Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | Tmpfs is a file system which keeps all files in virtual memory. |
| 2 | |
| 3 | |
| 4 | Everything in tmpfs is temporary in the sense that no files will be |
| 5 | created on your hard drive. If you unmount a tmpfs instance, |
| 6 | everything stored therein is lost. |
| 7 | |
| 8 | tmpfs puts everything into the kernel internal caches and grows and |
| 9 | shrinks to accommodate the files it contains and is able to swap |
| 10 | unneeded pages out to swap space. It has maximum size limits which can |
| 11 | be adjusted on the fly via 'mount -o remount ...' |
| 12 | |
| 13 | If you compare it to ramfs (which was the template to create tmpfs) |
| 14 | you gain swapping and limit checking. Another similar thing is the RAM |
| 15 | disk (/dev/ram*), which simulates a fixed size hard disk in physical |
| 16 | RAM, where you have to create an ordinary filesystem on top. Ramdisks |
| 17 | cannot swap and you do not have the possibility to resize them. |
| 18 | |
| 19 | Since tmpfs lives completely in the page cache and on swap, all tmpfs |
Rodrigo Freire | 0bc126d | 2016-01-14 15:21:58 -0800 | [diff] [blame] | 20 | pages will be shown as "Shmem" in /proc/meminfo and "Shared" in |
| 21 | free(1). Notice that these counters also include shared memory |
| 22 | (shmem, see ipcs(1)). The most reliable way to get the count is |
| 23 | using df(1) and du(1). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | |
| 25 | tmpfs has the following uses: |
| 26 | |
| 27 | 1) There is always a kernel internal mount which you will not see at |
| 28 | all. This is used for shared anonymous mappings and SYSV shared |
| 29 | memory. |
| 30 | |
| 31 | This mount does not depend on CONFIG_TMPFS. If CONFIG_TMPFS is not |
| 32 | set, the user visible part of tmpfs is not build. But the internal |
| 33 | mechanisms are always present. |
| 34 | |
| 35 | 2) glibc 2.2 and above expects tmpfs to be mounted at /dev/shm for |
| 36 | POSIX shared memory (shm_open, shm_unlink). Adding the following |
| 37 | line to /etc/fstab should take care of this: |
| 38 | |
| 39 | tmpfs /dev/shm tmpfs defaults 0 0 |
| 40 | |
| 41 | Remember to create the directory that you intend to mount tmpfs on |
Adrian Bunk | bf6ee0a | 2006-10-03 22:17:48 +0200 | [diff] [blame] | 42 | if necessary. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | |
| 44 | This mount is _not_ needed for SYSV shared memory. The internal |
| 45 | mount is used for that. (In the 2.3 kernel versions it was |
| 46 | necessary to mount the predecessor of tmpfs (shm fs) to use SYSV |
| 47 | shared memory) |
| 48 | |
| 49 | 3) Some people (including me) find it very convenient to mount it |
| 50 | e.g. on /tmp and /var/tmp and have a big swap partition. And now |
| 51 | loop mounts of tmpfs files do work, so mkinitrd shipped by most |
| 52 | distributions should succeed with a tmpfs /tmp. |
| 53 | |
| 54 | 4) And probably a lot more I do not know about :-) |
| 55 | |
| 56 | |
| 57 | tmpfs has three mount options for sizing: |
| 58 | |
| 59 | size: The limit of allocated bytes for this tmpfs instance. The |
| 60 | default is half of your physical RAM without swap. If you |
| 61 | oversize your tmpfs instances the machine will deadlock |
| 62 | since the OOM handler will not be able to free that memory. |
Kirill A. Shutemov | ea1754a | 2016-04-01 15:29:48 +0300 | [diff] [blame] | 63 | nr_blocks: The same as size, but in blocks of PAGE_SIZE. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | nr_inodes: The maximum number of inodes for this instance. The default |
| 65 | is half of the number of your physical RAM pages, or (on a |
Paolo Ornati | 670e9f3 | 2006-10-03 22:57:56 +0200 | [diff] [blame] | 66 | machine with highmem) the number of lowmem RAM pages, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | whichever is the lower. |
| 68 | |
| 69 | These parameters accept a suffix k, m or g for kilo, mega and giga and |
| 70 | can be changed on remount. The size parameter also accepts a suffix % |
| 71 | to limit this tmpfs instance to that percentage of your physical RAM: |
| 72 | the default, when neither size nor nr_blocks is specified, is size=50% |
| 73 | |
Hugh Dickins | 0edd73b | 2005-06-21 17:15:04 -0700 | [diff] [blame] | 74 | If nr_blocks=0 (or size=0), blocks will not be limited in that instance; |
| 75 | if nr_inodes=0, inodes will not be limited. It is generally unwise to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | mount with such options, since it allows any user with write access to |
| 77 | use up all the memory on the machine; but enhances the scalability of |
| 78 | that instance in a system with many cpus making intensive use of it. |
| 79 | |
| 80 | |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 81 | tmpfs has a mount option to set the NUMA memory allocation policy for |
Hugh Dickins | b00dc3a | 2006-02-21 23:49:47 +0000 | [diff] [blame] | 82 | all files in that instance (if CONFIG_NUMA is enabled) - which can be |
| 83 | adjusted on the fly via 'mount -o remount ...' |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 84 | |
KOSAKI Motohiro | 5574169 | 2010-03-23 13:35:33 -0700 | [diff] [blame] | 85 | mpol=default use the process allocation policy |
| 86 | (see set_mempolicy(2)) |
Hugh Dickins | b00dc3a | 2006-02-21 23:49:47 +0000 | [diff] [blame] | 87 | mpol=prefer:Node prefers to allocate memory from the given Node |
| 88 | mpol=bind:NodeList allocates memory only from nodes in NodeList |
| 89 | mpol=interleave prefers to allocate from each node in turn |
| 90 | mpol=interleave:NodeList allocates from each node of NodeList in turn |
KOSAKI Motohiro | 5574169 | 2010-03-23 13:35:33 -0700 | [diff] [blame] | 91 | mpol=local prefers to allocate memory from the local node |
Hugh Dickins | b00dc3a | 2006-02-21 23:49:47 +0000 | [diff] [blame] | 92 | |
| 93 | NodeList format is a comma-separated list of decimal numbers and ranges, |
| 94 | a range being two hyphen-separated decimal numbers, the smallest and |
| 95 | largest node numbers in the range. For example, mpol=bind:0-3,5,7,9-15 |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 96 | |
Lee Schermerhorn | 971ada0 | 2010-05-24 14:32:05 -0700 | [diff] [blame] | 97 | A memory policy with a valid NodeList will be saved, as specified, for |
| 98 | use at file creation time. When a task allocates a file in the file |
| 99 | system, the mount option memory policy will be applied with a NodeList, |
| 100 | if any, modified by the calling task's cpuset constraints |
| 101 | [See Documentation/cgroups/cpusets.txt] and any optional flags, listed |
| 102 | below. If the resulting NodeLists is the empty set, the effective memory |
| 103 | policy for the file will revert to "default" policy. |
| 104 | |
David Rientjes | 65d66fc | 2008-04-28 02:12:31 -0700 | [diff] [blame] | 105 | NUMA memory allocation policies have optional flags that can be used in |
| 106 | conjunction with their modes. These optional flags can be specified |
| 107 | when tmpfs is mounted by appending them to the mode before the NodeList. |
| 108 | See Documentation/vm/numa_memory_policy.txt for a list of all available |
Lee Schermerhorn | 971ada0 | 2010-05-24 14:32:05 -0700 | [diff] [blame] | 109 | memory allocation policy mode flags and their effect on memory policy. |
David Rientjes | 65d66fc | 2008-04-28 02:12:31 -0700 | [diff] [blame] | 110 | |
| 111 | =static is equivalent to MPOL_F_STATIC_NODES |
| 112 | =relative is equivalent to MPOL_F_RELATIVE_NODES |
| 113 | |
| 114 | For example, mpol=bind=static:NodeList, is the equivalent of an |
| 115 | allocation policy of MPOL_BIND | MPOL_F_STATIC_NODES. |
| 116 | |
Hugh Dickins | ad329b15 | 2006-02-24 13:04:26 -0800 | [diff] [blame] | 117 | Note that trying to mount a tmpfs with an mpol option will fail if the |
| 118 | running kernel does not support NUMA; and will fail if its nodelist |
Hugh Dickins | a210906 | 2007-06-08 13:46:46 -0700 | [diff] [blame] | 119 | specifies a node which is not online. If your system relies on that |
| 120 | tmpfs being mounted, but from time to time runs a kernel built without |
| 121 | NUMA capability (perhaps a safe recovery kernel), or with fewer nodes |
| 122 | online, then it is advisable to omit the mpol option from automatic |
Hugh Dickins | ad329b15 | 2006-02-24 13:04:26 -0800 | [diff] [blame] | 123 | mount options. It can be added later, when the tmpfs is already mounted |
| 124 | on MountPoint, by 'mount -o remount,mpol=Policy:NodeList MountPoint'. |
| 125 | |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 126 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | To specify the initial root directory you can use the following mount |
| 128 | options: |
| 129 | |
| 130 | mode: The permissions as an octal number |
| 131 | uid: The user id |
| 132 | gid: The group id |
| 133 | |
| 134 | These options do not have any effect on remount. You can change these |
| 135 | parameters with chmod(1), chown(1) and chgrp(1) on a mounted filesystem. |
| 136 | |
| 137 | |
| 138 | So 'mount -t tmpfs -o size=10G,nr_inodes=10k,mode=700 tmpfs /mytmpfs' |
| 139 | will give you tmpfs instance on /mytmpfs which can allocate 10GB |
| 140 | RAM/SWAP in 10240 inodes and it is only accessible by root. |
| 141 | |
| 142 | |
| 143 | Author: |
| 144 | Christoph Rohland <cr@sap.com>, 1.12.01 |
| 145 | Updated: |
Hugh Dickins | 98f3260 | 2009-05-21 20:33:58 +0100 | [diff] [blame] | 146 | Hugh Dickins, 4 June 2007 |
KOSAKI Motohiro | 5574169 | 2010-03-23 13:35:33 -0700 | [diff] [blame] | 147 | Updated: |
| 148 | KOSAKI Motohiro, 16 Mar 2010 |