You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

2753 lines
114 KiB

11 years ago
4 years ago
5 years ago
5 years ago
5 years ago
2 years ago
5 years ago
  1. % mergerfs(1) mergerfs user manual
  2. # NAME
  3. mergerfs - a featureful union filesystem
  4. # SYNOPSIS
  5. mergerfs -o<options> <branches> <mountpoint>
  6. # DESCRIPTION
  7. **mergerfs** is a union filesystem geared towards simplifying storage
  8. and management of files across numerous commodity storage devices. It
  9. is similar to **mhddfs**, **unionfs**, and **aufs**.
  10. # FEATURES
  11. * Configurable behaviors / file placement
  12. * Ability to add or remove filesystems at will
  13. * Resistance to individual filesystem failure
  14. * Support for extended attributes (xattrs)
  15. * Support for file attributes (chattr)
  16. * Runtime configurable (via xattrs)
  17. * Works with heterogeneous filesystem types
  18. * Moving of file when filesystem runs out of space while writing
  19. * Ignore read-only filesystems when creating files
  20. * Turn read-only files into symlinks to underlying file
  21. * Hard link copy-on-write / CoW
  22. * Support for POSIX ACLs
  23. * Misc other things
  24. # HOW IT WORKS
  25. mergerfs logically merges multiple paths together. Think a union of
  26. sets. The file/s or directory/s acted on or presented through mergerfs
  27. are based on the policy chosen for that particular action. Read more
  28. about policies below.
  29. ```
  30. A + B = C
  31. /disk1 /disk2 /merged
  32. | | |
  33. +-- /dir1 +-- /dir1 +-- /dir1
  34. | | | | | |
  35. | +-- file1 | +-- file2 | +-- file1
  36. | | +-- file3 | +-- file2
  37. +-- /dir2 | | +-- file3
  38. | | +-- /dir3 |
  39. | +-- file4 | +-- /dir2
  40. | +-- file5 | |
  41. +-- file6 | +-- file4
  42. |
  43. +-- /dir3
  44. | |
  45. | +-- file5
  46. |
  47. +-- file6
  48. ```
  49. mergerfs does **not** support the copy-on-write (CoW) or whiteout
  50. behaviors found in **aufs** and **overlayfs**. You can **not** mount a
  51. read-only filesystem and write to it. However, mergerfs will ignore
  52. read-only filesystems when creating new files so you can mix
  53. read-write and read-only filesystems. It also does **not** split data
  54. across filesystems. It is not RAID0 / striping. It is simply a union of
  55. other filesystems.
  56. # TERMINOLOGY
  57. * branch: A base path used in the pool.
  58. * pool: The mergerfs mount. The union of the branches.
  59. * relative path: The path in the pool relative to the branch and mount.
  60. * function: A filesystem call (open, unlink, create, getattr, rmdir, etc.)
  61. * category: A collection of functions based on basic behavior (action, create, search).
  62. * policy: The algorithm used to select a file when performing a function.
  63. * path preservation: Aspect of some policies which includes checking the path for which a file would be created.
  64. # BASIC SETUP
  65. If you don't already know that you have a special use case then just
  66. start with one of the following option sets.
  67. #### You need `mmap` (used by rtorrent and many sqlite3 base software)
  68. `cache.files=auto-full,dropcacheonclose=true,category.create=mfs`
  69. or if you are on a Linux kernel >= 6.6.x mergerfs will enable a mode
  70. that allows shared mmap when `cache.files=off`. To be sure of the best
  71. performance between `cache.files=off` and `cache.files=auto-full`
  72. you'll need to do your own benchmarking but often `off` is faster.
  73. #### You don't need `mmap`
  74. `cache.files=off,dropcacheonclose=true,category.create=mfs`
  75. ### Command Line
  76. `mergerfs -o cache.files=auto-full,dropcacheonclose=true,category.create=mfs /mnt/hdd0:/mnt/hdd1 /media`
  77. ### /etc/fstab
  78. `/mnt/hdd0:/mnt/hdd1 /media mergerfs cache.files=auto-full,dropcacheonclose=true,category.create=mfs 0 0`
  79. ### systemd mount
  80. https://github.com/trapexit/mergerfs/wiki/systemd
  81. ```
  82. [Unit]
  83. Description=mergerfs service
  84. [Service]
  85. Type=simple
  86. KillMode=none
  87. ExecStart=/usr/bin/mergerfs \
  88. -f \
  89. -o cache.files=auto-full \
  90. -o dropcacheonclose=true \
  91. -o category.create=mfs \
  92. /mnt/hdd0:/mnt/hdd1 \
  93. /media
  94. ExecStop=/bin/fusermount -uz /media
  95. Restart=on-failure
  96. [Install]
  97. WantedBy=default.target
  98. ```
  99. See the mergerfs [wiki for real world
  100. deployments](https://github.com/trapexit/mergerfs/wiki/Real-World-Deployments)
  101. for comparisons / ideas.
  102. # OPTIONS
  103. These options are the same regardless of whether you use them with the
  104. `mergerfs` commandline program, in fstab, or in a config file.
  105. ### mount options
  106. * **config**: Path to a config file. Same arguments as below in
  107. key=val / ini style format.
  108. * **branches**: Colon delimited list of branches.
  109. * **minfreespace=SIZE**: The minimum space value used for creation
  110. policies. Can be overridden by branch specific option. Understands
  111. 'K', 'M', and 'G' to represent kilobyte, megabyte, and gigabyte
  112. respectively. (default: 4G)
  113. * **moveonenospc=BOOL|POLICY**: When enabled if a **write** fails with
  114. **ENOSPC** (no space left on device) or **EDQUOT** (disk quota
  115. exceeded) the policy selected will run to find a new location for
  116. the file. An attempt to move the file to that branch will occur
  117. (keeping all metadata possible) and if successful the original is
  118. unlinked and the write retried. (default: false, true = mfs)
  119. * **inodecalc=passthrough|path-hash|devino-hash|hybrid-hash**: Selects
  120. the inode calculation algorithm. (default: hybrid-hash)
  121. * **dropcacheonclose=BOOL**: When a file is requested to be closed
  122. call `posix_fadvise` on it first to instruct the kernel that we no
  123. longer need the data and it can drop its cache. Recommended when
  124. **cache.files=partial|full|auto-full|per-process** to limit double
  125. caching. (default: false)
  126. * **direct-io-allow-mmap=BOOL**: On newer kernels (>= 6.6) it is
  127. possible to disable file page caching while still allowing for
  128. shared mmap support. mergerfs will enable this feature if available
  129. but an option is provided to turn it off for testing and debugging
  130. purposes. (default: true)
  131. * **symlinkify=BOOL**: When enabled and a file is not writable and its
  132. mtime or ctime is older than **symlinkify_timeout** files will be
  133. reported as symlinks to the original files. Please read more below
  134. before using. (default: false)
  135. * **symlinkify_timeout=UINT**: Time to wait, in seconds, to activate
  136. the **symlinkify** behavior. (default: 3600)
  137. * **nullrw=BOOL**: Turns reads and writes into no-ops. The request
  138. will succeed but do nothing. Useful for benchmarking
  139. mergerfs. (default: false)
  140. * **lazy-umount-mountpoint=BOOL**: mergerfs will attempt to "lazy
  141. umount" the mountpoint before mounting itself. Useful when
  142. performing live upgrades of mergerfs. (default: false)
  143. * **ignorepponrename=BOOL**: Ignore path preserving on
  144. rename. Typically rename and link act differently depending on the
  145. policy of `create` (read below). Enabling this will cause rename and
  146. link to always use the non-path preserving behavior. This means
  147. files, when renamed or linked, will stay on the same
  148. filesystem. (default: false)
  149. * **export-support=BOOL**: Sets a low-level FUSE feature intended to
  150. indicate the filesystem can support being exported via
  151. NFS. (default: true)
  152. * **security_capability=BOOL**: If false return ENOATTR when xattr
  153. security.capability is queried. (default: true)
  154. * **xattr=passthrough|noattr|nosys**: Runtime control of
  155. xattrs. Default is to passthrough xattr requests. 'noattr' will
  156. short circuit as if nothing exists. 'nosys' will respond with ENOSYS
  157. as if xattrs are not supported or disabled. (default: passthrough)
  158. * **link_cow=BOOL**: When enabled if a regular file is opened which
  159. has a link count > 1 it will copy the file to a temporary file and
  160. rename over the original. Breaking the link and providing a basic
  161. copy-on-write function similar to cow-shell. (default: false)
  162. * **statfs=base|full**: Controls how statfs works. 'base' means it
  163. will always use all branches in statfs calculations. 'full' is in
  164. effect path preserving and only includes branches where the path
  165. exists. (default: base)
  166. * **statfs_ignore=none|ro|nc**: 'ro' will cause statfs calculations to
  167. ignore available space for branches mounted or tagged as 'read-only'
  168. or 'no create'. 'nc' will ignore available space for branches tagged
  169. as 'no create'. (default: none)
  170. * **nfsopenhack=off|git|all**: A workaround for exporting mergerfs
  171. over NFS where there are issues with creating files for write while
  172. setting the mode to read-only. (default: off)
  173. * **branches-mount-timeout=UINT**: Number of seconds to wait at
  174. startup for branches to be a mount other than the mountpoint's
  175. filesystem. (default: 0)
  176. * **follow-symlinks=never|directory|regular|all**: Turns symlinks into
  177. what they point to. (default: never)
  178. * **link-exdev=passthrough|rel-symlink|abs-base-symlink|abs-pool-symlink**:
  179. When a link fails with EXDEV optionally create a symlink to the file
  180. instead.
  181. * **rename-exdev=passthrough|rel-symlink|abs-symlink**: When a rename
  182. fails with EXDEV optionally move the file to a special directory and
  183. symlink to it.
  184. * **readahead=UINT**: Set readahead (in kilobytes) for mergerfs and
  185. branches if greater than 0. (default: 0)
  186. * **posix_acl=BOOL**: Enable POSIX ACL support (if supported by kernel
  187. and underlying filesystem). (default: false)
  188. * **async_read=BOOL**: Perform reads asynchronously. If disabled or
  189. unavailable the kernel will ensure there is at most one pending read
  190. request per file handle and will attempt to order requests by
  191. offset. (default: true)
  192. * **fuse_msg_size=UINT**: Set the max number of pages per FUSE
  193. message. Only available on Linux >= 4.20 and ignored
  194. otherwise. (min: 1; max: 256; default: 256)
  195. * **threads=INT**: Number of threads to use. When used alone
  196. (`process-thread-count=-1`) it sets the number of threads reading
  197. and processing FUSE messages. When used together it sets the number
  198. of threads reading from FUSE. When set to zero it will attempt to
  199. discover and use the number of logical cores. If the thread count is
  200. set negative it will look up the number of cores then divide by the
  201. absolute value. ie. threads=-2 on an 8 core machine will result in 8
  202. / 2 = 4 threads. There will always be at least 1 thread. If set to
  203. -1 in combination with `process-thread-count` then it will try to
  204. pick reasonable values based on CPU thread count. NOTE: higher
  205. number of threads increases parallelism but usually decreases
  206. throughput. (default: 0)
  207. * **read-thread-count=INT**: Alias for `threads`.
  208. * **process-thread-count=INT**: Enables separate thread pool to
  209. asynchronously process FUSE requests. In this mode
  210. `read-thread-count` refers to the number of threads reading FUSE
  211. messages which are dispatched to process threads. -1 means disabled
  212. otherwise acts like `read-thread-count`. (default: -1)
  213. * **process-thread-queue-depth=UINT**: Sets the number of requests any
  214. single process thread can have queued up at one time. Meaning the
  215. total memory usage of the queues is queue depth multiplied by the
  216. number of process threads plus read thread count. 0 sets the depth
  217. to the same as the process thread count. (default: 0)
  218. * **pin-threads=STR**: Selects a strategy to pin threads to CPUs
  219. (default: unset)
  220. * **flush-on-close=never|always|opened-for-write**: Flush data cache
  221. on file close. Mostly for when writeback is enabled or merging
  222. network filesystems. (default: opened-for-write)
  223. * **scheduling-priority=INT**: Set mergerfs' scheduling
  224. priority. Valid values range from -20 to 19. See `setpriority` man
  225. page for more details. (default: -10)
  226. * **fsname=STR**: Sets the name of the filesystem as seen in
  227. **mount**, **df**, etc. Defaults to a list of the source paths
  228. concatenated together with the longest common prefix removed.
  229. * **func.FUNC=POLICY**: Sets the specific FUSE function's policy. See
  230. below for the list of value types. Example: **func.getattr=newest**
  231. * **func.readdir=seq|cosr|cor|cosr:INT|cor:INT**: Sets `readdir`
  232. policy. INT value sets the number of threads to use for
  233. concurrency. (default: seq)
  234. * **category.action=POLICY**: Sets policy of all FUSE functions in the
  235. action category. (default: epall)
  236. * **category.create=POLICY**: Sets policy of all FUSE functions in the
  237. create category. (default: epmfs)
  238. * **category.search=POLICY**: Sets policy of all FUSE functions in the
  239. search category. (default: ff)
  240. * **cache.open=UINT**: 'open' policy cache timeout in
  241. seconds. (default: 0)
  242. * **cache.statfs=UINT**: 'statfs' cache timeout in seconds. (default:
  243. 0)
  244. * **cache.attr=UINT**: File attribute cache timeout in
  245. seconds. (default: 1)
  246. * **cache.entry=UINT**: File name lookup cache timeout in
  247. seconds. (default: 1)
  248. * **cache.negative_entry=UINT**: Negative file name lookup cache
  249. timeout in seconds. (default: 0)
  250. * **cache.files=libfuse|off|partial|full|auto-full|per-process**: File
  251. page caching mode (default: libfuse)
  252. * **cache.files.process-names=LIST**: A pipe | delimited list of
  253. process [comm](https://man7.org/linux/man-pages/man5/proc.5.html)
  254. names to enable page caching for when
  255. `cache.files=per-process`. (default: "rtorrent|qbittorrent-nox")
  256. * **cache.writeback=BOOL**: Enable kernel writeback caching (default:
  257. false)
  258. * **cache.symlinks=BOOL**: Cache symlinks (if supported by kernel)
  259. (default: false)
  260. * **cache.readdir=BOOL**: Cache readdir (if supported by kernel)
  261. (default: false)
  262. * **parallel-direct-writes=BOOL**: Allow the kernel to dispatch
  263. multiple, parallel (non-extending) write requests for files opened
  264. with `cache.files=per-process` (if the process is not in `process-names`)
  265. or `cache.files=off`. (This requires kernel support, and was added in v6.2)
  266. * **direct_io**: deprecated - Bypass page cache. Use `cache.files=off`
  267. instead. (default: false)
  268. * **kernel_cache**: deprecated - Do not invalidate data cache on file
  269. open. Use `cache.files=full` instead. (default: false)
  270. * **auto_cache**: deprecated - Invalidate data cache if file mtime or
  271. size change. Use `cache.files=auto-full` instead. (default: false)
  272. * **async_read**: deprecated - Perform reads asynchronously. Use
  273. `async_read=true` instead.
  274. * **sync_read**: deprecated - Perform reads synchronously. Use
  275. `async_read=false` instead.
  276. * **splice_read**: deprecated - Does nothing.
  277. * **splice_write**: deprecated - Does nothing.
  278. * **splice_move**: deprecated - Does nothing.
  279. * **allow_other**: deprecated - mergerfs v2.35.0 and newer sets this FUSE option
  280. automatically if running as root.
  281. * **use_ino**: deprecated - mergerfs should always control inode
  282. calculation so this is enabled all the time.
  283. **NOTE:** Options are evaluated in the order listed so if the options
  284. are **func.rmdir=rand,category.action=ff** the **action** category
  285. setting will override the **rmdir** setting.
  286. **NOTE:** Always look at the documentation for the version of mergerfs
  287. you're using. Not all features are available in older releases. Use
  288. `man mergerfs` or find the docs as linked in the release.
  289. #### Value Types
  290. * BOOL = 'true' | 'false'
  291. * INT = [MIN_INT,MAX_INT]
  292. * UINT = [0,MAX_INT]
  293. * SIZE = 'NNM'; NN = INT, M = 'K' | 'M' | 'G' | 'T'
  294. * STR = string (may refer to an enumerated value, see details of
  295. argument)
  296. * FUNC = filesystem function
  297. * CATEGORY = function category
  298. * POLICY = mergerfs function policy
  299. ### branches
  300. The 'branches' argument is a colon (':') delimited list of paths to be
  301. pooled together. It does not matter if the paths are on the same or
  302. different filesystems nor does it matter the filesystem type (within
  303. reason). Used and available space will not be duplicated for paths on
  304. the same filesystem and any features which aren't supported by the
  305. underlying filesystem (such as file attributes or extended attributes)
  306. will return the appropriate errors.
  307. Branches currently have two options which can be set. A type which
  308. impacts whether or not the branch is included in a policy calculation
  309. and a individual minfreespace value. The values are set by prepending
  310. an `=` at the end of a branch designation and using commas as
  311. delimiters. Example: `/mnt/drive=RW,1234`
  312. #### branch mode
  313. * RW: (read/write) - Default behavior. Will be eligible in all policy
  314. categories.
  315. * RO: (read-only) - Will be excluded from `create` and `action`
  316. policies. Same as a read-only mounted filesystem would be (though
  317. faster to process).
  318. * NC: (no-create) - Will be excluded from `create` policies. You can't
  319. create on that branch but you can change or delete.
  320. #### minfreespace
  321. Same purpose and syntax as the global option but specific to the
  322. branch. If not set the global value is used.
  323. #### globbing
  324. To make it easier to include multiple branches mergerfs supports
  325. [globbing](http://linux.die.net/man/7/glob). **The globbing tokens
  326. MUST be escaped when using via the shell else the shell itself will
  327. apply the glob itself.**
  328. ```
  329. # mergerfs /mnt/hdd\*:/mnt/ssd /media
  330. ```
  331. The above line will use all mount points in /mnt prefixed with **hdd** and **ssd**.
  332. To have the pool mounted at boot or otherwise accessible from related tools use **/etc/fstab**.
  333. ```
  334. # <file system> <mount point> <type> <options> <dump> <pass>
  335. /mnt/hdd*:/mnt/ssd /media mergerfs minfreespace=16G 0 0
  336. ```
  337. **NOTE:** the globbing is done at mount or when updated using the
  338. runtime API. If a new directory is added matching the glob after the
  339. fact it will not be automatically included.
  340. **NOTE:** for mounting via **fstab** to work you must have
  341. **mount.fuse** installed. For Ubuntu/Debian it is included in the
  342. **fuse** package.
  343. ### inodecalc
  344. Inodes (st_ino) are unique identifiers within a filesystem. Each
  345. mounted filesystem has device ID (st_dev) as well and together they
  346. can uniquely identify a file on the whole of the system. Entries on
  347. the same device with the same inode are in fact references to the same
  348. underlying file. It is a many to one relationship between names and an
  349. inode. Directories, however, do not have multiple links on most
  350. systems due to the complexity they add.
  351. FUSE allows the server (mergerfs) to set inode values but not device
  352. IDs. Creating an inode value is somewhat complex in mergerfs' case as
  353. files aren't really in its control. If a policy changes what directory
  354. or file is to be selected or something changes out of band it becomes
  355. unclear what value should be used. Most software does not to care what
  356. the values are but those that do often break if a value changes
  357. unexpectedly. The tool `find` will abort a directory walk if it sees a
  358. directory inode change. NFS can return stale handle errors if the
  359. inode changes out of band. File dedup tools will usually leverage
  360. device ids and inodes as a shortcut in searching for duplicate files
  361. and would resort to full file comparisons should it find different
  362. inode values.
  363. mergerfs offers multiple ways to calculate the inode in hopes of
  364. covering different usecases.
  365. * passthrough: Passes through the underlying inode value. Mostly
  366. intended for testing as using this does not address any of the
  367. problems mentioned above and could confuse file deduplication
  368. software as inodes from different filesystems can be the same.
  369. * path-hash: Hashes the relative path of the entry in question. The
  370. underlying file's values are completely ignored. This means the
  371. inode value will always be the same for that file path. This is
  372. useful when using NFS and you make changes out of band such as copy
  373. data between branches. This also means that entries that do point to
  374. the same file will not be recognizable via inodes. That **does not**
  375. mean hard links don't work. They will.
  376. * path-hash32: 32bit version of path-hash.
  377. * devino-hash: Hashes the device id and inode of the underlying
  378. entry. This won't prevent issues with NFS should the policy pick a
  379. different file or files move out of band but will present the same
  380. inode for underlying files that do too.
  381. * devino-hash32: 32bit version of devino-hash.
  382. * hybrid-hash: Performs `path-hash` on directories and `devino-hash`
  383. on other file types. Since directories can't have hard links the
  384. static value won't make a difference and the files will get values
  385. useful for finding duplicates. Probably the best to use if not using
  386. NFS. As such it is the default.
  387. * hybrid-hash32: 32bit version of hybrid-hash.
  388. 32bit versions are provided as there is some software which does not
  389. handle 64bit inodes well.
  390. While there is a risk of hash collision in tests of a couple of million
  391. entries there were zero collisions. Unlike a typical filesystem FUSE
  392. filesystems can reuse inodes and not refer to the same entry. The
  393. internal identifier used to reference a file in FUSE is different from
  394. the inode value presented. The former is the `nodeid` and is actually
  395. a tuple of 2 64bit values: `nodeid` and `generation`. This tuple is
  396. not client facing. The inode that is presented to the client is passed
  397. through the kernel uninterpreted.
  398. From FUSE docs for `use_ino`:
  399. ```
  400. Honor the st_ino field in the functions getattr() and
  401. fill_dir(). This value is used to fill in the st_ino field
  402. in the stat(2), lstat(2), fstat(2) functions and the d_ino
  403. field in the readdir(2) function. The filesystem does not
  404. have to guarantee uniqueness, however some applications
  405. rely on this value being unique for the whole filesystem.
  406. Note that this does *not* affect the inode that libfuse
  407. and the kernel use internally (also called the "nodeid").
  408. ```
  409. As of version 2.35.0 the `use_ino` option has been removed. mergerfs
  410. should always be managing inode values.
  411. ### pin-threads
  412. Simple strategies for pinning read and/or process threads. If process
  413. threads are not enabled then the strategy simply works on the read
  414. threads. Invalid values are ignored.
  415. * R1L: All read threads pinned to a single logical CPU.
  416. * R1P: All read threads pinned to a single physical CPU.
  417. * RP1L: All read and process threads pinned to a single logical CPU.
  418. * RP1P: All read and process threads pinned to a single physical CPU.
  419. * R1LP1L: All read threads pinned to a single logical CPU, all process
  420. threads pinned to a (if possible) different logical CPU.
  421. * R1PP1P: All read threads pinned to a single physical CPU, all
  422. process threads pinned to a (if possible) different logical CPU.
  423. * RPSL: All read and process threads are spread across all logical CPUs.
  424. * RPSP: All read and process threads are spread across all physical CPUs.
  425. * R1PPSP: All read threads are pinned to a single physical CPU while
  426. process threads are spread across all other physical CPUs.
  427. ### fuse_msg_size
  428. FUSE applications communicate with the kernel over a special character
  429. device: `/dev/fuse`. A large portion of the overhead associated with
  430. FUSE is the cost of going back and forth between user space and kernel
  431. space over that device. Generally speaking, the fewer trips needed the
  432. better the performance will be. Reducing the number of trips can be
  433. done a number of ways. Kernel level caching and increasing message
  434. sizes being two significant ones. When it comes to reads and writes if
  435. the message size is doubled the number of trips are approximately
  436. halved.
  437. In Linux 4.20 a new feature was added allowing the negotiation of the
  438. max message size. Since the size is in multiples of
  439. [pages](https://en.wikipedia.org/wiki/Page_(computer_memory)) the
  440. feature is called `max_pages`. There is a maximum `max_pages` value of
  441. 256 (1MiB) and minimum of 1 (4KiB). The default used by Linux >=4.20,
  442. and hardcoded value used before 4.20, is 32 (128KiB). In mergerfs it's
  443. referred to as `fuse_msg_size` to make it clear what it impacts and
  444. provide some abstraction.
  445. Since there should be no downsides to increasing `fuse_msg_size` /
  446. `max_pages`, outside a minor bump in RAM usage due to larger message
  447. buffers, mergerfs defaults the value to 256. On kernels before 4.20
  448. the value has no effect. The reason the value is configurable is to
  449. enable experimentation and benchmarking. See the BENCHMARKING section
  450. for examples.
  451. ### follow-symlinks
  452. This feature, when enabled, will cause symlinks to be interpreted by
  453. mergerfs as their target (depending on the mode).
  454. When there is a getattr/stat request for a file mergerfs will check if
  455. the file is a symlink and depending on the `follow-symlinks` setting
  456. will replace the information about the symlink with that of that which
  457. it points to.
  458. When unlink'ing or rmdir'ing the followed symlink it will remove the
  459. symlink itself and not that which it points to.
  460. * never: Behave as normal. Symlinks are treated as such.
  461. * directory: Resolve symlinks only which point to directories.
  462. * regular: Resolve symlinks only which point to regular files.
  463. * all: Resolve all symlinks to that which they point to.
  464. Symlinks which do not point to anything are left as is.
  465. WARNING: This feature works but there might be edge cases yet
  466. found. If you find any odd behaviors please file a ticket on
  467. [github](https://github.com/trapexit/mergerfs/issues).
  468. ### link-exdev
  469. If using path preservation and a `link` fails with EXDEV make a call
  470. to `symlink` where the `target` is the `oldlink` and the `linkpath` is
  471. the `newpath`. The `target` value is determined by the value of
  472. `link-exdev`.
  473. * passthrough: Return EXDEV as normal.
  474. * rel-symlink: A relative path from the `newpath`.
  475. * abs-base-symlink: An absolute value using the underlying branch.
  476. * abs-pool-symlink: An absolute value using the mergerfs mount point.
  477. NOTE: It is possible that some applications check the file they
  478. link. In those cases, it is possible it will error or complain.
  479. ### rename-exdev
  480. If using path preservation and a `rename` fails with EXDEV:
  481. 1. Move file from **/branch/a/b/c** to **/branch/.mergerfs_rename_exdev/a/b/c**.
  482. 2. symlink the rename's `newpath` to the moved file.
  483. The `target` value is determined by the value of `rename-exdev`.
  484. * passthrough: Return EXDEV as normal.
  485. * rel-symlink: A relative path from the `newpath`.
  486. * abs-symlink: An absolute value using the mergerfs mount point.
  487. NOTE: It is possible that some applications check the file they
  488. rename. In those cases it is possible it will error or complain.
  489. NOTE: The reason `abs-symlink` is not split into two like `link-exdev`
  490. is due to the complexities in managing absolute base symlinks when
  491. multiple `oldpaths` exist.
  492. ### symlinkify
  493. Due to the levels of indirection introduced by mergerfs and the
  494. underlying technology FUSE there can be varying levels of performance
  495. degradation. This feature will turn non-directories which are not
  496. writable into symlinks to the original file found by the `readlink`
  497. policy after the mtime and ctime are older than the timeout.
  498. **WARNING:** The current implementation has a known issue in which if
  499. the file is open and being used when the file is converted to a
  500. symlink then the application which has that file open will receive an
  501. error when using it. This is unlikely to occur in practice but is
  502. something to keep in mind.
  503. **WARNING:** Some backup solutions, such as CrashPlan, do not backup
  504. the target of a symlink. If using this feature it will be necessary to
  505. point any backup software to the original filesystems or configure the
  506. software to follow symlinks if such an option is available.
  507. Alternatively, create two mounts. One for backup and one for general
  508. consumption.
  509. ### nullrw
  510. Due to how FUSE works there is an overhead to all requests made to a
  511. FUSE filesystem that wouldn't exist for an in kernel one. Meaning that
  512. even a simple passthrough will have some slowdown. However, generally
  513. the overhead is minimal in comparison to the cost of the underlying
  514. I/O. By disabling the underlying I/O we can test the theoretical
  515. performance boundaries.
  516. By enabling `nullrw` mergerfs will work as it always does **except**
  517. that all reads and writes will be no-ops. A write will succeed (the
  518. size of the write will be returned as if it were successful) but
  519. mergerfs does nothing with the data it was given. Similarly a read
  520. will return the size requested but won't touch the buffer.
  521. See the BENCHMARKING section for suggestions on how to test.
  522. ### xattr
  523. Runtime extended attribute support can be managed via the `xattr`
  524. option. By default it will passthrough any xattr calls. Given xattr
  525. support is rarely used and can have significant performance
  526. implications mergerfs allows it to be disabled at runtime. The
  527. performance problems mostly comes when file caching is enabled. The
  528. kernel will send a `getxattr` for `security.capability` *before every
  529. single write*. It doesn't cache the responses to any `getxattr`. This
  530. might be addressed in the future but for now mergerfs can really only
  531. offer the following workarounds.
  532. `noattr` will cause mergerfs to short circuit all xattr calls and
  533. return ENOATTR where appropriate. mergerfs still gets all the requests
  534. but they will not be forwarded on to the underlying filesystems. The
  535. runtime control will still function in this mode.
  536. `nosys` will cause mergerfs to return ENOSYS for any xattr call. The
  537. difference with `noattr` is that the kernel will cache this fact and
  538. itself short circuit future calls. This is more efficient than
  539. `noattr` but will cause mergerfs' runtime control via the hidden file
  540. to stop working.
  541. ### nfsopenhack
  542. NFS is not fully POSIX compliant and historically certain behaviors,
  543. such as opening files with O_EXCL, are not or not well supported. When
  544. mergerfs (or any FUSE filesystem) is exported over NFS some of these
  545. issues come up due to how NFS and FUSE interact.
  546. This hack addresses the issue where the creation of a file with a
  547. read-only mode but with a read/write or write only flag. Normally this
  548. is perfectly valid but NFS chops the one open call into multiple
  549. calls. Exactly how it is translated depends on the configuration and
  550. versions of the NFS server and clients but it results in a permission
  551. error because a normal user is not allowed to open a read-only file as
  552. writable.
  553. Even though it's a more niche situation this hack breaks normal
  554. security and behavior and as such is `off` by default. If set to `git`
  555. it will only perform the hack when the path in question includes
  556. `/.git/`. `all` will result in it applying anytime a read-only file which
  557. is empty is opened for writing.
  558. ### export-support
  559. In theory, this flag should not be exposed to the end user. It is a
  560. low-level FUSE flag which indicates whether or not the kernel can send
  561. certain kinds of messages to it for the purposes of using it with
  562. NFS. mergerfs does support these messages but due to bugs and quirks
  563. found in the kernel and mergerfs this option is provided just in case
  564. it is needed for debugging.
  565. Given that this flag is set when the FUSE connection is first
  566. initiated it is not possible to change during run time.
  567. # FUNCTIONS, CATEGORIES and POLICIES
  568. The POSIX filesystem API is made up of a number of
  569. functions. **creat**, **stat**, **chown**, etc. For ease of
  570. configuration in mergerfs, most of the core functions are grouped into
  571. 3 categories: **action**, **create**, and **search**. These functions
  572. and categories can be assigned a policy which dictates which branch is
  573. chosen when performing that function.
  574. Some functions, listed in the category `N/A` below, can not be
  575. assigned the normal policies. These functions work with file handles,
  576. rather than file paths, which were created by `open` or `create`. That
  577. said many times the current FUSE kernel driver will not always provide
  578. the file handle when a client calls `fgetattr`, `fchown`, `fchmod`,
  579. `futimens`, `ftruncate`, etc. This means it will call the regular,
  580. path based, versions. `statfs`'s behavior can be modified via other
  581. options.
  582. When using policies which are based on a branch's available space the
  583. base path provided is used. Not the full path to the file in
  584. question. Meaning that mounts in the branch won't be considered in the
  585. space calculations. The reason is that it doesn't really work for
  586. non-path preserving policies and can lead to non-obvious behaviors.
  587. NOTE: While any policy can be assigned to a function or category,
  588. some may not be very useful in practice. For instance: **rand**
  589. (random) may be useful for file creation (create) but could lead to
  590. very odd behavior if used for `chmod` if there were more than one copy
  591. of the file.
  592. ### Functions and their Category classifications
  593. | Category | FUSE Functions |
  594. |----------|-------------------------------------------------------------------------------------|
  595. | action | chmod, chown, link, removexattr, rename, rmdir, setxattr, truncate, unlink, utimens |
  596. | create | create, mkdir, mknod, symlink |
  597. | search | access, getattr, getxattr, ioctl (directories), listxattr, open, readlink |
  598. | N/A | fchmod, fchown, futimens, ftruncate, fallocate, fgetattr, fsync, ioctl (files), read, readdir, release, statfs, write, copy_file_range |
  599. In cases where something may be searched for (such as a path to clone)
  600. **getattr** will usually be used.
  601. ### Policies
  602. A policy is the algorithm used to choose a branch or branches for a
  603. function to work on or generally how the function behaves.
  604. Any function in the `create` category will clone the relative path if
  605. needed. Some other functions (`rename`,`link`,`ioctl`) have special
  606. requirements or behaviors which you can read more about below.
  607. #### Filtering
  608. Most policies basically search branches and create a list of files / paths
  609. for functions to work on. The policy is responsible for filtering and
  610. sorting the branches. Filters include **minfreespace**, whether or not
  611. a branch is mounted read-only, and the branch tagging
  612. (RO,NC,RW). These filters are applied across most policies.
  613. * No **search** function policies filter.
  614. * All **action** function policies filter out branches which are
  615. mounted **read-only** or tagged as **RO (read-only)**.
  616. * All **create** function policies filter out branches which are
  617. mounted **read-only**, tagged **RO (read-only)** or **NC (no
  618. create)**, or has available space less than `minfreespace`.
  619. Policies may have their own additional filtering such as those that
  620. require existing paths to be present.
  621. If all branches are filtered an error will be returned. Typically
  622. **EROFS** (read-only filesystem) or **ENOSPC** (no space left on
  623. device) depending on the most recent reason for filtering a
  624. branch. **ENOENT** will be returned if no eligible branch is found.
  625. If **create**, **mkdir**, **mknod**, or **symlink** fail with `EROFS`
  626. or other fundamental errors then mergerfs will mark any branch found
  627. to be read-only as such (IE will set the mode `RO`) and will rerun the
  628. policy and try again. This is mostly for `ext4` filesystems that can
  629. suddenly become read-only when it encounters an error.
  630. #### Path Preservation
  631. Policies, as described below, are of two basic classifications. `path
  632. preserving` and `non-path preserving`.
  633. All policies which start with `ep` (**epff**, **eplfs**, **eplus**,
  634. **epmfs**, **eprand**) are `path preserving`. `ep` stands for
  635. `existing path`.
  636. A path preserving policy will only consider branches where the relative
  637. path being accessed already exists.
  638. When using non-path preserving policies paths will be cloned to target
  639. branches as necessary.
  640. With the `msp` or `most shared path` policies they are defined as
  641. `path preserving` for the purpose of controlling `link` and `rename`'s
  642. behaviors since `ignorepponrename` is available to disable that
  643. behavior.
  644. #### Policy descriptions
  645. A policy's behavior differs, as mentioned above, based on the function
  646. it is used with. Sometimes it really might not make sense to even
  647. offer certain policies because they are literally the same as others
  648. but it makes things a bit more uniform.
  649. | Policy | Description |
  650. |------------------|------------------------------------------------------------|
  651. | all | Search: For **mkdir**, **mknod**, and **symlink** it will apply to all branches. **create** works like **ff**. |
  652. | epall (existing path, all) | For **mkdir**, **mknod**, and **symlink** it will apply to all found. **create** works like **epff** (but more expensive because it doesn't stop after finding a valid branch). |
  653. | epff (existing path, first found) | Given the order of the branches, as defined at mount time or configured at runtime, act on the first one found where the relative path exists. |
  654. | eplfs (existing path, least free space) | Of all the branches on which the relative path exists choose the branch with the least free space. |
  655. | eplus (existing path, least used space) | Of all the branches on which the relative path exists choose the branch with the least used space. |
  656. | epmfs (existing path, most free space) | Of all the branches on which the relative path exists choose the branch with the most free space. |
  657. | eppfrd (existing path, percentage free random distribution) | Like **pfrd** but limited to existing paths. |
  658. | eprand (existing path, random) | Calls **epall** and then randomizes. Returns 1. |
  659. | ff (first found) | Given the order of the branches, as defined at mount time or configured at runtime, act on the first one found. |
  660. | lfs (least free space) | Pick the branch with the least available free space. |
  661. | lus (least used space) | Pick the branch with the least used space. |
  662. | mfs (most free space) | Pick the branch with the most available free space. |
  663. | msplfs (most shared path, least free space) | Like **eplfs** but if it fails to find a branch it will try again with the parent directory. Continues this pattern till finding one. |
  664. | msplus (most shared path, least used space) | Like **eplus** but if it fails to find a branch it will try again with the parent directory. Continues this pattern till finding one. |
  665. | mspmfs (most shared path, most free space) | Like **epmfs** but if it fails to find a branch it will try again with the parent directory. Continues this pattern till finding one. |
  666. | msppfrd (most shared path, percentage free random distribution) | Like **eppfrd** but if it fails to find a branch it will try again with the parent directory. Continues this pattern till finding one. |
  667. | newest | Pick the file / directory with the largest mtime. |
  668. | pfrd (percentage free random distribution) | Chooses a branch at random with the likelihood of selection based on a branch's available space relative to the total. |
  669. | rand (random) | Calls **all** and then randomizes. Returns 1 branch. |
  670. **NOTE:** If you are using an underlying filesystem that reserves
  671. blocks such as ext2, ext3, or ext4 be aware that mergerfs respects the
  672. reservation by using `f_bavail` (number of free blocks for
  673. unprivileged users) rather than `f_bfree` (number of free blocks) in
  674. policy calculations. **df** does NOT use `f_bavail`, it uses
  675. `f_bfree`, so direct comparisons between **df** output and mergerfs'
  676. policies is not appropriate.
  677. #### Defaults
  678. | Category | Policy |
  679. |----------|--------|
  680. | action | epall |
  681. | create | epmfs |
  682. | search | ff |
  683. #### func.readdir
  684. examples: `func.readdir=seq`, `func.readdir=cor:4`
  685. `readdir` has policies to control how it manages reading directory
  686. content.
  687. | Policy | Description |
  688. |--------|-------------|
  689. | seq | "sequential" : Iterate over branches in the order defined. This is the default and traditional behavior found prior to the readdir policy introduction. |
  690. | cosr | "concurrent open, sequential read" : Concurrently open branch directories using a thread pool and process them in order of definition. This keeps memory and CPU usage low while also reducing the time spent waiting on branches to respond. Number of threads defaults to the number of logical cores. Can be overwritten via the syntax `func.readdir=cosr:N` where `N` is the number of threads. |
  691. | cor | "concurrent open and read" : Concurrently open branch directories and immediately start reading their contents using a thread pool. This will result in slightly higher memory and CPU usage but reduced latency. Particularly when using higher latency / slower speed network filesystem branches. Unlike `seq` and `cosr` the order of files could change due the async nature of the thread pool. Number of threads defaults to the number of logical cores. Can be overwritten via the syntax `func.readdir=cor:N` where `N` is the number of threads.
  692. Keep in mind that `readdir` mostly just provides a list of file names
  693. in a directory and possibly some basic metadata about said files. To
  694. know details about the files, as one would see from commands like
  695. `find` or `ls`, it is required to call `stat` on the file which is
  696. controlled by `fuse.getattr`.
  697. #### ioctl
  698. When `ioctl` is used with an open file then it will use the file
  699. handle which was created at the original `open` call. However, when
  700. using `ioctl` with a directory mergerfs will use the `open` policy to
  701. find the directory to act on.
  702. #### rename & link ####
  703. **NOTE:** If you're receiving errors from software when files are
  704. moved / renamed / linked then you should consider changing the create
  705. policy to one which is **not** path preserving, enabling
  706. `ignorepponrename`, or contacting the author of the offending software
  707. and requesting that `EXDEV` (cross device / improper link) be properly
  708. handled.
  709. `rename` and `link` are tricky functions in a union
  710. filesystem. `rename` only works within a single filesystem or
  711. device. If a rename can't be done atomically due to the source and
  712. destination paths existing on different mount points it will return
  713. **-1** with **errno = EXDEV** (cross device / improper link). So if a
  714. `rename`'s source and target are on different filesystems within the pool
  715. it creates an issue.
  716. Originally mergerfs would return EXDEV whenever a rename was requested
  717. which was cross directory in any way. This made the code simple and
  718. was technically compliant with POSIX requirements. However, many
  719. applications fail to handle EXDEV at all and treat it as a normal
  720. error or otherwise handle it poorly. Such apps include: gvfsd-fuse
  721. v1.20.3 and prior, Finder / CIFS/SMB client in Apple OSX 10.9+,
  722. NZBGet, Samba's recycling bin feature.
  723. As a result a compromise was made in order to get most software to
  724. work while still obeying mergerfs' policies. Below is the basic logic.
  725. * If using a **create** policy which tries to preserve directory paths (epff,eplfs,eplus,epmfs)
  726. * Using the **rename** policy get the list of files to rename
  727. * For each file attempt rename:
  728. * If failure with ENOENT (no such file or directory) run **create** policy
  729. * If create policy returns the same branch as currently evaluating then clone the path
  730. * Re-attempt rename
  731. * If **any** of the renames succeed the higher level rename is considered a success
  732. * If **no** renames succeed the first error encountered will be returned
  733. * On success:
  734. * Remove the target from all branches with no source file
  735. * Remove the source from all branches which failed to rename
  736. * If using a **create** policy which does **not** try to preserve directory paths
  737. * Using the **rename** policy get the list of files to rename
  738. * Using the **getattr** policy get the target path
  739. * For each file attempt rename:
  740. * If the source branch != target branch:
  741. * Clone target path from target branch to source branch
  742. * Rename
  743. * If **any** of the renames succeed the higher level rename is considered a success
  744. * If **no** renames succeed the first error encountered will be returned
  745. * On success:
  746. * Remove the target from all branches with no source file
  747. * Remove the source from all branches which failed to rename
  748. The removals are subject to normal entitlement checks.
  749. The above behavior will help minimize the likelihood of EXDEV being
  750. returned but it will still be possible.
  751. **link** uses the same strategy but without the removals.
  752. #### statfs / statvfs ####
  753. [statvfs](http://linux.die.net/man/2/statvfs) normalizes the source
  754. filesystems based on the fragment size and sums the number of adjusted
  755. blocks and inodes. This means you will see the combined space of all
  756. sources. Total, used, and free. The sources however are dedupped based
  757. on the filesystem so multiple sources on the same drive will not result in
  758. double counting its space. Other filesystems mounted further down the tree
  759. of the branch will not be included when checking the mount's stats.
  760. The options `statfs` and `statfs_ignore` can be used to modify
  761. `statfs` behavior.
  762. #### flush-on-close
  763. https://lkml.kernel.org/linux-fsdevel/20211024132607.1636952-1-amir73il@gmail.com/T/
  764. By default, FUSE would issue a flush before the release of a file
  765. descriptor. This was considered a bit aggressive and a feature added
  766. to give the FUSE server the ability to choose when that happens.
  767. Options:
  768. * always
  769. * never
  770. * opened-for-write
  771. For now it defaults to "opened-for-write" which is less aggressive
  772. than the behavior before this feature was added. It should not be a
  773. problem because the flush is really only relevant when a file is
  774. written to. Given flush is irrelevant for many filesystems in the
  775. future a branch specific flag may be added so only files opened on a
  776. specific branch would be flushed on close.
  777. # ERROR HANDLING
  778. POSIX filesystem functions offer a single return code meaning that
  779. there is some complication regarding the handling of multiple branches
  780. as mergerfs does. It tries to handle errors in a way that would
  781. generally return meaningful values for that particular function.
  782. ### chmod, chown, removexattr, setxattr, truncate, utimens
  783. 1) if no error: return 0 (success)
  784. 2) if no successes: return first error
  785. 3) if one of the files acted on was the same as the related search function: return its value
  786. 4) return 0 (success)
  787. While doing this increases the complexity and cost of error handling,
  788. particularly step 3, this provides probably the most reasonable return
  789. value.
  790. ### unlink, rmdir
  791. 1) if no errors: return 0 (success)
  792. 2) return first error
  793. Older versions of mergerfs would return success if any success occurred
  794. but for unlink and rmdir there are downstream assumptions that, while
  795. not impossible to occur, can confuse some software.
  796. ### others
  797. For search functions, there is always a single thing acted on and as
  798. such whatever return value that comes from the single function call is
  799. returned.
  800. For create functions `mkdir`, `mknod`, and `symlink` which don't
  801. return a file descriptor and therefore can have `all` or `epall`
  802. policies it will return success if any of the calls succeed and an
  803. error otherwise.
  804. # INSTALL
  805. https://github.com/trapexit/mergerfs/releases
  806. If your distribution's package manager includes mergerfs check if the
  807. version is up to date. If out of date it is recommended to use
  808. the latest release found on the release page. Details for common
  809. distros are below.
  810. #### Debian
  811. Most Debian installs are of a stable branch and therefore do not have
  812. the most up to date software. While mergerfs is available via `apt` it
  813. is suggested that users install the most recent version available from
  814. the [releases page](https://github.com/trapexit/mergerfs/releases).
  815. #### prebuilt deb
  816. ```
  817. wget https://github.com/trapexit/mergerfs/releases/download/<ver>/mergerfs_<ver>.debian-<rel>_<arch>.deb
  818. dpkg -i mergerfs_<ver>.debian-<rel>_<arch>.deb
  819. ```
  820. #### apt
  821. ```
  822. sudo apt install -y mergerfs
  823. ```
  824. #### Ubuntu
  825. Most Ubuntu installs are of a stable branch and therefore do not have
  826. the most up to date software. While mergerfs is available via `apt` it
  827. is suggested that users install the most recent version available from
  828. the [releases page](https://github.com/trapexit/mergerfs/releases).
  829. #### prebuilt deb
  830. ```
  831. wget https://github.com/trapexit/mergerfs/releases/download/<version>/mergerfs_<ver>.ubuntu-<rel>_<arch>.deb
  832. dpkg -i mergerfs_<ver>.ubuntu-<rel>_<arch>.deb
  833. ```
  834. #### apt
  835. ```
  836. sudo apt install -y mergerfs
  837. ```
  838. #### Raspberry Pi OS
  839. Effectively the same as Debian or Ubuntu.
  840. #### Fedora
  841. ```
  842. wget https://github.com/trapexit/mergerfs/releases/download/<ver>/mergerfs-<ver>.fc<rel>.<arch>.rpm
  843. sudo rpm -i mergerfs-<ver>.fc<rel>.<arch>.rpm
  844. ```
  845. #### CentOS / Rocky
  846. ```
  847. wget https://github.com/trapexit/mergerfs/releases/download/<ver>/mergerfs-<ver>.el<rel>.<arch>.rpm
  848. sudo rpm -i mergerfs-<ver>.el<rel>.<arch>.rpm
  849. ```
  850. #### ArchLinux
  851. 1. Setup AUR
  852. 2. Install `mergerfs`
  853. #### Other
  854. Static binaries are provided for situations where native packages are
  855. unavailable.
  856. ```
  857. wget https://github.com/trapexit/mergerfs/releases/download/<ver>/mergerfs-static-linux_<arch>.tar.gz
  858. sudo tar xvf mergerfs-static-linux_<arch>.tar.gz -C /
  859. ```
  860. # BUILD
  861. **NOTE:** Prebuilt packages can be found at and recommended for most
  862. users: https://github.com/trapexit/mergerfs/releases
  863. **NOTE:** Only tagged releases are supported. `master` and other
  864. branches should be considered works in progress.
  865. First, get the code from [github](https://github.com/trapexit/mergerfs).
  866. ```
  867. $ git clone https://github.com/trapexit/mergerfs.git
  868. $ # or
  869. $ wget https://github.com/trapexit/mergerfs/releases/download/<ver>/mergerfs-<ver>.tar.gz
  870. ```
  871. #### Debian / Ubuntu
  872. ```
  873. $ cd mergerfs
  874. $ sudo tools/install-build-pkgs
  875. $ make deb
  876. $ sudo dpkg -i ../mergerfs_<version>_<arch>.deb
  877. ```
  878. #### RHEL / CentOS / Rocky / Fedora
  879. ```
  880. $ su -
  881. # cd mergerfs
  882. # tools/install-build-pkgs
  883. # make rpm
  884. # rpm -i rpmbuild/RPMS/<arch>/mergerfs-<version>.<arch>.rpm
  885. ```
  886. #### Generic
  887. Have git, g++, make, python installed.
  888. ```
  889. $ cd mergerfs
  890. $ make
  891. $ sudo make install
  892. ```
  893. #### Build options
  894. ```
  895. $ make help
  896. usage: make
  897. make USE_XATTR=0 - build program without xattrs functionality
  898. make STATIC=1 - build static binary
  899. make LTO=1 - build with link time optimization
  900. ```
  901. # UPGRADE
  902. mergerfs can be upgraded live by mounting on top of the previous
  903. instance. Simply install the new version of mergerfs and follow the
  904. instructions below.
  905. Run mergerfs again or if using `/etc/fstab` call for it to mount
  906. again. Existing open files and such will continue to work fine though
  907. they won't see runtime changes since any such change would be the new
  908. mount. If you plan on changing settings with the new mount you should
  909. / could apply those before mounting the new version.
  910. ```
  911. $ sudo mount /mnt/mergerfs
  912. $ mount | grep mergerfs
  913. media on /mnt/mergerfs type mergerfs (rw,relatime,user_id=0,group_id=0,default_permissions,allow_other)
  914. media on /mnt/mergerfs type mergerfs (rw,relatime,user_id=0,group_id=0,default_permissions,allow_other)
  915. ```
  916. A problem with this approach is that the underlying instance will
  917. continue to run even if the software using it stop or are
  918. restarted. To work around this you can use a "lazy umount". Before
  919. mounting over top the mount point with the new instance of mergerfs
  920. issue: `umount -l <mergerfs_mountpoint>`. Or you can let mergerfs do
  921. it by setting the option `lazy-umount-mountpoint=true`.
  922. # RUNTIME INTERFACES
  923. ## RUNTIME CONFIG
  924. #### .mergerfs pseudo file ####
  925. ```
  926. <mountpoint>/.mergerfs
  927. ```
  928. There is a pseudo file available at the mount point which allows for
  929. the runtime modification of certain **mergerfs** options. The file
  930. will not show up in **readdir** but can be **stat**'ed and manipulated
  931. via [{list,get,set}xattrs](http://linux.die.net/man/2/listxattr)
  932. calls.
  933. Any changes made at runtime are **not** persisted. If you wish for
  934. values to persist they must be included as options wherever you
  935. configure the mounting of mergerfs (/etc/fstab).
  936. ##### Keys #####
  937. Use `getfattr -d /mountpoint/.mergerfs` or `xattr -l
  938. /mountpoint/.mergerfs` to see all supported keys. Some are
  939. informational and therefore read-only. `setxattr` will return EINVAL
  940. (invalid argument) on read-only keys.
  941. ##### Values #####
  942. Same as the command line.
  943. ###### user.mergerfs.branches ######
  944. Used to query or modify the list of branches. When modifying there are
  945. several shortcuts to easy manipulation of the list.
  946. | Value | Description |
  947. |--------------|-------------|
  948. | [list] | set |
  949. | +<[list] | prepend |
  950. | +>[list] | append |
  951. | -[list] | remove all values provided |
  952. | -< | remove first in list |
  953. | -> | remove last in list |
  954. `xattr -w user.mergerfs.branches +</mnt/drive3 /mnt/pool/.mergerfs`
  955. The `=NC`, `=RO`, `=RW` syntax works just as on the command line.
  956. ##### Example #####
  957. ```
  958. [trapexit:/mnt/mergerfs] $ getfattr -d .mergerfs
  959. user.mergerfs.branches="/mnt/a=RW:/mnt/b=RW"
  960. user.mergerfs.minfreespace="4294967295"
  961. user.mergerfs.moveonenospc="false"
  962. ...
  963. [trapexit:/mnt/mergerfs] $ getfattr -n user.mergerfs.category.search .mergerfs
  964. user.mergerfs.category.search="ff"
  965. [trapexit:/mnt/mergerfs] $ setfattr -n user.mergerfs.category.search -v newest .mergerfs
  966. [trapexit:/mnt/mergerfs] $ getfattr -n user.mergerfs.category.search .mergerfs
  967. user.mergerfs.category.search="newest"
  968. ```
  969. #### file / directory xattrs ####
  970. While they won't show up when using `getfattr` **mergerfs** offers a
  971. number of special xattrs to query information about the files
  972. served. To access the values you will need to issue a
  973. [getxattr](http://linux.die.net/man/2/getxattr) for one of the
  974. following:
  975. * **user.mergerfs.basepath**: the base mount point for the file given the current getattr policy
  976. * **user.mergerfs.relpath**: the relative path of the file from the perspective of the mount point
  977. * **user.mergerfs.fullpath**: the full path of the original file given the getattr policy
  978. * **user.mergerfs.allpaths**: a NUL ('\0') separated list of full paths to all files found
  979. ## SIGNALS
  980. * USR1: This will cause mergerfs to send invalidation notifications to
  981. the kernel for all files. This will cause all unused files to be
  982. released from memory.
  983. * USR2: Trigger a general cleanup of currently unused memory. A more
  984. thorough version of what happens every ~15 minutes.
  985. ## IOCTLS
  986. Found in `fuse_ioctl.cpp`:
  987. ```C++
  988. typedef char IOCTL_BUF[4096];
  989. #define IOCTL_APP_TYPE 0xDF
  990. #define IOCTL_FILE_INFO _IOWR(IOCTL_APP_TYPE,0,IOCTL_BUF)
  991. #define IOCTL_GC _IO(IOCTL_APP_TYPE,1)
  992. #define IOCTL_GC1 _IO(IOCTL_APP_TYPE,2)
  993. #define IOCTL_INVALIDATE_ALL_NODES _IO(IOCTL_APP_TYPE,3)
  994. ```
  995. * IOCTL\_FILE\_INFO: Same as the "file / directory xattrs" mentioned
  996. above. Use a buffer size of 4096 bytes. Pass in a string of
  997. "basepath", "relpath", "fullpath", or "allpaths". Receive details in
  998. same buffer.
  999. * IOCTL\_GC: Triggers a thorough garbage collection of excess
  1000. memory. Same as SIGUSR2.
  1001. * IOCTL\_GC1: Triggers a simple garbage collection of excess
  1002. memory. Same as what happens every 15 minutes normally.
  1003. * IOCTL\_INVALIDATE\_ALL\_NODES: Same as SIGUSR1. Send invalidation
  1004. notifications to the kernel for all files causing unused files to be
  1005. released from memory.
  1006. # TOOLING
  1007. ## preload.so
  1008. EXPERIMENTAL
  1009. For some time there has been work to enable passthrough IO in
  1010. FUSE. Passthrough IO would allow for near native performance with
  1011. regards to reads and writes (at the expense of certain mergerfs
  1012. features.) However, there have been several complications which have
  1013. kept the feature from making it into the mainline Linux kernel. Until
  1014. that feature is available there are two methods to provide similar
  1015. functionality. One method is using the LD_PRELOAD feature of the
  1016. dynamic linker. The other leveraging ptrace to intercept
  1017. syscalls. Each has their disadvantages. At the moment only a preload
  1018. based tool is available. A ptrace based tool may be developed later if
  1019. there is a need.
  1020. `/usr/lib/mergerfs/preload.so`
  1021. This [preloadable
  1022. library](https://man7.org/linux/man-pages/man8/ld.so.8.html#ENVIRONMENT)
  1023. overrides the creation and opening of files in order to simulate
  1024. passthrough file IO. It catches the open/creat/fopen calls, has
  1025. mergerfs do the call, queries mergerfs for the branch the file exists
  1026. on, reopens the file on the underlying filesystem and returns that
  1027. instead. Meaning that you will get native read/write performance
  1028. because mergerfs is no longer part of the workflow. Keep in mind that
  1029. this also means certain mergerfs features that work by interrupting
  1030. the read/write workflow, such as `moveonenospc`, will no longer work.
  1031. Also, understand that this will only work on dynamically linked
  1032. software. Anything statically compiled will not work. Many GoLang and
  1033. Rust apps are statically compiled.
  1034. The library will not interfere with non-mergerfs filesystems. The
  1035. library is written to always fallback to returning the mergerfs opened
  1036. file on error.
  1037. While the library was written to account for a number of edgecases
  1038. there could be some yet accounted for so please report any oddities.
  1039. Thank you to
  1040. [nohajc](https://github.com/nohajc/mergerfs-io-passthrough) for
  1041. prototyping the idea.
  1042. ### general usage
  1043. ```sh
  1044. LD_PRELOAD=/usr/lib/mergerfs/preload.so touch /mnt/mergerfs/filename
  1045. ```
  1046. ### Docker usage
  1047. Assume `/mnt/fs0` and `/mnt/fs1` are pooled with mergerfs at `/media`.
  1048. All mergerfs branch paths *must* be bind mounted into the container at
  1049. the same path as found on the host so the preload library can see them.
  1050. ```sh
  1051. docker run \
  1052. -e LD_PRELOAD=/usr/lib/mergerfs/preload.so \
  1053. -v /usr/lib/mergerfs/preload.so:/usr/lib/mergerfs/preload.so:ro \
  1054. -v /media:/data \
  1055. -v /mnt:/mnt \
  1056. ubuntu:latest \
  1057. bash
  1058. ```
  1059. or more explicitly
  1060. ```sh
  1061. docker run \
  1062. -e LD_PRELOAD=/usr/lib/mergerfs/preload.so \
  1063. -v /usr/lib/mergerfs/preload.so:/usr/lib/mergerfs/preload.so:ro \
  1064. -v /media:/data \
  1065. -v /mnt/fs0:/mnt/fs0 \
  1066. -v /mnt/fs1:/mnt/fs1 \
  1067. ubuntu:latest \
  1068. bash
  1069. ```
  1070. ### systemd unit
  1071. Use the `Environment` option to set the LD_PRELOAD variable.
  1072. * https://www.freedesktop.org/software/systemd/man/latest/systemd.service.html#Command%20lines
  1073. * https://serverfault.com/questions/413397/how-to-set-environment-variable-in-systemd-service
  1074. ```
  1075. [Service]
  1076. Environment=LD_PRELOAD=/usr/lib/mergerfs/preload.so
  1077. ```
  1078. ## Misc
  1079. * https://github.com/trapexit/mergerfs-tools
  1080. * mergerfs.ctl: A tool to make it easier to query and configure mergerfs at runtime
  1081. * mergerfs.fsck: Provides permissions and ownership auditing and the ability to fix them
  1082. * mergerfs.dedup: Will help identify and optionally remove duplicate files
  1083. * mergerfs.dup: Ensure there are at least N copies of a file across the pool
  1084. * mergerfs.balance: Rebalance files across filesystems by moving them from the most filled to the least filled
  1085. * mergerfs.consolidate: move files within a single mergerfs directory to the filesystem with most free space
  1086. * https://github.com/trapexit/scorch
  1087. * scorch: A tool to help discover silent corruption of files and keep track of files
  1088. * https://github.com/trapexit/bbf
  1089. * bbf (bad block finder): a tool to scan for and 'fix' hard drive bad blocks and find the files using those blocks
  1090. # CACHING
  1091. #### page caching
  1092. https://en.wikipedia.org/wiki/Page_cache
  1093. * cache.files=off: Disables page caching. Underlying files cached,
  1094. mergerfs files are not.
  1095. * cache.files=partial: Enables page caching. Underlying files cached,
  1096. mergerfs files cached while open.
  1097. * cache.files=full: Enables page caching. Underlying files cached,
  1098. mergerfs files cached across opens.
  1099. * cache.files=auto-full: Enables page caching. Underlying files
  1100. cached, mergerfs files cached across opens if mtime and size are
  1101. unchanged since previous open.
  1102. * cache.files=libfuse: follow traditional libfuse `direct_io`,
  1103. `kernel_cache`, and `auto_cache` arguments.
  1104. * cache.files=per-process: Enable page caching (equivalent to
  1105. `cache.files=partial`) only for processes whose 'comm' name matches
  1106. one of the values defined in `cache.files.process-names`. If the
  1107. name does not match the file open is equivalent to
  1108. `cache.files=off`.
  1109. FUSE, which mergerfs uses, offers a number of page caching modes. mergerfs tries to simplify their use via the `cache.files`
  1110. option. It can and should replace usage of `direct_io`,
  1111. `kernel_cache`, and `auto_cache`.
  1112. Due to mergerfs using FUSE and therefore being a userland process
  1113. proxying existing filesystems the kernel will double cache the content
  1114. being read and written through mergerfs. Once from the underlying
  1115. filesystem and once from mergerfs (it sees them as two separate
  1116. entities). Using `cache.files=off` will keep the double caching from
  1117. happening by disabling caching of mergerfs but this has the side
  1118. effect that *all* read and write calls will be passed to mergerfs
  1119. which may be slower than enabling caching, you lose shared `mmap`
  1120. support which can affect apps such as rtorrent, and no read-ahead will
  1121. take place. The kernel will still cache the underlying filesystem data
  1122. but that only helps so much given mergerfs will still process all
  1123. requests.
  1124. If you do enable file page caching,
  1125. `cache.files=partial|full|auto-full`, you should also enable
  1126. `dropcacheonclose` which will cause mergerfs to instruct the kernel to
  1127. flush the underlying file's page cache when the file is closed. This
  1128. behavior is the same as the rsync fadvise / drop cache patch and Feh's
  1129. nocache project.
  1130. If most files are read once through and closed (like media) it is best
  1131. to enable `dropcacheonclose` regardless of caching mode in order to
  1132. minimize buffer bloat.
  1133. It is difficult to balance memory usage, cache bloat & duplication,
  1134. and performance. Ideally, mergerfs would be able to disable caching for
  1135. the files it reads/writes but allow page caching for itself. That
  1136. would limit the FUSE overhead. However, there isn't a good way to
  1137. achieve this. It would need to open all files with O_DIRECT which
  1138. places limitations on what the underlying filesystems would be
  1139. supported and complicates the code.
  1140. kernel documentation: https://www.kernel.org/doc/Documentation/filesystems/fuse-io.txt
  1141. #### entry & attribute caching
  1142. Given the relatively high cost of FUSE due to the kernel <-> userspace
  1143. round trips there are kernel side caches for file entries and
  1144. attributes. The entry cache limits the `lookup` calls to mergerfs
  1145. which ask if a file exists. The attribute cache limits the need to
  1146. make `getattr` calls to mergerfs which provide file attributes (mode,
  1147. size, type, etc.). As with the page cache these should not be used if
  1148. the underlying filesystems are being manipulated at the same time as
  1149. it could lead to odd behavior or data corruption. The options for
  1150. setting these are `cache.entry` and `cache.negative_entry` for the
  1151. entry cache and `cache.attr` for the attributes
  1152. cache. `cache.negative_entry` refers to the timeout for negative
  1153. responses to lookups (non-existent files).
  1154. #### writeback caching
  1155. When `cache.files` is enabled the default is for it to perform
  1156. writethrough caching. This behavior won't help improve performance as
  1157. each write still goes one for one through the filesystem. By enabling
  1158. the FUSE writeback cache small writes may be aggregated by the kernel
  1159. and then sent to mergerfs as one larger request. This can greatly
  1160. improve the throughput for apps which write to files
  1161. inefficiently. The amount the kernel can aggregate is limited by the
  1162. size of a FUSE message. Read the `fuse_msg_size` section for more
  1163. details.
  1164. There is a small side effect as a result of enabling writeback
  1165. caching. Underlying files won't ever be opened with O_APPEND or
  1166. O_WRONLY. The former because the kernel then manages append mode and
  1167. the latter because the kernel may request file data from mergerfs to
  1168. populate the write cache. The O_APPEND change means that if a file is
  1169. changed outside of mergerfs it could lead to corruption as the kernel
  1170. won't know the end of the file has changed. That said any time you use
  1171. caching you should keep from using the same file outside of mergerfs
  1172. at the same time.
  1173. Note that if an application is properly sizing writes then writeback
  1174. caching will have little or no effect. It will only help with writes
  1175. of sizes below the FUSE message size (128K on older kernels, 1M on
  1176. newer).
  1177. #### statfs caching
  1178. Of the syscalls used by mergerfs in policies the `statfs` / `statvfs`
  1179. call is perhaps the most expensive. It's used to find out the
  1180. available space of a filesystem and whether it is mounted
  1181. read-only. Depending on the setup and usage pattern these queries can
  1182. be relatively costly. When `cache.statfs` is enabled all calls to
  1183. `statfs` by a policy will be cached for the number of seconds its set
  1184. to.
  1185. Example: If the create policy is `mfs` and the timeout is 60 then for
  1186. that 60 seconds the same filesystem will be returned as the target for
  1187. creates because the available space won't be updated for that time.
  1188. #### symlink caching
  1189. As of version 4.20 Linux supports symlink caching. Significant
  1190. performance increases can be had in workloads which use a lot of
  1191. symlinks. Setting `cache.symlinks=true` will result in requesting
  1192. symlink caching from the kernel only if supported. As a result it's
  1193. safe to enable it on systems prior to 4.20. That said it is disabled
  1194. by default for now. You can see if caching is enabled by querying the
  1195. xattr `user.mergerfs.cache.symlinks` but given it must be requested at
  1196. startup you can not change it at runtime.
  1197. #### readdir caching
  1198. As of version 4.20 Linux supports readdir caching. This can have a
  1199. significant impact on directory traversal. Especially when combined
  1200. with entry (`cache.entry`) and attribute (`cache.attr`)
  1201. caching. Setting `cache.readdir=true` will result in requesting
  1202. readdir caching from the kernel on each `opendir`. If the kernel
  1203. doesn't support readdir caching setting the option to `true` has no
  1204. effect. This option is configurable at runtime via xattr
  1205. `user.mergerfs.cache.readdir`.
  1206. #### tiered caching
  1207. Some storage technologies support what some call "tiered" caching. The
  1208. placing of usually smaller, faster storage as a transparent cache to
  1209. larger, slower storage. NVMe, SSD, Optane in front of traditional HDDs
  1210. for instance.
  1211. mergerfs does not natively support any sort of tiered caching. Most
  1212. users have no use for such a feature and its inclusion would
  1213. complicate the code. However, there are a few situations where a cache
  1214. filesystem could help with a typical mergerfs setup.
  1215. 1. Fast network, slow filesystems, many readers: You've a 10+Gbps network
  1216. with many readers and your regular filesystems can't keep up.
  1217. 2. Fast network, slow filesystems, small'ish bursty writes: You have a
  1218. 10+Gbps network and wish to transfer amounts of data less than your
  1219. cache filesystem but wish to do so quickly.
  1220. With #1 it's arguable if you should be using mergerfs at all. RAID
  1221. would probably be the better solution. If you're going to use mergerfs
  1222. there are other tactics that may help: spreading the data across
  1223. filesystems (see the mergerfs.dup tool) and setting `func.open=rand`,
  1224. using `symlinkify`, or using dm-cache or a similar technology to add
  1225. tiered cache to the underlying device.
  1226. With #2 one could use dm-cache as well but there is another solution
  1227. which requires only mergerfs and a cronjob.
  1228. 1. Create 2 mergerfs pools. One which includes just the slow devices
  1229. and one which has both the fast devices (SSD,NVME,etc.) and slow
  1230. devices.
  1231. 2. The 'cache' pool should have the cache filesystems listed first.
  1232. 3. The best `create` policies to use for the 'cache' pool would
  1233. probably be `ff`, `epff`, `lfs`, or `eplfs`. The latter two under
  1234. the assumption that the cache filesystem(s) are far smaller than the
  1235. backing filesystems. If using path preserving policies remember that
  1236. you'll need to manually create the core directories of those paths
  1237. you wish to be cached. Be sure the permissions are in sync. Use
  1238. `mergerfs.fsck` to check / correct them. You could also set the
  1239. slow filesystems mode to `NC` though that'd mean if the cache
  1240. filesystems fill you'd get "out of space" errors.
  1241. 4. Enable `moveonenospc` and set `minfreespace` appropriately. To make
  1242. sure there is enough room on the "slow" pool you might want to set
  1243. `minfreespace` to at least as large as the size of the largest
  1244. cache filesystem if not larger. This way in the worst case the
  1245. whole of the cache filesystem(s) can be moved to the other drives.
  1246. 5. Set your programs to use the cache pool.
  1247. 6. Save one of the below scripts or create you're own.
  1248. 7. Use `cron` (as root) to schedule the command at whatever frequency
  1249. is appropriate for your workflow.
  1250. ##### time based expiring
  1251. Move files from cache to backing pool based only on the last time the
  1252. file was accessed. Replace `-atime` with `-amin` if you want minutes
  1253. rather than days. May want to use the `fadvise` / `--drop-cache`
  1254. version of rsync or run rsync with the tool "nocache".
  1255. *NOTE:* The arguments to these scripts include the cache
  1256. **filesystem** itself. Not the pool with the cache filesystem. You
  1257. could have data loss if the source is the cache pool.
  1258. [mergerfs.time-based-mover](tools/mergerfs.time-based-mover?raw=1)
  1259. ##### percentage full expiring
  1260. Move the oldest file from the cache to the backing pool. Continue till
  1261. below percentage threshold.
  1262. *NOTE:* The arguments to these scripts include the cache
  1263. **filesystem** itself. Not the pool with the cache filesystem. You
  1264. could have data loss if the source is the cache pool.
  1265. [mergerfs.percent-full-mover](tools/mergerfs.percent-full-mover?raw=1)
  1266. # PERFORMANCE
  1267. mergerfs is at its core just a proxy and therefore its theoretical max
  1268. performance is that of the underlying devices. However, given it is a
  1269. FUSE filesystem working from userspace there is an increase in
  1270. overhead relative to kernel based solutions. That said the performance
  1271. can match the theoretical max but it depends greatly on the system's
  1272. configuration. Especially when adding network filesystems into the mix
  1273. there are many variables which can impact performance. Device speeds
  1274. and latency, network speeds and latency, general concurrency,
  1275. read/write sizes, etc. Unfortunately, given the number of variables it
  1276. has been difficult to find a single set of settings which provide
  1277. optimal performance. If you're having performance issues please look
  1278. over the suggestions below (including the benchmarking section.)
  1279. NOTE: be sure to read about these features before changing them to
  1280. understand what behaviors it may impact
  1281. * disable `security_capability` and/or `xattr`
  1282. * increase cache timeouts `cache.attr`, `cache.entry`, `cache.negative_entry`
  1283. * enable (or disable) page caching (`cache.files`)
  1284. * enable `parallel-direct-writes`
  1285. * enable `cache.writeback`
  1286. * enable `cache.statfs`
  1287. * enable `cache.symlinks`
  1288. * enable `cache.readdir`
  1289. * change the number of worker threads
  1290. * disable `posix_acl`
  1291. * disable `async_read`
  1292. * test theoretical performance using `nullrw` or mounting a ram disk
  1293. * use `symlinkify` if your data is largely static and read-only
  1294. * use tiered cache devices
  1295. * use LVM and LVM cache to place a SSD in front of your HDDs
  1296. * increase readahead: `readahead=1024`
  1297. If you come across a setting that significantly impacts performance
  1298. please contact trapexit so he may investigate further. Please test
  1299. both against your normal setup, a singular branch, and with
  1300. `nullrw=true`
  1301. # BENCHMARKING
  1302. Filesystems are complicated. They do many things and many of those are
  1303. interconnected. Additionally, the OS, drivers, hardware, etc. can all
  1304. impact performance. Therefore, when benchmarking, it is **necessary**
  1305. that the test focuses as narrowly as possible.
  1306. For most throughput is the key benchmark. To test throughput `dd` is
  1307. useful but **must** be used with the correct settings in order to
  1308. ensure the filesystem or device is actually being tested. The OS can
  1309. and will cache data. Without forcing synchronous reads and writes
  1310. and/or disabling caching the values returned will not be
  1311. representative of the device's true performance.
  1312. When benchmarking through mergerfs ensure you only use 1 branch to
  1313. remove any possibility of the policies complicating the
  1314. situation. Benchmark the underlying filesystem first and then mount
  1315. mergerfs over it and test again. If you're experiencing speeds below
  1316. your expectation you will need to narrow down precisely which
  1317. component is leading to the slowdown. Preferably test the following in
  1318. the order listed (but not combined).
  1319. 1. Enable `nullrw` mode with `nullrw=true`. This will effectively make
  1320. reads and writes no-ops. Removing the underlying device /
  1321. filesystem from the equation. This will give us the top theoretical
  1322. speeds.
  1323. 2. Mount mergerfs over `tmpfs`. `tmpfs` is a RAM disk. Extremely high
  1324. speed and very low latency. This is a more realistic best case
  1325. scenario. Example: `mount -t tmpfs -o size=2G tmpfs /tmp/tmpfs`
  1326. 3. Mount mergerfs over a local device. NVMe, SSD, HDD, etc. If you
  1327. have more than one I'd suggest testing each of them as drives
  1328. and/or controllers (their drivers) could impact performance.
  1329. 4. Finally, if you intend to use mergerfs with a network filesystem,
  1330. either as the source of data or to combine with another through
  1331. mergerfs, test each of those alone as above.
  1332. Once you find the component which has the performance issue you can do
  1333. further testing with different options to see if they impact
  1334. performance. For reads and writes the most relevant would be:
  1335. `cache.files`, `async_read`. Less likely but relevant when using NFS
  1336. or with certain filesystems would be `security_capability`, `xattr`,
  1337. and `posix_acl`. If you find a specific system, device, filesystem,
  1338. controller, etc. that performs poorly contact trapexit so he may
  1339. investigate further.
  1340. Sometimes the problem is really the application accessing or writing
  1341. data through mergerfs. Some software use small buffer sizes which can
  1342. lead to more requests and therefore greater overhead. You can test
  1343. this out yourself by replacing `bs=1M` in the examples below with `ibs`
  1344. or `obs` and using a size of `512` instead of `1M`. In one example
  1345. test using `nullrw` the write speed dropped from 4.9GB/s to 69.7MB/s
  1346. when moving from `1M` to `512`. Similar results were had when testing
  1347. reads. Small writes overhead may be improved by leveraging a write
  1348. cache but in casual tests little gain was found. More tests will need
  1349. to be done before this feature would become available. If you have an
  1350. app that appears slow with mergerfs it could be due to this. Contact
  1351. trapexit so he may investigate further.
  1352. ### write benchmark
  1353. ```
  1354. $ dd if=/dev/zero of=/mnt/mergerfs/1GB.file bs=1M count=1024 oflag=nocache conv=fdatasync status=progress
  1355. ```
  1356. ### read benchmark
  1357. ```
  1358. $ dd if=/mnt/mergerfs/1GB.file of=/dev/null bs=1M count=1024 iflag=nocache conv=fdatasync status=progress
  1359. ```
  1360. ### other benchmarks
  1361. If you are attempting to benchmark other behaviors you must ensure you
  1362. clear kernel caches before runs. In fact it would be a good deal to
  1363. run before the read and write benchmarks as well just in case.
  1364. ```
  1365. sync
  1366. echo 3 | sudo tee /proc/sys/vm/drop_caches
  1367. ```
  1368. # TIPS / NOTES
  1369. * This document is literal and thorough. If a suspected feature isn't
  1370. mentioned it doesn't exist. If certain libfuse arguments aren't
  1371. listed they probably shouldn't be used.
  1372. * Ensure you're using the latest version.
  1373. * Run mergerfs as `root`. mergerfs is designed and intended to be run
  1374. as `root` and may exibit incorrect behavior if run otherwise..
  1375. * If you don't see some directories and files you expect, policies
  1376. seem to skip branches, you get strange permission errors, etc. be
  1377. sure the underlying filesystems' permissions are all the same. Use
  1378. `mergerfs.fsck` to audit the filesystem for out of sync permissions.
  1379. * If you still have permission issues be sure you are using POSIX ACL
  1380. compliant filesystems. mergerfs doesn't generally make exceptions
  1381. for FAT, NTFS, or other non-POSIX filesystem.
  1382. * Do **not** use `cache.files=off` if you expect applications (such as
  1383. rtorrent) to use [mmap](http://linux.die.net/man/2/mmap)
  1384. files. Shared mmap is not currently supported in FUSE w/ page
  1385. caching disabled. Enabling `dropcacheonclose` is recommended when
  1386. `cache.files=partial|full|auto-full`.
  1387. * [Kodi](http://kodi.tv), [Plex](http://plex.tv),
  1388. [Subsonic](http://subsonic.org), etc. can use directory
  1389. [mtime](http://linux.die.net/man/2/stat) to more efficiently
  1390. determine whether to scan for new content rather than simply
  1391. performing a full scan. If using the default **getattr** policy of
  1392. **ff** it's possible those programs will miss an update on account
  1393. of it returning the first directory found's **stat** info and it's a
  1394. later directory on another mount which had the **mtime** recently
  1395. updated. To fix this you will want to set
  1396. **func.getattr=newest**. Remember though that this is just
  1397. **stat**. If the file is later **open**'ed or **unlink**'ed and the
  1398. policy is different for those then a completely different file or
  1399. directory could be acted on.
  1400. * Some policies mixed with some functions may result in strange
  1401. behaviors. Not that some of these behaviors and race conditions
  1402. couldn't happen outside **mergerfs** but that they are far more
  1403. likely to occur on account of the attempt to merge multiple sources
  1404. of data which could be out of sync due to the different policies.
  1405. * For consistency it's generally best to set **category** wide policies
  1406. rather than individual **func**'s. This will help limit the
  1407. confusion of tools such as
  1408. [rsync](http://linux.die.net/man/1/rsync). However, the flexibility
  1409. is there if needed.
  1410. # KNOWN ISSUES / BUGS
  1411. #### kernel issues & bugs
  1412. [https://github.com/trapexit/mergerfs/wiki/Kernel-Issues-&-Bugs](https://github.com/trapexit/mergerfs/wiki/Kernel-Issues-&-Bugs)
  1413. #### directory mtime is not being updated
  1414. Remember that the default policy for `getattr` is `ff`. The
  1415. information for the first directory found will be returned. If it
  1416. wasn't the directory which had been updated then it will appear
  1417. outdated.
  1418. The reason this is the default is because any other policy would be
  1419. more expensive and for many applications it is unnecessary. To always
  1420. return the directory with the most recent mtime or a faked value based
  1421. on all found would require a scan of all filesystems.
  1422. If you always want the directory information from the one with the
  1423. most recent mtime then use the `newest` policy for `getattr`.
  1424. #### 'mv /mnt/pool/foo /mnt/disk1/foo' removes 'foo'
  1425. This is not a bug.
  1426. Run in verbose mode to better understand what's happening:
  1427. ```
  1428. $ mv -v /mnt/pool/foo /mnt/disk1/foo
  1429. copied '/mnt/pool/foo' -> '/mnt/disk1/foo'
  1430. removed '/mnt/pool/foo'
  1431. $ ls /mnt/pool/foo
  1432. ls: cannot access '/mnt/pool/foo': No such file or directory
  1433. ```
  1434. `mv`, when working across devices, is copying the source to target and
  1435. then removing the source. Since the source **is** the target in this
  1436. case, depending on the unlink policy, it will remove the just copied
  1437. file and other files across the branches.
  1438. If you want to move files to one filesystem just copy them there and
  1439. use mergerfs.dedup to clean up the old paths or manually remove them
  1440. from the branches directly.
  1441. #### cached memory appears greater than it should be
  1442. Use `cache.files=off` and/or `dropcacheonclose=true`. See the section
  1443. on page caching.
  1444. #### NFS clients returning ESTALE / Stale file handle
  1445. NFS generally does not like out of band changes. Take a look at the
  1446. section on NFS in the [#remote-filesystems](#remote-filesystems) for
  1447. more details.
  1448. #### rtorrent fails with ENODEV (No such device)
  1449. Be sure to set
  1450. `cache.files=partial|full|auto-full|per-processe`. rtorrent and some
  1451. other applications use [mmap](http://linux.die.net/man/2/mmap) to read
  1452. and write to files and offer no fallback to traditional methods. FUSE
  1453. does not currently support mmap while using `direct_io`. There may be
  1454. a performance penalty on writes with `direct_io` off as well as the
  1455. problem of double caching but it's the only way to get such
  1456. applications to work. If the performance loss is too high for other
  1457. apps you can mount mergerfs twice. Once with `direct_io` enabled and
  1458. one without it. Be sure to set `dropcacheonclose=true` if not using
  1459. `direct_io`.
  1460. #### Plex doesn't work with mergerfs
  1461. It does. If you're trying to put Plex's config / metadata / database
  1462. on mergerfs you can't set `cache.files=off` because Plex is using
  1463. sqlite3 with mmap enabled. Shared mmap is not supported by Linux's
  1464. FUSE implementation when page caching is disabled. To fix this place
  1465. the data elsewhere (preferable) or enable `cache.files` (with
  1466. `dropcacheonclose=true`). Sqlite3 does not need mmap but the developer
  1467. needs to fall back to standard IO if mmap fails.
  1468. This applies to other software: Radarr, Sonarr, Lidarr, Jellyfin, etc.
  1469. I would recommend reaching out to the developers of the software
  1470. you're having troubles with and asking them to add a fallback to
  1471. regular file IO when mmap is unavailable.
  1472. If the issue is that scanning doesn't seem to pick up media then be
  1473. sure to set `func.getattr=newest`, though generally, a full scan will
  1474. pick up all media anyway.
  1475. #### When a program tries to move or rename a file it fails
  1476. Please read the section above regarding [rename & link](#rename--link).
  1477. The problem is that many applications do not properly handle `EXDEV`
  1478. errors which `rename` and `link` may return even though they are
  1479. perfectly valid situations which do not indicate actual device,
  1480. filesystem, or OS errors. The error will only be returned by mergerfs
  1481. if using a path preserving policy as described in the policy section
  1482. above. If you do not care about path preservation simply change the
  1483. mergerfs policy to the non-path preserving version. For example: `-o
  1484. category.create=mfs` Ideally the offending software would be fixed and
  1485. it is recommended that if you run into this problem you contact the
  1486. software's author and request proper handling of `EXDEV` errors.
  1487. #### my 32bit software has problems
  1488. Some software have problems with 64bit inode values. The symptoms can
  1489. include EOVERFLOW errors when trying to list files. You can address
  1490. this by setting `inodecalc` to one of the 32bit based algos as
  1491. described in the relevant section.
  1492. #### Samba: Moving files / directories fails
  1493. Workaround: Copy the file/directory and then remove the original
  1494. rather than move.
  1495. This isn't an issue with Samba but some SMB clients. GVFS-fuse v1.20.3
  1496. and prior (found in Ubuntu 14.04 among others) failed to handle
  1497. certain error codes correctly. Particularly **STATUS_NOT_SAME_DEVICE**
  1498. which comes from the **EXDEV** which is returned by **rename** when
  1499. the call is crossing mount points. When a program gets an **EXDEV** it
  1500. needs to explicitly take an alternate action to accomplish its
  1501. goal. In the case of **mv** or similar it tries **rename** and on
  1502. **EXDEV** falls back to a manual copying of data between the two
  1503. locations and unlinking the source. In these older versions of
  1504. GVFS-fuse if it received **EXDEV** it would translate that into
  1505. **EIO**. This would cause **mv** or most any application attempting to
  1506. move files around on that SMB share to fail with a IO error.
  1507. [GVFS-fuse v1.22.0](https://bugzilla.gnome.org/show_bug.cgi?id=734568)
  1508. and above fixed this issue but a large number of systems use the older
  1509. release. On Ubuntu, the version can be checked by issuing `apt-cache
  1510. showpkg gvfs-fuse`. Most distros released in 2015 seem to have the
  1511. updated release and will work fine but older systems may
  1512. not. Upgrading gvfs-fuse or the distro in general will address the
  1513. problem.
  1514. In Apple's MacOSX 10.9 they replaced Samba (client and server) with
  1515. their own product. It appears their new client does not handle
  1516. **EXDEV** either and responds similarly to older releases of gvfs on
  1517. Linux.
  1518. #### Trashing files occasionally fails
  1519. This is the same issue as with Samba. `rename` returns `EXDEV` (in our
  1520. case that will really only happen with path preserving policies like
  1521. `epmfs`) and the software doesn't handle the situation well. This is
  1522. unfortunately a common failure of software which moves files
  1523. around. The standard indicates that an implementation `MAY` choose to
  1524. support non-user home directory trashing of files (which is a
  1525. `MUST`). The implementation `MAY` also support "top directory trashes"
  1526. which many probably do.
  1527. To create a `$topdir/.Trash` directory as defined in the standard use
  1528. the [mergerfs-tools](https://github.com/trapexit/mergerfs-tools) tool
  1529. `mergerfs.mktrash`.
  1530. #### Supplemental user groups
  1531. Due to the overhead of
  1532. [getgroups/setgroups](http://linux.die.net/man/2/setgroups) mergerfs
  1533. utilizes a cache. This cache is opportunistic and per thread. Each
  1534. thread will query the supplemental groups for a user when that
  1535. particular thread needs to change credentials and will keep that data
  1536. for the lifetime of the thread. This means that if a user is added to
  1537. a group it may not be picked up without the restart of
  1538. mergerfs. However, since the high level FUSE API's (at least the
  1539. standard version) thread pool dynamically grows and shrinks it's
  1540. possible that over time a thread will be killed and later a new thread
  1541. with no cache will start and query the new data.
  1542. The gid cache uses fixed storage to simplify the design and be
  1543. compatible with older systems which may not have C++11
  1544. compilers. There is enough storage for 256 users' supplemental
  1545. groups. Each user is allowed up to 32 supplemental groups. Linux >=
  1546. 2.6.3 allows up to 65535 groups per user but most other *nixs allow
  1547. far less. NFS allows only 16. The system does handle overflow
  1548. gracefully. If the user has more than 32 supplemental groups only the
  1549. first 32 will be used. If more than 256 users are using the system
  1550. when an uncached user is found it will evict an existing user's cache
  1551. at random. So long as there aren't more than 256 active users this
  1552. should be fine. If either value is too low for your needs you will
  1553. have to modify `gidcache.hpp` to increase the values. Note that doing
  1554. so will increase the memory needed by each thread.
  1555. While not a bug some users have found when using containers that
  1556. supplemental groups defined inside the container don't work properly
  1557. with regard to permissions. This is expected as mergerfs lives outside
  1558. the container and therefore is querying the host's group
  1559. database. There might be a hack to work around this (make mergerfs
  1560. read the /etc/group file in the container) but it is not yet
  1561. implemented and would be limited to Linux and the /etc/group
  1562. DB. Preferably users would mount in the host group file into the
  1563. containers or use a standard shared user & groups technology like NIS
  1564. or LDAP.
  1565. # Remote Filesystems
  1566. Many users ask about compatibility with remote filesystems. This
  1567. section is to describe any known issues or quirks when using mergerfs
  1568. with common remote filesystems.
  1569. Keep in mind that, like with caching, it is not a good idea to change
  1570. the contents of the remote filesystem
  1571. [out-of-band](https://en.wikipedia.org/wiki/Out-of-band). Meaning that
  1572. you really shouldn't change the contents of the underlying
  1573. filesystems or mergerfs on the server hosting the remote
  1574. filesystem. Doing so can lead to weird behavior, inconsistency,
  1575. errors, and even data corruption should multiple programs try to write
  1576. or read the same data at the same time. This isn't to say you can't do
  1577. it or that data corruption is likely but it *could* happen. It is
  1578. better to always use the remote filesystem. Even on the machine
  1579. serving it.
  1580. ## NFS
  1581. [NFS](https://en.wikipedia.org/wiki/Network_File_System) is a common
  1582. remote filesystem on Unix/POSIX systems. Due to how NFS works there
  1583. are some settings which need to be set in order for mergerfs to work
  1584. with it.
  1585. It should be noted that NFS and FUSE (the technology mergerfs uses) do
  1586. not work perfectly with one another due to certain design choices in
  1587. FUSE (and mergerfs.) Due to these issues, it is generally recommended
  1588. to use SMB when possible till situations change. That said mergerfs
  1589. should generally work as an export of NFS and issues discovered should
  1590. still be reported.
  1591. To ensure compatibility between mergerfs and NFS use the following
  1592. settings.
  1593. mergerfs settings:
  1594. * noforget
  1595. * inodecalc=path-hash
  1596. NFS export settings:
  1597. * fsid=UUID
  1598. * no\_root\_squash
  1599. `noforget` is needed because NFS uses the `name_to_handle_at` and
  1600. `open_by_handle_at` functions which allow a program to keep a
  1601. reference to a file without technically having it open in the typical
  1602. sense. The problem is that FUSE has no way to know that NFS has a
  1603. handle that it will later use to open the file again. As a result, it
  1604. is possible for the kernel to tell mergerfs to forget about the node
  1605. and should NFS ever ask for that node's details in the future it would
  1606. have nothing to respond with. Keeping nodes around forever is not
  1607. ideal but at the moment the only way to manage the situation.
  1608. `inodecalc=path-hash` is needed because NFS is sensitive to
  1609. out-of-band changes. FUSE doesn't care if a file's inode value changes
  1610. but NFS, being stateful, does. So if you used the default inode
  1611. calculation algorithm then it is possible that if you changed a file
  1612. or updated a directory the file mergerfs will use will be on a
  1613. different branch and therefore the inode would change. This isn't an
  1614. ideal solution and others are being considered but it works for most
  1615. situations.
  1616. `fsid=UUID` is needed because FUSE filesystems don't have different
  1617. `st_dev` values which can cause issues when exporting. The easiest
  1618. thing to do is set each mergerfs export `fsid` to some random
  1619. value. An easy way to generate a random value is to use the command
  1620. line tool `uuid` or `uuidgen` or through a website such as
  1621. [uuidgenerator.net](https://www.uuidgenerator.net/).
  1622. `no_root_squash` is not strictly necessary but can lead to confusing
  1623. permission and ownership issues if root squashing is enabled.
  1624. ## SMB / CIFS
  1625. [SMB](https://en.wikipedia.org/wiki/Server_Message_Block) is a
  1626. protocol most used by Microsoft Windows systems to share file shares,
  1627. printers, etc. However, due to the popularity of Windows, it is also
  1628. supported on many other platforms including Linux. The most popular
  1629. way of supporting SMB on Linux is via the software Samba.
  1630. [Samba](https://en.wikipedia.org/wiki/Samba_(software)), and other
  1631. ways of serving Linux filesystems, via SMB should work fine with
  1632. mergerfs. The services do not tend to use the same technologies which
  1633. NFS uses and therefore don't have the same issues. There should not be
  1634. special settings required to use mergerfs with Samba. However,
  1635. [CIFSD](https://en.wikipedia.org/wiki/CIFSD) and other programs have
  1636. not been extensively tested. If you use mergerfs with CIFSD or other
  1637. SMB servers please submit your experiences so these docs can be
  1638. updated.
  1639. ## SSHFS
  1640. [SSHFS](https://en.wikipedia.org/wiki/SSHFS) is a FUSE filesystem
  1641. leveraging SSH as the connection and transport layer. While often
  1642. simpler to setup when compared to NFS or Samba the performance can be
  1643. lacking and the project is very much in maintenance mode.
  1644. There are no known issues using sshfs with mergerfs. You may want to
  1645. use the following arguments to improve performance but your millage
  1646. may vary.
  1647. * `-o Ciphers=arcfour`
  1648. * `-o Compression=no`
  1649. More info can be found
  1650. [here](https://ideatrash.net/2016/08/odds-and-ends-optimizing-sshfs-moving.html).
  1651. ## Other
  1652. There are other remote filesystems but none popularly used to serve
  1653. mergerfs. If you use something not listed above feel free to reach out
  1654. and I will add it to the list.
  1655. # FAQ
  1656. #### How well does mergerfs scale? Is it "production ready?"
  1657. Users have reported running mergerfs on everything from a Raspberry Pi
  1658. to dual socket Xeon systems with >20 cores. I'm aware of at least a
  1659. few companies which use mergerfs in production. [Open Media
  1660. Vault](https://www.openmediavault.org) includes mergerfs as its sole
  1661. solution for pooling filesystems. The author of mergerfs had it
  1662. running for over 300 days managing 16+ devices with reasonably heavy
  1663. 24/7 read and write usage. Stopping only after the machine's power
  1664. supply died.
  1665. Most serious issues (crashes or data corruption) have been due to
  1666. [kernel
  1667. bugs](https://github.com/trapexit/mergerfs/wiki/Kernel-Issues-&-Bugs). All
  1668. of which are fixed in stable releases.
  1669. #### Can mergerfs be used with filesystems which already have data / are in use?
  1670. Yes. mergerfs is really just a proxy and does **NOT** interfere with
  1671. the normal form or function of the filesystems / mounts / paths it
  1672. manages. It is just another userland application that is acting as a
  1673. man-in-the-middle. It can't do anything that any other random piece of
  1674. software can't do.
  1675. mergerfs is **not** a traditional filesystem that takes control over
  1676. the underlying block device. mergerfs is **not** RAID. It does **not**
  1677. manipulate the data that passes through it. It does **not** shard data
  1678. across filesystems. It merely shards some **behavior** and aggregates
  1679. others.
  1680. #### Can drives/filesystems be removed from the pool at will?
  1681. Yes. See previous question's answer.
  1682. #### Can mergerfs be removed without affecting the data?
  1683. Yes. See the previous question's answer.
  1684. #### Can drives/filesystems be moved to another pool?
  1685. Yes. See the previous question's answer.
  1686. #### How do I migrate data into or out of the pool when adding/removing drives/filesystems?
  1687. You don't need to. See the previous question's answer.
  1688. #### How do I remove a drive/filesystem but keep the data in the pool?
  1689. Nothing special needs to be done. Remove the branch from mergerfs'
  1690. config and copy (rsync) the data from the removed filesystem into the
  1691. pool. Effectively the same as if it were you transfering data from one
  1692. filesystem to another.
  1693. If you wish to continue using the pool while performing the transfer
  1694. simply create another, temporary pool without the filesystem in
  1695. question and then copy the data. It would probably be a good idea to
  1696. set the branch to `RO` prior to doing this to ensure no new content is
  1697. written to the filesystem while performing the copy.
  1698. #### What policies should I use?
  1699. Unless you're doing something more niche the average user is probably
  1700. best off using `mfs` for `category.create`. It will spread files out
  1701. across your branches based on available space. Use `mspmfs` if you
  1702. want to try to colocate the data a bit more. You may want to use `lus`
  1703. if you prefer a slightly different distribution of data if you have a
  1704. mix of smaller and larger filesystems. Generally though `mfs`, `lus`,
  1705. or even `rand` are good for the general use case. If you are starting
  1706. with an imbalanced pool you can use the tool **mergerfs.balance** to
  1707. redistribute files across the pool.
  1708. If you really wish to try to colocate files based on directory you can
  1709. set `func.create` to `epmfs` or similar and `func.mkdir` to `rand` or
  1710. `eprand` depending on if you just want to colocate generally or on
  1711. specific branches. Either way the *need* to colocate is rare. For
  1712. instance: if you wish to remove the device regularly and want the data
  1713. to predictably be on that device or if you don't use backup at all and
  1714. don't wish to replace that data piecemeal. In which case using path
  1715. preservation can help but will require some manual
  1716. attention. Colocating after the fact can be accomplished using the
  1717. **mergerfs.consolidate** tool. If you don't need strict colocation
  1718. which the `ep` policies provide then you can use the `msp` based
  1719. policies which will walk back the path till finding a branch that
  1720. works.
  1721. Ultimately there is no correct answer. It is a preference or based on
  1722. some particular need. mergerfs is very easy to test and experiment
  1723. with. I suggest creating a test setup and experimenting to get a sense
  1724. of what you want.
  1725. `epmfs` is the default `category.create` policy because `ep` policies
  1726. are not going to change the general layout of the branches. It won't
  1727. place files/dirs on branches that don't already have the relative
  1728. branch. So it keeps the system in a known state. It's much easier to
  1729. stop using `epmfs` or redistribute files around the filesystem than it
  1730. is to consolidate them back.
  1731. #### What settings should I use?
  1732. Depends on what features you want. Generally speaking, there are no
  1733. "wrong" settings. All settings are performance or feature related. The
  1734. best bet is to read over the available options and choose what fits
  1735. your situation. If something isn't clear from the documentation please
  1736. reach out and the documentation will be improved.
  1737. That said, for the average person, the following should be fine:
  1738. `cache.files=off,dropcacheonclose=true,category.create=mfs`
  1739. #### Why are all my files ending up on 1 filesystem?!
  1740. Did you start with empty filesystems? Did you explicitly configure a
  1741. `category.create` policy? Are you using an `existing path` / `path
  1742. preserving` policy?
  1743. The default create policy is `epmfs`. That is a path preserving
  1744. algorithm. With such a policy for `mkdir` and `create` with a set of
  1745. empty filesystems it will select only 1 filesystem when the first
  1746. directory is created. Anything, files or directories, created in that
  1747. first directory will be placed on the same branch because it is
  1748. preserving paths.
  1749. This catches a lot of new users off guard but changing the default
  1750. would break the setup for many existing users and this policy is the
  1751. safest policy as it will not change the general layout of the existing
  1752. filesystems. If you do not care about path preservation and wish your
  1753. files to be spread across all your filesystems change to `mfs` or
  1754. similar policy as described above. If you do want path preservation
  1755. you'll need to perform the manual act of creating paths on the
  1756. filesystems you want the data to land on before transferring your
  1757. data. Setting `func.mkdir=epall` can simplify managing path
  1758. preservation for `create`. Or use `func.mkdir=rand` if you're
  1759. interested in just grouping directory content by filesystem.
  1760. #### Do hardlinks work?
  1761. Yes. See also the option `inodecalc` for how inode values are
  1762. calculated.
  1763. What mergerfs does not do is fake hard links across branches. Read
  1764. the section "rename & link" for how it works.
  1765. Remember that hardlinks will NOT work across devices. That includes
  1766. between the original filesystem and a mergerfs pool, between two
  1767. separate pools of the same underlying filesystems, or bind mounts of
  1768. paths within the mergerfs pool. The latter is common when using Docker
  1769. or Podman. Multiple volumes (bind mounts) to the same underlying
  1770. filesystem are considered different devices. There is no way to link
  1771. between them. You should mount in the highest directory in the
  1772. mergerfs pool that includes all the paths you need if you want links
  1773. to work.
  1774. #### Does FICLONE or FICLONERANGE work?
  1775. Unfortunately not. FUSE, the technology mergerfs is based on, does not
  1776. support the `clone_file_range` feature needed for it to work. mergerfs
  1777. won't even know such a request is made. The kernel will simply return
  1778. an error back to the application making the request.
  1779. Should FUSE gain the ability mergerfs will be updated to support it.
  1780. #### Can I use mergerfs without SnapRAID? SnapRAID without mergerfs?
  1781. Yes. They are completely unrelated pieces of software.
  1782. #### Can mergerfs run via Docker, Podman, Kubernetes, etc.
  1783. Yes. With Docker you'll need to include `--cap-add=SYS_ADMIN
  1784. --device=/dev/fuse --security-opt=apparmor:unconfined` or similar with
  1785. other container runtimes. You should also be running it as root or
  1786. given sufficient caps to change user and group identity as well as
  1787. have root like filesystem permissions.
  1788. Keep in mind that you **MUST** consider identity when using
  1789. containers. For example: supplemental groups will be picked up from
  1790. the container unless you properly manage users and groups by sharing
  1791. relevant /etc files or by using some other means to share identity
  1792. across containers. Similarly, if you use "rootless" containers and user
  1793. namespaces to do uid/gid translations you **MUST** consider that while
  1794. managing shared files.
  1795. Also, as mentioned by [hotio](https://hotio.dev/containers/mergerfs),
  1796. with Docker you should probably be mounting with `bind-propagation`
  1797. set to `slave`.
  1798. #### Does mergerfs support CoW / copy-on-write / writes to read-only filesystems?
  1799. Not in the sense of a filesystem like BTRFS or ZFS nor in the
  1800. overlayfs or aufs sense. It does offer a
  1801. [cow-shell](http://manpages.ubuntu.com/manpages/bionic/man1/cow-shell.1.html)
  1802. like hard link breaking (copy to temp file then rename over original)
  1803. which can be useful when wanting to save space by hardlinking
  1804. duplicate files but wish to treat each name as if it were a unique and
  1805. separate file.
  1806. If you want to write to a read-only filesystem you should look at
  1807. overlayfs. You can always include the overlayfs mount into a mergerfs
  1808. pool.
  1809. #### Why can't I see my files / directories?
  1810. It's almost always a permissions issue. Unlike mhddfs and
  1811. unionfs-fuse, which runs as root and attempts to access content as
  1812. such, mergerfs always changes its credentials to that of the
  1813. caller. This means that if the user does not have access to a file or
  1814. directory than neither will mergerfs. However, because mergerfs is
  1815. creating a union of paths it may be able to read some files and
  1816. directories on one filesystem but not another resulting in an
  1817. incomplete set.
  1818. Whenever you run into a split permission issue (seeing some but not
  1819. all files) try using
  1820. [mergerfs.fsck](https://github.com/trapexit/mergerfs-tools) tool to
  1821. check for and fix the mismatch. If you aren't seeing anything at all
  1822. be sure that the basic permissions are correct. The user and group
  1823. values are correct and that directories have their executable bit
  1824. set. A common mistake by users new to Linux is to `chmod -R 644` when
  1825. they should have `chmod -R u=rwX,go=rX`.
  1826. If using a network filesystem such as NFS or SMB (Samba) be sure to
  1827. pay close attention to anything regarding permissioning and
  1828. users. Root squashing and user translation for instance has bitten a
  1829. few mergerfs users. Some of these also affect the use of mergerfs from
  1830. container platforms such as Docker.
  1831. #### Why use FUSE? Why not a kernel based solution?
  1832. As with any solution to a problem, there are advantages and
  1833. disadvantages to each one.
  1834. A FUSE based solution has all the downsides of FUSE:
  1835. * Higher IO latency due to the trips in and out of kernel space
  1836. * Higher general overhead due to trips in and out of kernel space
  1837. * Double caching when using page caching
  1838. * Misc limitations due to FUSE's design
  1839. But FUSE also has a lot of upsides:
  1840. * Easier to offer a cross platform solution
  1841. * Easier forward and backward compatibility
  1842. * Easier updates for users
  1843. * Easier and faster release cadence
  1844. * Allows more flexibility in design and features
  1845. * Overall easier to write, secure, and maintain
  1846. * Much lower barrier to entry (getting code into the kernel takes a
  1847. lot of time and effort initially)
  1848. FUSE was chosen because of all the advantages listed above. The
  1849. negatives of FUSE do not outweigh the positives.
  1850. #### Is my OS's libfuse needed for mergerfs to work?
  1851. No. Normally `mount.fuse` is needed to get mergerfs (or any FUSE
  1852. filesystem to mount using the `mount` command but in vendoring the
  1853. libfuse library the `mount.fuse` app has been renamed to
  1854. `mount.mergerfs` meaning the filesystem type in `fstab` can simply be
  1855. `mergerfs`. That said there should be no harm in having it installed
  1856. and continuing to using `fuse.mergerfs` as the type in `/etc/fstab`.
  1857. If `mergerfs` doesn't work as a type it could be due to how the
  1858. `mount.mergerfs` tool was installed. Must be in `/sbin/` with proper
  1859. permissions.
  1860. #### Why was splice support removed?
  1861. After a lot of testing over the years, splicing always appeared to
  1862. at best, provide equivalent performance, and in some cases, worse
  1863. performance. Splice is not supported on other platforms forcing a
  1864. traditional read/write fallback to be provided. The splice code was
  1865. removed to simplify the codebase.
  1866. #### What should mergerfs NOT be used for?
  1867. * databases: Even if the database stored data in separate files
  1868. (mergerfs wouldn't offer much otherwise) the higher latency of the
  1869. indirection will kill performance. If it is a lightly used SQLITE
  1870. database then it may be fine but you'll need to test.
  1871. * VM images: For the same reasons as databases. VM images are accessed
  1872. very aggressively and mergerfs will introduce too much latency (if
  1873. it works at all).
  1874. * As replacement for RAID: mergerfs is just for pooling branches. If
  1875. you need that kind of device performance aggregation or high
  1876. availability you should stick with RAID.
  1877. #### Can filesystems be written to directly? Outside of mergerfs while pooled?
  1878. Yes, however, it's not recommended to use the same file from within the
  1879. pool and from without at the same time (particularly
  1880. writing). Especially if using caching of any kind (cache.files,
  1881. cache.entry, cache.attr, cache.negative_entry, cache.symlinks,
  1882. cache.readdir, etc.) as there could be a conflict between cached data
  1883. and not.
  1884. #### Why do I get an "out of space" / "no space left on device" / ENOSPC error even though there appears to be lots of space available?
  1885. First make sure you've read the sections above about policies, path
  1886. preservation, branch filtering, and the options **minfreespace**,
  1887. **moveonenospc**, **statfs**, and **statfs_ignore**.
  1888. mergerfs is simply presenting a union of the content within multiple
  1889. branches. The reported free space is an aggregate of space available
  1890. within the pool (behavior modified by **statfs** and
  1891. **statfs_ignore**). It does not represent a contiguous space. In the
  1892. same way that read-only filesystems, those with quotas, or reserved
  1893. space report the full theoretical space available.
  1894. Due to path preservation, branch tagging, read-only status, and
  1895. **minfreespace** settings it is perfectly valid that `ENOSPC` / "out
  1896. of space" / "no space left on device" be returned. It is doing what
  1897. was asked of it: filtering possible branches due to those
  1898. settings. Only one error can be returned and if one of the reasons for
  1899. filtering a branch was **minfreespace** then it will be returned as
  1900. such. **moveonenospc** is only relevant to writing a file which is too
  1901. large for the filesystem it's currently on.
  1902. It is also possible that the filesystem selected has run out of
  1903. inodes. Use `df -i` to list the total and available inodes per
  1904. filesystem.
  1905. If you don't care about path preservation then simply change the
  1906. `create` policy to one which isn't. `mfs` is probably what most are
  1907. looking for. The reason it's not default is because it was originally
  1908. set to `epmfs` and changing it now would change people's setup. Such a
  1909. setting change will likely occur in mergerfs 3.
  1910. #### Why does the total available space in mergerfs not equal outside?
  1911. Are you using ext2/3/4? With reserve for root? mergerfs uses available
  1912. space for statfs calculations. If you've reserved space for root then
  1913. it won't show up.
  1914. You can remove the reserve by running: `tune2fs -m 0 <device>`
  1915. #### I notice massive slowdowns of writes when enabling cache.files.
  1916. When file caching is enabled in any form (`cache.files!=off`) it will
  1917. issue `getxattr` requests for `security.capability` prior to *every
  1918. single write*. This will usually result in performance degradation,
  1919. especially when using a network filesystem (such as NFS or SMB.)
  1920. Unfortunately at this moment, the kernel is not caching the response.
  1921. To work around this situation mergerfs offers a few solutions.
  1922. 1. Set `security_capability=false`. It will short circuit any call and
  1923. return `ENOATTR`. This still means though that mergerfs will
  1924. receive the request before every write but at least it doesn't get
  1925. passed through to the underlying filesystem.
  1926. 2. Set `xattr=noattr`. Same as above but applies to *all* calls to
  1927. getxattr. Not just `security.capability`. This will not be cached
  1928. by the kernel either but mergerfs' runtime config system will still
  1929. function.
  1930. 3. Set `xattr=nosys`. Results in mergerfs returning `ENOSYS` which
  1931. *will* be cached by the kernel. No future xattr calls will be
  1932. forwarded to mergerfs. The downside is that also means the xattr
  1933. based config and query functionality won't work either.
  1934. 4. Disable file caching. If you aren't using applications which use
  1935. `mmap` it's probably simpler to just disable it altogether. The
  1936. kernel won't send the requests when caching is disabled.
  1937. #### It's mentioned that there are some security issues with mhddfs. What are they? How does mergerfs address them?
  1938. [mhddfs](https://github.com/trapexit/mhddfs) manages running as
  1939. **root** by calling
  1940. [getuid()](https://github.com/trapexit/mhddfs/blob/cae96e6251dd91e2bdc24800b4a18a74044f6672/src/main.c#L319)
  1941. and if it returns **0** then it will
  1942. [chown](http://linux.die.net/man/1/chown) the file. Not only is that a
  1943. race condition but it doesn't handle other situations. Rather than
  1944. attempting to simulate POSIX ACL behavior the proper way to manage
  1945. this is to use [seteuid](http://linux.die.net/man/2/seteuid) and
  1946. [setegid](http://linux.die.net/man/2/setegid), in effect, becoming the
  1947. user making the original call, and perform the action as them. This is
  1948. what mergerfs does and why mergerfs should always run as root.
  1949. In Linux setreuid syscalls apply only to the thread. GLIBC hides this
  1950. away by using realtime signals to inform all threads to change
  1951. credentials. Taking after **Samba**, mergerfs uses
  1952. **syscall(SYS_setreuid,...)** to set the callers credentials for that
  1953. thread only. Jumping back to **root** as necessary should escalated
  1954. privileges be needed (for instance: to clone paths between
  1955. filesystems).
  1956. For non-Linux systems, mergerfs uses a read-write lock and changes
  1957. credentials only when necessary. If multiple threads are to be user X
  1958. then only the first one will need to change the processes
  1959. credentials. So long as the other threads need to be user X they will
  1960. take a readlock allowing multiple threads to share the
  1961. credentials. Once a request comes in to run as user Y that thread will
  1962. attempt a write lock and change to Y's credentials when it can. If the
  1963. ability to give writers priority is supported then that flag will be
  1964. used so threads trying to change credentials don't starve. This isn't
  1965. the best solution but should work reasonably well assuming there are
  1966. few users.
  1967. # mergerfs versus X
  1968. #### mhddfs
  1969. mhddfs had not been maintained for some time and has some known
  1970. stability and security issues. mergerfs provides a superset of mhddfs'
  1971. features and should offer the same or better performance.
  1972. Below is an example of mhddfs and mergerfs setup to work similarly.
  1973. `mhddfs -o mlimit=4G,allow_other /mnt/drive1,/mnt/drive2 /mnt/pool`
  1974. `mergerfs -o minfreespace=4G,category.create=ff /mnt/drive1:/mnt/drive2 /mnt/pool`
  1975. #### aufs
  1976. aufs is mostly abandoned and no longer available in most Linux distros.
  1977. While aufs can offer better peak performance mergerfs provides more
  1978. configurability and is generally easier to use. mergerfs however does
  1979. not offer the overlay / copy-on-write (CoW) features which aufs has.
  1980. #### unionfs-fuse
  1981. unionfs-fuse is more like aufs than mergerfs in that it offers overlay /
  1982. copy-on-write (CoW) features. If you're just looking to create a union
  1983. of filesystems and want flexibility in file/directory placement then
  1984. mergerfs offers that whereas unionfs is more for overlaying read/write
  1985. filesystems over read-only ones.
  1986. #### overlayfs
  1987. overlayfs is similar to aufs and unionfs-fuse in that it also is
  1988. primarily used to layer a read/write filesystem over one or more
  1989. read-only filesystems. It does not have the ability to spread
  1990. files/directories across numerous filesystems.
  1991. #### RAID0, JBOD, drive concatenation, striping
  1992. With simple JBOD / drive concatenation / stripping / RAID0 a single
  1993. drive failure will result in full pool failure. mergerfs performs a
  1994. similar function without the possibility of catastrophic failure and
  1995. the difficulties in recovery. Drives may fail but all other
  1996. filesystems and their data will continue to be accessible.
  1997. The main practical difference with mergerfs is the fact you don't
  1998. actually have contiguous space as large as if you used those other
  1999. technologies. Meaning you can't create a 2TB file on a pool of 2 1TB
  2000. filesystems.
  2001. When combined with something like [SnapRaid](http://www.snapraid.it)
  2002. and/or an offsite backup solution you can have the flexibility of JBOD
  2003. without the single point of failure.
  2004. #### UnRAID
  2005. UnRAID is a full OS and its storage layer, as I understand, is
  2006. proprietary and closed source. Users who have experience with both
  2007. have often said they prefer the flexibility offered by mergerfs and
  2008. for some the fact it is open source is important.
  2009. There are a number of UnRAID users who use mergerfs as well though I'm
  2010. not entirely familiar with the use case.
  2011. For semi-static data mergerfs + [SnapRaid](http://www.snapraid.it)
  2012. provides a similar solution.
  2013. #### ZFS
  2014. mergerfs is very different from ZFS. mergerfs is intended to provide
  2015. flexible pooling of arbitrary filesystems (local or remote), of
  2016. arbitrary sizes, and arbitrary filesystems. For `write once, read
  2017. many` usecases such as bulk media storage. Where data integrity and
  2018. backup is managed in other ways. In those usecases ZFS can introduce a
  2019. number of costs and limitations as described
  2020. [here](http://louwrentius.com/the-hidden-cost-of-using-zfs-for-your-home-nas.html),
  2021. [here](https://markmcb.com/2020/01/07/five-years-of-btrfs/), and
  2022. [here](https://utcc.utoronto.ca/~cks/space/blog/solaris/ZFSWhyNoRealReshaping).
  2023. #### StableBit's DrivePool
  2024. DrivePool works only on Windows so not as common an alternative as
  2025. other Linux solutions. If you want to use Windows then DrivePool is a
  2026. good option. Functionally the two projects work a bit
  2027. differently. DrivePool always writes to the filesystem with the most
  2028. free space and later rebalances. mergerfs does not offer rebalance but
  2029. chooses a branch at file/directory create time. DrivePool's
  2030. rebalancing can be done differently in any directory and has file
  2031. pattern matching to further customize the behavior. mergerfs, not
  2032. having rebalancing does not have these features, but similar features
  2033. are planned for mergerfs v3. DrivePool has builtin file duplication
  2034. which mergerfs does not natively support (but can be done via an
  2035. external script.)
  2036. There are a lot of misc differences between the two projects but most
  2037. features in DrivePool can be replicated with external tools in
  2038. combination with mergerfs.
  2039. Additionally, DrivePool is a closed source commercial product vs
  2040. mergerfs a ISC licensed OSS project.
  2041. # SUPPORT
  2042. Filesystems are complex and difficult to debug. mergerfs, while being
  2043. just a proxy of sorts, can be difficult to debug given the large
  2044. number of possible settings it can have itself and the number of
  2045. environments it can run in. When reporting on a suspected issue
  2046. **please** include as much of the below information as possible
  2047. otherwise it will be difficult or impossible to diagnose. Also please
  2048. read the above documentation as it provides details on many previously
  2049. encountered questions/issues.
  2050. **Please make sure you are using the [latest
  2051. release](https://github.com/trapexit/mergerfs/releases) or have tried
  2052. it in comparison. Old versions, which are often included in distros
  2053. like Debian and Ubuntu, are not ever going to be updated and the issue
  2054. you are encountering may have been addressed already.**
  2055. **For commercial support or feature requests please [contact me
  2056. directly.](mailto:support@spawn.link)**
  2057. #### Information to include in bug reports
  2058. * [Information about the broader problem along with any attempted
  2059. solutions.](https://xyproblem.info)
  2060. * Solution already ruled out and why.
  2061. * Version of mergerfs: `mergerfs --version`
  2062. * mergerfs settings / arguments: from fstab, systemd unit, command
  2063. line, OMV plugin, etc.
  2064. * Version of the OS: `uname -a` and `lsb_release -a`
  2065. * List of branches, their filesystem types, sizes (before and after issue): `df -h`
  2066. * **All** information about the relevant paths and files: permissions, ownership, etc.
  2067. * **All** information about the client app making the requests: version, uid/gid
  2068. * Runtime environment:
  2069. * Is mergerfs running within a container?
  2070. * Are the client apps using mergerfs running in a container?
  2071. * A `strace` of the app having problems:
  2072. * `strace -fvTtt -s 256 -o /tmp/app.strace.txt <cmd>`
  2073. * A `strace` of mergerfs while the program is trying to do whatever it is failing to do:
  2074. * `strace -fvTtt -s 256 -p <mergerfsPID> -o /tmp/mergerfs.strace.txt`
  2075. * **Precise** directions on replicating the issue. Do not leave **anything** out.
  2076. * Try to recreate the problem in the simplest way using standard programs: `ln`, `mv`, `cp`, `ls`, `dd`, etc.
  2077. #### Contact / Issue submission
  2078. * github.com: https://github.com/trapexit/mergerfs/issues
  2079. * discord: https://discord.gg/MpAr69V
  2080. * reddit: https://www.reddit.com/r/mergerfs
  2081. #### Donations
  2082. https://github.com/trapexit/support
  2083. Development and support of a project like mergerfs requires a
  2084. significant amount of time and effort. The software is released under
  2085. the very liberal ISC license and is therefore free to use for personal
  2086. or commercial uses.
  2087. If you are a personal user and find mergerfs and its support valuable
  2088. and would like to support the project financially it would be very
  2089. much appreciated.
  2090. If you are using mergerfs commercially please consider sponsoring the
  2091. project to ensure it continues to be maintained and receive
  2092. updates. If custom features are needed feel free to [contact me
  2093. directly](mailto:support@spawn.link).
  2094. # LINKS
  2095. * https://spawn.link
  2096. * https://github.com/trapexit/mergerfs
  2097. * https://github.com/trapexit/mergerfs/wiki
  2098. * https://github.com/trapexit/mergerfs-tools
  2099. * https://github.com/trapexit/scorch
  2100. * https://github.com/trapexit/bbf