linux/super: match vfs_t lifetime to fs_context

vfs_t is initially just parameters for the mount or remount operation,
so match them to the lifetime of the fs_context that represents that
operation.

When we actually execute the operation (calling .get_tree or .reconfigure),
transfer ownership of those options to the associated zfsvfs_t.

Sponsored-by: TrueNAS
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@truenas.com>
Closes #18377

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

@@ -76,7 +76,7 @@ typedef struct vfs {
 
 typedef struct zfs_mnt {
 	const char	*mnt_osname;	/* Objset name */
-	char		*mnt_data;	/* Raw mount options */
+	vfs_t		*mnt_opts;	/* Parsed options */
 } zfs_mnt_t;
 
 struct zfsvfs {

module/os/linux/zfs/zfs_vfsops.c

@@ -1323,7 +1323,7 @@ zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
 	uint64_t recordsize;
 	int error = 0;
 	zfsvfs_t *zfsvfs = NULL;
-	vfs_t *vfs = NULL;
+	vfs_t *vfs = zm->mnt_opts;
 	int canwrite;
 	int dataset_visible_zone;
 
@@ -1341,9 +1341,6 @@ zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
 		return (SET_ERROR(EPERM));
 	}
 
-	/* XXX temp keep things working during params upgrade */
-	vfs = zfsvfs_vfs_alloc();
-
 	/*
 	 * If a non-writable filesystem is being mounted without the
 	 * read-only flag, pretend it was set, as done for snapshots.
@@ -1352,16 +1349,12 @@ zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
 		vfs->vfs_readonly = B_TRUE;
 
 	error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs);
-	if (error) {
+	if (error)
-		zfsvfs_vfs_free(vfs);
 		goto out;
-	}
 
 	if ((error = dsl_prop_get_integer(osname, "recordsize",
-	    &recordsize, NULL))) {
+	    &recordsize, NULL)))
-		zfsvfs_vfs_free(vfs);
 		goto out;
-	}
 
 	vfs->vfs_data = zfsvfs;
 	zfsvfs->z_vfs = vfs;
@@ -1443,6 +1436,13 @@ zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
 out:
 	if (error) {
 		if (zfsvfs != NULL) {
+			/*
+			 * We're returning error, so the caller still owns
+			 * the mount options vfs_t. Remove them from zfsvfs
+			 * so we don't try to free them.
+			 */
+			zfsvfs->z_vfs = NULL;
+
 			dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
 			zfsvfs_free(zfsvfs);
 		}
@@ -1536,7 +1536,7 @@ int
 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
 {
 	zfsvfs_t *zfsvfs = sb->s_fs_info;
-	vfs_t *vfsp;
+	vfs_t *vfsp = zm->mnt_opts;
 	boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
 
 	if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
@@ -1545,9 +1545,6 @@ zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
 		return (EROFS);
 	}
 
-	/* XXX temp keep things working during params upgrade */
-	vfsp = zfsvfs_vfs_alloc();
-
 	if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY))
 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
 

module/os/linux/zfs/zpl_super.c

@@ -448,7 +448,7 @@ zpl_get_tree(struct fs_context *fc)
 	if (sb->s_root == NULL) {
 		zfs_mnt_t zm = {
 		    .mnt_osname = fc->source,
-		    .mnt_data = NULL,
+		    .mnt_opts = fc->fs_private,
 		};
 
 		fstrans_cookie_t cookie = spl_fstrans_mark();
@@ -460,6 +460,12 @@ zpl_get_tree(struct fs_context *fc)
 			return (-err);
 		}
 
+		/*
+		 * zfsvfs has taken ownership of the mount options, so we
+		 * need to ensure we don't free them.
+		 */
+		fc->fs_private = NULL;
+
 		sb->s_flags |= SB_ACTIVE;
 	} else if (!issnap && ((fc->sb_flags ^ sb->s_flags) & SB_RDONLY)) {
 		/*
@@ -481,7 +487,7 @@ zpl_get_tree(struct fs_context *fc)
 static int
 zpl_reconfigure(struct fs_context *fc)
 {
-	zfs_mnt_t zm = { .mnt_osname = NULL, .mnt_data = NULL };
+	zfs_mnt_t zm = { .mnt_osname = NULL, .mnt_opts = fc->fs_private };
 	fstrans_cookie_t cookie;
 	int error;
 
@@ -490,18 +496,80 @@ zpl_reconfigure(struct fs_context *fc)
 	spl_fstrans_unmark(cookie);
 	ASSERT3S(error, <=, 0);
 
+	if (error == 0) {
+		/*
+		 * zfsvfs has taken ownership of the mount options, so we
+		 * need to ensure we don't free them.
+		 */
+		fc->fs_private = NULL;
+	}
+
 	return (error);
 }
 
+static int
+zpl_dup_fc(struct fs_context *fc, struct fs_context *src_fc)
+{
+	vfs_t *src_vfs = src_fc->fs_private;
+	if (src_vfs == NULL)
+		return (0);
+
+	vfs_t *vfs = zfsvfs_vfs_alloc();
+	if (vfs == NULL)
+		return (-SET_ERROR(ENOMEM));
+
+	/*
+	 * This is annoying, but a straight memcpy() would require us to
+	 * reinitialise the lock.
+	 */
+	vfs->vfs_xattr = src_vfs->vfs_xattr;
+	vfs->vfs_readonly = src_vfs->vfs_readonly;
+	vfs->vfs_do_readonly = src_vfs->vfs_do_readonly;
+	vfs->vfs_setuid = src_vfs->vfs_setuid;
+	vfs->vfs_do_setuid = src_vfs->vfs_do_setuid;
+	vfs->vfs_exec = src_vfs->vfs_exec;
+	vfs->vfs_do_exec = src_vfs->vfs_do_exec;
+	vfs->vfs_devices = src_vfs->vfs_devices;
+	vfs->vfs_do_devices = src_vfs->vfs_do_devices;
+	vfs->vfs_do_xattr = src_vfs->vfs_do_xattr;
+	vfs->vfs_atime = src_vfs->vfs_atime;
+	vfs->vfs_do_atime = src_vfs->vfs_do_atime;
+	vfs->vfs_relatime = src_vfs->vfs_relatime;
+	vfs->vfs_do_relatime = src_vfs->vfs_do_relatime;
+	vfs->vfs_nbmand = src_vfs->vfs_nbmand;
+	vfs->vfs_do_nbmand = src_vfs->vfs_do_nbmand;
+
+	mutex_enter(&src_vfs->vfs_mntpt_lock);
+	if (src_vfs->vfs_mntpoint != NULL)
+		vfs->vfs_mntpoint = kmem_strdup(src_vfs->vfs_mntpoint);
+	mutex_exit(&src_vfs->vfs_mntpt_lock);
+
+	fc->fs_private = vfs;
+	return (0);
+}
+
+static void
+zpl_free_fc(struct fs_context *fc)
+{
+	zfsvfs_vfs_free(fc->fs_private);
+}
+
 const struct fs_context_operations zpl_fs_context_operations = {
 	.get_tree		= zpl_get_tree,
 	.reconfigure		= zpl_reconfigure,
+	.dup			= zpl_dup_fc,
+	.free			= zpl_free_fc,
 };
 
 static int
 zpl_init_fs_context(struct fs_context *fc)
 {
+	fc->fs_private = zfsvfs_vfs_alloc();
+	if (fc->fs_private == NULL)
+		return (-SET_ERROR(ENOMEM));
+
 	fc->ops = &zpl_fs_context_operations;
+
 	return (0);
 }