[lxc-devel] [PATCH v3] Make overlayfs mounts work directly

Serge Hallyn serge.hallyn at ubuntu.com
Mon Oct 5 13:45:53 UTC 2015


Quoting Christian Brauner (christianvanbrauner at gmail.com):
> When users wanted to mount overlay directories with lxc.mount.entry they had to
> create upperdirs and workdirs beforehand in order to mount them. To create it
> for them we add the functions mount_entry_create_overlay_dirs() and
> mount_entry_create_aufs_dirs() which do this for them. User can now simply
> specify e.g.:
> 
>         lxc.mount.entry = /lower merged overlay lowerdir=/lower,upper=/upper,workdir=/workdir,create=dir
> 
> and /upper and /workdir will be created for them. /upper and /workdir need to
> be absolute paths to directories which are created under the containerdir (e.g.
> $lxcpath/container/). Relative mountpoints and mountpoints within the
> container's rootfs and outside the containerdir are ignored. (The latter
> *might* change in the future should it be considered safe/useful.)
> 
> Specifying
> 
>         lxc.mount.entry = /lower merged overlay lowerdir=/lower:/lower2,create=dir
> 
> will lead to a read-only overlay mount in accordance with the
> kernel-documentation.
> 
> Specifying
> 
>         lxc.mount.entry = /lower merged overlay lowerdir=/lower,create=dir
> 
> will fail when no upperdir and workdir options are given.
> 
> Signed-off-by: Christian Brauner <christianvanbrauner at gmail.com>
> ---
>  src/lxc/conf.c | 153 ++++++++++++++++++++++++++++++++++++++++++++++++++++-----
>  1 file changed, 141 insertions(+), 12 deletions(-)
> 
> diff --git a/src/lxc/conf.c b/src/lxc/conf.c
> index bb4c19f..bb415ba 100644
> --- a/src/lxc/conf.c
> +++ b/src/lxc/conf.c
> @@ -1813,13 +1813,142 @@ static void cull_mntent_opt(struct mntent *mntent)
>  	}
>  }
>  
> +static int mount_entry_create_overlay_dirs(const struct mntent *mntent,
> +					   const struct lxc_rootfs *rootfs)
> +{
> +	char *containerdir = NULL;
> +	char *del = NULL;
> +	char *upperdir = NULL;
> +	char *workdir = NULL;
> +	char **opts = NULL;
> +	size_t arrlen = 0;
> +	size_t dirlen = 0;
> +	size_t i;
> +	size_t len = 0;
> +	size_t rootfslen = 0;
> +
> +	opts = lxc_string_split(mntent->mnt_opts, ',');
> +	if (opts)
> +		arrlen = lxc_array_len((void **)opts);
> +	else
> +		return -1;
> +
> +	for (i = 0; i < arrlen; i++) {
> +		if (strstr(opts[i], "upperdir=") && (strlen(opts[i]) > (len = strlen("upperdir="))))
> +				upperdir = opts[i] + len;
> +		else if (strstr(opts[i], "workdir=") && (strlen(opts[i]) > (len = strlen("workdir="))))
> +				workdir = opts[i] + len;
> +	}
> +
> +	containerdir = strdup(rootfs->path);
> +	if (!containerdir) {
> +		lxc_free_array((void **)opts, free);
> +		return -1;
> +	}
> +
> +	if ((del = strstr(containerdir, "/rootfs"))) {
> +		memmove(del, del + 7, strlen(del) - 7 + 1);

if rootfs->path happens to be /a/b/c/rootfs/d/e, you'll end up
making containerpath = /a/b/c/d/e right?  Is that what you want, or
do you just want to *del = 0; ?

> +	}
> +
> +	dirlen = strlen(containerdir);
> +	rootfslen = strlen(rootfs->path);
> +
> +	/* We neither allow users to create upperdirs and workdirs outside the
> +	 * containerdir nor inside the rootfs. The latter might be debatable. */
> +	if (upperdir)
> +		if ((strncmp(upperdir, containerdir, dirlen) == 0) && (strncmp(upperdir, rootfs->path, rootfslen) != 0))
> +			if (mkdir_p(upperdir, 0755) < 0) {
> +				WARN("Failed to create upperdir");
> +			}
> +
> +
> +	if (workdir)
> +		if ((strncmp(workdir, containerdir, dirlen) == 0) && (strncmp(workdir, rootfs->path, rootfslen) != 0))
> +			if (mkdir_p(workdir, 0755) < 0) {
> +				WARN("Failed to create workdir");
> +			}
> +
> +	free(containerdir);
> +	lxc_free_array((void **)opts, free);
> +	return 0;
> +}
> +
> +static int mount_entry_create_aufs_dirs(const struct mntent *mntent,
> +					const struct lxc_rootfs *rootfs)
> +{
> +	char *containerdir = NULL;
> +	char *del = NULL;
> +	char *scratch = NULL;
> +	char *tmp = NULL;
> +	char *upperdir = NULL;
> +	char **opts = NULL;
> +	size_t arrlen = 0;
> +	size_t dirlen = 0;
> +	size_t i;
> +	size_t len = 0;
> +	size_t rootfslen = 0;
> +
> +	opts = lxc_string_split(mntent->mnt_opts, ',');
> +	if (opts)
> +		arrlen = lxc_array_len((void **)opts);
> +	else
> +		return -1;
> +
> +	for (i = 0; i < arrlen; i++) {
> +		if (strstr(opts[i], "br=") && (strlen(opts[i]) > (len = strlen("br="))))
> +			tmp = opts[i] + len;
> +	}
> +	if (!tmp) {
> +		lxc_free_array((void **)opts, free);
> +		return -1;
> +	}
> +
> +	upperdir = strtok_r(tmp, ":=", &scratch);
> +	if (!upperdir) {
> +		lxc_free_array((void **)opts, free);
> +		return -1;
> +	}
> +
> +	containerdir = strdup(rootfs->path);
> +	if (!containerdir) {
> +		lxc_free_array((void **)opts, free);
> +		return -1;
> +	}
> +
> +	if ((del = strstr(containerdir, "/rootfs"))) {
> +		memmove(del, del + 7, strlen(del) - 7 + 1);
> +	}
> +
> +	dirlen = strlen(containerdir);
> +	rootfslen = strlen(rootfs->path);
> +
> +	/* We neither allow users to create upperdirs outside the containerdir
> +	 * nor inside the rootfs. The latter might be debatable. */
> +	if ((strncmp(upperdir, containerdir, dirlen) == 0) && (strncmp(upperdir, rootfs->path, rootfslen) != 0))
> +		if (mkdir_p(upperdir, 0755) < 0) {
> +			WARN("Failed to create upperdir");
> +		}
> +
> +	free(containerdir);
> +	lxc_free_array((void **)opts, free);
> +	return 0;
> +}
> +
>  static int mount_entry_create_dir_file(const struct mntent *mntent,
> -				       const char* path)
> +				       const char* path, const struct lxc_rootfs *rootfs)
>  {
>  	char *pathdirname = NULL;
>  	int ret = 0;
>  	FILE *pathfile = NULL;
>  
> +	if (strncmp(mntent->mnt_type, "overlay", 7) == 0) {
> +		if (mount_entry_create_overlay_dirs(mntent, rootfs) < 0)
> +			ret = -1;
> +	} else if (strncmp(mntent->mnt_type, "aufs", 4) == 0) {
> +		if (mount_entry_create_aufs_dirs(mntent, rootfs) < 0)
> +			ret = -1;
> +	}

Why if mount_entry_create_*_dirs()  failed to you continue to try to
create the file?  Seems better to simply return -1 right there.

> +
>  	if (hasmntopt(mntent, "create=dir")) {
>  		if (mkdir_p(path, 0755) < 0) {
>  			WARN("Failed to create mount target '%s'", path);
> @@ -1837,23 +1966,23 @@ static int mount_entry_create_dir_file(const struct mntent *mntent,
>  		if (!pathfile) {
>  			WARN("Failed to create mount target '%s'", path);
>  			ret = -1;
> -		}
> -		else
> +		} else {
>  			fclose(pathfile);
> +		}
>  	}
>  	free(pathdirname);
>  	return ret;
>  }
>  
>  static inline int mount_entry_on_generic(struct mntent *mntent,
> -                 const char* path, const char *rootfs)
> +                 const char* path, const struct lxc_rootfs *rootfs)
>  {
>  	unsigned long mntflags;
>  	char *mntdata;
>  	int ret;
>  	bool optional = hasmntopt(mntent, "optional") != NULL;
>  
> -	ret = mount_entry_create_dir_file(mntent, path);
> +	ret = mount_entry_create_dir_file(mntent, path, rootfs);
>  
>  	if (ret < 0)
>  		return optional ? 0 : -1;
> @@ -1865,11 +1994,11 @@ static inline int mount_entry_on_generic(struct mntent *mntent,
>  		return -1;
>  	}
>  
> -	ret = mount_entry(mntent->mnt_fsname, path, mntent->mnt_type,
> -			  mntflags, mntdata, optional, rootfs);
> +	ret = mount_entry(mntent->mnt_fsname, path, mntent->mnt_type, mntflags,
> +			  mntdata, optional,
> +			  rootfs->mount ? rootfs->mount : NULL);

This isn't right.  Normally it's done as

	rootfs->path ? rootfs->mount : NULL

because rootfs->mount is always allocated.  If no rootfs was
specified it will be LXCROOTFSMOUNT which is something like
$libdir/lxc/rootfs.

This is obviously begging for a cleanup, but this patch isn't
the place to do that.

>  
>  	free(mntdata);
> -
>  	return ret;
>  }
>  
> @@ -1922,17 +2051,17 @@ skipabs:
>  		return -1;
>  	}
>  
> -	return mount_entry_on_generic(mntent, path, rootfs->mount);
> +	return mount_entry_on_generic(mntent, path, rootfs);
>  }
>  
>  static int mount_entry_on_relative_rootfs(struct mntent *mntent,
> -					  const char *rootfs)
> +					  const struct lxc_rootfs *rootfs)
>  {
>  	char path[MAXPATHLEN];
>  	int ret;
>  
>  	/* relative to root mount point */
> -	ret = snprintf(path, sizeof(path), "%s/%s", rootfs, mntent->mnt_dir);
> +	ret = snprintf(path, sizeof(path), "%s/%s", rootfs->mount, mntent->mnt_dir);
>  	if (ret >= sizeof(path)) {
>  		ERROR("path name too long");
>  		return -1;
> @@ -1959,7 +2088,7 @@ static int mount_file_entries(const struct lxc_rootfs *rootfs, FILE *file,
>  		/* We have a separate root, mounts are relative to it */
>  		if (mntent.mnt_dir[0] != '/') {
>  			if (mount_entry_on_relative_rootfs(&mntent,
> -							   rootfs->mount))
> +							   rootfs))
>  				goto out;
>  			continue;
>  		}
> -- 
> 2.6.0
> 
> _______________________________________________
> lxc-devel mailing list
> lxc-devel at lists.linuxcontainers.org
> http://lists.linuxcontainers.org/listinfo/lxc-devel


More information about the lxc-devel mailing list