Linux 6.3 compat: idmapped mount API changes

Linux kernel 6.3 changed a bunch of APIs to use the dedicated idmap 
type for mounts (struct mnt_idmap), we need to detect these changes 
and make zfs work with the new APIs.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Youzhong Yang <yyang@mathworks.com>
Closes #14682

include/os/freebsd/spl/sys/types.h

@@ -105,7 +105,7 @@ typedef	u_longlong_t	len_t;
 
 typedef	longlong_t	diskaddr_t;
 
-typedef void		zuserns_t;
+typedef void		zidmap_t;
 
 #include <sys/debug.h>
 #endif	/* !_OPENSOLARIS_SYS_TYPES_H_ */

include/os/freebsd/zfs/sys/zfs_vnops_os.h

@@ -35,23 +35,23 @@ int dmu_read_pages(objset_t *os, uint64_t object, vm_page_t *ma, int count,
     int *rbehind, int *rahead, int last_size);
 extern int zfs_remove(znode_t *dzp, const char *name, cred_t *cr, int flags);
 extern int zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap,
-    znode_t **zpp, cred_t *cr, int flags, vsecattr_t *vsecp, zuserns_t *mnt_ns);
+    znode_t **zpp, cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns);
 extern int zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd,
     cred_t *cr, int flags);
 extern int zfs_setattr(znode_t *zp, vattr_t *vap, int flag, cred_t *cr,
-    zuserns_t *mnt_ns);
+    zidmap_t *mnt_ns);
 extern int zfs_rename(znode_t *sdzp, const char *snm, znode_t *tdzp,
     const char *tnm, cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap,
-    zuserns_t *mnt_ns);
+    zidmap_t *mnt_ns);
 extern int zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap,
-    const char *link, znode_t **zpp, cred_t *cr, int flags, zuserns_t *mnt_ns);
+    const char *link, znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns);
 extern int zfs_link(znode_t *tdzp, znode_t *sp,
     const char *name, cred_t *cr, int flags);
 extern int zfs_space(znode_t *zp, int cmd, struct flock *bfp, int flag,
     offset_t offset, cred_t *cr);
 extern int zfs_create(znode_t *dzp, const char *name, vattr_t *vap, int excl,
     int mode, znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp,
-    zuserns_t *mnt_ns);
+    zidmap_t *mnt_ns);
 extern int zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag,
     cred_t *cr);
 extern int zfs_write_simple(znode_t *zp, const void *data, size_t len,

include/os/linux/kernel/linux/vfs_compat.h

@@ -341,7 +341,8 @@ static inline void zfs_gid_write(struct inode *ip, gid_t gid)
  * 4.9 API change
  */
 #if !(defined(HAVE_SETATTR_PREPARE_NO_USERNS) || \
-    defined(HAVE_SETATTR_PREPARE_USERNS))
+    defined(HAVE_SETATTR_PREPARE_USERNS) || \
+    defined(HAVE_SETATTR_PREPARE_IDMAP))
 static inline int
 setattr_prepare(struct dentry *dentry, struct iattr *ia)
 {
@@ -396,6 +397,15 @@ func(struct user_namespace *user_ns, const struct path *path,	\
 	return (func##_impl(user_ns, path, stat, request_mask, \
 	    query_flags));	\
 }
+#elif defined(HAVE_IDMAP_IOPS_GETATTR)
+#define	ZPL_GETATTR_WRAPPER(func)					\
+static int								\
+func(struct mnt_idmap *user_ns, const struct path *path,	\
+    struct kstat *stat, u32 request_mask, unsigned int query_flags)	\
+{									\
+	return (func##_impl(user_ns, path, stat, request_mask,	\
+	    query_flags));	\
+}
 #else
 #error
 #endif
@@ -447,8 +457,15 @@ zpl_is_32bit_api(void)
  * 5.12 API change
  * To support id-mapped mounts, generic_fillattr() was modified to
  * accept a new struct user_namespace* as its first arg.
+ *
+ * 6.3 API change
+ * generic_fillattr() first arg is changed to struct mnt_idmap *
+ *
  */
-#ifdef HAVE_GENERIC_FILLATTR_USERNS
+#ifdef HAVE_GENERIC_FILLATTR_IDMAP
+#define	zpl_generic_fillattr(idmap, ip, sp)	\
+    generic_fillattr(idmap, ip, sp)
+#elif defined(HAVE_GENERIC_FILLATTR_USERNS)
 #define	zpl_generic_fillattr(user_ns, ip, sp)	\
     generic_fillattr(user_ns, ip, sp)
 #else

include/os/linux/kernel/linux/xattr_compat.h

@@ -133,13 +133,28 @@ fn(const struct xattr_handler *handler, struct dentry *dentry,		\
 #error "Unsupported kernel"
 #endif
 
+/*
+ * 6.3 API change,
+ * The xattr_handler->set() callback was changed to take the
+ * struct mnt_idmap* as the first arg, to support idmapped
+ * mounts.
+ */
+#if defined(HAVE_XATTR_SET_IDMAP)
+#define	ZPL_XATTR_SET_WRAPPER(fn)					\
+static int								\
+fn(const struct xattr_handler *handler, struct mnt_idmap *user_ns,	\
+    struct dentry *dentry, struct inode *inode, const char *name,	\
+    const void *buffer, size_t size, int flags)	\
+{									\
+	return (__ ## fn(user_ns, inode, name, buffer, size, flags));	\
+}
 /*
  * 5.12 API change,
  * The xattr_handler->set() callback was changed to take the
  * struct user_namespace* as the first arg, to support idmapped
  * mounts.
  */
-#if defined(HAVE_XATTR_SET_USERNS)
+#elif defined(HAVE_XATTR_SET_USERNS)
 #define	ZPL_XATTR_SET_WRAPPER(fn)					\
 static int								\
 fn(const struct xattr_handler *handler, struct user_namespace *user_ns, \

include/os/linux/spl/sys/cred.h

@@ -48,6 +48,8 @@ extern struct task_struct init_task;
 #define	SGID_TO_KGID(x)		(KGIDT_INIT(x))
 #define	KGIDP_TO_SGIDP(x)	(&(x)->val)
 
+extern zidmap_t *zfs_get_init_idmap(void);
+
 /* Check if the user ns is the initial one */
 static inline boolean_t
 zfs_is_init_userns(struct user_namespace *user_ns)
@@ -74,36 +76,39 @@ static inline boolean_t zfs_no_idmapping(struct user_namespace *mnt_userns,
 	return (zfs_is_init_userns(mnt_userns) || mnt_userns == fs_userns);
 }
 
-static inline uid_t zfs_uid_to_vfsuid(struct user_namespace *mnt_userns,
+static inline uid_t zfs_uid_to_vfsuid(zidmap_t *mnt_userns,
     struct user_namespace *fs_userns, uid_t uid)
 {
-	if (zfs_no_idmapping(mnt_userns, fs_userns))
+	struct user_namespace *owner = idmap_owner(mnt_userns);
+	if (zfs_no_idmapping(owner, fs_userns))
 		return (uid);
 	if (!zfs_is_init_userns(fs_userns))
 		uid = from_kuid(fs_userns, KUIDT_INIT(uid));
 	if (uid == (uid_t)-1)
 		return (uid);
-	return (__kuid_val(make_kuid(mnt_userns, uid)));
+	return (__kuid_val(make_kuid(owner, uid)));
 }
 
-static inline gid_t zfs_gid_to_vfsgid(struct user_namespace *mnt_userns,
+static inline gid_t zfs_gid_to_vfsgid(zidmap_t *mnt_userns,
     struct user_namespace *fs_userns, gid_t gid)
 {
-	if (zfs_no_idmapping(mnt_userns, fs_userns))
+	struct user_namespace *owner = idmap_owner(mnt_userns);
+	if (zfs_no_idmapping(owner, fs_userns))
 		return (gid);
 	if (!zfs_is_init_userns(fs_userns))
 		gid = from_kgid(fs_userns, KGIDT_INIT(gid));
 	if (gid == (gid_t)-1)
 		return (gid);
-	return (__kgid_val(make_kgid(mnt_userns, gid)));
+	return (__kgid_val(make_kgid(owner, gid)));
 }
 
-static inline uid_t zfs_vfsuid_to_uid(struct user_namespace *mnt_userns,
+static inline uid_t zfs_vfsuid_to_uid(zidmap_t *mnt_userns,
     struct user_namespace *fs_userns, uid_t uid)
 {
-	if (zfs_no_idmapping(mnt_userns, fs_userns))
+	struct user_namespace *owner = idmap_owner(mnt_userns);
+	if (zfs_no_idmapping(owner, fs_userns))
 		return (uid);
-	uid = from_kuid(mnt_userns, KUIDT_INIT(uid));
+	uid = from_kuid(owner, KUIDT_INIT(uid));
 	if (uid == (uid_t)-1)
 		return (uid);
 	if (zfs_is_init_userns(fs_userns))
@@ -111,12 +116,13 @@ static inline uid_t zfs_vfsuid_to_uid(struct user_namespace *mnt_userns,
 	return (__kuid_val(make_kuid(fs_userns, uid)));
 }
 
-static inline gid_t zfs_vfsgid_to_gid(struct user_namespace *mnt_userns,
+static inline gid_t zfs_vfsgid_to_gid(zidmap_t *mnt_userns,
     struct user_namespace *fs_userns, gid_t gid)
 {
-	if (zfs_no_idmapping(mnt_userns, fs_userns))
+	struct user_namespace *owner = idmap_owner(mnt_userns);
+	if (zfs_no_idmapping(owner, fs_userns))
 		return (gid);
-	gid = from_kgid(mnt_userns, KGIDT_INIT(gid));
+	gid = from_kgid(owner, KGIDT_INIT(gid));
 	if (gid == (gid_t)-1)
 		return (gid);
 	if (zfs_is_init_userns(fs_userns))

include/os/linux/spl/sys/types.h

@@ -55,6 +55,19 @@ typedef int			major_t;
 typedef int			minor_t;
 
 struct user_namespace;
-typedef struct user_namespace	zuserns_t;
+#ifdef HAVE_IOPS_CREATE_IDMAP
+#include <linux/refcount.h>
+struct mnt_idmap {
+	struct user_namespace *owner;
+	refcount_t count;
+};
+typedef struct mnt_idmap	zidmap_t;
+#define	idmap_owner(p)	(((struct mnt_idmap *)p)->owner)
+#else
+typedef struct user_namespace	zidmap_t;
+#define	idmap_owner(p)	((struct user_namespace *)p)
+#endif
+
+extern zidmap_t *zfs_init_idmap;
 
 #endif	/* _SPL_TYPES_H */

include/os/linux/zfs/sys/policy.h

@@ -47,14 +47,14 @@ int secpolicy_vnode_create_gid(const cred_t *);
 int secpolicy_vnode_remove(const cred_t *);
 int secpolicy_vnode_setdac(const cred_t *, uid_t);
 int secpolicy_vnode_setid_retain(struct znode *, const cred_t *, boolean_t);
-int secpolicy_vnode_setids_setgids(const cred_t *, gid_t, zuserns_t *,
-    zuserns_t *);
+int secpolicy_vnode_setids_setgids(const cred_t *, gid_t, zidmap_t *,
+    struct user_namespace *);
 int secpolicy_zinject(const cred_t *);
 int secpolicy_zfs(const cred_t *);
 int secpolicy_zfs_proc(const cred_t *, proc_t *);
 void secpolicy_setid_clear(vattr_t *, cred_t *);
 int secpolicy_setid_setsticky_clear(struct inode *, vattr_t *,
-    const vattr_t *, cred_t *, zuserns_t *, zuserns_t *);
+    const vattr_t *, cred_t *, zidmap_t *, struct user_namespace *);
 int secpolicy_xvattr(xvattr_t *, uid_t, cred_t *, mode_t);
 int secpolicy_vnode_setattr(cred_t *, struct inode *, struct vattr *,
     const struct vattr *, int, int (void *, int, cred_t *), void *);

include/os/linux/zfs/sys/zfs_vnops_os.h

@@ -46,25 +46,24 @@ extern int zfs_lookup(znode_t *dzp, char *nm, znode_t **zpp, int flags,
     cred_t *cr, int *direntflags, pathname_t *realpnp);
 extern int zfs_create(znode_t *dzp, char *name, vattr_t *vap, int excl,
     int mode, znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp,
-    zuserns_t *mnt_ns);
+    zidmap_t *mnt_ns);
 extern int zfs_tmpfile(struct inode *dip, vattr_t *vapzfs, int excl,
     int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp,
-    zuserns_t *mnt_ns);
+    zidmap_t *mnt_ns);
 extern int zfs_remove(znode_t *dzp, char *name, cred_t *cr, int flags);
 extern int zfs_mkdir(znode_t *dzp, char *dirname, vattr_t *vap,
-    znode_t **zpp, cred_t *cr, int flags, vsecattr_t *vsecp, zuserns_t *mnt_ns);
+    znode_t **zpp, cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns);
 extern int zfs_rmdir(znode_t *dzp, char *name, znode_t *cwd,
     cred_t *cr, int flags);
 extern int zfs_readdir(struct inode *ip, zpl_dir_context_t *ctx, cred_t *cr);
-extern int zfs_getattr_fast(struct user_namespace *, struct inode *ip,
-	struct kstat *sp);
+extern int zfs_getattr_fast(zidmap_t *, struct inode *ip, struct kstat *sp);
 extern int zfs_setattr(znode_t *zp, vattr_t *vap, int flag, cred_t *cr,
-    zuserns_t *mnt_ns);
+    zidmap_t *mnt_ns);
 extern int zfs_rename(znode_t *sdzp, char *snm, znode_t *tdzp,
     char *tnm, cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap,
-    zuserns_t *mnt_ns);
+    zidmap_t *mnt_ns);
 extern int zfs_symlink(znode_t *dzp, char *name, vattr_t *vap,
-    char *link, znode_t **zpp, cred_t *cr, int flags, zuserns_t *mnt_ns);
+    char *link, znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns);
 extern int zfs_readlink(struct inode *ip, zfs_uio_t *uio, cred_t *cr);
 extern int zfs_link(znode_t *tdzp, znode_t *szp,
     char *name, cred_t *cr, int flags);

include/os/linux/zfs/sys/zpl.h

@@ -39,7 +39,7 @@
 
 /* zpl_inode.c */
 extern void zpl_vap_init(vattr_t *vap, struct inode *dir,
-    umode_t mode, cred_t *cr, zuserns_t *mnt_ns);
+    umode_t mode, cred_t *cr, zidmap_t *mnt_ns);
 
 extern const struct inode_operations zpl_inode_operations;
 #ifdef HAVE_RENAME2_OPERATIONS_WRAPPER
@@ -68,7 +68,10 @@ extern int zpl_xattr_security_init(struct inode *ip, struct inode *dip,
     const struct qstr *qstr);
 #if defined(CONFIG_FS_POSIX_ACL)
 #if defined(HAVE_SET_ACL)
-#if defined(HAVE_SET_ACL_USERNS)
+#if defined(HAVE_SET_ACL_IDMAP_DENTRY)
+extern int zpl_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
+    struct posix_acl *acl, int type);
+#elif defined(HAVE_SET_ACL_USERNS)
 extern int zpl_set_acl(struct user_namespace *userns, struct inode *ip,
     struct posix_acl *acl, int type);
 #elif defined(HAVE_SET_ACL_USERNS_DENTRY_ARG2)
@@ -189,13 +192,15 @@ zpl_dir_emit_dots(struct file *file, zpl_dir_context_t *ctx)
 
 #if defined(HAVE_INODE_OWNER_OR_CAPABLE)
 #define	zpl_inode_owner_or_capable(ns, ip)	inode_owner_or_capable(ip)
-#elif defined(HAVE_INODE_OWNER_OR_CAPABLE_IDMAPPED)
+#elif defined(HAVE_INODE_OWNER_OR_CAPABLE_USERNS)
 #define	zpl_inode_owner_or_capable(ns, ip)	inode_owner_or_capable(ns, ip)
+#elif defined(HAVE_INODE_OWNER_OR_CAPABLE_IDMAP)
+#define	zpl_inode_owner_or_capable(idmap, ip) inode_owner_or_capable(idmap, ip)
 #else
 #error "Unsupported kernel"
 #endif
 
-#ifdef HAVE_SETATTR_PREPARE_USERNS
+#if defined(HAVE_SETATTR_PREPARE_USERNS) || defined(HAVE_SETATTR_PREPARE_IDMAP)
 #define	zpl_setattr_prepare(ns, dentry, ia)	setattr_prepare(ns, dentry, ia)
 #else
 /*

include/sys/zfs_acl.h

@@ -206,7 +206,7 @@ struct zfsvfs;
 
 #ifdef _KERNEL
 int zfs_acl_ids_create(struct znode *, int, vattr_t *,
-    cred_t *, vsecattr_t *, zfs_acl_ids_t *, zuserns_t *);
+    cred_t *, vsecattr_t *, zfs_acl_ids_t *, zidmap_t *);
 void zfs_acl_ids_free(zfs_acl_ids_t *);
 boolean_t zfs_acl_ids_overquota(struct zfsvfs *, zfs_acl_ids_t *, uint64_t);
 int zfs_getacl(struct znode *, vsecattr_t *, boolean_t, cred_t *);
@@ -216,15 +216,15 @@ void zfs_oldace_byteswap(ace_t *, int);
 void zfs_ace_byteswap(void *, size_t, boolean_t);
 extern boolean_t zfs_has_access(struct znode *zp, cred_t *cr);
 extern int zfs_zaccess(struct znode *, int, int, boolean_t, cred_t *,
-    zuserns_t *);
+    zidmap_t *);
 int zfs_fastaccesschk_execute(struct znode *, cred_t *);
-extern int zfs_zaccess_rwx(struct znode *, mode_t, int, cred_t *, zuserns_t *);
+extern int zfs_zaccess_rwx(struct znode *, mode_t, int, cred_t *, zidmap_t *);
 extern int zfs_zaccess_unix(void *, int, cred_t *);
 extern int zfs_acl_access(struct znode *, int, cred_t *);
 int zfs_acl_chmod_setattr(struct znode *, zfs_acl_t **, uint64_t);
-int zfs_zaccess_delete(struct znode *, struct znode *, cred_t *, zuserns_t *);
+int zfs_zaccess_delete(struct znode *, struct znode *, cred_t *, zidmap_t *);
 int zfs_zaccess_rename(struct znode *, struct znode *,
-    struct znode *, struct znode *, cred_t *cr, zuserns_t *mnt_ns);
+    struct znode *, struct znode *, cred_t *cr, zidmap_t *mnt_ns);
 void zfs_acl_free(zfs_acl_t *);
 int zfs_vsec_2_aclp(struct zfsvfs *, umode_t, vsecattr_t *, cred_t *,
     struct zfs_fuid_info **, zfs_acl_t **);