mm/mremap: fix vm_pgoff in vma_merge() case 3

After upgrading build guests to v6.3, rpm started segfaulting for
specific packages, which was bisected to commit 0503ea8f ("mm/mmap:
remove __vma_adjust()"). rpm is doing many mremap() operations with file
mappings of its db. The problem is that in vma_merge() case 3 (we merge
with the next vma, expanding it downwards) vm_pgoff is not adjusted as
it should when vm_start changes. As a result the rpm process most likely
sees data from the wrong offset of the file. Fix the vm_pgoff
calculation.

For case 8 this is a non-functional change as the resulting vm_pgoff is
the same.
Reported-and-bisected-by: Jiri Slaby <jirislaby@kernel.org>
Reported-and-tested-by: Fabian Vogt <fvogt@suse.com>
Link: https://bugzilla.suse.com/show_bug.cgi?id=1210903
Fixes: 0503ea8f ("mm/mmap: remove __vma_adjust()")
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Cc: <stable@vger.kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mmap.c

@@ -978,7 +978,7 @@ struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm,
 			vma = next;			/* case 3 */
 			vma_start = addr;
 			vma_end = next->vm_end;
-			vma_pgoff = mid->vm_pgoff;
+			vma_pgoff = next->vm_pgoff - pglen;
 			err = 0;
 			if (mid != next) {		/* case 8 */
 				remove = mid;