Where does uname get its information from?

Question

Where does uname really get its information from?

I figure this is something that should be straightforward. Unfortunately, I can't find any header containing just that information.

Say someone wanted to change the basic output of uname/uname -s from Linux to something else (essentially, renaming the kernel).

How would he/she go about doing that the proper way (that is, changing the source)?

Solution 1

The data is stored in init/version.c:

struct uts_namespace init_uts_ns = {
        .kref = {
                .refcount       = ATOMIC_INIT(2),
        },
        .name = {
                .sysname        = UTS_SYSNAME,
                .nodename       = UTS_NODENAME,
                .release        = UTS_RELEASE,
                .version        = UTS_VERSION,
                .machine        = UTS_MACHINE,
                .domainname     = UTS_DOMAINNAME,
        },
        .user_ns = &init_user_ns,
        .proc_inum = PROC_UTS_INIT_INO,
};
EXPORT_SYMBOL_GPL(init_uts_ns);

The strings themselves are in include/generated/compile.h:

#define UTS_MACHINE "x86_64"
#define UTS_VERSION "#30 SMP Fri Apr 11 00:24:23 BST 2014"

and in include/generated/utsrelease.h:

#define UTS_RELEASE "3.14.0-v2-v"

UTS_SYSNAME may be defined in include/linux/uts.h

#ifndef UTS_SYSNAME
#define UTS_SYSNAME "Linux"
#endif

or as a #define in makefiles

Finally, the hostname and domainname can be controlled by /proc/sys/kernel/{hostname,domainname}. These are per UTS namespace:

# hostname
hell
# unshare --uts /bin/bash
# echo test > /proc/sys/kernel/hostname 
# hostname
test
# exit
# hostname
hell

Solution 2

The uname utility gets its information from the uname() system call. It populates a struct like this (see man 2 uname):

       struct utsname {
           char sysname[];    /* Operating system name (e.g., "Linux") */
           char nodename[];   /* Name within "some implementation-defined
                                 network" */
           char release[];    /* Operating system release (e.g., "2.6.28") */
           char version[];    /* Operating system version */
           char machine[];    /* Hardware identifier */
       #ifdef _GNU_SOURCE
           char domainname[]; /* NIS or YP domain name */
       #endif
       };

This comes directly from the running kernel. I would assume all of the information is hard-coded into it, except perhaps domainname (and as it turns out, also nodename, machine, and release, see comments). The release string, from uname -r, can be set via configuration at compile time, but I doubt very much the sysname field can -- it's the Linux kernel and there's no conceivable reason for it to use anything else.

However, since it is open source, you could change the source code and recompile the kernel to use whatever sysname you want.

Solution 3

With the help of a Linux Cross Reference and your mention of /proc/sys/kernel/ostype, I tracked ostype to include/linux/sysctl.h, where a comment says that names are added by calling register_sysctl_table.

So where is that called from? One place is kernel/utsname_sysctl.c, which includes include/linux/uts.h, where we find:

/*
 * Defines for what uname() should return 
 */
#ifndef UTS_SYSNAME
#define UTS_SYSNAME "Linux"
#endif

So, as the kernel documentation states:

The only way to tune these values is to rebuild the kernel

:-)

Solution 4

As commented elsewhere, the information come with the uname syscall, which information is hard-coded in the running kernel.

The version part is normally set when compiling a new kernel by the Makefile:

VERSION = 3
PATCHLEVEL = 15
SUBLEVEL = 0
EXTRAVERSION =

when I had time to play compiling my kernels, I used to add things over there in EXTRAVERSION; that gave you uname -r with things like 3.4.1-mytestkernel.

I do not fully understand it, but I think that the rest of the information is setup in the Makefile also around line 944:

# ---------------------------------------------------------------------------

# KERNELRELEASE can change from a few different places, meaning version.h
# needs to be updated, so this check is forced on all builds

uts_len := 64
define filechk_utsrelease.h
    if [ `echo -n "$(KERNELRELEASE)" | wc -c ` -gt $(uts_len) ]; then \
      echo '"$(KERNELRELEASE)" exceeds $(uts_len) characters' >&2;    \
      exit 1;                                                         \
    fi;                                                               \
    (echo \#define UTS_RELEASE \"$(KERNELRELEASE)\";)
endef

define filechk_version.h
    (echo \#define LINUX_VERSION_CODE $(shell                         \
    expr $(VERSION) \* 65536 + 0$(PATCHLEVEL) \* 256 + 0$(SUBLEVEL)); \
    echo '#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))';)
endef

$(version_h): $(srctree)/Makefile FORCE
    $(call filechk,version.h)

include/generated/utsrelease.h: include/config/kernel.release FORCE
    $(call filechk,utsrelease.h)

PHONY += headerdep
headerdep:
    $(Q)find $(srctree)/include/ -name '*.h' | xargs --max-args 1 \
    $(srctree)/scripts/headerdep.pl -I$(srctree)/include

For the rest of the data, the sys_uname syscall is generated using macros (in a quite convoluted way), you can start from here if you feel adventurous.

Probably the best way to change such information is writing a kernel module to override the uname syscall; I never did that but you can find info in this page at section 4.2 (sorry, no direct link). Notice however that that code is referring to a quite old kernel (now Linux kernel has uts namespaces, whatever they mean) so you will need to change it probably a lot.

Solution 5

scripts/mkcompile_h

In v4.19, this is the file that generates include/generated/compile.h, and contains several interesting parts of /proc/version: https://github.com/torvalds/linux/blob/v4.19/scripts/mkcompile_h

• the #<version> part comes from the .version file on the build tree, which gets incremented whenever link happens (requires file / config changes) by scripts/link-vmlinux.sh.

  It can be overridden by the KBUILD_BUILD_VERSION environment variable:

  if [ -z "$KBUILD_BUILD_VERSION" ]; then
      VERSION=$(cat .version 2>/dev/null || echo 1)
  else
      VERSION=$KBUILD_BUILD_VERSION
  fi
  

• the date is just a raw date call:

  if [ -z "$KBUILD_BUILD_TIMESTAMP" ]; then
      TIMESTAMP=`date`
  else
      TIMESTAMP=$KBUILD_BUILD_TIMESTAMP
  fi
  

  and similarly the username comes from whoami (KBUILD_BUILD_USER) and hostname from hostname (KBUILD_BUILD_HOST)

• The compiler version comes from gcc -v, and cannot be controlled it seems.

Here is a how to change stuff version of the question: https://stackoverflow.com/questions/23424174/how-to-customize-or-remove-extra-linux-kernel-version-details-shown-at-boot

View more solutions

Comments

[Graeme]

If you do strace uname, it will confirm that the uname system call is used.

[user237251]

Thanks, everyone. I already knew it had something to do with uname. However, what I can't fathom is how and where inside the source the string "Linux" is defined. All I know is where I can find that information during runtime (it's contained inside /proc/sys/kernel/ostype). Finding out how exactly the kernel itself knows it's proper name would be one of the more interesting things, I'd say.

[Scott - Слава Україні]

The domainname field is set by the domainname command, using the setdomainname system call. Similarly, the nodename field is set by the hostname command, using the sethostname system call. (The nodename / hostname value may be stored in /etc/nodename.)

[goldilocks]

@Gilles: What's irrelevent? If the answer is "provided by other posters here: it's hard-coded into the kernel..." note I've said the exact same thing: "This comes directly from the running kernel. I would assume all of the information is hard-coded into it...since it is open source, you could change the source code and recompile the kernel to use whatever sysname you want. It is not a config option.

[Gilles 'SO- stop being evil']

Some fields there (nodename, obviously, and also machine) can be changed at runtime, on a per-process basis. The assumption that your conclusion rests on is wrong. It happens to be true for sysname, but you need to dig deeper to find this out.

[goldilocks]

@Gilles O_O? The assumption is obviously correct, as the other answers bare out: it is a hard coded define which could be altered via make or editing the source itself.

[goldilocks]

@Gilles : Because the question was about sysname, and I felt safe about that. It says Linux because that's what it is, why would it say anything else? +1 to the other answers that confirmed this (I would have done so if prodded by the OP; "it's hard-coded" seems generally sufficient).

[Gilles 'SO- stop being evil']

@goldilocks Why would machine ever change? It might not be hardcoded into the kernel because it might adapt to the hardware, but surely then it would be set at boot time and wouldn't change after that. But no: it can be set per process (e.g. to report i686 to 32-bit processed on x86_64). By the way, release can also be customized per process to some extent (try setarch i686 --uname-2.6 uname -a).

[Alen Milakovic]

This is generally a good and complete answer, but it may be worth while answering the poster's question directl. I believe this would amount to - change the relevant entry in the relevant file and recompile. You wrote "or as a #define in makefiles". Can you elaborate?

[user237251]

@JAB Way too many. Fortunately, someone on kernelnewbies.org helped me solve the "mystery". Linux gets its sys name from /include/Linux/uts.h. See here: lxr.free-electrons.com/source/include/linux/uts.h?v=3.10

[invot]

+1 for unshare. Somehow I managed to miss this command until today. Thanks!

[Ciro Santilli Путлер Капут 六四事]

And include/generated/compile.h is generated by scripts/mkcompile_h: unix.stackexchange.com/a/485962/32558