Re: [PATCH v2] mm/secretmem: Keep locked memory accounting until inode eviction

From: Mike Rapoport

Date: Sun Nov 22 2026 - 03:30:37 EST


Hi,

Please don't send vN+1 as a reply to vN next time please.

On Sun, Nov 22, 2026 at 01:40:20AM +0900, Baek Kyu, Kim wrote:
> A secretmem inode can outlive its original file. An O_PATH descriptor
> opened through /proc/self/fd keeps the inode and its unevictable folios
> alive after the mapping and original descriptor are closed. The file
> release path nevertheless drops the locked-memory charge, allowing the
> same user to reuse RLIMIT_MEMLOCK while those folios remain allocated.
>
> Move private-state and locked-memory accounting cleanup to inode eviction,
> after truncating the page cache. Keep secretmem_users active over the same
> lifetime so that the hibernation guard covers retained secret pages.
>
> Allocate private state before inode creation and release it if the
> secure-inode helper fails. Attach it immediately after creation so
> later failures can use iput() for cleanup. An LSM denial inside the
> helper can call iput() before i_private is set; skip private cleanup
> at eviction when i_private is NULL.
>
> On arm64 QEMU with an 8 KiB RLIMIT_MEMLOCK, the original kernel allowed
> an additional 4 KiB secretmem fault while O_PATH retained two pages.
> v2 rejected that fault with SIGBUS and allowed it after O_PATH closed.
> Injected LSM denial and file-allocation failure completed without a
> crash; file-allocation cleanup returned secretmem_users to zero.
> KASAN, lockdep, hibernation and long-running concurrency were not
> tested in this v2 run.
>
> Fixes: 97d34aa65c29 ("mm/secretmem: properly account locked pages")
> Signed-off-by: Baek Kyu, Kim <pocj8ur4in@xxxxxxxxx>
> ---
> Changes in v2:
> - Allocate state before inode creation, as Mike Rapoport suggested,
> and free it if the secure-inode helper fails.
> - Guard private-state cleanup in secretmem_evict_inode() because an
> LSM denial can call iput() before i_private is set.
>
> Link to v1: https://lists.openwall.net/linux-kernel/2026/10/30/1883
>
> mm/secretmem.c | 58 ++++++++++++++++++++++++++++------------------------------
> 1 file changed, 28 insertions(+), 30 deletions(-)
>
> diff --git a/mm/secretmem.c b/mm/secretmem.c
> index b3f79ad56790..5a7378b62bb2 100644
> --- a/mm/secretmem.c
> +++ b/mm/secretmem.c
> @@ -189,14 +189,7 @@ static void secretmem_destroy_inode_priv(struct inode *inode)
> free_uid(state->user);
> kfree(state);
> inode->i_private = NULL;
> -}
> -
> -static int secretmem_release(struct inode *inode, struct file *file)
> -{
> atomic_dec(&secretmem_users);
> - secretmem_destroy_inode_priv(inode);
> -
> - return 0;
> }
>
> static int secretmem_mmap_prepare(struct vm_area_desc *desc)
> @@ -216,7 +209,6 @@ bool vma_is_secretmem(struct vm_area_struct *vma)
> }
>
> static const struct file_operations secretmem_fops = {
> - .release = secretmem_release,
> .mmap_prepare = secretmem_mmap_prepare,
> };
>
> @@ -264,39 +256,34 @@ static const struct inode_operations secretmem_iops = {
>
> static struct vfsmount *secretmem_mnt;
>
> -static int secretmem_init_inode_priv(struct inode *inode)
> -{
> - struct secretmem_inode_state *state;
> -
> - state = kzalloc_obj(*state);
> - if (!state)
> - return -ENOMEM;
> -
> - state->user = get_uid(current_user());
> - inode->i_private = state;
> - return 0;
> -}
> -
> static struct file *secretmem_file_create(unsigned long flags)
> {
> + struct secretmem_inode_state *state;
> struct file *file;
> struct inode *inode;
> const char *anon_name = "[secretmem]";
> int err;
>
> + state = kzalloc_obj(*state);
> + if (!state)
> + return ERR_PTR(-ENOMEM);
> + state->user = get_uid(current_user());
> +
> inode = anon_inode_make_secure_inode(secretmem_mnt->mnt_sb, anon_name, NULL);
> - if (IS_ERR(inode))
> + if (IS_ERR(inode)) {
> + free_uid(state->user);
> + kfree(state);

Please don't do cleanups inside if (err), use goto.

See https://docs.kernel.org/process/coding-style.html#centralized-exiting-of-functions

> return ERR_CAST(inode);
> + }
>
> - err = secretmem_init_inode_priv(inode);
> - if (err)
> - goto err_free_inode;
> + inode->i_private = state;
> + atomic_inc(&secretmem_users);
>
> file = alloc_file_pseudo(inode, secretmem_mnt, "secretmem",
> O_RDWR | O_LARGEFILE, &secretmem_fops);
> if (IS_ERR(file)) {
> err = PTR_ERR(file);
> - goto err_free_priv;
> + goto err_free_inode;

I think here we lost free_uid().

> }
>
> mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
> @@ -309,11 +296,7 @@ static struct file *secretmem_file_create(unsigned long flags)
> inode->i_mode |= S_IFREG;
> inode->i_size = 0;
>
> - atomic_inc(&secretmem_users);
> -
> return file;
> -err_free_priv:
> - secretmem_destroy_inode_priv(inode);

What prevents moving this between iput() and return err?

> err_free_inode:
> iput(inode);
> return ERR_PTR(err);
> @@ -335,6 +318,20 @@ SYSCALL_DEFINE1(memfd_secret, unsigned int, flags)
> return FD_ADD(flags & O_CLOEXEC, secretmem_file_create(flags));
> }
>
> +static void secretmem_evict_inode(struct inode *inode)
> +{
> + truncate_inode_pages_final(&inode->i_data);
> + clear_inode(inode);
> + /* The filesystem root and failed secure inodes have no private state. */
> + if (inode->i_private)
> + secretmem_destroy_inode_priv(inode);
> +}
> +
> +static const struct super_operations secretmem_super_ops = {
> + .statfs = simple_statfs,
> + .evict_inode = secretmem_evict_inode,
> +};
> +
> static int secretmem_init_fs_context(struct fs_context *fc)
> {
> struct pseudo_fs_context *ctx;
> @@ -342,6 +339,7 @@ static int secretmem_init_fs_context(struct fs_context *fc)
> ctx = init_pseudo(fc, SECRETMEM_MAGIC);
> if (!ctx)
> return -ENOMEM;
> + ctx->ops = &secretmem_super_ops;
>
> return 0;
> }

--
Sincerely yours,
Mike.