| 1 |
.\" Copyright 2016, Michael Kerrisk <[email protected]> |
| 2 |
.\" Copyright, the authors of the Linux man-pages project |
| 3 |
.\" |
| 4 |
.\" SPDX-License-Identifier: Linux-man-pages-copyleft |
| 5 |
.\" |
| 6 |
.TH tmpfs 5 2026-02-08 "Linux man-pages 6.18" |
| 7 |
.SH NAME |
| 8 |
tmpfs \- a virtual memory filesystem |
| 9 |
.SH DESCRIPTION |
| 10 |
The |
| 11 |
.B tmpfs |
| 12 |
facility allows the creation of filesystems whose contents reside |
| 13 |
in virtual memory. |
| 14 |
Since the files on such filesystems typically reside in RAM, |
| 15 |
file access is extremely fast. |
| 16 |
.P |
| 17 |
The filesystem is automatically created when mounting |
| 18 |
a filesystem with the type |
| 19 |
.B tmpfs |
| 20 |
via a command such as the following: |
| 21 |
.P |
| 22 |
.in +4n |
| 23 |
.EX |
| 24 |
$ sudo mount \-t tmpfs \-o size=10M tmpfs /mnt/mytmpfs |
| 25 |
.EE |
| 26 |
.in |
| 27 |
.P |
| 28 |
A |
| 29 |
.B tmpfs |
| 30 |
filesystem has the following properties: |
| 31 |
.IP \[bu] 3 |
| 32 |
The filesystem can employ swap space when physical memory pressure |
| 33 |
demands it. |
| 34 |
.IP \[bu] |
| 35 |
The filesystem consumes only as much physical memory and swap space |
| 36 |
as is required to store the current contents of the filesystem. |
| 37 |
.IP \[bu] |
| 38 |
During a remount operation |
| 39 |
.RI ( "mount\ \-o\ remount" ), |
| 40 |
the filesystem size can be changed |
| 41 |
(without losing the existing contents of the filesystem). |
| 42 |
.P |
| 43 |
If a |
| 44 |
.B tmpfs |
| 45 |
filesystem is unmounted, its contents are discarded (lost). |
| 46 |
.\" See mm/shmem.c:shmem_parse_options for options it supports. |
| 47 |
.SS Mount options |
| 48 |
The |
| 49 |
.B tmpfs |
| 50 |
filesystem supports the following mount options: |
| 51 |
.TP |
| 52 |
.BR size =\f[I]bytes\f[] |
| 53 |
Specify an upper limit on the size of the filesystem. |
| 54 |
The size is given in bytes, and rounded up to entire pages. |
| 55 |
The limit is removed if the size is |
| 56 |
.BR 0 . |
| 57 |
.IP |
| 58 |
The size may have a |
| 59 |
.BR k , |
| 60 |
.BR m , |
| 61 |
or |
| 62 |
.B g |
| 63 |
suffix for Ki, Mi, Gi (binary kilo (kibi), binary mega (mebi), and binary giga |
| 64 |
(gibi)). |
| 65 |
.IP |
| 66 |
The size may also have a % suffix to limit this instance to a percentage of |
| 67 |
physical RAM. |
| 68 |
.IP |
| 69 |
The default, when neither |
| 70 |
.B size |
| 71 |
nor |
| 72 |
.B nr_blocks |
| 73 |
is specified, is |
| 74 |
.IR size=50% . |
| 75 |
.TP |
| 76 |
.BR nr_blocks =\f[I]blocks\f[] |
| 77 |
The same as |
| 78 |
.BR size , |
| 79 |
but in blocks of |
| 80 |
.BR PAGE_CACHE_SIZE . |
| 81 |
.IP |
| 82 |
Blocks may be specified with |
| 83 |
.BR k , |
| 84 |
.BR m , |
| 85 |
or |
| 86 |
.B g |
| 87 |
suffixes like |
| 88 |
.BR size , |
| 89 |
but not a % suffix. |
| 90 |
.TP |
| 91 |
.BR nr_inodes =\f[I]inodes\f[] |
| 92 |
The maximum number of inodes for this instance. |
| 93 |
The default is half of the number of your physical RAM pages, or (on a |
| 94 |
machine with highmem) the number of lowmem RAM pages, whichever is smaller. |
| 95 |
The limit is removed if the number is |
| 96 |
.BR 0 . |
| 97 |
.IP |
| 98 |
Inodes may be specified with |
| 99 |
.BR k , |
| 100 |
.BR m , |
| 101 |
or |
| 102 |
.B g |
| 103 |
suffixes like |
| 104 |
.BR size , |
| 105 |
but not a % suffix. |
| 106 |
.TP |
| 107 |
.BR noswap "(since Linux 6.4)" |
| 108 |
.\" commit 2c6efe9cf2d7841b75fe38ed1adbd41a90f51ba0 |
| 109 |
Disables swap. |
| 110 |
Remounts must respect the original settings. |
| 111 |
By default swap is enabled. |
| 112 |
.TP |
| 113 |
.BR mode =\f[I]mode\f[] |
| 114 |
Set initial permissions of the root directory. |
| 115 |
.TP |
| 116 |
.BR gid "=\f[I]gid\f[] (since Linux 2.5.7)" |
| 117 |
.\" Technically this is also in some version of Linux 2.4. |
| 118 |
.\" commit 099445b489625b80b1d6687c9b6072dbeaca4096 |
| 119 |
Set the initial group ID of the root directory. |
| 120 |
.TP |
| 121 |
.BR uid "=\f[I]uid\f[] (since Linux 2.5.7)" |
| 122 |
.\" Technically this is also in some version of Linux 2.4. |
| 123 |
.\" commit 099445b489625b80b1d6687c9b6072dbeaca4096 |
| 124 |
Set the initial user ID of the root directory. |
| 125 |
.TP |
| 126 |
.BR huge "=\f[I]huge_option\f[] (since Linux 4.7.0)" |
| 127 |
.\" commit 5a6e75f8110c97e2a5488894d4e922187e6cb343 |
| 128 |
Set the huge table memory allocation policy for all files in this instance (if |
| 129 |
.B CONFIG_TRANSPARENT_HUGEPAGE |
| 130 |
is enabled). |
| 131 |
.IP |
| 132 |
The |
| 133 |
.I huge_option |
| 134 |
value is one of the following: |
| 135 |
.RS |
| 136 |
.TP |
| 137 |
.B never |
| 138 |
Do not allocate huge pages. |
| 139 |
This is the default. |
| 140 |
.TP |
| 141 |
.B always |
| 142 |
Attempt to allocate huge pages every time a new page is needed. |
| 143 |
.TP |
| 144 |
.B within_size |
| 145 |
Only allocate huge page if it will be fully within |
| 146 |
.IR i_size . |
| 147 |
Also respect |
| 148 |
.BR fadvise (2) |
| 149 |
and |
| 150 |
.BR madvise (2) |
| 151 |
hints |
| 152 |
.TP |
| 153 |
.B advise |
| 154 |
Only allocate huge pages if requested with |
| 155 |
.BR fadvise (2) |
| 156 |
or |
| 157 |
.BR madvise (2). |
| 158 |
.TP |
| 159 |
.B deny |
| 160 |
For use in emergencies, to force the huge option off from all mounts. |
| 161 |
.TP |
| 162 |
.B force |
| 163 |
Force the huge option on for all mounts; |
| 164 |
useful for testing. |
| 165 |
.RE |
| 166 |
.TP |
| 167 |
.BR mpol "=\f[I]mpol_option\f[] (since Linux 2.6.15)" |
| 168 |
.\" commit 7339ff8302fd70aabf5f1ae26e0c4905fa74a495 |
| 169 |
Set the NUMA memory allocation policy for all files in this instance (if |
| 170 |
.B CONFIG_NUMA |
| 171 |
is enabled). |
| 172 |
.IP |
| 173 |
The |
| 174 |
.I mpol_option |
| 175 |
value is one of the following: |
| 176 |
.RS |
| 177 |
.TP |
| 178 |
.B default |
| 179 |
Use the process allocation policy (see |
| 180 |
.BR set_mempolicy (2)). |
| 181 |
.TP |
| 182 |
.BR prefer :\f[I]node\f[] |
| 183 |
Preferably allocate memory from the given |
| 184 |
.IR node . |
| 185 |
.TP |
| 186 |
.BR bind :\f[I]nodelist\f[] |
| 187 |
Allocate memory only from nodes in |
| 188 |
.IR nodelist . |
| 189 |
.TP |
| 190 |
.B interleave |
| 191 |
Allocate from each node in turn. |
| 192 |
.TP |
| 193 |
.BR interleave :\f[I]nodelist\f[] |
| 194 |
Allocate from each node of |
| 195 |
.I in |
| 196 |
turn. |
| 197 |
.TP |
| 198 |
.B local |
| 199 |
Preferably allocate memory from the local node. |
| 200 |
.RE |
| 201 |
.IP |
| 202 |
In the above, |
| 203 |
.I nodelist |
| 204 |
is a comma-separated list of decimal numbers and ranges |
| 205 |
that specify NUMA nodes. |
| 206 |
A range is a pair of hyphen-separated decimal numbers, |
| 207 |
the smallest and largest node numbers in the range. |
| 208 |
For example, |
| 209 |
.IR mpol=bind:0\-3,5,7,9\-15 . |
| 210 |
.SH VERSIONS |
| 211 |
The |
| 212 |
.B tmpfs |
| 213 |
facility was added in Linux 2.4, as a successor to the older |
| 214 |
.B ramfs |
| 215 |
facility, which did not provide limit checking or |
| 216 |
allow for the use of swap space. |
| 217 |
.SH NOTES |
| 218 |
In order for user-space tools and applications to create |
| 219 |
.B tmpfs |
| 220 |
filesystems, the kernel must be configured with the |
| 221 |
.B CONFIG_TMPFS |
| 222 |
option. |
| 223 |
.P |
| 224 |
The |
| 225 |
.B tmpfs |
| 226 |
filesystem supports extended attributes (see |
| 227 |
.BR xattr (7)), |
| 228 |
but |
| 229 |
.I user |
| 230 |
extended attributes are not permitted. |
| 231 |
.P |
| 232 |
An internal shared memory filesystem is used for |
| 233 |
System V shared memory |
| 234 |
.RB ( shmget (2)) |
| 235 |
and shared anonymous mappings |
| 236 |
.RB ( mmap (2) |
| 237 |
with the |
| 238 |
.B MAP_SHARED |
| 239 |
and |
| 240 |
.B MAP_ANONYMOUS |
| 241 |
flags). |
| 242 |
This filesystem is available regardless of whether |
| 243 |
the kernel was configured with the |
| 244 |
.B CONFIG_TMPFS |
| 245 |
option. |
| 246 |
.P |
| 247 |
A |
| 248 |
.B tmpfs |
| 249 |
filesystem mounted at |
| 250 |
.I /dev/shm |
| 251 |
is used for the implementation of POSIX shared memory |
| 252 |
.RB ( shm_overview (7)) |
| 253 |
and POSIX semaphores |
| 254 |
.RB ( sem_overview (7)). |
| 255 |
.P |
| 256 |
The amount of memory consumed by all |
| 257 |
.B tmpfs |
| 258 |
filesystems is shown in the |
| 259 |
.I Shmem |
| 260 |
field of |
| 261 |
.I /proc/meminfo |
| 262 |
and in the |
| 263 |
.I shared |
| 264 |
field displayed by |
| 265 |
.BR free (1). |
| 266 |
.P |
| 267 |
The |
| 268 |
.B tmpfs |
| 269 |
facility was formerly called |
| 270 |
.BR shmfs . |
| 271 |
.SH SEE ALSO |
| 272 |
.BR df (1), |
| 273 |
.BR du (1), |
| 274 |
.BR memfd_create (2), |
| 275 |
.BR mmap (2), |
| 276 |
.BR set_mempolicy (2), |
| 277 |
.BR shm_open (3), |
| 278 |
.BR mount (8) |
| 279 |
.P |
| 280 |
The kernel source files |
| 281 |
.I Documentation/filesystems/tmpfs.rst |
| 282 |
and |
| 283 |
.IR Documentation/admin\-guide/mm/transhuge.rst . |