mm/mremap: Add comment explaining the untagging behaviour of mremap()
authorWill Deacon <will@kernel.org>
Wed, 25 Mar 2020 11:13:46 +0000 (11:13 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 10 May 2020 08:31:33 +0000 (10:31 +0200)
commit b2a84de2a2deb76a6a51609845341f508c518c03 upstream.

Commit dcde237319e6 ("mm: Avoid creating virtual address aliases in
brk()/mmap()/mremap()") changed mremap() so that only the 'old' address
is untagged, leaving the 'new' address in the form it was passed from
userspace. This prevents the unexpected creation of aliasing virtual
mappings in userspace, but looks a bit odd when you read the code.

Add a comment justifying the untagging behaviour in mremap().

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm/mremap.c

index 1d98281f7204e61e6cf9907cc46b5a6c005ec6f5..245bf9c555b27578dfce38baaaad04e9d59b0941 100644 (file)
@@ -606,6 +606,16 @@ SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
        LIST_HEAD(uf_unmap_early);
        LIST_HEAD(uf_unmap);
 
+       /*
+        * There is a deliberate asymmetry here: we strip the pointer tag
+        * from the old address but leave the new address alone. This is
+        * for consistency with mmap(), where we prevent the creation of
+        * aliasing mappings in userspace by leaving the tag bits of the
+        * mapping address intact. A non-zero tag will cause the subsequent
+        * range checks to reject the address as invalid.
+        *
+        * See Documentation/arm64/tagged-address-abi.rst for more information.
+        */
        addr = untagged_addr(addr);
 
        if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))