]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
KVM: guest_memfd: Add slab-allocated inode cache
authorShivank Garg <shivankg@amd.com>
Thu, 16 Oct 2025 17:28:45 +0000 (10:28 -0700)
committerSean Christopherson <seanjc@google.com>
Mon, 20 Oct 2025 13:30:40 +0000 (06:30 -0700)
Add a dedicated gmem_inode structure and a slab-allocated inode cache for
guest memory backing, similar to how shmem handles inodes.

This adds the necessary allocation/destruction functions and prepares
for upcoming guest_memfd NUMA policy support changes.  Using a dedicated
structure will also allow for additional cleanups, e.g. to track flags in
gmem_inode instead of i_private.

Signed-off-by: Shivank Garg <shivankg@amd.com>
Tested-by: Ashish Kalra <ashish.kalra@amd.com>
[sean: s/kvm_gmem_inode_info/gmem_inode, name init_once()]
Reviewed-by: Ackerley Tng <ackerleytng@google.com>
Tested-by: Ackerley Tng <ackerleytng@google.com>
Link: https://lore.kernel.org/r/20251016172853.52451-5-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
virt/kvm/guest_memfd.c

index ce04fc85e6312964b7dee6a9490f61962870544b..88fd812f0f3138ac652127a6b6b06e423d70f432 100644 (file)
@@ -26,6 +26,15 @@ struct gmem_file {
        struct list_head entry;
 };
 
+struct gmem_inode {
+       struct inode vfs_inode;
+};
+
+static __always_inline struct gmem_inode *GMEM_I(struct inode *inode)
+{
+       return container_of(inode, struct gmem_inode, vfs_inode);
+}
+
 #define kvm_gmem_for_each_file(f, mapping) \
        list_for_each_entry(f, &(mapping)->i_private_list, entry)
 
@@ -830,13 +839,61 @@ put_folio_and_exit:
 EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_gmem_populate);
 #endif
 
+static struct kmem_cache *kvm_gmem_inode_cachep;
+
+static void kvm_gmem_init_inode_once(void *__gi)
+{
+       struct gmem_inode *gi = __gi;
+
+       /*
+        * Note!  Don't initialize the inode with anything specific to the
+        * guest_memfd instance, or that might be specific to how the inode is
+        * used (from the VFS-layer's perspective).  This hook is called only
+        * during the initial slab allocation, i.e. only fields/state that are
+        * idempotent across _all_ use of the inode _object_ can be initialized
+        * at this time!
+        */
+       inode_init_once(&gi->vfs_inode);
+}
+
+static struct inode *kvm_gmem_alloc_inode(struct super_block *sb)
+{
+       struct gmem_inode *gi;
+
+       gi = alloc_inode_sb(sb, kvm_gmem_inode_cachep, GFP_KERNEL);
+       if (!gi)
+               return NULL;
+
+       return &gi->vfs_inode;
+}
+
+static void kvm_gmem_destroy_inode(struct inode *inode)
+{
+}
+
+static void kvm_gmem_free_inode(struct inode *inode)
+{
+       kmem_cache_free(kvm_gmem_inode_cachep, GMEM_I(inode));
+}
+
+static const struct super_operations kvm_gmem_super_operations = {
+       .statfs         = simple_statfs,
+       .alloc_inode    = kvm_gmem_alloc_inode,
+       .destroy_inode  = kvm_gmem_destroy_inode,
+       .free_inode     = kvm_gmem_free_inode,
+};
+
 static int kvm_gmem_init_fs_context(struct fs_context *fc)
 {
+       struct pseudo_fs_context *ctx;
+
        if (!init_pseudo(fc, GUEST_MEMFD_MAGIC))
                return -ENOMEM;
 
        fc->s_iflags |= SB_I_NOEXEC;
        fc->s_iflags |= SB_I_NODEV;
+       ctx = fc->fs_private;
+       ctx->ops = &kvm_gmem_super_operations;
 
        return 0;
 }
@@ -860,13 +917,31 @@ static int kvm_gmem_init_mount(void)
 
 int kvm_gmem_init(struct module *module)
 {
+       struct kmem_cache_args args = {
+               .align = 0,
+               .ctor = kvm_gmem_init_inode_once,
+       };
+       int ret;
+
        kvm_gmem_fops.owner = module;
+       kvm_gmem_inode_cachep = kmem_cache_create("kvm_gmem_inode_cache",
+                                                 sizeof(struct gmem_inode),
+                                                 &args, SLAB_ACCOUNT);
+       if (!kvm_gmem_inode_cachep)
+               return -ENOMEM;
 
-       return kvm_gmem_init_mount();
+       ret = kvm_gmem_init_mount();
+       if (ret) {
+               kmem_cache_destroy(kvm_gmem_inode_cachep);
+               return ret;
+       }
+       return 0;
 }
 
 void kvm_gmem_exit(void)
 {
        kern_unmount(kvm_gmem_mnt);
        kvm_gmem_mnt = NULL;
+       rcu_barrier();
+       kmem_cache_destroy(kvm_gmem_inode_cachep);
 }