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.

2743 lines
114 KiB

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