blob: 94b8766eeae598bc10b2157bfd979163db4109b2 [file] [log] [blame]
/*--------------------------------------------------------------------*/
/*--- Redirections, etc. pub_tool_redir.h ---*/
/*--------------------------------------------------------------------*/
/*
This file is part of Valgrind, a dynamic binary instrumentation
framework.
Copyright (C) 2000-2013 Julian Seward
jseward@acm.org
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of the
License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
02111-1307, USA.
The GNU General Public License is contained in the file COPYING.
*/
#ifndef __PUB_TOOL_REDIR_H
#define __PUB_TOOL_REDIR_H
#include "config.h" /* DARWIN_VERS */
/* The following macros facilitate function replacement and wrapping.
Function wrapping and function replacement are similar but not
identical.
A replacement for some function F simply diverts all calls to F
to the stated replacement. There is no way to get back to F itself
from the replacement.
A wrapper for a function F causes all calls to F to instead go to
the wrapper. However, from inside the wrapper, it is possible
(with some difficulty) to get to F itself.
You may notice that replacement is a special case of wrapping, in
which the call to the original is omitted. For implementation
reasons, though, it is important to use the following macros
correctly: in particular, if you want to write a replacement, make
sure you use the VG_REPLACE_FN_ macros and not the VG_WRAP_FN_
macros.
Finally there is the concept of prioritised behavioural equivalence
tags. A tag is a 5-digit decimal number (00000 to 99999) encoded
in the name. The top 4 digits are the equivalence class number,
and the last digit is a priority.
When processing redirections at library load time, if the set of
available specifications yields more than one replacement or
wrapper function for a given address, the system will try to
resolve the situation by examining the tags on the
replacements/wrappers.
If two replacement/wrapper functions have the same tag and
priority, then the redirection machinery will assume they have
identical behaviour and can choose between them arbitrarily. If
they have the same tag but different priorities, then the one with
higher priority will be chosen. If neither case holds, then the
redirection is ambiguous and the system will ignore one of them
arbitrarily, but print a warning when running at -v or above.
The tag is mandatory and must comprise 5 decimal digits. The tag
00000 is special and means "does not have behaviour identical to any
other replacement/wrapper function". Hence if you wish to write a
wrap/replacement function that is not subject to the above
resolution rules, use 00000 for the tag. Tags 00001 through 00009
may not be used for any purpose.
Replacement
~~~~~~~~~~~
To write a replacement function, do this:
ret_type
VG_REPLACE_FUNCTION_ZU(zEncodedSoname,fnname) ( .. args .. )
{
... body ...
}
zEncodedSoname should be a Z-encoded soname (see below for
Z-encoding details) and fnname should be an unencoded fn name. A
default-safe equivalence tag of 00000 is assumed (see comments
above). The resulting name is
_vgr00000ZU_zEncodedSoname_fnname
The "_vgr00000ZU_" is a prefix that gets discarded upon decoding.
It identifies this function as a replacement and specifies its
equivalence tag.
It is also possible to write
ret_type
VG_REPLACE_FUNCTION_ZZ(zEncodedSoname,zEncodedFnname) ( .. args .. )
{
... body ...
}
which means precisely the same, but the function name is also
Z-encoded. This can sometimes be necessary. In this case the
resulting function name is
_vgr00000ZZ_zEncodedSoname_zEncodedFnname
When it sees this either such name, the core's symbol-table reading
machinery and redirection machinery first Z-decode the soname and
if necessary the fnname. They are encoded so that they may include
arbitrary characters, and in particular they may contain '*', which
acts as a wildcard.
They then will conspire to cause calls to any function matching
'fnname' in any object whose soname matches 'soname' to actually be
routed to this function. This is used in Valgrind to define dozens
of replacements of malloc, free, etc.
The soname must be a Z-encoded bit of text because sonames can
contain dots etc which are not valid symbol names. The function
name may or may not be Z-encoded: to include wildcards it has to be,
but Z-encoding C++ function names which are themselves already mangled
using Zs in some way is tedious and error prone, so the _ZU variant
allows them not to be Z-encoded.
Note that the soname "NONE" is specially interpreted to match any
shared object which doesn't have a soname.
Note also that the replacement function should probably (must be?) in
client space, so it runs on the simulated CPU. So it must be in
either vgpreload_<tool>.so or vgpreload_core.so. It also only works
with functions in shared objects, I think.
It is important that the Z-encoded names contain no unencoded
underscores, since the intercept-handlers in m_redir.c detect the
end of the soname by looking for the first trailing underscore.
To write function names which explicitly state the equivalence class
tag, use
VG_REPLACE_FUNCTION_EZU(5-digit-tag,zEncodedSoname,fnname)
or
VG_REPLACE_FUNCTION_EZZ(5-digit-tag,zEncodedSoname,zEncodedFnname)
As per comments above, the tag must be a 5 digit decimal number,
padded with leading zeroes, in the range 00010 to 99999 inclusive.
Wrapping
~~~~~~~~
This is identical to replacement, except that you should use the
macro names
VG_WRAP_FUNCTION_ZU
VG_WRAP_FUNCTION_ZZ
VG_WRAP_FUNCTION_EZU
VG_WRAP_FUNCTION_EZZ
instead.
Z-encoding
~~~~~~~~~~
Z-encoding details: the scheme is like GHC's. It is just about
readable enough to make a preprocessor unnecessary. First the
"_vgrZU_" or "_vgrZZ_" prefix is added, and then the following
characters are transformed.
* --> Za (asterisk)
: --> Zc (colon)
. --> Zd (dot)
- --> Zh (hyphen)
+ --> Zp (plus)
(space) --> Zs (space)
_ --> Zu (underscore)
@ --> ZA (at)
$ --> ZD (dollar)
( --> ZL (left)
) --> ZR (right)
Z --> ZZ (Z)
Everything else is left unchanged.
*/
/* If you change these, the code in VG_(maybe_Z_demangle) needs to be
changed accordingly. NOTE: duplicates
I_{WRAP,REPLACE}_SONAME_FNNAME_Z{U,Z} in valgrind.h. */
/* Use an extra level of macroisation so as to ensure the soname/fnname
args are fully macro-expanded before pasting them together. */
#define VG_CONCAT4(_aa,_bb,_cc,_dd) _aa##_bb##_cc##_dd
#define VG_CONCAT6(_aa,_bb,_cc,_dd,_ee,_ff) _aa##_bb##_cc##_dd##_ee##_ff
/* The 4 basic macros. */
#define VG_REPLACE_FUNCTION_EZU(_eclasstag,_soname,_fnname) \
VG_CONCAT6(_vgr,_eclasstag,ZU_,_soname,_,_fnname)
#define VG_REPLACE_FUNCTION_EZZ(_eclasstag,_soname,_fnname) \
VG_CONCAT6(_vgr,_eclasstag,ZZ_,_soname,_,_fnname)
#define VG_WRAP_FUNCTION_EZU(_eclasstag,_soname,_fnname) \
VG_CONCAT6(_vgw,_eclasstag,ZU_,_soname,_,_fnname)
#define VG_WRAP_FUNCTION_EZZ(_eclasstag,_soname,_fnname) \
VG_CONCAT6(_vgw,_eclasstag,ZZ_,_soname,_,_fnname)
/* Convenience macros defined in terms of the above 4. */
#define VG_REPLACE_FUNCTION_ZU(_soname,_fnname) \
VG_CONCAT6(_vgr,00000,ZU_,_soname,_,_fnname)
#define VG_REPLACE_FUNCTION_ZZ(_soname,_fnname) \
VG_CONCAT6(_vgr,00000,ZZ_,_soname,_,_fnname)
#define VG_WRAP_FUNCTION_ZU(_soname,_fnname) \
VG_CONCAT6(_vgw,00000,ZU_,_soname,_,_fnname)
#define VG_WRAP_FUNCTION_ZZ(_soname,_fnname) \
VG_CONCAT6(_vgw,00000,ZZ_,_soname,_,_fnname)
/* --------- Some handy Z-encoded names. --------- */
// Nb: ALL THESE NAMES MUST BEGIN WITH "VG_Z_". Why? If we applied
// conditional compilation inconsistently we could accidentally use an
// undefined constant like VG_Z_LIBC_DOT_A, resulting in a bogus Z-encoded
// name like "_vgrZU_VG_Z_LIBC_DOT_A_foo". This can't be detected at
// compile-time, because both the constant's name and its value are
// identifiers. However, by always using "VG_Z_" as a prefix, we can do a
// run-time check and abort if any name has "VG_Z_" in it, because that
// indicates that the constant has been used without being defined.
/* --- Soname of the standard C library. --- */
#if defined(VGO_linux)
# define VG_Z_LIBC_SONAME libcZdsoZa // libc.so*
#elif defined(VGO_darwin) && (DARWIN_VERS <= DARWIN_10_6)
# define VG_Z_LIBC_SONAME libSystemZdZaZddylib // libSystem.*.dylib
#elif defined(VGO_darwin) && (DARWIN_VERS == DARWIN_10_7 \
|| DARWIN_VERS == DARWIN_10_8)
# define VG_Z_LIBC_SONAME libsystemZucZaZddylib // libsystem_c*.dylib
/* Note that the idea of a single name for the C library falls
apart on more recent Darwins (10.8 and later) since the
functionality (malloc, free, str*) is split between
libsystem_c.dylib, libsystem_malloc.dylib and
libsystem_platform.dylib. This makes VG_Z_LIBC_SONAME somewhat useless
at least inside vg_replace_strmem.c, and that hardwires some dylib
names directly, for OSX 10.9. */
#elif defined(VGO_darwin) && (DARWIN_VERS >= DARWIN_10_9)
# define VG_Z_LIBC_SONAME libsystemZumallocZddylib // libsystem_malloc.dylib
#else
# error "Unknown platform"
#endif
/* --- Soname of the GNU C++ library. --- */
// Valid on all platforms(?)
#define VG_Z_LIBSTDCXX_SONAME libstdcZpZpZa // libstdc++*
/* --- Soname of the pthreads library. --- */
#if defined(VGO_linux)
# define VG_Z_LIBPTHREAD_SONAME libpthreadZdsoZd0 // libpthread.so.0
#elif defined(VGO_darwin)
# define VG_Z_LIBPTHREAD_SONAME libSystemZdZaZddylib // libSystem.*.dylib
#else
# error "Unknown platform"
#endif
/* --- Sonames for Linux ELF linkers, plus unencoded versions. --- */
#if defined(VGO_linux)
#define VG_Z_LD_LINUX_SO_3 ldZhlinuxZdsoZd3 // ld-linux.so.3
#define VG_U_LD_LINUX_SO_3 "ld-linux.so.3"
#define VG_Z_LD_LINUX_SO_2 ldZhlinuxZdsoZd2 // ld-linux.so.2
#define VG_U_LD_LINUX_SO_2 "ld-linux.so.2"
#define VG_Z_LD_LINUX_X86_64_SO_2 ldZhlinuxZhx86Zh64ZdsoZd2
// ld-linux-x86-64.so.2
#define VG_U_LD_LINUX_X86_64_SO_2 "ld-linux-x86-64.so.2"
#define VG_Z_LD64_SO_1 ld64ZdsoZd1 // ld64.so.1
#define VG_U_LD64_SO_1 "ld64.so.1"
#define VG_Z_LD_SO_1 ldZdsoZd1 // ld.so.1
#define VG_U_LD_SO_1 "ld.so.1"
#endif
/* --- Executable name for Darwin Mach-O linker. --- */
#if defined(VGO_darwin)
#define VG_Z_DYLD dyld // dyld
#define VG_U_DYLD "dyld"
#endif
// Prefix for synonym soname synonym handling
#define VG_SO_SYN(name) VgSoSyn##name
#define VG_SO_SYN_PREFIX "VgSoSyn"
#define VG_SO_SYN_PREFIX_LEN 7
#endif // __PUB_TOOL_REDIR_H
/*--------------------------------------------------------------------*/
/*--- end ---*/
/*--------------------------------------------------------------------*/