% mergerfs(1) mergerfs user manual % Antonio SJ Musumeci trapexit@spawn.link % 2015-09-14

mergerfs - another FUSE union filesystem

mergerfs -o<options> <srcpoints> <mountpoint>

mergerfs is similar to mhddfs, unionfs, and aufs. Like mhddfs in that it too uses FUSE. Like aufs in that it provides multiple policies for how to handle behavior.

Why create mergerfs when those exist? mhddfs has not been updated in some time nor very flexible. There are also security issues when with running as root. aufs is more flexible than mhddfs but contains some hard to debug inconsistencies in behavior on account of it being a kernel driver. Neither support file attributes (chattr).

###options###

• defaults is a shortcut for FUSE's atomic_o_trunc, auto_cache, big_writes, default_permissions, splice_move, splice_read, and splice_write. These options seem to provide the best performance.
• minfreespace (defaults to 4G) is the minimum space value used for the lfs, fwfs, and epmfs policies. Understands 'K', 'M', and 'G' to represent kilobyte, megabyte, and gigabyte respectively.
• All FUSE functions which have a category (see below) are option keys. The syntax being func.<func>=<policy>. Example: func.getattr=newest.
• To set all function policies in a category use category.<category>=<policy>. Example: category.create=mfs.
• Options are evaluated in the order listed so if the options are func.rmdir=rand,category.action=ff the action category setting will override the rmdir setting.

###srcpoints###

The source points argument is a colon (':') delimited list of paths. To make it simpler to include multiple source points without having to modify your fstab we also support globbing.

$ mergerfs /mnt/disk\*:/mnt/cdrom /media/drives

The above line will use all points in /mnt prefixed with disk and the directory cdrom. The * (or any glob tokens) must be escaped on the command line else the shell resolve them.

In /etc/fstab it'd look like the following:

# <file system>        <mount point>  <type>         <options>             <dump>  <pass>
/mnt/disk*:/mnt/cdrom  /media/drives  fuse.mergerfs  defaults,allow_other  0       0

NOTE: the globbing is done at mount or xattr update time. If a new directory is added matching the glob after the fact it will not be included.

Filesystem calls are broken up into 3 categories: action, create, search. There are also some calls which have no policy attached due to state being kept between calls. These categories can be assigned a policy which dictates how mergerfs behaves. Any policy can be assigned to a category though some aren't terribly practical. For instance: rand (Random) may be useful for create but could lead to very odd behavior if used for search.

ioctl behaves differently if its acting on a directory. It'll use the getattr policy to find and open the directory before issuing the ioctl. In other cases where something may be searched (to confirm a directory exists across all source mounts) then getattr will be used.

rename is a tricky function in a merged system. Normally if a rename can't be done atomically due to the from and to paths existing on different mount points it will return -1 with errno = EXDEV. The atomic rename is most critical for replacing files in place atomically (such as securing writing to a temp file and then replacing a target). The problem is that by merging multiple paths you can have N instances of the source and destinations on different drives. Meaning that if you just renamed each source locally you could end up with the destination files not overwriten / replaced. To address this mergerfs works in the following way. If the source and destination exist in different directories it will immediately return EXDEV. Generally it's not expected for cross directory renames to work so it should be fine for most instances (mv,rsync,etc.). If they do belong to the same directory it then runs the rename policy to get the files to rename. It iterates through and renames each file while keeping track of those paths which have not been renamed. If all the renames succeed it will then unlink or rmdir the other paths to clean up any preexisting target files. This allows the new file to be found without the file itself ever disappearing. There may still be some issues with this behavior. Particularly on error. At the moment however this seems the best policy.

readdir is very different from most functions in this realm. It certainly could have it's own set of policies to tweak its behavior. At this time it provides a simple first found merging of directories and file found. That is: only the first file or directory found for a directory is returned. Given how FUSE works though the data representing the returned entry comes from getattr.

It could be extended to offer the ability to see all files found. Perhaps concatenating # and a number to the name. But to really be useful you'd need to be able to access them which would complicate file lookup.

statvfs normalizes the source drives based on the fragment size and sums the number of adjusted blocks and inodes. This means you will see the combined space of all sources. Total, used, and free. The sources however are dedupped based on the drive so multiple points on the same drive will not result in double counting it's space.

NOTE: Since we can not (easily) replicate the atomicity of an mkdir or mknod without side effects those calls will first do a scan to see if the file exists and then attempts a create. This means there is a slight race condition. Worse case you'd end up with the directory or file on more than one mount.

First get the code from github.

$ git clone https://github.com/trapexit/mergerfs.git
$ # or
$ wget https://github.com/trapexit/mergerfs/archive/master.zip

$ sudo apt-get install g++ pkg-config git git-buildpackage pandoc debhelper libfuse-dev libattr1-dev
$ cd mergerfs
$ make deb
$ sudo dpkg -i ../mergerfs_version_arch.deb

$ su -
# dnf install fuse-devel libattr-devel pandoc gcc-c++
# cd mergerfs
# make rpm
# rpm -i rpmbuild/RPMS/<arch>/mergerfs-<verion>.<arch>.rpm

Have pkg-config, pandoc, libfuse, libattr1 installed.

$ cd mergerfs
$ make
$ make man
$ sudo make install

<mountpoint>/.mergerfs

There is a pseudo file available at the mount point which allows for the runtime modification of certain mergerfs options. The file will not show up in readdir but can be stat'ed and manipulated via {list,get,set}xattrs calls.

Even if xattrs are disabled the {list,get,set}xattrs calls will still work.

Use xattr -l /mount/point/.mergerfs to see all supported keys.

[trapexit:/tmp/mount] $ xattr -l .mergerfs
user.mergerfs.srcmounts: /tmp/a:/tmp/b
user.mergerfs.minfreespace: 4294967295
user.mergerfs.policies: all,einval,enosys,enotsup,epmfs,erofs,exdev,ff,ffwp,fwfs,lfs,mfs,newest,rand
user.mergerfs.version: 2.5.0
user.mergerfs.category.action: all
user.mergerfs.category.create: epmfs
user.mergerfs.category.search: ff
user.mergerfs.func.access: ff
user.mergerfs.func.chmod: all
user.mergerfs.func.chown: all
user.mergerfs.func.create: epmfs
user.mergerfs.func.getattr: ff
user.mergerfs.func.getxattr: ff
user.mergerfs.func.link: all
user.mergerfs.func.listxattr: ff
user.mergerfs.func.mkdir: epmfs
user.mergerfs.func.mknod: epmfs
user.mergerfs.func.open: ff
user.mergerfs.func.readlink: ff
user.mergerfs.func.removexattr: all
user.mergerfs.func.rename: all
user.mergerfs.func.rmdir: all
user.mergerfs.func.setxattr: all
user.mergerfs.func.symlink: epmfs
user.mergerfs.func.truncate: all
user.mergerfs.func.unlink: all
user.mergerfs.func.utimens: all

[trapexit:/tmp/mount] $ xattr -p user.mergerfs.category.search .mergerfs
ff

[trapexit:/tmp/mount] $ xattr -w user.mergerfs.category.search ffwp .mergerfs
[trapexit:/tmp/mount] $ xattr -p user.mergerfs.category.search .mergerfs
ffwp

[trapexit:/tmp/mount] $ xattr -w user.mergerfs.srcmounts +/tmp/c .mergerfs
[trapexit:/tmp/mount] $ xattr -p user.mergerfs.srcmounts .mergerfs
/tmp/a:/tmp/b:/tmp/c

[trapexit:/tmp/mount] $ xattr -w user.mergerfs.srcmounts =/tmp/c .mergerfs
[trapexit:/tmp/mount] $ xattr -p user.mergerfs.srcmounts .mergerfs
/tmp/c

[trapexit:/tmp/mount] $ xattr -w user.mergerfs.srcmounts '+</tmp/a:/tmp/b' .mergerfs
[trapexit:/tmp/mount] $ xattr -p user.mergerfs.srcmounts .mergerfs
/tmp/a:/tmp/b:/tmp/c

For user.mergerfs.srcmounts there are several instructions available for manipulating the list. The value provided is just as the value used at mount time. A colon (':') delimited list of full path globs.

Categories and funcs take a policy as described in the previous section. When reading funcs you'll get the policy string. However, with categories you'll get a comma separated list of policies for each type found. For example: if all search functions are ff except for access which is ffwp the value for user.mergerfs.category.search will be ff,ffwp.

While they won't show up when using listxattr mergerfs offers a number of special xattrs to query information about the files served. To access the values you will need to issue a getxattr for one of the following:

• user.mergerfs.basepath: the base mount point for the file given the current search policy
• user.mergerfs.relpath: the relative path of the file from the perspective of the mount point
• user.mergerfs.fullpath: the full path of the original file given the search policy
• user.mergerfs.allpaths: a NUL ('\0') separated list of full paths to all files found

[trapexit:/tmp/mount] $ ls
A B C
[trapexit:/tmp/mount] $ xattr -p user.mergerfs.fullpath A
/mnt/a/full/path/to/A
[trapexit:/tmp/mount] $ xattr -p user.mergerfs.basepath A
/mnt/a
[trapexit:/tmp/mount] $ xattr -p user.mergerfs.relpath A
/full/path/to/A
[trapexit:/tmp/mount] $ xattr -p user.mergerfs.allpaths A | tr '\0' '\n'
/mnt/a/full/path/to/A
/mnt/b/full/path/to/A

• If you don't see some directories / files you expect in a merged point be sure the user has permission to all the underlying directories. If /drive0/a has is owned by root:root with ACLs set to 0700 and /drive1/a is root:root and 0755 you'll see only /drive1/a.
• Since POSIX gives you only error or success on calls its difficult to determine the proper behavior when applying the behavior to multiple targets. Generally if something succeeds when reading it returns the data it can. If something fails when making an action we continue on and return the last error.
• The recommended options are defaults,allow_other. The allow_other is to allow users who are not the one which executed mergerfs access to the mountpoint. defaults is described above and should offer the best performance. It's possible that if you're running on an older platform the splice features aren't available and could error. In that case simply use the other options manually.
• Remember that some policies mixed with some functions may result in strange behaviors. Not that some of these behaviors and race conditions couldn't happen outside mergerfs but that they are far more likely to occur on account of attempt to merge together multiple sources of data which could be out of sync due to the different policies.
• An example: Kodi and Plex can apparently use directory mtime to more efficiently determine whether or not to scan for new content rather than simply performing a full scan. If using the current default getattr policy of ff its possible Kodi will miss an update on account of it returning the first directory found's stat info and its a later directory on another mount which had the mtime recently updated. To fix this you will want to set func.getattr=newest. Remember though that this is just stat. If the file is later open'ed or unlink'ed and the policy is different for those then a completely different file or directory could be acted on.
• Due to previously mentioned issues its generally best to set category wide policies rather than individual func's. This will help limit the confusion of tools such as rsync.

• Due to the overhead of getgroups/setgroups mergerfs utilizes a cache. This cache is opportunistic and per thread. This means each thread will query the supplemental groups for a user when that particular thread needs to change credentials and currently it will keep that data for the lifetime of the mount or thread. This means that if a user is added to a group it may not be picked up without the restart of mergerfs. However, since the high level FUSE API's (at least the standard version) thread pool dynamically grows and shrinks it's possible that over time a thread will be killed and later a new thread with no cache will start and query the new data.

It's mentioned that there are some security issues with mhddfs. What are they? How does mergerfs address them?

mhddfs tries to handle being run as root by calling getuid() and if it returns 0 then it will chown the file. Not only is that a race condition but it doesn't handle many other situations. Rather than attempting to simulate POSIX ACL behaviors the proper behavior is to use seteuid and setegid, become the user making the original call and perform the action as them. This is how mergerfs handles things.

If you are familiar with POSIX standards you'll know that this behavior poses a problem. seteuid and setegid affect the whole process and libfuse is multithreaded by default. We'd need to lock access to seteuid and setegid with a mutex so that the several threads aren't stepping on one another and files end up with weird permissions and ownership. This however wouldn't scale well. With lots of calls the contention on that mutex would be extremely high. Thankfully on Linux and OSX we have a better solution.

OSX has a non-portable pthread extension for per-thread user and group impersonation.

Linux does not support pthread_setugid_np but user and group IDs are a per-thread attribute though documentation on that fact or how to manipulate them is not well distributed. From the 4.00 release of the Linux man-pages project for setuid

At the kernel level, user IDs and group IDs are a per-thread attribute. However, POSIX requires that all threads in a process share the same credentials. The NPTL threading implementation handles the POSIX requirements by providing wrapper functions for the various system calls that change process UIDs and GIDs. These wrapper functions (including the one for setuid()) employ a signal-based technique to ensure that when one thread changes credentials, all of the other threads in the process also change their credentials. For details, see nptl(7).

Turns out the setreuid syscalls apply only to the thread. GLIBC hides this away using RT signals to inform all threads to change credentials. Taking after Samba mergerfs uses syscall(SYS_setreuid,...) to set the callers credentials for that thread only. Jumping back to root as necessary should escalated privileges be needed (for instance: to clone paths).

For non-Linux systems mergerfs uses a read-write lock and changes credentials only when necessary. If multiple threads are to be user X then only the first one will need to change the processes credentials. So long as the other threads need to be user X they will take a readlock allow multiple threads to share the credentials. Once a request comes in to run as user Y that thread will attempt a write lock and change to Y's credentials when it can. If the ability to give writers priority is supported then that flag will be used so threads trying to change credentials don't starve. This isn't the best solution but should work reasonably well. As new platforms are supported if they offer per thread credentials those APIs will be adopted.