reviewed by aivanov (Reviewer).

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Sergey Bylokhov-2]

On Tue, 9 Mar 2021 20:56:15 GMT, Alexey Ivanov <[hidden email]> wrote:

>> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe
>
> Marked as reviewed by aivanov (Reviewer).

> I don't know why making this volatile is doing anything.
> Let's discuss this further.

It will prevent the possibility that one thread will store the non-null value to the SunFontManager.platformFontMap and only after that complete initialization of object stored in that field due to code optimizations. It is similar to the volatile DCL:
see "broken multithreaded version" example:
https://en.wikipedia.org/wiki/Double-checked_locking#Usage_in_Java

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Alexander Zuev-3]

On Sat, 27 Feb 2021 18:33:52 GMT, Andrey Turbanov <[hidden email]> wrote:

> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe

Marked as reviewed by kizune (Reviewer).

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Sergey Bylokhov-2]

On Sat, 27 Feb 2021 18:33:52 GMT, Andrey Turbanov <[hidden email]> wrote:

> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe

Marked as reviewed by serb (Reviewer).

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Sergey Bylokhov-2]

On Wed, 10 Mar 2021 23:13:09 GMT, Sergey Bylokhov <[hidden email]> wrote:

>> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe
>
> Marked as reviewed by serb (Reviewer).

@prrace Do you have any objections to integrating this change?

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762

[Andrey Turbanov-2]

On Sat, 27 Feb 2021 18:33:52 GMT, Andrey Turbanov <[hidden email]> wrote:

> 8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe

This pull request has now been integrated.

Changeset: ab66d699
Author:    Andrey Turbanov <[hidden email]>
Committer: Sergey Bylokhov <[hidden email]>
URL:       https://git.openjdk.java.net/jdk/commit/ab66d699
Stats:     7 lines in 1 file changed: 3 ins; 0 del; 4 mod

8263138: Initialization of sun.font.SunFontManager.platformFontMap is not thread safe

Reviewed-by: aivanov, kizune, serb

-------------

PR: https://git.openjdk.java.net/jdk/pull/2762