[PATCH] KVM: guest_memfd: Reject large folios until support is implemented

[PATCH] KVM: guest_memfd: Reject large folios until support is implemented

[Deepanshu Kartikey]

Large folios are not yet supported in guest_memfd (see TODO comment
in kvm_gmem_get_folio()), but can still be allocated if userspace
uses madvise(MADV_HUGEPAGE), which overrides the folio order
restrictions set by mapping_set_folio_order_range().

When a large folio is allocated, it triggers WARN_ON_ONCE() at line
416 in kvm_gmem_fault_user_mapping(), causing a kernel panic if
panic_on_warn is enabled.

Add mapping_set_folio_order_range(0, 0) as defense in depth, and
actively check for large folios in kvm_gmem_get_folio() on both
the fast-path (existing folio) and slow-path (newly created folio).
If a large folio is found, unlock it, drop the reference, and return
-E2BIG to prevent the WARNING from triggering.

This avoids kernel panics when panic_on_warn is enabled.

Reported-by: syzbot+33a04338019ac7e43a44@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=33a04338019ac7e43a44
Fixes: b85524314a3d ("KVM: guest_memfd: delay kvm_gmem_prepare_folio() until the memory is passed to the guest")
Tested-by: syzbot+33a04338019ac7e43a44@syzkaller.appspotmail.com
Signed-off-by: Deepanshu Kartikey <Kartikey406@gmail.com>
---
 virt/kvm/guest_memfd.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/virt/kvm/guest_memfd.c b/virt/kvm/guest_memfd.c
index fdaea3422c30..ee5bcf238f98 100644
--- a/virt/kvm/guest_memfd.c
+++ b/virt/kvm/guest_memfd.c
@@ -143,13 +143,29 @@ static struct folio *kvm_gmem_get_folio(struct inode *inode, pgoff_t index)
 	folio = __filemap_get_folio(inode->i_mapping, index,
 				    FGP_LOCK | FGP_ACCESSED, 0);
 	if (!IS_ERR(folio))
-		return folio;
+		goto check_folio;
 
 	policy = mpol_shared_policy_lookup(&GMEM_I(inode)->policy, index);
 	folio = __filemap_get_folio_mpol(inode->i_mapping, index,
 					 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
 					 mapping_gfp_mask(inode->i_mapping), policy);
 	mpol_cond_put(policy);
+	if (IS_ERR(folio))
+		return folio;
+check_folio:
+	/*
+	 * Large folios are not supported yet. This can still happen
+	 * despite mapping_set_folio_order_range() if userspace uses
+	 * madvise(MADV_HUGEPAGE) which can override the folio order
+	 * restrictions. Reject the large folio and remove it from
+	 * the page cache so the next fault can allocate a order-0
+	 * page instead.
+	 */
+	if (folio_test_large(folio)) {
+		folio_unlock(folio);
+		folio_put(folio);
+		return ERR_PTR(-E2BIG);
+	}
 
 	return folio;
 }
@@ -596,6 +612,7 @@ static int __kvm_gmem_create(struct kvm *kvm, loff_t size, u64 flags)
 	inode->i_mode |= S_IFREG;
 	inode->i_size = size;
 	mapping_set_gfp_mask(inode->i_mapping, GFP_HIGHUSER);
+	mapping_set_folio_order_range(inode->i_mapping, 0, 0);
 	mapping_set_inaccessible(inode->i_mapping);
 	/* Unmovable mappings are supposed to be marked unevictable as well. */
 	WARN_ON_ONCE(!mapping_unevictable(inode->i_mapping));
-- 
2.43.0