[PATCH v22 08/12] landlock: Add syscall implementations

[Jann Horn]

On Tue, Oct 27, 2020 at 9:04 PM Mickaël Salaün <mic at digikod.net> wrote:
> These 3 system calls are designed to be used by unprivileged processes
> to sandbox themselves:
> * landlock_create_ruleset(2): Creates a ruleset and returns its file
>   descriptor.
> * landlock_add_rule(2): Adds a rule (e.g. file hierarchy access) to a
>   ruleset, identified by the dedicated file descriptor.
> * landlock_enforce_ruleset_current(2): Enforces a ruleset on the current
>   thread and its future children (similar to seccomp).  This syscall has
>   the same usage restrictions as seccomp(2): the caller must have the
>   no_new_privs attribute set or have CAP_SYS_ADMIN in the current user
>   namespace.
[...]
> Cc: Arnd Bergmann <arnd at arndb.de>
> Cc: James Morris <jmorris at namei.org>
> Cc: Jann Horn <jannh at google.com>
> Cc: Kees Cook <keescook at chromium.org>
> Cc: Serge E. Hallyn <serge at hallyn.com>
> Signed-off-by: Mickaël Salaün <mic at linux.microsoft.com>
[...]
> diff --git a/include/uapi/linux/landlock.h b/include/uapi/linux/landlock.h
[...]
> +/**
> + * struct landlock_path_beneath_attr - Path hierarchy definition
> + *
> + * Argument of sys_landlock_add_rule().
> + */
> +struct landlock_path_beneath_attr {
> +       /**
> +        * @allowed_access: Bitmask of allowed actions for this file hierarchy
> +        * (cf. `Filesystem flags`_).
> +        */
> +       __u64 allowed_access;
> +       /**
> +        * @parent_fd: File descriptor, open with ``O_PATH``, which identify

nit: "identifies"

> +        * the parent directory of a file hierarchy, or just a file.
> +        */
> +       __s32 parent_fd;
> +       /*
> +        * This struct is packed to avoid trailing reserved members.
> +        * Cf. security/landlock/syscall.c:build_check_abi()
> +        */
> +} __attribute__((packed));
[...]
> diff --git a/security/landlock/syscall.c b/security/landlock/syscall.c
[...]
> +static int copy_min_struct_from_user(void *const dst, const size_t ksize,
> +               const size_t ksize_min, const void __user *const src,
> +               const size_t usize)
> +{
> +       /* Checks buffer inconsistencies. */
> +       BUILD_BUG_ON(!dst);
> +       if (!src)
> +               return -EFAULT;
> +
> +       /* Checks size ranges. */
> +       BUILD_BUG_ON(ksize <= 0);
> +       BUILD_BUG_ON(ksize < ksize_min);

To make these checks work reliably, you should add __always_inline to
the function.

> +       if (usize < ksize_min)
> +               return -EINVAL;
> +       if (usize > PAGE_SIZE)
> +               return -E2BIG;
> +
> +       /* Copies user buffer and fills with zeros. */
> +       return copy_struct_from_user(dst, ksize, src, usize);
> +}
[...]
> +static int get_path_from_fd(const s32 fd, struct path *const path)
> +{
> +       struct fd f;
> +       int err = 0;
> +
> +       BUILD_BUG_ON(!__same_type(fd,
> +               ((struct landlock_path_beneath_attr *)NULL)->parent_fd));
> +
> +       /* Handles O_PATH. */
> +       f = fdget_raw(fd);
> +       if (!f.file)
> +               return -EBADF;
> +       /*
> +        * Only allows O_PATH file descriptor: enables to restrict ambient
> +        * filesystem access without requiring to open and risk leaking or
> +        * misusing a file descriptor.  Forbid internal filesystems (e.g.
> +        * nsfs), including pseudo filesystems that will never be mountable
> +        * (e.g. sockfs, pipefs).
> +        */
> +       if (!(f.file->f_mode & FMODE_PATH) ||
> +                       (f.file->f_path.mnt->mnt_flags & MNT_INTERNAL) ||
> +                       (f.file->f_path.dentry->d_sb->s_flags & SB_NOUSER) ||
> +                       d_is_negative(f.file->f_path.dentry) ||
> +                       IS_PRIVATE(d_backing_inode(f.file->f_path.dentry))) {
> +               err = -EBADFD;
> +               goto out_fdput;
> +       }
> +       path->mnt = f.file->f_path.mnt;
> +       path->dentry = f.file->f_path.dentry;

those two lines can be replaced with "*path = f.file->f_path"

> +       path_get(path);
> +
> +out_fdput:
> +       fdput(f);
> +       return err;
> +}
[...]
> +/**
> + * sys_landlock_enforce_ruleset_current - Enforce a ruleset on the current task
> + *
> + * @ruleset_fd: File descriptor tied to the ruleset to merge with the target.
> + * @flags: Must be 0.
> + *
> + * This system call enables to enforce a Landlock ruleset on the current
> + * thread.  Enforcing a ruleset requires that the task has CAP_SYS_ADMIN in its
> + * namespace or be running with no_new_privs.  This avoids scenarios where

s/be/is/

> + * unprivileged tasks can affect the behavior of privileged children.
> + *
> + * Possible returned errors are:
> + *
> + * - EOPNOTSUPP: Landlock is supported by the kernel but disabled at boot time;
> + * - EINVAL: @flags is not 0.
> + * - EBADF: @ruleset_fd is not a file descriptor for the current thread;
> + * - EBADFD: @ruleset_fd is not a ruleset file descriptor;
> + * - EPERM: @ruleset_fd has no read access to the underlying ruleset, or the
> + *   current thread is not running with no_new_privs (or doesn't have
> + *   CAP_SYS_ADMIN in its namespace).
> + */
> +SYSCALL_DEFINE2(landlock_enforce_ruleset_current,
> +               const int, ruleset_fd, const __u32, flags)
> +{
> +       struct landlock_ruleset *new_dom, *ruleset;
> +       struct cred *new_cred;
> +       struct landlock_cred_security *new_llcred;
> +       int err;
> +
> +       if (!landlock_initialized)
> +               return -EOPNOTSUPP;
> +
> +       /* No flag for now. */
> +       if (flags)
> +               return -EINVAL;
> +
> +       /*
> +        * Similar checks as for seccomp(2), except that an -EPERM may be
> +        * returned.
> +        */
> +       if (!task_no_new_privs(current)) {
> +               err = security_capable(current_cred(), current_user_ns(),
> +                               CAP_SYS_ADMIN, CAP_OPT_NOAUDIT);

I think this should be ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN)?

> +               if (err)
> +                       return err;
> +       }
> +
> +       /* Gets and checks the ruleset. */
> +       ruleset = get_ruleset_from_fd(ruleset_fd, FMODE_CAN_READ);
> +       if (IS_ERR(ruleset))
> +               return PTR_ERR(ruleset);
> +
> +       /* Prepares new credentials. */
> +       new_cred = prepare_creds();
> +       if (!new_cred) {
> +               err = -ENOMEM;
> +               goto out_put_ruleset;
> +       }
> +       new_llcred = landlock_cred(new_cred);
> +
> +       /*
> +        * There is no possible race condition while copying and manipulating
> +        * the current credentials because they are dedicated per thread.
> +        */
> +       new_dom = landlock_merge_ruleset(new_llcred->domain, ruleset);
> +       if (IS_ERR(new_dom)) {
> +               err = PTR_ERR(new_dom);
> +               goto out_put_creds;
> +       }
> +
> +       /* Replaces the old (prepared) domain. */
> +       landlock_put_ruleset(new_llcred->domain);
> +       new_llcred->domain = new_dom;
> +
> +       landlock_put_ruleset(ruleset);
> +       return commit_creds(new_cred);
> +
> +out_put_creds:
> +       abort_creds(new_cred);
> +       return err;

I think this "return err" is wrong - don't we still have to put "ruleset" here?


> +out_put_ruleset:
> +       landlock_put_ruleset(ruleset);
> +       return err;
> +}
> --
> 2.28.0
>