mm, mremap: fix mremap() expanding for vma's with vm_ops->close()
authorVlastimil Babka <vbabka@suse.cz>
Tue, 17 Jan 2023 10:19:39 +0000 (11:19 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 9 Feb 2023 10:28:22 +0000 (11:28 +0100)
commit d014cd7c1c358edc3ea82ebf327a036a42ed0164 upstream.

Fabian has reported another regression in 6.1 due to ca3d76b0aa80 ("mm:
add merging after mremap resize").  The problem is that vma_merge() can
fail when vma has a vm_ops->close() method, causing is_mergeable_vma()
test to be negative.  This was happening for vma mapping a file from
fuse-overlayfs, which does have the method.  But when we are simply
expanding the vma, we never remove it due to the "merge" with the added
area, so the test should not prevent the expansion.

As a quick fix, check for such vmas and expand them using vma_adjust()
directly as was done before commit ca3d76b0aa80.  For a more robust long
term solution we should try to limit the check for vma_ops->close only to
cases that actually result in vma removal, so that no merge would be
prevented unnecessarily.

[akpm@linux-foundation.org: fix indenting whitespace, reflow comment]
Link: https://lkml.kernel.org/r/20230117101939.9753-1-vbabka@suse.cz
Fixes: ca3d76b0aa80 ("mm: add merging after mremap resize")
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Fabian Vogt <fvogt@suse.com>
Link: https://bugzilla.suse.com/show_bug.cgi?id=1206359#c35
Tested-by: Fabian Vogt <fvogt@suse.com>
Cc: Jakub Matěna <matenajakub@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm/mremap.c

index fe587c5d65913ab19efc27ff0f7261bd58516dd5..930f65c315c02469bd72f1608e69a4f4131256f1 100644 (file)
@@ -1027,16 +1027,29 @@ SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
                        }
 
                        /*
-                        * Function vma_merge() is called on the extension we are adding to
-                        * the already existing vma, vma_merge() will merge this extension with
-                        * the already existing vma (expand operation itself) and possibly also
-                        * with the next vma if it becomes adjacent to the expanded vma and
-                        * otherwise compatible.
+                        * Function vma_merge() is called on the extension we
+                        * are adding to the already existing vma, vma_merge()
+                        * will merge this extension with the already existing
+                        * vma (expand operation itself) and possibly also with
+                        * the next vma if it becomes adjacent to the expanded
+                        * vma and  otherwise compatible.
+                        *
+                        * However, vma_merge() can currently fail due to
+                        * is_mergeable_vma() check for vm_ops->close (see the
+                        * comment there). Yet this should not prevent vma
+                        * expanding, so perform a simple expand for such vma.
+                        * Ideally the check for close op should be only done
+                        * when a vma would be actually removed due to a merge.
                         */
-                       vma = vma_merge(mm, vma, extension_start, extension_end,
+                       if (!vma->vm_ops || !vma->vm_ops->close) {
+                               vma = vma_merge(mm, vma, extension_start, extension_end,
                                        vma->vm_flags, vma->anon_vma, vma->vm_file,
                                        extension_pgoff, vma_policy(vma),
                                        vma->vm_userfaultfd_ctx, anon_vma_name(vma));
+                       } else if (vma_adjust(vma, vma->vm_start, addr + new_len,
+                                  vma->vm_pgoff, NULL)) {
+                               vma = NULL;
+                       }
                        if (!vma) {
                                vm_unacct_memory(pages);
                                ret = -ENOMEM;