fix: lookup for page hole in all ancestors

[fntlnz]

Changed the check pagehole logic to go through the entire parent chain to find missing pages. More details on how to reproduce in the linked issue.

Fixes #2738

[fntlnz]

I figured y'all might want me to add a zdtm test for this, I see that there are tests for predumps already but they don't trigger the behavior. I see that it consistently happens when the target process uses at least 3.6G of memory, maybe we can add a test that does that or modify the existing ones.

[fntlnz]

force push to rebase criu-dev

[fntlnz]

@avagin @rst0git if there is any more info I can provide to help you folks when reviewing this please let me know. Thanks

[avagin]

@fntlnz this patch isn't right. By design, criu has to read just previous pagemap image to figure out what pages has to be dumped. The issue in pagemap_len is an integer overflow when calculating a region size. Your reproducer should pass with this patch:

diff --git a/criu/include/pagemap.h b/criu/include/pagemap.h
index 3ae15deb9..fae110108 100644
--- a/criu/include/pagemap.h
+++ b/criu/include/pagemap.h
@@ -121,7 +121,7 @@ extern int dedup_one_iovec(struct page_read *pr, unsigned long base, unsigned lo
 
 static inline unsigned long pagemap_len(PagemapEntry *pe)
 {
-	return pe->nr_pages * PAGE_SIZE;
+	return (unsigned long)pe->nr_pages * PAGE_SIZE;
 }
 
 static inline bool page_read_has_parent(struct page_read *pr)

The real fix should convert nr_pages to uint64, but we need to preserve restore compatibility with previous images.

[fntlnz]

@avagin thanks for taking a look, I suspected that something else might be wrong because it seemed strange that we would need to do that. I didn't realize it was supposed to read only the previous pagemap. I'll test and report back soon.

[fntlnz]

@avagin I tested the cast and it works with my reproducer. I'll close this one and send another patch with your fix.