1.. SPDX-License-Identifier: GPL-2.0 2 3===== 4Tmpfs 5===== 6 7Tmpfs is a file system which keeps all of its files in virtual memory. 8 9 10Everything in tmpfs is temporary in the sense that no files will be 11created on your hard drive. If you unmount a tmpfs instance, 12everything stored therein is lost. 13 14tmpfs puts everything into the kernel internal caches and grows and 15shrinks to accommodate the files it contains and is able to swap 16unneeded pages out to swap space, if swap was enabled for the tmpfs 17mount. tmpfs also supports THP. 18 19tmpfs extends ramfs with a few userspace configurable options listed and 20explained further below, some of which can be reconfigured dynamically on the 21fly using a remount ('mount -o remount ...') of the filesystem. A tmpfs 22filesystem can be resized but it cannot be resized to a size below its current 23usage. tmpfs also supports POSIX ACLs, and extended attributes for the 24trusted.*, security.* and user.* namespaces. ramfs does not use swap and you 25cannot modify any parameter for a ramfs filesystem. The size limit of a ramfs 26filesystem is how much memory you have available, and so care must be taken if 27used so to not run out of memory. 28 29An alternative to tmpfs and ramfs is to use brd to create RAM disks 30(/dev/ram*), which allows you to simulate a block device disk in physical RAM. 31To write data you would just then need to create an regular filesystem on top 32this ramdisk. As with ramfs, brd ramdisks cannot swap. brd ramdisks are also 33configured in size at initialization and you cannot dynamically resize them. 34Contrary to brd ramdisks, tmpfs has its own filesystem, it does not rely on the 35block layer at all. 36 37Since tmpfs lives completely in the page cache and optionally on swap, 38all tmpfs pages will be shown as "Shmem" in /proc/meminfo and "Shared" in 39free(1). Notice that these counters also include shared memory 40(shmem, see ipcs(1)). The most reliable way to get the count is 41using df(1) and du(1). 42 43tmpfs has the following uses: 44 451) There is always a kernel internal mount which you will not see at 46 all. This is used for shared anonymous mappings and SYSV shared 47 memory. 48 49 This mount does not depend on CONFIG_TMPFS. If CONFIG_TMPFS is not 50 set, the user visible part of tmpfs is not built. But the internal 51 mechanisms are always present. 52 532) glibc 2.2 and above expects tmpfs to be mounted at /dev/shm for 54 POSIX shared memory (shm_open, shm_unlink). Adding the following 55 line to /etc/fstab should take care of this:: 56 57 tmpfs /dev/shm tmpfs defaults 0 0 58 59 Remember to create the directory that you intend to mount tmpfs on 60 if necessary. 61 62 This mount is _not_ needed for SYSV shared memory. The internal 63 mount is used for that. (In the 2.3 kernel versions it was 64 necessary to mount the predecessor of tmpfs (shm fs) to use SYSV 65 shared memory.) 66 673) Some people (including me) find it very convenient to mount it 68 e.g. on /tmp and /var/tmp and have a big swap partition. And now 69 loop mounts of tmpfs files do work, so mkinitrd shipped by most 70 distributions should succeed with a tmpfs /tmp. 71 724) And probably a lot more I do not know about :-) 73 74 75tmpfs has three mount options for sizing: 76 77========= ============================================================ 78size The limit of allocated bytes for this tmpfs instance. The 79 default is half of your physical RAM without swap. If you 80 oversize your tmpfs instances the machine will deadlock 81 since the OOM handler will not be able to free that memory. 82nr_blocks The same as size, but in blocks of PAGE_SIZE. 83nr_inodes The maximum number of inodes for this instance. The default 84 is half of the number of your physical RAM pages, or (on a 85 machine with highmem) the number of lowmem RAM pages, 86 whichever is the lower. 87========= ============================================================ 88 89These parameters accept a suffix k, m or g for kilo, mega and giga and 90can be changed on remount. The size parameter also accepts a suffix % 91to limit this tmpfs instance to that percentage of your physical RAM: 92the default, when neither size nor nr_blocks is specified, is size=50% 93 94If nr_blocks=0 (or size=0), blocks will not be limited in that instance; 95if nr_inodes=0, inodes will not be limited. It is generally unwise to 96mount with such options, since it allows any user with write access to 97use up all the memory on the machine; but enhances the scalability of 98that instance in a system with many CPUs making intensive use of it. 99 100If nr_inodes is not 0, that limited space for inodes is also used up by 101extended attributes: "df -i"'s IUsed and IUse% increase, IFree decreases. 102 103tmpfs blocks may be swapped out, when there is a shortage of memory. 104tmpfs has a mount option to disable its use of swap: 105 106====== =========================================================== 107noswap Disables swap. Remounts must respect the original settings. 108 By default swap is enabled. 109====== =========================================================== 110 111tmpfs also supports Transparent Huge Pages which requires a kernel 112configured with CONFIG_TRANSPARENT_HUGEPAGE and with huge supported for 113your system (has_transparent_hugepage(), which is architecture specific). 114The mount options for this are: 115 116================ ============================================================== 117huge=never Do not allocate huge pages. This is the default. 118huge=always Attempt to allocate huge page every time a new page is needed. 119huge=within_size Only allocate huge page if it will be fully within i_size. 120 Also respect madvise(2) hints. 121huge=advise Only allocate huge page if requested with madvise(2). 122================ ============================================================== 123 124See also Documentation/admin-guide/mm/transhuge.rst, which describes the 125sysfs file /sys/kernel/mm/transparent_hugepage/shmem_enabled: which can 126be used to deny huge pages on all tmpfs mounts in an emergency, or to 127force huge pages on all tmpfs mounts for testing. 128 129tmpfs also supports quota with the following mount options 130 131======================== ================================================= 132quota User and group quota accounting and enforcement 133 is enabled on the mount. Tmpfs is using hidden 134 system quota files that are initialized on mount. 135usrquota User quota accounting and enforcement is enabled 136 on the mount. 137grpquota Group quota accounting and enforcement is enabled 138 on the mount. 139usrquota_block_hardlimit Set global user quota block hard limit. 140usrquota_inode_hardlimit Set global user quota inode hard limit. 141grpquota_block_hardlimit Set global group quota block hard limit. 142grpquota_inode_hardlimit Set global group quota inode hard limit. 143======================== ================================================= 144 145None of the quota related mount options can be set or changed on remount. 146 147Quota limit parameters accept a suffix k, m or g for kilo, mega and giga 148and can't be changed on remount. Default global quota limits are taking 149effect for any and all user/group/project except root the first time the 150quota entry for user/group/project id is being accessed - typically the 151first time an inode with a particular id ownership is being created after 152the mount. In other words, instead of the limits being initialized to zero, 153they are initialized with the particular value provided with these mount 154options. The limits can be changed for any user/group id at any time as they 155normally can be. 156 157Note that tmpfs quotas do not support user namespaces so no uid/gid 158translation is done if quotas are enabled inside user namespaces. 159 160tmpfs has a mount option to set the NUMA memory allocation policy for 161all files in that instance (if CONFIG_NUMA is enabled) - which can be 162adjusted on the fly via 'mount -o remount ...' 163 164======================== ============================================== 165mpol=default use the process allocation policy 166 (see set_mempolicy(2)) 167mpol=prefer:Node prefers to allocate memory from the given Node 168mpol=bind:NodeList allocates memory only from nodes in NodeList 169mpol=interleave prefers to allocate from each node in turn 170mpol=interleave:NodeList allocates from each node of NodeList in turn 171mpol=local prefers to allocate memory from the local node 172======================== ============================================== 173 174NodeList format is a comma-separated list of decimal numbers and ranges, 175a range being two hyphen-separated decimal numbers, the smallest and 176largest node numbers in the range. For example, mpol=bind:0-3,5,7,9-15 177 178A memory policy with a valid NodeList will be saved, as specified, for 179use at file creation time. When a task allocates a file in the file 180system, the mount option memory policy will be applied with a NodeList, 181if any, modified by the calling task's cpuset constraints 182[See Documentation/admin-guide/cgroup-v1/cpusets.rst] and any optional flags, 183listed below. If the resulting NodeLists is the empty set, the effective 184memory policy for the file will revert to "default" policy. 185 186NUMA memory allocation policies have optional flags that can be used in 187conjunction with their modes. These optional flags can be specified 188when tmpfs is mounted by appending them to the mode before the NodeList. 189See Documentation/admin-guide/mm/numa_memory_policy.rst for a list of 190all available memory allocation policy mode flags and their effect on 191memory policy. 192 193:: 194 195 =static is equivalent to MPOL_F_STATIC_NODES 196 =relative is equivalent to MPOL_F_RELATIVE_NODES 197 198For example, mpol=bind=static:NodeList, is the equivalent of an 199allocation policy of MPOL_BIND | MPOL_F_STATIC_NODES. 200 201Note that trying to mount a tmpfs with an mpol option will fail if the 202running kernel does not support NUMA; and will fail if its nodelist 203specifies a node which is not online. If your system relies on that 204tmpfs being mounted, but from time to time runs a kernel built without 205NUMA capability (perhaps a safe recovery kernel), or with fewer nodes 206online, then it is advisable to omit the mpol option from automatic 207mount options. It can be added later, when the tmpfs is already mounted 208on MountPoint, by 'mount -o remount,mpol=Policy:NodeList MountPoint'. 209 210 211To specify the initial root directory you can use the following mount 212options: 213 214==== ================================== 215mode The permissions as an octal number 216uid The user id 217gid The group id 218==== ================================== 219 220These options do not have any effect on remount. You can change these 221parameters with chmod(1), chown(1) and chgrp(1) on a mounted filesystem. 222 223 224tmpfs has a mount option to select whether it will wrap at 32- or 64-bit inode 225numbers: 226 227======= ======================== 228inode64 Use 64-bit inode numbers 229inode32 Use 32-bit inode numbers 230======= ======================== 231 232On a 32-bit kernel, inode32 is implicit, and inode64 is refused at mount time. 233On a 64-bit kernel, CONFIG_TMPFS_INODE64 sets the default. inode64 avoids the 234possibility of multiple files with the same inode number on a single device; 235but risks glibc failing with EOVERFLOW once 33-bit inode numbers are reached - 236if a long-lived tmpfs is accessed by 32-bit applications so ancient that 237opening a file larger than 2GiB fails with EINVAL. 238 239 240So 'mount -t tmpfs -o size=10G,nr_inodes=10k,mode=700 tmpfs /mytmpfs' 241will give you tmpfs instance on /mytmpfs which can allocate 10GB 242RAM/SWAP in 10240 inodes and it is only accessible by root. 243 244 245:Author: 246 Christoph Rohland <cr@sap.com>, 1.12.01 247:Updated: 248 Hugh Dickins, 4 June 2007 249:Updated: 250 KOSAKI Motohiro, 16 Mar 2010 251:Updated: 252 Chris Down, 13 July 2020 253