aboutsummaryrefslogtreecommitdiffstats
path: root/module/zfs/zpl_super.c
diff options
context:
space:
mode:
Diffstat (limited to 'module/zfs/zpl_super.c')
-rw-r--r--module/zfs/zpl_super.c226
1 files changed, 205 insertions, 21 deletions
diff --git a/module/zfs/zpl_super.c b/module/zfs/zpl_super.c
index ecdbc36d8..3edacba46 100644
--- a/module/zfs/zpl_super.c
+++ b/module/zfs/zpl_super.c
@@ -184,35 +184,204 @@ zpl_statfs(struct dentry *dentry, struct kstatfs *statp)
return (error);
}
+enum {
+ TOKEN_RO,
+ TOKEN_RW,
+ TOKEN_SETUID,
+ TOKEN_NOSETUID,
+ TOKEN_EXEC,
+ TOKEN_NOEXEC,
+ TOKEN_DEVICES,
+ TOKEN_NODEVICES,
+ TOKEN_XATTR,
+ TOKEN_NOXATTR,
+ TOKEN_ATIME,
+ TOKEN_NOATIME,
+ TOKEN_RELATIME,
+ TOKEN_NORELATIME,
+ TOKEN_NBMAND,
+ TOKEN_NONBMAND,
+ TOKEN_MNTPOINT,
+ TOKEN_LAST,
+};
+
+static const match_table_t zpl_tokens = {
+ { TOKEN_RO, MNTOPT_RO },
+ { TOKEN_RW, MNTOPT_RW },
+ { TOKEN_SETUID, MNTOPT_SETUID },
+ { TOKEN_NOSETUID, MNTOPT_NOSETUID },
+ { TOKEN_EXEC, MNTOPT_EXEC },
+ { TOKEN_NOEXEC, MNTOPT_NOEXEC },
+ { TOKEN_DEVICES, MNTOPT_DEVICES },
+ { TOKEN_NODEVICES, MNTOPT_NODEVICES },
+ { TOKEN_XATTR, MNTOPT_XATTR },
+ { TOKEN_NOXATTR, MNTOPT_NOXATTR },
+ { TOKEN_ATIME, MNTOPT_ATIME },
+ { TOKEN_NOATIME, MNTOPT_NOATIME },
+ { TOKEN_RELATIME, MNTOPT_RELATIME },
+ { TOKEN_NORELATIME, MNTOPT_NORELATIME },
+ { TOKEN_NBMAND, MNTOPT_NBMAND },
+ { TOKEN_NONBMAND, MNTOPT_NONBMAND },
+ { TOKEN_MNTPOINT, MNTOPT_MNTPOINT "=%s" },
+ { TOKEN_LAST, NULL },
+};
+
+static int
+zpl_parse_option(char *option, int token, substring_t *args,
+ zfs_mntopts_t *zmo, boolean_t isremount)
+{
+ switch (token) {
+ case TOKEN_RO:
+ zmo->z_readonly = B_TRUE;
+ zmo->z_do_readonly = B_TRUE;
+ break;
+ case TOKEN_RW:
+ zmo->z_readonly = B_FALSE;
+ zmo->z_do_readonly = B_TRUE;
+ break;
+ case TOKEN_SETUID:
+ zmo->z_setuid = B_TRUE;
+ zmo->z_do_setuid = B_TRUE;
+ break;
+ case TOKEN_NOSETUID:
+ zmo->z_setuid = B_FALSE;
+ zmo->z_do_setuid = B_TRUE;
+ break;
+ case TOKEN_EXEC:
+ zmo->z_exec = B_TRUE;
+ zmo->z_do_exec = B_TRUE;
+ break;
+ case TOKEN_NOEXEC:
+ zmo->z_exec = B_FALSE;
+ zmo->z_do_exec = B_TRUE;
+ break;
+ case TOKEN_DEVICES:
+ zmo->z_devices = B_TRUE;
+ zmo->z_do_devices = B_TRUE;
+ break;
+ case TOKEN_NODEVICES:
+ zmo->z_devices = B_FALSE;
+ zmo->z_do_devices = B_TRUE;
+ break;
+ case TOKEN_XATTR:
+ zmo->z_xattr = B_TRUE;
+ zmo->z_do_xattr = B_TRUE;
+ break;
+ case TOKEN_NOXATTR:
+ zmo->z_xattr = B_FALSE;
+ zmo->z_do_xattr = B_TRUE;
+ break;
+ case TOKEN_ATIME:
+ zmo->z_atime = B_TRUE;
+ zmo->z_do_atime = B_TRUE;
+ break;
+ case TOKEN_NOATIME:
+ zmo->z_atime = B_FALSE;
+ zmo->z_do_atime = B_TRUE;
+ break;
+ case TOKEN_RELATIME:
+ zmo->z_relatime = B_TRUE;
+ zmo->z_do_relatime = B_TRUE;
+ break;
+ case TOKEN_NORELATIME:
+ zmo->z_relatime = B_FALSE;
+ zmo->z_do_relatime = B_TRUE;
+ break;
+ case TOKEN_NBMAND:
+ zmo->z_nbmand = B_TRUE;
+ zmo->z_do_nbmand = B_TRUE;
+ break;
+ case TOKEN_NONBMAND:
+ zmo->z_nbmand = B_FALSE;
+ zmo->z_do_nbmand = B_TRUE;
+ break;
+ case TOKEN_MNTPOINT:
+ zmo->z_mntpoint = match_strdup(&args[0]);
+ if (zmo->z_mntpoint == NULL)
+ return (-ENOMEM);
+
+ break;
+ default:
+ break;
+ }
+
+ return (0);
+}
+
+/*
+ * Parse the mntopts string storing the results in provided zmo argument.
+ * If an error occurs the zmo argument will not be modified. The caller
+ * needs to set isremount when recycling an existing zfs_mntopts_t.
+ */
+static int
+zpl_parse_options(char *osname, char *mntopts, zfs_mntopts_t *zmo,
+ boolean_t isremount)
+{
+ zfs_mntopts_t *tmp_zmo;
+ substring_t args[MAX_OPT_ARGS];
+ char *tmp_mntopts, *p;
+ int error, token;
+
+ if (mntopts == NULL)
+ return (-EINVAL);
+
+ tmp_zmo = zfs_mntopts_alloc();
+ tmp_zmo->z_osname = strdup(osname);
+ tmp_mntopts = strdup(mntopts);
+
+ while ((p = strsep(&tmp_mntopts, ",")) != NULL) {
+ if (!*p)
+ continue;
+
+ args[0].to = args[0].from = NULL;
+ token = match_token(p, zpl_tokens, args);
+ error = zpl_parse_option(p, token, args, tmp_zmo, isremount);
+ if (error) {
+ zfs_mntopts_free(tmp_zmo);
+ strfree(tmp_mntopts);
+ return (error);
+ }
+ }
+
+ strfree(tmp_mntopts);
+
+ if (isremount == B_TRUE) {
+ if (zmo->z_osname)
+ strfree(zmo->z_osname);
+
+ if (zmo->z_mntpoint)
+ strfree(zmo->z_mntpoint);
+ } else {
+ ASSERT3P(zmo->z_osname, ==, NULL);
+ ASSERT3P(zmo->z_mntpoint, ==, NULL);
+ }
+
+ memcpy(zmo, tmp_zmo, sizeof (zfs_mntopts_t));
+ kmem_free(tmp_zmo, sizeof (zfs_mntopts_t));
+
+ return (0);
+}
+
static int
zpl_remount_fs(struct super_block *sb, int *flags, char *data)
{
+ zfs_sb_t *zsb = sb->s_fs_info;
fstrans_cookie_t cookie;
int error;
+ error = zpl_parse_options(zsb->z_mntopts->z_osname, data,
+ zsb->z_mntopts, B_TRUE);
+ if (error)
+ return (error);
+
cookie = spl_fstrans_mark();
- error = -zfs_remount(sb, flags, data);
+ error = -zfs_remount(sb, flags, zsb->z_mntopts);
spl_fstrans_unmark(cookie);
ASSERT3S(error, <=, 0);
return (error);
}
-/*
- * ZFS specific features must be explicitly handled here, the VFS will
- * automatically handled the following generic functionality.
- *
- * MNT_NOSUID,
- * MNT_NODEV,
- * MNT_NOEXEC,
- * MNT_NOATIME,
- * MNT_NODIRATIME,
- * MNT_READONLY,
- * MNT_STRICTATIME,
- * MS_SYNCHRONOUS,
- * MS_DIRSYNC,
- * MS_MANDLOCK.
- */
static int
__zpl_show_options(struct seq_file *seq, zfs_sb_t *zsb)
{
@@ -249,11 +418,12 @@ zpl_show_options(struct seq_file *seq, struct vfsmount *vfsp)
static int
zpl_fill_super(struct super_block *sb, void *data, int silent)
{
+ zfs_mntopts_t *zmo = (zfs_mntopts_t *)data;
fstrans_cookie_t cookie;
int error;
cookie = spl_fstrans_mark();
- error = -zfs_domount(sb, data, silent);
+ error = -zfs_domount(sb, zmo, silent);
spl_fstrans_unmark(cookie);
ASSERT3S(error, <=, 0);
@@ -265,18 +435,32 @@ static struct dentry *
zpl_mount(struct file_system_type *fs_type, int flags,
const char *osname, void *data)
{
- zpl_mount_data_t zmd = { osname, data };
+ zfs_mntopts_t *zmo = zfs_mntopts_alloc();
+ int error;
+
+ error = zpl_parse_options((char *)osname, (char *)data, zmo, B_FALSE);
+ if (error) {
+ zfs_mntopts_free(zmo);
+ return (ERR_PTR(error));
+ }
- return (mount_nodev(fs_type, flags, &zmd, zpl_fill_super));
+ return (mount_nodev(fs_type, flags, zmo, zpl_fill_super));
}
#else
static int
zpl_get_sb(struct file_system_type *fs_type, int flags,
const char *osname, void *data, struct vfsmount *mnt)
{
- zpl_mount_data_t zmd = { osname, data };
+ zfs_mntopts_t *zmo = zfs_mntopts_alloc();
+ int error;
+
+ error = zpl_parse_options((char *)osname, (char *)data, zmo, B_FALSE);
+ if (error) {
+ zfs_mntopts_free(zmo);
+ return (error);
+ }
- return (get_sb_nodev(fs_type, flags, &zmd, zpl_fill_super, mnt));
+ return (get_sb_nodev(fs_type, flags, zmo, zpl_fill_super, mnt));
}
#endif /* HAVE_MOUNT_NODEV */