api/docs/deployment.dox

/* ******************************************************************************
 * Copyright (c) 2010-2025 Google, Inc.  All rights reserved.
 * Copyright (c) 2011 Massachusetts Institute of Technology  All rights reserved.
 * Copyright (c) 2007-2010 VMware, Inc.  All rights reserved.
 * ******************************************************************************/

/*
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 * * Redistributions of source code must retain the above copyright notice,
 *   this list of conditions and the following disclaimer.
 *
 * * Redistributions in binary form must reproduce the above copyright notice,
 *   this list of conditions and the following disclaimer in the documentation
 *   and/or other materials provided with the distribution.
 *
 * * Neither the name of VMware, Inc. nor the names of its contributors may be
 *   used to endorse or promote products derived from this software without
 *   specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED. IN NO EVENT SHALL VMWARE, INC. OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
 * DAMAGE.
 */

/**

***************************************************************************
\page page_build_client How to Build a Tool

To use the DynamoRIO API, a tool, or "client" of DynamoRIO, should
include the main DynamoRIO header file:

\code
#include "dr_api.h"
\endcode

The client's target operating system and architecture must
be specified by setting pre-processor defines before including the
DynamoRIO header files.  The appropriate library must then be linked
with.  The define choices are:

-# \p WINDOWS, \p LINUX, or (coming soon) \p MACOS
-# \p X86_32, \p X86_64, \p ARM_32, or \p ARM_64

Currently we provide a private loader for both Windows and Linux.
With private loading, clients use a separate copy of each library
from any copy used by the application.

If the private loader is deliberately disabled, for transparency reasons
(see \ref transparency), clients should be
self-contained and should not share libraries with the application.
Without the private loader, 64-bit clients must take care to try and load
themselves within reachable range of DynamoRIO's code caches by setting a
preferred base address, although this may not always be honored by the
system loader.

The DynamoRIO release supplies <a href="http://www.cmake.org">CMake</a>
configuration files to facilitate building clients with the proper
compiler and linker flags.  CMake is a cross-platform build system that
generates Makefiles or other development system project files.  A \p
DynamoRIOConfig.cmake configuration file, along with supporting files, is
distributed in the \p cmake/ directory.

In its \p CMakeLists.txt file, a client should first invoke a \p
find_package(DynamoRIO) command.  This can optionally take a version
parameter.  This adds DynamoRIO as an imported target.  If found, the
client should then invoke the \p configure_DynamoRIO_client() function in
order to configure build settings.  Here is an example:

\code
add_library(myclient SHARED myclient.c)
find_package(DynamoRIO)
if (NOT DynamoRIO_FOUND)
  message(FATAL_ERROR "DynamoRIO package required to build")
endif(NOT DynamoRIO_FOUND)
configure_DynamoRIO_client(myclient)
\endcode

The \p samples/CMakeLists.txt file in the release package serves as another
example.  The top of \p DynamoRIOConfig.cmake contains detailed
instructions as well.

If your client uses DynamoRIO extensions like drmgr or drx be sure to call
use_DynamoRIO_extension() in your client's CMakeLists.txt file.

When configuring, the \p DynamoRIO_DIR CMake variable can be passed in to
identify the directory that contains the \p DynamoRIOConfig.cmake file.  For
example:

\code
mkdir ../build
cd ../build
cmake -DDynamoRIO_DIR=$DYNAMORIO_HOME/cmake ../myclient
make
\endcode

The compiler needs to be configured prior to invoking cmake.  If using gcc
with a non-default target platform, the \p CFLAGS and \p CXXFLAGS
environment variables should be set prior to invoking cmake.  For example,
to configure a 32-bit client when gcc's default is 64-bit:

\code
mkdir ../build
cd ../build
CFLAGS=-m32 cmake -DDynamoRIO_DIR=$DYNAMORIO_HOME/cmake ../myclient
make
\endcode

Note that \p CXXFLAGS should be set instead for a C++ client, and both should
be set when building both types of clients from the same configuration
(e.g., \p samples/CMakeLists.txt).

To improve clean call performance (see \ref sec_clean_call and \ref
op_cleancall "-opt_cleancall"), we recommend high levels of optimization
when building a client.

If a client is not using CMake, the appropriate compiler and linker flags
can be gleaned from \p DynamoRIOConfig.cmake.  One method is to invoke CMake to
generate a Makefile and then build with \p VERBOSE=1.  We also summarize
here the key flags required for 32-bit clients for \p gcc:

\code
gcc -fPIC -shared -lgcc -DLINUX -DX86_32 -I$DYNAMORIO_HOME/include my-client.c
\endcode

And for \p cl:

\code
cl my-client.c /I$DYNAMORIO_HOME/include /GS- /DWINDOWS /DX86_32
   /link /libpath:$DYNAMORIO_HOME/bin dynamorio.lib /dll /out:my-client.dll
\endcode

For a 64-bit client with \p cl:

\code
cl my-client.c /I$DYNAMORIO_HOME/include /GS- /DWINDOWS /DX86_64
   /link /libpath:$DYNAMORIO_HOME/bin dynamorio.lib /dll /out:my-client.dll
   /base:0x72000000 /fixed
\endcode

For 64-bit Linux clients, setting the preferred base takes several steps.
Refer to \p DynamoRIOConfig.cmake for details.

To make clean call sequences more likely to be optimized, it is recommended
to compile the client with optimizations, \p -O2 for gcc or \p /O2 for cl.



***************************************************************************
\page page_deploy How to Run

Once the DynamoRIO distribution contents are unpacked (see \ref
sec_package), configuration and execution of applications under DynamoRIO
is handled by a set of libraries and tools.  On Windows, the
tools are \c drconfig.exe, \c drrun.exe, and \c drinject.exe.  The
corresponding libraries (whose APIs are exposed by the tools) are \c
drconfiglib.dll and \c drinjectlib.dll with header files \c dr_config.h and
\c dr_inject.h.  On Linux, the tools are named \c drconfig, \c drrun, and \c
drinject, and the libraries are \c libdrconfiglib.a and \c
libdrinjectlib.a.

When using DynamoRIO as a third-party disassembly library (see \ref
page_standalone), no deployment is needed, as DynamoRIO does not control a
target application when used as a regular library.

\section win_deploy Windows Deployment

There are two methods for running a process under DynamoRIO: the one-time
configure-and-run, and the two-step separate configuration and execution.
The \c drrun.exe tool supports the first, simpler model, while the \c
drconfig.exe and \c drinject.exe tools support the second, more powerful
model.  The \c drconfig.exe tool, or the corresponding the \c
drconfiglib.dll library, can also be used to \ref sec_comm "nudge" running
processes.

Configuration information is stored in files in the current user's profile
directory, which is obtained from the environment variable \c USERPROFILE.
Thus, configurations are persistent across reboots and are private to each
user.  If the <tt>DYNAMORIO_CONFIGDIR</tt> environment variable is set,
its value is used instead of \c USERPROFILE.
If neither is set, a temp directory will be used when creating new
configuration files for configure-and-run execution.

DynamoRIO also supports global configurations, which are stored in
the "config" subdirectory of the directory specified by the \c
DYNAMORIO_HOME registry value in the registry key \c
\\HKLM\\SOFTWARE\\DynamoRIO\\DynamoRIO (or for 32-bit on 64-bit Windows
(WOW64) \c \\HKLM\\SOFTWARE\\Wow6432Node\\DynamoRIO\\DynamoRIO).  Setting
that \c DYNAMORIO_HOME value and creating the directory it points to must
be done manually.  The provided tools support reading and writing both
local and global configuration files, and automatically creating the local
directory.  DynamoRIO gives local files precedence when both exist.  Note
that applications that do not have a \c USEPROFILE environment variable can
be controlled using <tt>DYNAMORIO_CONFIGDIR</tt> or global configurations.
Also note that by default \c USERPROFILE is not set over cygwin ssh and
must be explicitly set in the shell startup files.

Configurations are per-process, with the basename of the process used for
identification (e.g., \c notepad.exe).  One-time configuration also uses the
process id to specify that the configuration is for that process instance
only.

As an example, assume you have unpacked the DynamoRIO distribution and
your current directory is its base directory.  Run \c notepad.exe with the
bbsize sample client using the following configure-and-run command:
\code
bin32/drrun.exe -c samples/bin32/bbsize.dll -- notepad
\endcode

Alternatively, you can first run \c notepad.exe, and then use \c drrun
to attach to it, but please note that attaching is an experimental
feature and is not yet as well-supported as launching a new process:
\code
bin32/drrun.exe -attach <notepad_pid> -c samples/bin32/bbsize.dll
\endcode

Then, you can also detach DynamoRIO from the process of \c notepad.exe
without affecting the normal execution of \c notepad.exe.
\code
bin32/drconfig.exe -detach <notepad_pid>
\endcode

To use system-wide injection, allowing for an application to be run
under DynamoRIO regardless of how it is invoked, configure the application
first (-syswide_on requires administrative privileges):
\code
bin32/drconfig.exe -reg notepad.exe -syswide_on -c samples/bin32/bbsize.dll
\endcode

The next time \c notepad.exe is started by the current user, it will run under
DynamoRIO with the bbsize client.

To unregister \c notepad.exe, issue the following command:
\code
bin32/drconfig.exe -unreg notepad.exe
\endcode

Invoke any of the \c drconfig.exe, \c drrun.exe, or \c drinject.exe tools
with no arguments to see the full list of options available.

By default, DynamoRIO follows into all child processes, with the parent's
settings inherited by the child if there is no configuration set up ahead
of time for the child application.  To instead only
follow children that are configured (via \c drconfig.exe), use the
\ref op_children "-no_follow_children" runtime option.

To ensure a client is loaded into a child process of a different
bitwidth (i.e., a 32-bit child created by a 64-bit parent), use the \c
-c32 and \c -c64 options to \c drconfig or \c drrun, with \c -- ending
the first client's options:

bin32/drrun.exe -c32 samples/bin32/bbsize.dll -- -c64 samples/bin64/bbsize.dll -- notepad

The order matters: \c -c32 must come first.

To \ref sec_comm "nudge" all instances of \c notepad.exe running under
DynamoRIO with argument "5", use:
\code
bin32/drconfig.exe -nudge notepad.exe 0 5
\endcode
This will result in a nudge event with argument=5 delivered to the
client callback registered with dr_register_nudge_event() in all
\c notepad.exe processes running under DynamoRIO.  The third argument,
0, is an ID supplied at registration which uniquely identifies the
target client (see dr_deploy.h for details).

To view 32-bit or WOW64 processes running under DynamoRIO the
\c drview.exe tool can be used.  The bin64 version will display both 32-bit
and 64-bit processes and will indicate which are 32-bit.  The bin32 version
will display 64-bit processes but is unable to determine whether DynamoRIO
is present.

\attention
Note that on Windows NT a reboot is required after using -syswide_on or -syswide_off.

DynamoRIO uses the
<tt>\\HKLM\\SOFTWARE\\Microsoft\\Windows\\Windows NT\\CurrentVersion\\AppInit_DLLs</tt>
key
(for 32-bit on 64-bit Windows (WOW64),
<tt>\\HKLM\\SOFTWARE\\Wow6432Node\\Microsoft\\Windows NT\\CurrentVersion\\Windows\\AppInit_DLLs</tt>)
for -syswide_on to inject into new processes without having to directly
launch them \c drrun.exe or \c drinject.exe.  For injection to work, the
registered process must statically link to user32.dll (only a few small
non-graphical windows applications don't link user32.dll).  If a target
application does not link to user32.dll, DynamoRIO can still inject if the
process is launched with \c drinject.exe or if the parent process (usually
cmd.exe or explorer.exe for user launched processes) is running under
DynamoRIO.  The drinject.exe tool uses the configuration information set
by \c drconfig.exe for the target application.

\attention
The -syswide_on, -syswide_off, use of global configuration files, and
nudging certain processes may require administrative privileges.  On
Windows Vista or higher, if UAC is enabled, use an elevated (runas admin)
process. When using \c drconfig.exe and \c drrun.exe in these scenarios, be
sure that the cmd shell being used was started with elevated permissions.

An alternative method to run an application under DynamoRIO is the \ref
sec_startstop "app_start()/app_stop()" interface, which requires modifying
application source code.

\section lin_deploy Linux Deployment

Once DynamoRIO has been unpacked, the same set of helper binaries as on
Windows provide flexibility in configuring and executing applications.

There are two methods for invoking an application under DynamoRIO:
 -# Configure and launch in one step via \p drrun
 -# Configure via \p drconfig and launch via \p drinject

As an example of the simpler method, the following command runs \c ls
under DynamoRIO with the bbsize sample client:
\code
% bin64/drrun -c samples/bin64/libbbsize.so -- ls
\endcode

Alternatively, you can first run the target, and then use \c drrun
to attach to it, but please note that attaching is an experimental
feature and is not yet as well-supported as launching a new process.
In particular, if the application is in the middle of a blocking syscall,
DynamoRIO will wait for that to finish.  To instead force interruption of
the syscall, additionally pass -skip_syscall.
\code
% bin64/drrun -attach <target_pid> -c samples/bin64/libbbsize.so
\endcode

This attach feature requires ptrace capabilities, which can be enabled
with this command:
\code
% echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
\endcode

Then, you can also detach DynamoRIO from the target process
without affecting the normal execution of the application.
\code
% bin64/drconfig -detach <target_pid>
\endcode

Run \c drrun with no options to get a list of the options and
environment variable shortcuts it supports.  To disable following across
child execve calls, use the \ref op_children "-no_follow_children" runtime
option.

Use the tools in \c bin32/ for 32-bit applications and the tools in \c
bin64/ for 64-bit applications.

The two-step method allows for greater control over child processes.  The
\p drconfig tool writes a configuration file for a given application
name.  DynamoRIO reads its options from the configuration file at runtime.
Once each process name is configured, the \p drinject tool can be used to
invoke the parent process.  The \p drrun tool can also be used but it
creates a temporary configuration file that will override settings
requested via \p drconfig.  The configuration file for each application is
stored in <tt>$DYNAMORIO_CONFIGDIR/.dynamorio/&lt;appname&gt;.config32</tt>
(or a \p config64 suffix for 64-bit). If <tt>DYNAMORIO_CONFIGDIR</tt> is
not set, <tt>$HOME/.dynamorio/&lt;appname&gt;.config32</tt> is used;
if neither is set, a temp directory will be used when creating new
configuration files for configure-and-run execution.  On Android, if
neither <tt>/data/local/tmp</tt> nor the current working directory are
writable, you will need to specify a writable directory by setting the
<tt>DYNAMORIO_CONFIGDIR</tt> environment variable.

DynamoRIO also supports global configuration
files in <tt>/etc/dynamorio/&lt;appname&gt;.config32</tt> when a local
configuration file is not found.  \p drconfig does not support directly
writing a global config file but such files can be copied from or modeled
on local files.

If a target application executes an \c execve that discards the \c HOME
environment variable, the resulting process will still run under DynamoRIO
control with the same settings as the parent process.
Use <tt>DYNAMORIO_CONFIGDIR</tt> or global configuration files to
specify separate options for such a child process.

When running scripts it is best to explicitly invoke the interpreter rather
than invoking the script directly:

\code
% bin64/drrun -- /bin/bash myscript.sh
\endcode

To \ref sec_comm "nudge" a process with pid \c targetpid running under
DynamoRIO and pass argument "5" to the nudge callback, use the \c
drnudgeunix tool:
\code
bin64/drnudgeunix -pid targetpid -client 0 5
\endcode
This will result in a nudge event with argument=5 delivered to the
client callback registered with dr_register_nudge_event() in the
target process.  The 0 argument is an ID supplied at registration which
uniquely identifies the target client (see dr_deploy.h for details).
If you used the -c argument to drrun or drconfig to register the client, the
client's id defaults to 0.

An alternative method to run an application under DynamoRIO is the \ref
sec_startstop "app_start()/app_stop()" interface, which requires modifying
application source code.


\section android_deploy Android Deployment

Android deployment is generally the same as \ref lin_deploy except for the
following differences.

For pure native applications, the default configuration file location (if
<tt>DYNAMORIO_CONFIGDIR</tt> is not explicitly set) is usually (depending
on whether <tt>$HOME</tt> happens to be writable) <tt>/data/local/tmp</tt>.

For Android applications on recent versions of Android, SELinux prevents
writing to <tt>/data/local/tmp</tt>.  The application's data directory
should be pointed at in the \p TMPDIR or <tt>DYNAMORIO_CONFIGDIR</tt>
environment variables, as shown below in the example wrapper script.  We
recommend using \p TMPDIR as its value will also be used by tools such as
Dr. Memory for their log files.

To launch an Android application under DynamoRIO, use a wrapper script and
point at the script via the \p logwrapper property set on your
application's name prefixed by \p <tt>wrap.</tt>.  For example, if your
application's name is <tt>com.myco.appname</tt>, set the property for
<tt>wrap.com.myco.appname</tt>, truncating to 31 characters:

\code
setprop wrap.com.myco.appname "logwrapper /system/xbin/wrap.sh"
\endcode

The wrapper shell script should contain the command line prefix you wish to
use to launch your application under DynamoRIO:

\code
#!/system/bin/sh
export TMPDIR=/data/data/com.myco.appname
exec /system/xbin/dynamorio/bin32/drrun -- $@
\endcode

Be sure to place the DynamoRIO binaries and the wrapper script in an
executable location, such as <tt>/system/xbin</tt>.  Alternatively, disable
SELinux via <tt>setenforce 0</tt>.

If you run into problems with configuration files being created due to
SELinux denials (look in the logs for such denials), it may be simplest to
disable SELinux via <tt>setenforce 0</tt>.  We have attempted to get
everything to work without this step, but we are not able to test on all
versions or configurations of Android.


\section qemu_deploy Running Under QEMU

To run DynamoRIO under QEMU, use the \ref op_xarch_root "-xarch_root"
option to point at the base directory of the guest system libraries.
This is the same path passed to QEMU's `-L` option:

```
$ qemu-arm -L /usr/arm-linux-gnueabihf bin32/drrun -xarch_root /usr/arm-linux-gnueabihf -- hello_world
```

Be sure to install both the `qemu-user` and `qemu-user-binfmt`
packages in order for QEMU to impose itself across an execve:

```
$ sudo apt-get install qemu-user qemu-user-binfmt
```

If using a custom build of QEMU, be sure to update the binfmt rule to
ensure the proper build is used across execve: QEMU does *not* ensure
that the same build is preserved and instead relies on the global
binfmt rule.  On an Ubuntu system, edit `/usr/share/binfmts/qemu-arm`
and replace the `interpreter` path.  Then run:

```
$ sudo update-binfmts --import /usr/share/binfmts/qemu-arm
```

Confirm the change took effect:
```
$ update-binfmts --find bin32/drrun
```

On Fedora, edit `/usr/lib/binfmt.d/qemu-arm-dynamic.conf` and then run:
```
$ sudo systemctl restart systemd-binfmt.service
```


\section client_ops Passing Options to Clients

All of the earlier examples did not need to pass any arguments to the client.
When using the -c argument to set the client, all arguments between the client
path and the double dash are passed to the client.  When using the -client
argument to drrun, the third argument following -client is passed through to
the client.  For example, all these invocations of drrun pass '-op1
-op2 "value with spaces"' to the client:

\code
bin32/drrun.exe -c libmyclient.dll -op1 -op2 \"value with spaces\" -- notepad
bin32/drrun.exe -client myclient.dll 0 '-op1 -op2 "value with spaces"' -- notepad
\endcode

On Linux:

\code
bin32/drrun -c libmyclient.so -op1 -op2 \"value with spaces\" -- ls
bin32/drrun -client libmyclient.so 0 '-op1 -op2 "value with spaces"' -- ls
\endcode

When using a two-step model, the options are passed to \p drconfig:

\code
bin32/drconfig.exe -reg notepad.exe -c myclient.dll -op1 -op2 \"value with spaces\"
bin32/drconfig.exe -reg notepad.exe -client myclient.dll 0 '-op1 -op2 "value with spaces"'
\endcode

The client's options are passed directly to its \p dr_client_main()
initialization routine, in the same manner as arguments are passed to a
regular application's \p main() routine.  To match standalone application
conventions, \p argv[0] is set to the client library path, with the actual
parameters starting at index 1.  The client can also optionally call
dr_get_option_array() to retrieve the options passed to it.  C++ clients
can use the convenience of the \ref page_droption.

Client options are not allowed to contain semicolons.  Additionally, the
client option string combined with the path to the client library cannot
contain all three quote characters (\', \", \`) simultaneously.

To ensure a client is loaded into a child process of a different
bitwidth (i.e., a 32-bit child created by a 64-bit parent), use the \c
-c32 and \c -c64 options to \c drconfig or \c drrun, with \c -- ending
the first client's options:

bin32/drrun -c32 samples/bin32/bbsize.dll -- -c64 samples/bin64/bbsize.dll -- notepad

The order matters: \c -c32 must come first.

\section multi_client Multiple Clients

DynamoRIO does support multiple clients.  It is each client's
responsibility, however, to ensure compatibility with other clients.
DynamoRIO makes no attempt to force cooperation among clients.  For
example, instruction stream modifcations made by one client are
visible to other clients.  Systems employing multiple clients must
be aware of such interactions and design accordingly.

Client registration requires users to specify the \em priority of each
client.  DynamoRIO calls each client's
dr_client_main() routine sequentially according to this priority.  Clients
with a numerically lower priority value are called first and therefore
given the first opportunity to register callbacks (the client with
priority 0 is called first).  Since DynamoRIO delivers event
callbacks sequentially, client priority and the order of event
registration is important.  For a given event, the \em first
registered callback is called \em last.  This scheme gives precedence
to the first registered callback since that callback is given the final
opportunity to modify the instruction stream or influence DynamoRIO's
operation.


\section tool_frontend End-User Tools

A client can be packaged up with DynamoRIO to create an end-user tool.  For
many tools, a separate front-end executable is not necessary, and \p drrun
is sufficient.  Using \p drrun for a tool is made simpler by the \p -t
option.  To use the option, first create a file in the \p tools
subdirectory of the root of the DynamoRIO installation called \p
toolname.drrun32 or \p toolname.drrun64, depending on the target
architecture.  Here, \p toolname is the desired external name of the tool.
This file should contain one of the following lines, or two if
they are a pair of CLIENT32_* and CLIENT64_* specifiers:

\code
CLIENT_ABS=/absolute/path/to/client
CLIENT32_ABS=/absolute/path/to/32-bit-client
CLIENT64_ABS=/absolute/path/to/64-bit-client
\endcode

or

\code
CLIENT_REL=relative/path/to/client/from/DynamoRIO/root
CLIENT32_REL=relative/path/to/32-bit-client/from/DynamoRIO/root
CLIENT64_REL=relative/path/to/64-bit-client/from/DynamoRIO/root
\endcode

This enables \p drrun to locate the tool's client library.  The 32 and
64 specifiers allow pointing at alternate-bitwidth paths for use if
the target application creates a child process of a different bitwidth.

The file can also modify the default DynamoRIO runtime options (see \ref
sec_options) via \p DR_OP= lines.  Each line contains only one option string
token.  For example:

\code
DR_OP=-persist
DR_OP=-persist_dir
DR_OP=c:\\path with spaces\\subdir
\endcode

Tool options can also be specified, but normally the defaults should be set
up appropriately in the client itself:

\code
TOOL_OP=-custom_op1
TOOL_OP=-custom_op2
\endcode

Lines beginning with \p # are considered comments.

When \p drrun is passed the option string <tt>-t toolname</tt>, it looks
for <tt>tools/toolname.drrun64</tt> or <tt>tools/toolname.drrun32</tt> and
reads the file to determine the client library to use and the default
DynamoRIO options.  This makes for a simpler launching command, rather than
the end user needing to name the exact location of the client library.  For
example, this command:

\code
bin64/drrun -t mytool -tool_option1 -tool_option2 -- myapp
\endcode

can be made to expand to this equivalent command:

\code
bin64/drrun -mytool_dr_option1 -mytool_dr_option2 -c tools/mytool/libmytool.so -tool_option1 -tool_option2 -- myapp
\endcode

For more extensive actions on launching the tool, a custom front-end
executable can be created that replaces \p drrun by using \p drinjectlib,
\p drconfiglib, and \p drfrontendlib.  These three libraries facilitate
creating cross-platform tools for configuring and launching applications
under DynamoRIO.  For more information about the interfaces they provide,
see their header files: dr_inject.h, dr_config.h, dr_frontend.h.

A custom front-end executable can be invoked via a \p drrun \p -t
configuration file using one of the following lines:

\code
FRONTEND_ABS=/absolute/path/to/front-end
\endcode

or

\code
FRONTEND_REL=relative/path/to/front-end/from/DynamoRIO/root
\endcode

This will cause \p drrun to transfer control to the specified front-end
executable, passing any tool arguments (including client paths, if \p
CLIENT{,32,64}_{ABS,REL} appears after the \p FRONTEND_* command)
followed by "--" and the target application command line.

The path to the DynamoRIO install base can be included in the front-end
options via this line

\code
TOOL_OP_DR_PATH
\endcode

The DynamoRIO runtime options can be included in a single token, preceded by a prefix,
via this line, using "-dr_ops" as an example prefix:

\code
TOOL_OP_DR_BUNDLE=-dr_ops
\endcode

A warning message can be presented up front to the user with:

\code
USER_NOTICE=This tool is currently experimental.  Please report issues to mytool.com/issues.
\endcode


\section sec_startstop Running a Subset of an Application

An alternative to running an entire application under DynamoRIO control is to
use the Application Interface to specify a portion of the application to
run.  This interface consists of the following routines:

\code
dr_app_setup()
dr_app_start()
dr_app_stop()
dr_app_cleanup()
dr_app_take_over()
dr_app_setup_and_start()
dr_app_stop_and_cleanup()
dr_app_stop_and_cleanup_with_stats()
dr_app_running_under_dynamorio()
\endcode

When building an executable that uses DynamoRIO's Application Interface,
follow the steps for \ref page_build_client to include the header files and link
with the DynamoRIO library, but omit the linker flags requesting no
standard libraries or startup files.  DynamoRIO's CMake support does this
automatically, as the linker flags for shared libraries are separate from
those for executables.


\section sec_static_DR Statically Linking DynamoRIO

Limited support is provided to statically link DynamoRIO and clients
into the application itself.  This generally requires the start-stop
interface (see \ref sec_startstop) in order to trigger takeover.

The configure_DynamoRIO_static and use_DynamoRIO_static_client CMake
utilities are provided to help in setting up the link steps.

One significant downside of statically linking is that DynamoRIO is
not able to provide library isolation in this mode.  This means that
clients cannot safely use third-party library code at runtime.  It is
considered safe to use libraries during process-wide initialization,
as that occurs in just one thread which is at a relatively safe point
having called something like dr_app_start().  The
#DR_DISALLOW_UNSAFE_STATIC and dr_allow_unsafe_static_behavior()
features are available to help detect violations of this rule, but
they only detect these violations when using DynamoRIO in its dynamic
library form.  Thus, the recommendation is to support both modes and
employ regression tests which set #DR_DISALLOW_UNSAFE_STATIC in
dynamic mode in order to catch unsafe library code being added later.

***************************************************************************
\section sec_options DynamoRIO Runtime Options

DynamoRIO's behavior can be fine-tuned using runtime parameters.  Options
are specified via \c drconfig, \c drrun, or dr_register_process().  See
\ref page_deploy.

 - \b -no_follow_children: \anchor op_children
   By default, DynamoRIO follows all child processes.  When this option
   is disabled via \p -no_follow_children, DynamoRIO follows only
   into child processes for which a configuration file exists (typically
   created by \c drconfig; see \ref page_deploy).  On Linux,
   forked children are always followed and this option only affects execve.

   To follow all children in general but exclude certain children, leave \p
   -follow_children on (the default) and create config files that exclude
   the desired applications by running \c drconfig with the \c -norun
   option.

 - \b -xarch_root: \anchor op_xarch_root
   Support for running under an emulator, in particular QEMU.  This option takes
   a string pointing at the base directory for guest system libraries, which DynamoRIO
   uses to prefix the application executable's interpreter, client library paths,
   and SYS_openat.  The option also enables workarounds for problems with QEMU's
   threads which would otherwise cause a hang while DynamoRIO tries to take them over.

 - \b -opt_memory: \anchor op_memory
   Reduce memory usage, but potentially at the cost of performance.  This
   option can result in memory savings as high as 20%, and usually incurs
   no noticable performance degradation.  However, it conflicts with the
   \ref op_enable "-enable_full_api option" and cannot be used with
   dr_unlink_flush_region().

 - \b -opt_cleancall \e \<number\>: \anchor op_cleancall
   Optimize (shrink or inline) the clean call sequences (see \ref sec_clean_call).
   When DynamoRIO analyzes the callee and optimizes each clean call invocation,
   it assumes that a client will not modify the clean call callee or application
   instructions after the inserted clean call.
   If a client changes application instructions after an inserted clean call,
   the client may need to reduce the -opt_cleancall level to preserve correct
   execution.
   There are four optimization levels:
   - 0: no optimization.
   - 1: callee register usage analysis and optimization on context switch.
   - 2: simple callee inline optimization, callee-save register analysis,
        and aflags usage analysis on the instruction list to be inserted.
   - 3: more aggressive, but potentially unsafe, optimizations.
   By default, the clean call optimization level is 2.
   The clean call will only be optimized if it is a leaf function.
   Currently, the callee will be inlined only if it is small, has at most
   one argument, and has no control flow other than for the PIC base.
   Compiling the client with optimizations makes clean call sequences more likely
   to be optimized.
   The optimization results (e.g. whether the inserted clean call is inlined or not,
   and which registers were saved on each context switch) are logged.
   Users can run DynamoRIO debug build with the runtime option
   "-loglevel 2 -logmask 0x02000000" (the logmask is optional but reduces the
   logfile size significantly) and grep for
   "CLEANCALL" in the log file to retrieve the information
   about clean call optimization.

 - \b -opt_speed: \anchor op_speed
   By default, DynamoRIO provides a more straightforward code stream to
   clients in lieu of performance optimizations.  This option attempts
   to obtain higher performance with potential loss of client simplicity.
   In particular, unconditional branches (both jumps and calls) and in some
   cases indirect calls may be elided in basic blocks.  See also \ref sec_limit_perf.
   Note that dr_insert_mbr_instrumentation() is not supported when -opt_speed
   is specified.

 - \b -stack_size \e \<number\>: \anchor op_stack_size
   DynamoRIO's per-thread stack is limited to 56KB by default on systems
   with a 4K page size (this may seem small, but this is much larger than its size
   when no client is present). For systems with different page sizes, it defaults
   to 2 * page size. This parameter can be used to increase the size;
   however, larger stack sizes use significantly more memory when targeting
   applications with hundreds of threads.  The parameter can take a 'K'
   suffix, and must be a multiple of the page size (4K, 16K, or 64K depending on
   the system).  This stack is used by the routines dr_insert_clean_call(),
   dr_swap_to_clean_stack(), dr_prepare_for_call(),
   dr_insert_call_instrumentation(), dr_insert_mbr_instrumentation(),
   dr_insert_cbr_instrumentation(), and dr_insert_ubr_instrumentation().
   The stack is started fresh for each use, so <em>no persistent state may be
   stored on it</em>.
 - \b -signal_stack_size \e \<number\>:
   DynamoRIO uses an alternate signal handling stack which is smaller than
   the regular stack by default, as signal handlers are called in
   relatively fragile contexts and cannot safely use arbitrary library
   routines: it is 32K for systems with a 4K page size. For systems with
   different page sizes, it defaults to 2 * page size. This parameter can
   be used to change the size.  The parameter can take a 'K' suffix, and
   must be a multiple of the page size (4K, 16K, or 64K depending on the
   system).
 - \b -thread_private: \anchor op_thread_priv
   By default, DynamoRIO's code caches are shared across threads.  This
   option requests code caches that are private to each thread.  For
   applications with many threads, thread-private code caches use more
   memory.  However, they can be more efficient, particularly when
   inserting thread-specific instrumentation.

 - \b -disable_traces:
   By default, DynamoRIO builds both a <em>basic block</em> code cache and
   a <em>trace</em> code cache (see \ref sec_IR).  This option disables
   trace building, which can have a negative performance impact.
   When running large, short-running applications, however, disabling
   traces can improve performance.
   When traces are disabled, dr_register_trace_event() has no effect.
   DynamoRIO tries to keep traces transparent to a client who is
   interested in all code and not only hot code, so there is rarely a
   reason to disable traces.
   \if internal_comment
     if we expose -enable_traces, note that it must be specified BEFORE
     -thread_private as today it turns on -shared_traces
   \endif

 - \b -enable_full_api: \anchor op_enable
   DynamoRIO's default internal options balance performance with API
   usability.  A few API functions, such as dr_unlink_flush_region(),
   are incompatible with this default mode.  Client users can gain
   access to the entire set of API functions with -enable_full_api.
   Note that this option may result in a small performance degradation.

 - \b -reachable_heap: \anchor op_reachable_heap
   By default, DynamoRIO guarantees that heap allocated directly through
   its API routines dr_thread_alloc(), dr_global_alloc(),
   dr_nonheap_alloc(), or dr_custom_alloc() with #DR_ALLOC_CACHE_REACHABLE
   is reachable by a 32-bit displacement from the code cache.  However, it
   does not guarantee that memory allocated through system libraries
   (including malloc, operator new, and HeapAlloc) or DynamoRIO's own
   internal memory is reachable.  Turning this option on combines all of
   the heap memory such that it is all guaranteed to be reachable from the
   code cache, at the risk of running out memory due to the smaller space
   of available memory.

 - \b -reachable_client: \anchor op_reachable_client
   By default, DynamoRIO guarantees that client libraries are reachable
   from its code caches by a 32-bit displacement (except for clients
   statically linked into the application).  Disabling this option
   removes that guarantee and allows a client to be located elsewhere.
   The option is automatically turned off when DynamoRIO is statically
   linked into the application.

 - \b -max_bb_instrs:
   DynamoRIO stops building a basic block if it hits this application
   instruction count limit before hitting control flow or other block
   termination conditions.  The default value is 1024; lower it if
   extensive client instrumentation is running into code cache size
   limit asserts.

 - \b -max_trace_bbs:
   DynamoRIO will not build a trace with larger than this number of
   constituent basic block.  The default value is 128; lower it if
   extensive client instrumentation is running into code cache size
   limit asserts.

 - \b -sysenter_is_int80: \anchor op_sysenter
   This option only applies to Linux.  If sysenter is the system call
   gateway, DynamoRIO normally hooks the vsyscall vdso page when it can.
   This option requests that DynamoRIO convert sysenter into int 0x80
   instead.  See \ref sec_extlibs.

 - \b -disable_rseq: \anchor op_disable_rseq
   This option only applies to Linux.  It returns -ENOSYS from the SYS_rseq
   system call, forcing applications to fall back to code that does not
   use restartable sequences.  See \ref sec_rseq for more information.

 - \b -no_ldstex2cas: \anchor op_ldstex2cas
   This option only applies to ARM and AArch64.  It disables the conversion of
   exclusive monitor regions to compare-and-swap operations.  That conversion can
   cause behavior changes in applications, but disabling it can result in failure to
   add instrumentation.  See \ref sec_ldrex for more information.

 - \b -multi_thread_exit:
   By default, DynamoRIO synchronizes with all remaining threads
   at process exit time and the process exit event executes with only
   one live thread.  This option requests that in release build the
   synchronization be avoided.  The process exit event must be written
   in a thread-safe manner.  Note that if thread exit events are
   registered, to avoid the synchronization the -skip_thread_exit_at_exit
   option must also be set.  These options can also be enabled
   programmatically via dr_set_process_exit_behavior().

 - \b -skip_thread_exit_at_exit:
   By default, DynamoRIO synchronizes with all remaining threads at
   process exit time in order to safely call each thread exit event.  This
   option requests that in release build the synchronization be avoided by
   removing the invocation of thread exit events at process exit time.
   Note that if the process exit event is registered, to avoid the
   synchronization the -multi_thread_exit option must also be set.  These
   options can also be enabled programmatically via
   dr_set_process_exit_behavior().

 - \b -persist:\anchor op_persist
   Enables persisting of code caches to disk and re-use on subsequent runs.
   Caches are persisted in units that correspond to application libraries,
   or sometimes smaller units.  Each unit is persisted to its own file
   in a subdirectory of the base directory specified by \p -persist_dir.
   See \ref sec_pcache for more details.

 - \b -persist_dir \e \<path\>:
   Sets the base directory for persistent code cache files.  If unset,
   the default base directory is the log directory.  A different
   sub-directory will be created for each user inside the specified
   directory.

 - \b -translate_fpu_pc:\anchor op_translate_fpu_pc
   Enables translation of the last floating-point instruction address when
   the last floating-point instruction is not in the same basic block as
   the instruction saving the FPU state.  This is off by default as it
   incurs significant performance penalties and few applications require
   this feature.

\if cache_sizing
XXX: users may want control over adaptive wset cache management,
particularly for thread-private to avoid deletions, but also for shared if
they want to shrink memory usage
\endif

 - \b -syntax_intel: \anchor op_syntax_intel
    This option causes DynamoRIO to output all disassembly using Intel
    syntax rather than the default show-implicit-operands syntax.  This can also be set
    using disassemble_set_syntax().

 - \b -syntax_att: \anchor op_syntax_att
    This option causes DynamoRIO to output all disassembly using AT&T
    syntax rather than the default show-implicit-operands syntax.  This can also be set
    using disassemble_set_syntax().

 - \b -syntax_arm: \anchor op_syntax_arm
    This option causes DynamoRIO to output all disassembly using standard
    ARM assembler syntax rather than the default show-implicit-operands
    syntax.  This can also be set using disassemble_set_syntax().

 - \b -syntax_riscv: \anchor op_syntax_riscv
    This option causes DynamoRIO to output all disassembly using
    RISC-V assembler syntax rather than the default show-implicit-operands
    syntax.  This can also be set using disassemble_set_syntax().

 - \b -disasm_mask:
    This option sets the disassembly style to the specified bitmask of
    dr_disasm_flags_t values.  This option overlaps with -syntax_intel,
    -syntax_att, and -syntax_arm.  The style can also be set using
    disassemble_set_syntax().

 - \b -tracedump_text and \b -tracedump_binary:
    These options cause DynamoRIO to output all traces that were created
    to the log file \e traces-shared.0.TID.html, where \e
    TID is the thread id of the initial thread; any thread-private traces
    (see \ref op_thread_priv "-thread_private option") produce per-thread
    files \e traces.TID.html.
    Traces are logged whenever they are flushed from the cache (which can
    be during execution or at the latest at program termination).  The two
    options select either a text dump or a binary dump.  The text dump
    takes up considerable room and time to dump, while the binary dump
    requires more effort to examine.  The binary trace dump format is
    documented in dr_tools.h, and a sample reader is provided with this
    distribution.

 - \b -tracedump_origins
   When selected by itself with neither -tracedump_text nor
   -tracedump_binary, dumps only a text list of the constituent basic block
   tags of each trace to the trace log file.  When combined with either of
   -tracedump_text or -tracedump_binary, adds a full disassembly of the
   constituent basic blocks to the selected dump.

\if profiling
XXX PR 225255: profiling options
\endif

Options controlling notifications from DynamoRIO:

 - \b -msgbox_mask \e 0xN: \anchor op_msgbox_mask
   Controls whether DynamoRIO uses pop-up message boxes on Windows,
   or waits for a key press on Linux, when presenting information.
   The mask takes the following bitfields:
   - INFORMATION = 0x1
   - WARNING     = 0x2
   - ERROR       = 0x4
   - CRITICAL    = 0x8

   dr_messagebox() is not affected by -msgbox_mask. For the
   provided Windows debug build -msgbox_mask defaults to 0xC.
   On Linux the default is 0, as this feature reads from standard input and
   might conflict with some applications.  On Linux the pause can be
   changed to use an infinite loop rather than reading from standard input
   by passing the \b -pause_via_loop runtime option, which allows attaching
   a debugger.

\attention
   On Vista or higher most Windows services are currently unable to display
   message boxes (see \ref limits_vista_service_messagebox
   "Limitations"). Since these services also don't have an associated
   console for stderr printing, the \ref op_loglevel "-loglevel"
   and \ref op_logmask "-logmask" options should be used
   instead. For the messages that would be displayed by -msgbox_mask,
   setting any bit in -logmask is sufficient for the message to be
   included in the logfile.

 - \b -stderr_mask \e 0xN:
   Parallel to -msgbox_mask, but controls DynamoRIO's output to standard
   error.  This option takes the same bitfields as -msgbox_mask.  The API
   routine dr_is_notify_on() can be used to determine if -stderr_mask is
   non-zero.  Messages printed to stderr will only be visible for
   applications that have an attached console.  They will not be visible
   in the \p cmd console on Windows 7 or earlier or on any Windows version
   when running a graphical application in \p cmd (even with dr_enable_console_printing(),
   as that only affects clients calling dr_printf() or dr_fprintf()) but
   the output can be viewed from \p cmd by redirecting to a file.
   For the provided Linux
   debug builds, -stderr_mask defaults to 0xF; for the Linux release
   builds, its default is 0xE.  The default on Windows is 0.

Options aiding in debugging:

 - \b -no_hide: \anchor op_no_hide
    By default, DynamoRIO hides itself from the Windows module list, for
    transparency.  However, this makes it more difficult to debug a process
    under DynamoRIO's control.  The option -no_hide turns off this module
    hiding.  However, the client library and any libraries it imports from
    will still be hidden.  We provide a windbg script that can locate
    DynamoRIO, the client library, and all of its dependences, so this
    option should no longer be necessary (see \ref page_debugging).
    This option is for Windows only.

Options available only in the debug build of DynamoRIO:

\anchor op_loglevel
 - \b -loglevel \e N:
    If N is greater than 0, DynamoRIO prints out a log of its actions.
    The greater the value of N, the more information DynamoRIO prints.
    Useful ranges are from 1 to 6.  Verbosity is set to 0 by default, i.e.,
    no log written.  All log files are kept in a log directory.  There is
    one directory per address space per run.  The directories are named \e
    app.NNN, where \e app is the application name and \e NNN is a number
    that is incremented with each directory created.  On
    Windows the directories are located by default in
    a subdirectory \e logs of the DynamoRIO home directory as
    specified in the dr_register_process(), \c drconfig, or \c drrun
    configuration for the target application.
    The runtime option \ref op_logdir "-logdir" can be used to override the
    default directory.
    There is one main log file per directory named
    \e app.0.TID.html, where \e TID is the thread identifier of the initial
    thread.  There is also a log file per thread, named \e log.N.TID.html,
    where \e N is the thread's creation ordinal and \e TID is its thread
    identifier.  The loglevel may be changed during program execution, but
    if it began at 0 then it cannot be raised later.  The -logmask
    parameter can be used to control which DynamoRIO modules output data
    to the log files. dr_log() allows the client to write to the above
    logfiles.

\anchor op_logmask
 - \b -logmask \e 0xN:
    Selects which DynamoRIO modules print out logging information, at the
    -loglevel level.  The mask is a combination of the DR_LOG_ bitfields
    listed in dr_tools.h (#DR_LOG_ALL selects all modules).

\anchor op_logdir
 - \b -logdir \e \<path\>:
    Specifies the directory to use for log files.  See the documentation
    for \ref op_loglevel "-loglevel" for a description of the default
    log directory.

 - \b -ignore_assert_list \b '*': \anchor op_ignore_assert
    Ignores all DynamoRIO asserts of the form "<file>:1234". * may be
    replaced by a ; separated list of individual asserts to ignore
    "foo.c:333;bar.c:12".

*/