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.

2453 lines
102 KiB

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