Avoid size overflow when allocating memory #6771
Conversation
Guard against very large allocation sizes which overflow when memory tagging overhead is added. Signed-off-by: Devin Nakamura <devinn@ca.ibm.com>
Signed-off-by: Devin Nakamura <devinn@ca.ibm.com>
|
jenkins build all |
Known failure: #6571. |
|
A new assertion is noticed, which does not appear in previous PR builds. Why does the assertion occur? Does it prevent |
It does not stop the test from completing. Looking into it, it seems that we clear the allocaiton headers before calling the c-lib realloc, however we never consider the possibility that realloc can fail thus we end up with a still valid chunk of memory, but invalid allocation headers. Note: this isn't specific to this change, but would seem to come into play any time realloc fails for any reason. |
|
jenkins build all |
babsingh
left a comment
There was a problem hiding this comment.
The current approach modifies the common-path and stack allocates a variable. The error-path, which handles the realloc failure, will be rarely taken. My suggestion would be to only modify the error-path and avoid stack allocation to minimize any impact on perf.
Restore allocation headers in the case realloc fails Signed-off-by: Devin Nakamura <devinn@ca.ibm.com>
f0ee5b3 to
4808a14
Compare
|
jenkins build all |
There was a problem hiding this comment.
https://ci.eclipse.org/omr/job/PullRequest-linux_ppc-64_le_gcc/3851/
Known failure: #6571.
https://github.com/eclipse/omr/pull/6771/checks?check_run_id=8990482324
Known failure: #6516.
@dnakamura Can you please port these changes to https://github.com/eclipse-openj9/openj9-omr/tree/v0.35.0-release?
Commit 1 - Guard against very large allocation sizes which overflow
when memory tagging overhead is added.
Commit 2 - omrmemtest: check that large allocations don't overflow.
Commit 3 - Handle reallocation failures: Restore allocation headers in
the case realloc fails.
Signed-off-by: Devin Nakamura devinn@ca.ibm.com