tmpfs.rst (7813B)
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. It has maximum size limits which can 17be adjusted on the fly via 'mount -o remount ...' 18 19If you compare it to ramfs (which was the template to create tmpfs) 20you gain swapping and limit checking. Another similar thing is the RAM 21disk (/dev/ram*), which simulates a fixed size hard disk in physical 22RAM, where you have to create an ordinary filesystem on top. Ramdisks 23cannot swap and you do not have the possibility to resize them. 24 25Since tmpfs lives completely in the page cache and on swap, all tmpfs 26pages will be shown as "Shmem" in /proc/meminfo and "Shared" in 27free(1). Notice that these counters also include shared memory 28(shmem, see ipcs(1)). The most reliable way to get the count is 29using df(1) and du(1). 30 31tmpfs has the following uses: 32 331) There is always a kernel internal mount which you will not see at 34 all. This is used for shared anonymous mappings and SYSV shared 35 memory. 36 37 This mount does not depend on CONFIG_TMPFS. If CONFIG_TMPFS is not 38 set, the user visible part of tmpfs is not built. But the internal 39 mechanisms are always present. 40 412) glibc 2.2 and above expects tmpfs to be mounted at /dev/shm for 42 POSIX shared memory (shm_open, shm_unlink). Adding the following 43 line to /etc/fstab should take care of this:: 44 45 tmpfs /dev/shm tmpfs defaults 0 0 46 47 Remember to create the directory that you intend to mount tmpfs on 48 if necessary. 49 50 This mount is _not_ needed for SYSV shared memory. The internal 51 mount is used for that. (In the 2.3 kernel versions it was 52 necessary to mount the predecessor of tmpfs (shm fs) to use SYSV 53 shared memory.) 54 553) Some people (including me) find it very convenient to mount it 56 e.g. on /tmp and /var/tmp and have a big swap partition. And now 57 loop mounts of tmpfs files do work, so mkinitrd shipped by most 58 distributions should succeed with a tmpfs /tmp. 59 604) And probably a lot more I do not know about :-) 61 62 63tmpfs has three mount options for sizing: 64 65========= ============================================================ 66size The limit of allocated bytes for this tmpfs instance. The 67 default is half of your physical RAM without swap. If you 68 oversize your tmpfs instances the machine will deadlock 69 since the OOM handler will not be able to free that memory. 70nr_blocks The same as size, but in blocks of PAGE_SIZE. 71nr_inodes The maximum number of inodes for this instance. The default 72 is half of the number of your physical RAM pages, or (on a 73 machine with highmem) the number of lowmem RAM pages, 74 whichever is the lower. 75========= ============================================================ 76 77These parameters accept a suffix k, m or g for kilo, mega and giga and 78can be changed on remount. The size parameter also accepts a suffix % 79to limit this tmpfs instance to that percentage of your physical RAM: 80the default, when neither size nor nr_blocks is specified, is size=50% 81 82If nr_blocks=0 (or size=0), blocks will not be limited in that instance; 83if nr_inodes=0, inodes will not be limited. It is generally unwise to 84mount with such options, since it allows any user with write access to 85use up all the memory on the machine; but enhances the scalability of 86that instance in a system with many CPUs making intensive use of it. 87 88 89tmpfs has a mount option to set the NUMA memory allocation policy for 90all files in that instance (if CONFIG_NUMA is enabled) - which can be 91adjusted on the fly via 'mount -o remount ...' 92 93======================== ============================================== 94mpol=default use the process allocation policy 95 (see set_mempolicy(2)) 96mpol=prefer:Node prefers to allocate memory from the given Node 97mpol=bind:NodeList allocates memory only from nodes in NodeList 98mpol=interleave prefers to allocate from each node in turn 99mpol=interleave:NodeList allocates from each node of NodeList in turn 100mpol=local prefers to allocate memory from the local node 101======================== ============================================== 102 103NodeList format is a comma-separated list of decimal numbers and ranges, 104a range being two hyphen-separated decimal numbers, the smallest and 105largest node numbers in the range. For example, mpol=bind:0-3,5,7,9-15 106 107A memory policy with a valid NodeList will be saved, as specified, for 108use at file creation time. When a task allocates a file in the file 109system, the mount option memory policy will be applied with a NodeList, 110if any, modified by the calling task's cpuset constraints 111[See Documentation/admin-guide/cgroup-v1/cpusets.rst] and any optional flags, 112listed below. If the resulting NodeLists is the empty set, the effective 113memory policy for the file will revert to "default" policy. 114 115NUMA memory allocation policies have optional flags that can be used in 116conjunction with their modes. These optional flags can be specified 117when tmpfs is mounted by appending them to the mode before the NodeList. 118See Documentation/admin-guide/mm/numa_memory_policy.rst for a list of 119all available memory allocation policy mode flags and their effect on 120memory policy. 121 122:: 123 124 =static is equivalent to MPOL_F_STATIC_NODES 125 =relative is equivalent to MPOL_F_RELATIVE_NODES 126 127For example, mpol=bind=static:NodeList, is the equivalent of an 128allocation policy of MPOL_BIND | MPOL_F_STATIC_NODES. 129 130Note that trying to mount a tmpfs with an mpol option will fail if the 131running kernel does not support NUMA; and will fail if its nodelist 132specifies a node which is not online. If your system relies on that 133tmpfs being mounted, but from time to time runs a kernel built without 134NUMA capability (perhaps a safe recovery kernel), or with fewer nodes 135online, then it is advisable to omit the mpol option from automatic 136mount options. It can be added later, when the tmpfs is already mounted 137on MountPoint, by 'mount -o remount,mpol=Policy:NodeList MountPoint'. 138 139 140To specify the initial root directory you can use the following mount 141options: 142 143==== ================================== 144mode The permissions as an octal number 145uid The user id 146gid The group id 147==== ================================== 148 149These options do not have any effect on remount. You can change these 150parameters with chmod(1), chown(1) and chgrp(1) on a mounted filesystem. 151 152 153tmpfs has a mount option to select whether it will wrap at 32- or 64-bit inode 154numbers: 155 156======= ======================== 157inode64 Use 64-bit inode numbers 158inode32 Use 32-bit inode numbers 159======= ======================== 160 161On a 32-bit kernel, inode32 is implicit, and inode64 is refused at mount time. 162On a 64-bit kernel, CONFIG_TMPFS_INODE64 sets the default. inode64 avoids the 163possibility of multiple files with the same inode number on a single device; 164but risks glibc failing with EOVERFLOW once 33-bit inode numbers are reached - 165if a long-lived tmpfs is accessed by 32-bit applications so ancient that 166opening a file larger than 2GiB fails with EINVAL. 167 168 169So 'mount -t tmpfs -o size=10G,nr_inodes=10k,mode=700 tmpfs /mytmpfs' 170will give you tmpfs instance on /mytmpfs which can allocate 10GB 171RAM/SWAP in 10240 inodes and it is only accessible by root. 172 173 174:Author: 175 Christoph Rohland <cr@sap.com>, 1.12.01 176:Updated: 177 Hugh Dickins, 4 June 2007 178:Updated: 179 KOSAKI Motohiro, 16 Mar 2010 180:Updated: 181 Chris Down, 13 July 2020