libfaketime, version 0.9.10 (March 2022)
========================================


Content of this file:
---------------------

1. Introduction
2. Compatibility issues
3. Installation
4. Usage
   a) Basics
   b) Using absolute dates
   c) Using 'start at' dates
   d) Using offsets for relative dates
   e) Advanced features and caveats
   f) Faking the date and time system-wide
   g) Using the "faketime" wrapper script
   h) "Limiting" libfaketime based on elapsed time or number of calls
   i) "Limiting" libfaketime per process
   j) Spawning an external process
   k) Saving timestamps to file, loading them from file
   l) Replacing random numbers with deterministic number sequences
5. License
6. Contact


1. Introduction
---------------

libfaketime intercepts various system calls that programs use to retrieve the
current date and time. It then reports modified (faked) dates and times (as
specified by you, the user) to these programs. This means you can modify the
system time a program sees without having to change the time system-wide.

libfaketime allows you to specify both absolute dates (e.g., 01/01/2004) and
relative dates (e.g., 10 days ago).

libfaketime might be used for various purposes, for example

- deterministic build processes
- debugging time-related issues, such as expired SSL certificates
- testing software for year-2038 compliance

libfaketime ships with a command line wrapper called "faketime" that makes it
easier to use, but does not expose all of libfaketime's functionality. If your
use case is not covered by the faketime command, make sure to look in this
documentation whether it can be achieved by using libfaketime directly.


2. Compatibility issues
-----------------------

- libfaketime is supposed to work on Linux and macOS.
  Your mileage may vary; some other *NIXes have been reported to work as well.

- libfaketime uses the library preload mechanism of your operating system's
  linker (which is involved in starting programs) and thus cannot work with
  statically linked binaries or binaries that have the setuid-flag set (e.g.,
  suidroot programs like "ping" or "passwd"). Please see you system linker's
  manpage for further details.

- libfaketime supports the pthreads environment. A separate library is built
  (libfaketimeMT.so.1), which contains the pthread synchronization calls. This
  library also single-threads calls through the time() intercept because
  several variables are statically cached by the library and could cause issues
  when accessed without synchronization.

  However, the performance penalty for this might be an issue for some
  applications. If this is the case, you can try using an unsynchronized time()
  intercept by removing the -DPTHREAD_SINGLETHREADED_TIME from the Makefile and
  rebuilding libfaketimeMT.so.1

* Java-/JVM-based applications work but you need to pass in an extra argument
  (FAKETIME_DONT_FAKE_MONOTONIC).  See usage basics below for details.  Without
  this argument the java command usually hangs.

* libfaketime will eventually be bypassed by applications that dynamically load
  system libraries, such as librt, explicitly themselves instead of relying on
  the linker to do so at application start time. libfaketime will not work with
  those applications unless you can modify them.

  This apparently happens a lot in complex run-time environments, e.g., for
  programs written in golang, for some Java Virtual Machine implementations,
  etc. Since libfaketime is effectively bypassed in such situations, there's
  nothing we can do about it. Please consider asking the appropriate developers
  and vendors to implement their runtime environment in a way that supports
  intercepting selected system calls through LD_PRELOAD.

* Applications can explicitly be designed to prevent libfaketime from working,
  e.g., by checking whether certain environment variables are set or whether
  libfaketime-specific files are present.

* CLOCK_MONOTONIC test: Running "make test" performs a series of tests after
  successful compilation of the libfaketime library. On some platforms, the
  "CLOCK_MONOTONIC test" will apparently hang forever. If and only if this
  happens on your platform, add the CFLAG -DFORCE_MONOTONIC_FIX to
  src/Makefile and recompile libfaketime. Do not set FORCE_MONOTONIC_FIX on
  platforms where the test does not hang.

  If you observe hangs on the CLOCK_REALTIME test, add the CFLAG
  -DFORCE_PTHREAD_NONVER. Also set this FORCE_PTHREAD_NONVER flag in case
  FORCE_MONOTONIC_FIX alone does not solve the hang on the MONOTONIC_CLOCK
  test.

  If FORCE_MONOTONIC_FIX was not set as a compile-time flag, you can also
  set an environment variable FAKETIME_FORCE_MONOTONIC_FIX=1 if you want
  to enable the fix at run-time, or to 0 if you explicitly want to disable
  it. The fix is automatically enabled if libfaketime was compiled on a
  system with glibc as the underlying libc implementation, and a glibc
  version is detected at run-time that is assumed to need this workaround.
  Please use Github issues at https://github.com/wolfcw/libfaketime/issues
  to report any observed hangs during CLOCK_MONOTONIC tests and report
  your CPU architecture, libc implementation (e.g., glibc 2.30) and any
  other details that might help (e.g., Linux distribution, use within, e.g.,
  Docker containers etc.).

  Please try to avoid compiling with FORCE_MONOTONIC_FIX on platforms that
  do not need it. While it won't make a difference in most cases, depending
  on the specific FAKETIME settings in use, it would cause certain
  intercepted functions such as pthread_cond_timedwait() return with a
  time-out too early or too late, which could break some applications.
  Try compiling without FORCE_MONOTONIC_FIX first and check whether the
  tests appear to hang. If they do, you can either set the
  FAKETIME_FORCE_MONOTONIC_FIX environment variable to 1, or re-compile
  with FORCE_MONOTONIC_FIX set.


3. Installation
---------------

Running "make" compiles both library versions and a test program, which it then
also executes.

If the test works fine, you should copy the libfaketime libraries
(libfaketime.so.1, and libfaketimeMT.so.1) to the place you want them in.
Running "make install" will attempt to place them in /usr/local/lib/faketime
and will install the wrapper shell script "faketime" in /usr/local/bin, both of
which most likely will require root privileges. However, from a technical point
of view, there is no necessity for a system-wide installation, so you can use
libfaketime also on machines where you do not have root privileges. You may
want to adjust the PREFIX variable in the Makefiles accordingly.

By default, the Makefile compiles/links libfaketime for your default system
architecture. If you need to build, e.g., 32-bit files on a 64-bit platform,
please see the notes about CFLAGS and LDFLAGS in src/Makefile.

Since version 0.6, system calls to file timestamps are also intercepted,
thanks to a contribution by Philipp Hachtmann. This is especially useful in
combination with relative time offsets as explained in section 4d) below, if a
program writes and reads files whose timestamps also shall be faked. If you do
not need this feature or if it confuses the application you want to use FTPL
with, define the environment variable NO_FAKE_STAT, and the intercepted stat
calls will be passed through unaltered.

On macOS, it is necessary to compile differently, due to the different
behavior dyld has. Use the Makefile.OSX file provided to compile
libfaketime.1.dylib. Additionally, instead of using LD_PRELOAD,
the variable DYLD_INSERT_LIBRARIES should be set to the path to
libfaketime.1.dylib, and the variable DYLD_FORCE_FLAT_NAMESPACE should be
set (to anything). macOS users should read README.OSX for additional
details.


4. Usage
--------

4a) Usage basics
----------------

Using libfaketime on a program of your choice consists of two steps:

1. Making sure libfaketime gets loaded by the system's linker.
2. Specify the faked time.

As an example, we want the "date" command to report our faked time. To do so,
we could use the following command line on Linux:

user@host> date
Tue Nov 23 12:01:05 CEST 2016

user@host> LD_PRELOAD=/usr/local/lib/libfaketime.so.1 FAKETIME="-15d" date
Mon Nov  8 12:01:12 CEST 2016

user@host> LD_PRELOAD=/usr/local/lib/libfaketime.so.1 FAKETIME="-15d"
FAKETIME_DONT_FAKE_MONOTONIC=1 java -version
java version "1.8.0_111"
Java(TM) SE Runtime Environment (build 1.8.0_111-b14)
Java HotSpot(TM) 64-Bit Server VM (build 25.111-b14, mixed mode)

The basic way of running any command/program with libfaketime enabled is to
make sure the environment variable LD_PRELOAD contains the path and
filename of the libfaketime library. This can either be done by setting it once
beforehand:

export LD_PRELOAD=/path/to/libfaketime.so.1
(now run any command you want)

Or it can be done by specifying it on the command line itself:

LD_PRELOAD=/path/to/libfaketime.so.1 your_command_here

(These examples are for the bash shell; how environment variables are set may
vary on your system.)

On Linux, library search paths can be set as part of the linker configuration.
LD_PRELOAD then also works with relative paths. For example, when libfaketime.so.1
is installed as /path/to/libfaketime.so.1, you can add /path/to to an appropriate
linker configuration file, e.g., /etc/ld.so.conf.d/local.conf, and then run
the "ldconfig" command. Afterwards, using LD_PRELOAD=libfaketime.so.1 suffices.

However, also the faked time should be specified; otherwise, libfaketime will
be loaded, but just report the real system time. There are multiple ways to
specify the faked time:

a) By setting the environment variable FAKETIME.
b) By using the file given in the environment variable FAKETIME_TIMESTAMP_FILE
c) By using the file .faketimerc in your home directory.
d) By using the file /etc/faketimerc for a system-wide default.
e) By using FAKETIME_UPDATE_TIMESTAMP_FILE and date -s "<time>" or alike.

If you want to use b) c) or d), $HOME/.faketimerc or /etc/faketimerc consist of
only one line of text with exactly the same content as the FAKETIME environment
variable, which is described below. Note that /etc/faketimerc will only be used
if there is no $HOME/.faketimerc and no FAKETIME_TIMESTAMP_FILE file exists.
Also, the FAKETIME environment variable _always_ has priority over the files.
For FAKETIME_UPDATE_TIMESTAMP_FILE please see below.


4b) Using absolute dates
------------------------

The format that _must_ be used for _absolute_ dates is "YYYY-MM-DD hh:mm:ss".
For example, the 24th of December, 2020, 8:30 PM would have to be specified as
FAKETIME="2020-12-24 20:30:00".


4c) Using 'start at' dates
--------------------------

(Thanks to a major contribution by David North, TDI in version 0.7)

The format that _must_ be used for _start_at_ dates is "@YYYY-MM-DD hh:mm:ss".
For example, the 24th of December, 2020, 8:30 PM would have to be specified as
FAKETIME="@2020-12-24 20:30:00".

The absolute dates described in 4b) simulate a STOPPED system clock at the
specified absolute time. The 'start at' format allows a 'relative' clock
operation as described below in section 4d), but using a 'start at' time
instead of an offset time.

If the started process itself starts other (child) processes, they by default
will start with the specified start-at-date again. If this is not what you need,
set the environment variable FAKETIME_DONT_RESET=1. Try these examples to see
the difference:

LD_PRELOAD=src/libfaketime.so.1 FAKETIME="@2000-01-01 11:12:13" \
  FAKETIME_DONT_RESET=1 \
  /bin/bash -c 'while [ $SECONDS -lt 5 ]; do date; sleep 1; done'

LD_PRELOAD=src/libfaketime.so.1 FAKETIME="@2000-01-01 11:12:13" \
  /bin/bash -c 'while [ $SECONDS -lt 5 ]; do date; sleep 1; done'

In the second example, the "date" command will always print the same time,
while in the first example, with FAKETIME_DONT_RESET set, time will increment
even though all the "date" commands are new processes.


4d) Using offsets for relative dates
------------------------------------

Relative date offsets can be positive or negative, thus what you put into
FAKETIME _must_ either start with a + or a -, followed by a number, and
optionally followed by a multiplier:

- By default, the offset you specify is in seconds. Example:

  export FAKETIME="-120" will set the faked time 2 minutes (120 seconds) behind
  the real time.

- The multipliers "m", "h", "d" and "y" can be used to specify the offset in
  minutes, hours, days and years (365 days each), respectively. Examples:

  export FAKETIME="-10m" sets the faked time 10 minutes behind the real time.
  export FAKETIME="+14d" sets the faked time to 14 days in the future.

  Please note that if you need other multipliers (weeks, months etc.) or higher
  precision (e.g., correct leap year handling), you should use either the
  faketime wrapper or the GNU date command as shown in the first of the three
  examples below.

You now should understand the complete example we've used before:

LD_PRELOAD=/usr/local/lib/libfaketime.so.1 FAKETIME="-15d" date

This command line makes sure libfaketime gets loaded and sets the faked time to
15 days in the past.

Moreno Baricevic has contributed support for the FAKETIME_FMT environment
variable, which allows you to optionally set the strptime() format:

Some simple examples:
LD_PRELOAD=./libfaketime.so.1 FAKETIME_FMT=%s FAKETIME="`date +%s -d'1 year ago'`" date
LD_PRELOAD=./libfaketime.so.1 FAKETIME_FMT=%s FAKETIME="`stat -c %Y somefile`" date
LD_PRELOAD=./libfaketime.so.1 FAKETIME_FMT=%c FAKETIME="`date`" date


4e) Advanced features and caveats
---------------------------------

Advanced time specification options:
------------------------------------

Since version 0.8, thanks to a contribution by Karl Chen, fractions can be used
in the specification of time offsets. For example,

FAKETIME="+1,5h"

is equivalent to FAKETIME="+90m". Please be aware that the fraction delimiter
depends on your locale settings, so actually you might need to use

FAKETIME="+1.5h"

You should figure out the proper delimiter, e.g., by using libfaketime on
a command like /bin/date where you immediately can verify whether it worked as
expected.

Also contributed by Karl Chen in v0.8 is the option to speed up or slow
down the wall clock time for the program which is executed using libfaketime.
For example,

FAKETIME="+1y x2"

will set the faked time one year into the future and will make the clock run
twice as fast. Similarly,

FAKETIME="+1y x0,5"

will make the clock run only half as fast. As stated above, the fraction
delimiter depends on your locale. Furthermore,

FAKETIME="+1y i2,0"

will make the clock step two seconds per each time(), etc. call, being
completely independently of the system clock. It helps running programs
with some determinism. In this single case all spawned processes will use
the same global clock without restarting it at the start of each process.

Please note that using "x" or "i" in FAKETIME still requires giving an offset
(see section 4d). This means that "+1y x2" will work, but "x2" only will not.
If you do not want to fake the time, but just modify clock speed, use something
like "+0 x2", i.e., use an explicit zero offset as a prefix in your FAKETIME.

For testing, your should run a command like

LD_PRELOAD=./libfaketime.so.1 FAKETIME="+1,5y x10,0" \
/bin/bash -c 'while true; do echo $SECONDS ; sleep 1 ; done'

For each second that the endless loop sleeps, the executed bash shell will
think that 10 seconds have passed ($SECONDS is a bash-internal variable
measuring the time since the shell was started).

(Please note that replacing "echo $SECONDS" e.g. with a call to "/bin/date"
will not give the expected result, since /bin/date will always be started as a
new process for which also libfaketime will be re-initialized. It will show the
correct offset (1.5 years in the future), but no speed-ups or slow-downs.)

For applications that should use a different date & time each time they are
run, consider using the included timeprivacy wrapper shellscript (contributed
by adrelanos at riseup dot net).


Caveats:
--------

Whenever possible, you should use relative offsets or 'start at' dates,
and not use absolute dates.

Why? Because the absolute date/time you set is fixed, i.e., if a program
retrieves the current time, and retrieves the current time again 5 minutes
later, it will still get the same result twice. This is likely to break
programs which measure the time passing by (e.g., a mail program which checks
for new mail every X minutes).

Using relative offsets or 'start at' dates solves this problem.
libfaketime then will always report the faked time based on the real
current time and the offset you've specified.

Please also note that your default specification of the fake time is cached for
10 seconds in order to enhance the library's performance. Thus, if you change the
content of $HOME/.faketimerc or /etc/faketimerc while a program is running, it
may take up to 10 seconds before the new fake time is applied. If this is a
problem in your scenario, you can change number of seconds before the file is read
again with environment variable FAKETIME_CACHE_DURATION, or disable caching at all
with FAKETIME_NO_CACHE=1. Remember that disabling the cache may negatively
influence the performance (especially when not using FAKETIME environment
but configuration files, such as FAKETIME_TIMESTAMP_FILE).


Setting FAKETIME by means of a file timestamp
---------------------------------------------

Based on a proposal by Hitoshi Harada (umitanuki), the "start at" time can now be
set through any file in the file system by setting the FAKETIME environment variable
to "%" (a percent sign) and FAKETIME_FOLLOW_FILE to the name of the file whose
modification timestamp shall be used as source for the "start at" time.

Usage example:

# create any file with December 24th, 2009, 12:34:56 as timestamp
touch -t 0912241234.56 /tmp/my-demo-file.tmp

# run a bash shell with an endless loop printing the current time
LD_PRELOAD=/path/to/libfaketime.so.1 \
  FAKETIME='%' FAKETIME_FOLLOW_FILE=/tmp/my-demo-file.tmp \
  FAKETIME_DONT_RESET=1 \
  bash -c 'while true ; do date ; sleep 1 ; done'

# now, while the above is running, change the file's timestamp
# (in a different terminal window or whatever)
touch -t 2002290123.45 /tmp/my-demo-file.tmp


Changing the 'x' modifier during run-time
-----------------------------------------

Using FAKETIME_TIMESTAMP_FILE allows for easily changing the FAKETIME setting
while a program is running:

echo "+0 x1" > /tmp/my-faketime.rc
LD_PRELOAD=libfaketime.so.1 FAKETIME_TIMESTAMP_FILE="/tmp/my-faketime.rc" \
  FAKETIME_NO_CACHE=1 ./some-program &
sleep 10 ; echo "+0 x10" > /tmp/my-faketime.rc

Changing the speed of the wall clock time, i.e., using a different 'x' modifier
during run-time, by default can lead to greater jumps that may confuse the
program. For example, if the program has been running for 10 seconds on 'x1',
and then the setting is changed to 'x10', the faked time will look to the
program as if it has been running for more than 100 instead of just more than
10 seconds.

By setting the environment variable FAKETIME_XRESET to any value, transitions
between different 'x' modifier values will be significantly smoothed:

LD_PRELOAD=libfaketime.so.1 FAKETIME_TIMESTAMP_FILE="/tmp/my-faketime.rc" \
  FAKETIME_NO_CACHE=1 FAKETIME_XRESET=1 ./some-program &

Setting FAKETIME_XRESET ensures that wall clock time begins to run faster
only after the 'x' modifier has been changed (when increasing it) and also
ensures that the reported faked time does not jump back to past values
(when decreasing it).

Please note that FAKETIME_XRESET internally works by resetting libfaketime's
internal time-keeping data structures, which may have side effects on reported
file timestamps. Using FAKETIME_XRESET should be considered experimental at
the moment.


Cleaning up shared memory
-------------------------

libfaketime uses semaphores and shared memory on platforms that support it in
order to sync faketime settings across parent-child processes.

Please note that this does not share the time set by settimeofday (for that
see FAKETIME_UPDATE_TIMESTAMP_FILE below).

libfaketime will clean up when it exits properly.
However, when processes are terminated (e.g., by Ctrl-C on command line),
shared memory cannot be cleaned up properly. In such
cases, you should occasionally delete

  /dev/shm/faketime_shm_* and
  /dev/shm/sem.faketime_sem_*

manually (or properly automated). Leftover files there from processes that
already have been terminated are not a problem in general, but result in a
libfaketime error the next time a process is started with a process id for
which such a stale semaphore/shared memory exists. Thus, if you run across
the following error message

  libfaketime: In ft_shm_create(), shm_open failed: File exists

please cleanup /dev/shm as described above. This is especially relevant
for long-running systems (servers with high uptime) and systems on which
a lot of processes are started (e.g., servers handling many containers
or similar virtualization mechanisms).


Intercepting time-setting calls
-------------------------------

libfaketime can be compiled with the CFLAG "-DFAKE_SETTIME" in order
to also intercept time-setting functions, i.e., clock_settime(),
settimeofday(), and adjtime(). The FAKETIME environment
variable will be adjusted on each call.

When the environment variable FAKETIME_TIMESTAMP_FILE is set, points to a
writeable (creatable) custom config file and the environment variable
FAKETIME_UPDATE_TIMESTAMP_FILE is "1", then the file also is updated on each
call. By this, a common "virtual time" can be shared by several
processes, where each can adjust the time for all.


Sharing "virtual settable time" between independent processes
-------------------------------------------------------------

When libfaketime was compiled with FAKETIME_COMPILE_CFLAGS="-DFAKE_SETTIME",
it can be configured to support a common time offset for multiple processes.
This for example allows to use "ntpdate" as normal user without affecting
system clock, interactively testing software with different dates or testing
complex software with multiple independent processes that themself use
settime internally.

Examples:

  $ export LD_PRELOAD=libfaketime.so.1
  $ export FAKETIME_TIMESTAMP_FILE="/tmp/my-faketime.rc"
  $ export FAKETIME_UPDATE_TIMESTAMP_FILE=1
  $ export FAKETIME_CACHE_DURATION=1 # in seconds
  # or: export FAKETIME_NO_CACHE=1

  $ date -s "1999-12-24 16:00:00"
  Fri Dec 24 16:00:00 CET 1999
  $ LD_PRELOAD="" date
  Thu Apr  9 15:19:38 CEST 2020
  $ date
  Fri Dec 24 16:00:02 CET 1999
  $ /usr/sbin/ntpdate -u clock.isc.org
   9 Apr 15:18:37 ntpdate[718]: step time server xx offset 640390517.057257 sec
  $ date
  Thu Apr  9 15:18:40 CEST 2020

In another terminal, script or environmment the same variables could be set
and the same time would be printed.
This also avoid the need to directly update the rc config file to use
different times, but of course only supports time offsets.

Please note that this feature is not compatible with several other features,
such as FAKETIME_FOLLOW_FILE, FAKETIME_XRESET and maybe others. After first
settime, offsets will be used in FAKETIME_TIMESTAMP_FILE, even if it
initially used advanced time specification options.


4f) Faking the date and time system-wide
----------------------------------------

David Burley of SourceForge, Inc. reported an interesting use case of applying
libfaketime system-wide: Currently, all virtual machines running inside
an OpenVZ host have the same system date and time. In order to use multiple
sandboxes with different system dates, the libfaketime library can be put into
/etc/ld.so.preload; it will then be applied to all commands and programs
automatically. This is of course best used with a system-wide /etc/faketimerc
file. Kudos to SourceForge, Inc. for providing the patch!

Caveat: If you run a virtual machine, its real-time clock might be reset to the
real world date & time when you reboot. Depending on your FAKETIME setting,
this may lead to side effects, such as forced file system checks on each reboot.
System-wide faked time may also lead to unexpected side effects with software
auto-update tools, if the offset between real world time and faked system time
is too large. If in doubt, set your system's date to the faked time and try out
whether everything still works as expected before applying libfaketime
system-wide.


4g) Using the "faketime" wrapper
--------------------------------

As of version 0.8, libfaketime provides a command named "faketime", which is
placed into /usr/bin by "make install". It spares the hassle of setting
the LD_PRELOAD and FAKETIME environment variables manually, but only exposes
a subset of libfaketime's functionality. On the other hand, it uses the date
interpretation function by /bin/date in order to provide higher flexibility
regarding the specification of the faked date and time. For example, you
can use

faketime 'last Friday 5 pm' /your/command/here

Of course, also absolute dates can be used, such as in

faketime '2018-12-24 08:15:42' /bin/date

Thanks to Daniel Kahn Gillmor for providing these suggestions!

Balint Reczey has rewritten the wrapper in 0.9.5 from a simple shell script
to an efficient wrapper application.


4h) "Limiting" libfaketime based on elapsed time or number of calls
-------------------------------------------------------------------

Starting with version 0.9, libfaketime can be configured to not be continuously
active, but only during a certain time interval.

For example, you might want to start a program with the real current time, but
after 5 minutes of usage, you might want it to see a faked time, e.g., a year
in the future.

Dynamic changes to the faked time are alternatively possible by

- changing the FAKETIME environment variable at run-time; this is the preferred
  way if you use libfaketime for debugging and testing as a programmer, as it
  gives you the most direct control of libfaketime without any performance
  penalties.

- not using the FAKETIME environment variable, but specifying the fake time in
  a file (such as ~/.faketimerc). You can change the content of this file at
  run-time. This works best with caching disabled, but comes at a performance
  cost because this file has to be read and evaluated each time.

The feature described here works based on two pairs of environment variables,

  FAKETIME_START_AFTER_SECONDS and FAKETIME_STOP_AFTER_SECONDS, and
  FAKETIME_START_AFTER_NUMCALLS and FAKETIME_STOP_AFTER_NUMCALLS

The default value for each of these environment variables is -1, which means
"ignore this value".

If you want libfaketime to be only active during the run-time minutes 2 to 5
of your application, you would set

  FAKETIME_START_AFTER_SECONDS=60
  FAKETIME_STOP_AFTER_SECONDS=300

This means that your application will work with the real time from start (second
0) up to second 60. It will then see a faked time from run-time seconds 60 to
300 (minutes 2, 3, 4, and 5). After run-time second 600, it will again see the
real (not-faked) time.

This approach is not as flexible as changing the FAKETIME environment variable
during runtime, but may be easier to use, works on a per-program (and not a
per-user or system-wide) scope, and has only a minor performance overhead.

Using the other pair of environment variables, you can limit the activity time
of libfaketime not based on wall-clock seconds, but on the number of
time-related function calls the started program performs. This alternative is
probably only suitable for programmers who either know the code of the program
in order to determine useful start/stop values or want to perform fuzzing
tests.

Both pairs of environment variables can be combined to further restrict
libfaketime activity, although this is only useful in very few scenarios.

Limiting libfaketime activity in this way is not recommended in general. Many
programs will break when they are subject to sudden changes in time, especially
if they are started using the current (real) time and are then sent back into
the past after, e.g., 5 minutes. For example, they may appear to be frozen or
stuck because they are waiting until a certain point in time that, however, is
never reached due to the delayed libfaketime activity. Avoid using this
functionality unless you are sure you really need it and know what you are
doing.


4i) "Limiting" libfaketime per process
--------------------------------------

faketime can be instructed to fake time related calls only for selected
commands or to fake time for each command except for a certain subset of
commands.

The environment variables are FAKETIME_ONLY_CMDS and FAKETIME_SKIP_CMDS
respectively.

Example:
    FAKETIME_ONLY_CMDS=javadoc faketime '2008-12-24 08:15:42' make
will run the "make" command but the time faking will only be applied
to javadoc processes.

Multiple commands are separated by commas.

Example:
    FAKETIME_SKIP_CMDS="javadoc,ctags" faketime '2008-12-24 08:15:42' make
will run the "make" command and apply time faking for everything "make"
does except for javadoc and ctags processes.

FAKETIME_ONLY_CMDS and FAKETIME_SKIP_CMDS are mutually exclusive, i.e.,
you cannot set them both at the same time. faketime will terminate with
an error message if both environment variables are set.


4j) Spawning an external process
--------------------------------

From version 0.9 on, libfaketime can execute a shell command once after a) an
arbitrary number of seconds has passed or b) a number of time-related system
calls has been made by the program since it started. This has two limitations
one needs to be aware of:

* Spawning the external process happens during a time-related system call
  of the original program. If you want the external process to be started
  5 seconds after program start, but this program does not make any time-
  related system calls before run-time second 8, the start of your external
  process will be delayed until run-time second 8.

* The original program is blocked until the external process is finished,
  because the intercepting time-related system call will not return earlier. If
  you need to start a long-running external process, make sure it forks into the
  background.

Spawning the external process is controlled using three environment variables:
FAKETIME_SPAWN_TARGET, FAKETIME_SPAWN_SECONDS, FAKETIME_SPAWN_NUMCALLS.

Example (using bash on Linux):

(... usual libfaketime setup here, setting LD_PRELOAD and FAKETIME ...)
export FAKETIME_SPAWN_TARGET="/bin/echo 'Hello world'"
export FAKETIME_SPAWN_SECONDS=5
/opt/local/bin/myprogram

This will run the "echo" command with the given parameter during the first
time-related system function call that "myprogram" performs after running for 5
seconds.


4k) Saving timestamps to file, loading them from file
-----------------------------------------------------

To store and load timestamp _offsets_ using _one and the same_ file allowing
to share a common "virtual time" between independent processes, please see
FAKETIME_UPDATE_TIMESTAMP_FILE above. The FAKETIME_SAVE_FILE feature is
different.

faketime can save faked timestamps to a file specified by FAKETIME_SAVE_FILE
environment variable. It can also use the file specified by FAKETIME_LOAD_FILE
to replay timestamps from it. After consuming the whole file, libfaketime
returns to using the rule set in FAKETIME variable, but the timestamp processes
will start counting from will be the last timestamp in the file.

The file stores each timestamp in a stream of saved_timestamp structs
without any metadata or padding:

/* Storage format for timestamps written to file. Big endian. */
struct saved_timestamp {
  int64_t sec;
  uint64_t nsec;
};

faketime needs to be run using the faketime wrapper to use these files. This
functionality has been added by Balint Reczey in v0.9.5.


4l) Replacing random numbers with deterministic number sequences
----------------------------------------------------------------

libfaketime can be compiled with the CFLAG FAKE_RANDOM set (see src/Makefile).

When compiled this way, libfaketime additionally intercepts calls to the
function getrandom(), which currently is Linux-specific.

This functionality is intended to feed a sequence of deterministic, repeatable
numbers to applications, which use getrandom(), instead of the random numbers
provided by /dev/[u]random.

For creating the deterministic number sequence, libfaketime internally
uses Bernard Widynski's Middle Square Weyl Sequence Random Number Generator,
see https://mswsrng.wixsite.com/rand.

It requires a 64-bit seed value, which has to be passed via the environment
variable FAKERANDOM_SEED, as in, for example

  LD_PRELOAD=src/libfaketime.so.1 \
  FAKERANDOM_SEED="0x12345678DEADBEEF" \
  test/getrandom_test

Whenever the same seed value is used, the same sequence of "random-looking"
numbers is generated.

Please be aware that this definitely breaks any security properties that
may be attributed to random numbers delivered by getrandom(), e.g., in the
context of cryptographic operations. Use it for deterministic testing
purposes only. Never use it in production.

For a discussion on why this apparently not date-/time-related function
has been added to libfaketime and how it may evolve, see Github issue #275.


5. License
----------

libfaketime has been released under the GNU General Public License, GPL.
Please see the included COPYING file.


6. Contact
-----------

Bug reports, feature suggestions, success reports, and patches/pull
requests are highly appreciated:

            https://github.com/wolfcw/libfaketime