Fix erroneously large section size in ROM walk #18585
Conversation
If a ROM class has no methods, the CP NAS section should have size zero. Before, the section size would be calculated unconditionally based on the location of the J9ROMCLASS_ROMMETHODS of the ROM class. When the romMethods field is a null SRP, that would result in a negative pointer difference that would then wrap to a very large U_32 value. Signed-off-by: Christian Despres
|
Attn @mpirvu. I'm reasonably sure, but not totally confident, that the CP NAS section should have size zero when there are no ROM methods in a ROM class. |
| count = (U_32)(((UDATA)firstMethod - (UDATA)srpCursor) / (sizeof(J9SRP) * 2)); | ||
| } else { | ||
| count = 0; | ||
| } | ||
| rangeValid = callbacks->validateRangeCallback(romClass, srpCursor, count * sizeof(J9SRP) * 2, userData); |
There was a problem hiding this comment.
Most callbacks in this section of the walk still visit a section even if it's empty, so I kept that behaviour here. The one exception is the walk nest members SRPs block immediately above this one, which doesn't walk the nest member SRP block if it has a zero count.
Arguably it makes more sense to skip sections that are empty.
There was a problem hiding this comment.
For instance, this block
visits the enclosed inner classes SRP block even if it's empty, which has the odd consequence that the srpCursor for that section will point to the enclosedInnerClasses field of the ROM class, and not past the J9ROMClass header section like you might expect. That's because J9ROMCLASS_ENCLOSEDINNERCLASSES will return a pointer to enclosedInnerClasses when that field is zero. I don't think this ever messes anything up, but its an odd behaviour to have.
|
Just to be explicit about the problem: when |
There was a problem hiding this comment.
LGTM, but I am not an expert in this area.
@TobiAjila Could you please review or delegate? Thanks
|
@cjjdespres Which tests showed erroneously large section size in ROM walk? |
|
jenkins test sanity.functional,sanity.openjdk plinux jdk21 |
I noticed the problem while I was testing the code that revealed #18568. I'm not aware of existing failing tests. |
|
If a ROM class has no methods, the CP NAS section should have size zero. Before, the section size would be calculated unconditionally based on the location of the
J9ROMCLASS_ROMMETHODSof the ROM class. When theromMethodsfield is a null SRP, that would result in a negative pointer difference that would then wrap to a very largeU_32value.Signed-off-by: Christian Despres