Mixing PIC and non-PIC objects in a shared library - compiler-flags

This question is related to this one as well as its answer.
I just discovered some ugliness in a build I'm working on. The situation looks somewhat like the following (written in gmake format); note, this specifically applies to a 32-bit memory model on sparc and x86 hardware:
OBJ_SET1 := some objects
OBJ_SET2 := some objects
# note: OBJ_SET2 doesn't get this flag
${OBJ_SET1} : CCFLAGS += -PIC
${OBJ_SET1} ${OBJ_SET2} : %.o : %.cc
${CCC} ${CCFLAGS} -m32 -o ${#} -c ${<}
obj1.o : ${OBJ_SET1}
obj2.o : ${OBJ_SET2}
sharedlib.so : obj1.o obj2.o
obj1.o obj2.o sharedlib.so :
${LINK} ${LDFLAGS} -m32 -PIC -o ${#} ${^}
Clearly it can work to mix objects compiled with and without PIC in a shared object (this has been in use for years). I don't know enough about PIC to know whether it's a good idea/smart, and my guess is in this case it's not needed but rather it's happening because someone didn't care enough to find out the right way to do it when tacking on new stuff to the build.
My question is:
Is this safe
Is it a good idea
What potential problems can occur as a result
If I switch everything to PIC, are there any non-obvious gotchas that I might want to watch out for.

Forgot I even wrote this question.
Some explanations are in order first:
Non-PIC code may be loaded by the OS into any position in memory in [most?] modern OSs. After everything is loaded, it goes through a phase that fixes up the text segment (where the executable stuff ends up) so it correctly addresses global variables; to pull this off, the text segment must be writable.
PIC executable data can be loaded once by the OS and shared across multiple users/processes. For the OS to do this, however, the text segment must be read-only -- which means no fix-ups. The code is compiled to use a Global Offset Table (GOT) so it can address globals relative to the GOT, alleviating the need for fix-ups.
If a shared object is built without PIC, although it is strongly encouraged it doesn't appear that it's strictly necessary; if the OS must fix-up the text segment then it's forced to load it into memory that's marked read-write ... which prevents sharing across processes/users.
If an executable binary is built /with/ PIC, I don't know what goes wrong under the hood but I've witnessed a few tools become unstable (mysterious crashes & the like).
The answers:
Mixing PIC/non-PIC, or using PIC in executables can cause hard to predict and track down instabilities. I don't have a technical explanation for why.
... to include segfaults, bus errors, stack corruption, and probably more besides.
Non-PIC in shared objects is probably not going to cause any serious problems, though it can result in more RAM used if the library is used many times across processes and/or users.
update (4/17)
I've since discovered the cause of some of the crashes I had seen previously. To illustrate:
/*header.h*/
#include <map>
typedef std::map<std::string,std::string> StringMap;
StringMap asdf;
/*file1.cc*/
#include "header.h"
/*file2.cc*/
#include "header.h"
int main( int argc, char** argv ) {
for( int ii = 0; ii < argc; ++ii ) {
asdf[argv[ii]] = argv[ii];
}
return 0;
}
... then:
$ g++ file1.cc -shared -PIC -o libblah1.so
$ g++ file1.cc -shared -PIC -o libblah2.so
$ g++ file1.cc -shared -PIC -o libblah3.so
$ g++ file1.cc -shared -PIC -o libblah4.so
$ g++ file1.cc -shared -PIC -o libblah5.so
$ g++ -zmuldefs file2.cc -Wl,-{L,R}$(pwd) -lblah{1..5} -o fdsa
# ^^^^^^^^^
# This is the evil that made it possible
$ args=(this is the song that never ends);
$ eval ./fdsa $(for i in {1..100}; do echo -n ${args[*]}; done)
That particular example may not end up crashing, but it's basically the situation that had existed in that group's code. If it does crash it'll likely be in the destructor, usually a double-free error.
Many years previous they added -zmuldefs to their build to get rid of multiply defined symbol errors. The compiler emits code for running constructors/destructors on global objects. -zmuldefs forces them to live at the same location in memory but it still runs the constructors/destructors once for the exe and each library that included the offending header -- hence the double-free.

Related

conditional compilation in makefile so that command-line arguments are ignored

Using gcc compiler on linux, I have a C program that used command-line argument (one argument) like
./myprog 0
I want to write a makefile that uses conditional compilation so that if I compile like
make SPECIAL=1
then command-line argument is used.
if I compile without SPECIAL like
make
then command-line argument is ignored even if we enter it.
How to make it possible.
I am using following compilation command
gcc -o myprog myprog.c prog2.c prog3.c
The makefile can be trivial but the real work has to happen in the C code. Something like this, as a minimal example:
#include <stdio.h>
int main(int argc, char **argv, char **envp) {
int i;
#ifndef SPECIAL
argc=1;
argv[1] = NULL;
#endif
for(i=1; i<argc; ++i) {
printf(">>%s<<\n", argv[i]);
}
}
Now, you don't even really need a Makefile for this simple program.
bash$ make CFLAGS=-DSPECIAL=1 -f /dev/null myprog
cc -DSPECIAL=1 myprog.c -o myprog
Having said that, making your build nondeterministic by introducing variations which depend on ephemeral build-time whims is a recipe for insanity. Have two separate targets which create two separate binaries, one with the regular behavior, and the other with the foot-gun semantics.
DEPS := myprog.c prog2.c prog3.c # or whatever your dependencies are
myprog: $(DEPS)
myprog-footgun: CLAGS+=-DSPECIAL=1
myprog-footgun: $(DEPS) # same dependencies, different output file
$(CC) $(CFLAGS) -o $# $^
See Target-specific Variable Values in the GNU Make documentation for details of this syntax.
(Notice that Stack Overflow renders tabs in the Markdown source as spaces, so you will not be able to copy/paste this verbatim.)
I would perhaps in fact reverse the meaning of SPECIAL so that it enables the foot-gun version, rather than the other way around (the original version of this answer had this design, just because I had read your question that way originally).

linux gcc linking, duplicate symbols? [duplicate]

Is there any way we can get gcc to detect a duplicate symbol in static libraries vs the main code (Or another static library ?)
Here's the situation:
main.c erroneously contained a function definition, e.g. with the signature uint foohash(const char*)
foo.c also contains a function definition with the signature uint foohash(const char*)
foo.c and other source files are compiled to a static util library, which the main program links in, i.e. something like:
gcc -o main main.o util.o -L ./libs -lfooutils
So, now main.o and libs/libfooutils.a both contain a foohash function. Presumably the linker found that symbol in main.o and doesn't bother looking for it elsewhere.
Is there any way we can get gcc to detect such a situation ?
Indeed as Simon Richter stated, --whole-archive option can be useful. Try to change your command-line to:
gcc -o main main.o util.o -L ./libs -Wl,--whole-archive -lfooutils -Wl,--no-whole-archive
and you'll see a multiple definition error.
gcc calls the ld program for linking. The relevant ld options are:
--no-define-common
--traditional-format
--warn-common
See the man page for ld. These should be what you need to experiment with to get the warnings sought.
Short answer: no.
GCC does not actually do anything with libraries. It is the task of ld, the linker (called automatically by GCC) to pull in symbols from libraries, and that's really a fairly dumb tool.
The linker has lots of complex jiggery pokery for combining different types of data from different sources, and supporting different file formats, and all the evil little details of binary executables, but in the end, all it really does is look for undefined symbols and find the definitions.
What you can do is a link trace (pass -t to gcc) to see what comes from where. Or else run nm on all the object files and libraries in your system, and write a script to detect duplicates.

Why are there global offset tables and procedure linkage tables in statically linked executables?

I've done a bunch of reading on dynamic linker relocations and position independent code including procedure linkage tables and global offset tables. I don't understand why a statically linked executable needs a PLT and GOT. I compiled a hello world program on my ubuntu x86_64 machine and when I dump the section headers with readelf -S it shows PLT and GOT sections.
I also created a shared library with a simple increment function that I compiled with gcc -shared without -fpic and I also see PLT and GOT sections. I didn't expect this either.
I don't understand why a statically linked executable needs a PLT and GOT.
It doesn't.
I compiled a hello world program on my ubuntu x86_64 machine and when I dump the section headers with readelf -S it shows PLT and GOT sections.
This is an accident of implementation. The sections come from crt1.o, and there isn't a separate crt1s.o for fully-static linking, so you end up with .plt and .got entries from there.
You can strip these sections, and the binary will still work:
objcopy -R.got -R.plt a.out a.out2
Note: do not strip .rela.plt, as that section is still needed to implement IFUNCs.
I found that gcc generates a .got and .got.lpt when generating position independent code and taking the address of a function defined in another source file.
My test files were:
part1.c:
extern void afunc();
int _start()
{
return 0x55 & (__SIZE_TYPE__) afunc;
}
part2.c:
void afunc() {}
My test was (substitute your own gcc version):
for o in s 4 3 2 1 0
do
aarch64-linux-gnu-gcc-10 -fPIC part1.c part2.c -o static.elf -static -nostdlib -O$o &&
aarch64-linux-gnu-objdump -x static.elf | grep 'GLOBAL_OFFSET'
done
I get the following output for all optimization levels:
0000000000410fd8 l O .got 0000000000000000 _GLOBAL_OFFSET_TABLE_
Replacing -fPIC with -fno-PIC and the segment goes away.
You can tell if your compiler defaults to -fPIC by running this:
aarch64-linux-gnu-gcc-10 -mcmodel=large -x c - < /dev/null
From which, I get the error, if it does:
cc1: sorry, unimplemented: code model ‘large’ with ‘-fPIC’

why clang++ behaves differently from clang since the former is a symbol link of the latter?

I have a C program that tries to modify a const string literal. As now I learned that this is not allowed.
When I compile the code with clang test.c the compiler gives no warning. But when I compile it with clang++ test.c it gives a warning:
test.c:6:15: warning: conversion from string literal to 'char *' is deprecated
[-Wdeprecated-writable-strings]
char *s = "hello world";
^
The problem is that it turns out clang++ is just a symbol link of clang:
ll `which clang++`
lrwxr-xr-x 1 root admin 5 Jan 1 12:34 /usr/bin/clang++# -> clang
So my question is how could clang++ behaves differently from clang given that it's a symbol link of clang?
Clang is looking at its argv[0] and altering its behavior depending on what it sees. This is an uncommon and discouraged, but not rare, trick going at least as far back as 4.2BSD ex and vi, which were the same executable, and probably farther.
In this case, clang is compiling your .c file as C, and clang++ is compiling it as C++. This is a historical wart which you should not rely on; use the appropriate compiler command and make sure that your file extension reflects the true contents of the file.
By convention, the name by which a command is invoked is passed as argv[0]; it is not especially unusual for programs to change their behavior based on this. (Historically, ln, cp, and mv were hardlinks to the same executable on Research Unix and used argv[0] to decide which action to do. Also, most shells look for a leading - in argv[0] to decide if they should be a login shell.) Often there is also some other way to get the same effect (options, environment variables, etc.); you should in general use this instead of playing argv[0] games.
There are reasons to do this, but in most cases it's not a good idea to rely on it or to design programs around it.

Symbols from convenience library not getting exported in executable

I have a program, myprogram, which is linked with a static convenience library, call it libconvenience.a, which contains a function, func(). The function func() isn't called anywhere in myprogram; it needs to be able to be called from a plugin library, plugin.so.
The symbol func() is not getting exported dynamically in myprogram. If I run
nm myprogram | grep func
I get nothing. However, it isn't missing from libconvenience.a:
nm libconvenience/libconvenience.a | grep func
00000000 T func
I am using automake, but if I do the last linking step by hand on the command line instead, it doesn't work either:
gcc -Wl,--export-dynamic -o myprogram *.o libconvenience/libconvenience.a `pkg-config --libs somelibraries`
However, if I link the program like this, skipping the use of a convenience library and linking the object files that would have gone into libconvenience.a directly, func() shows up in myprogram's symbols as it should:
gcc -Wl,--export-dynamic -o myprogram *.o libconvenience/*.o `pkg-config --libs somelibraries`
If I add a dummy call to func() somewhere in myprogram, then func() also shows up in myprogram's symbols. But I thought that --export-dynamic was supposed to export all symbols regardless of whether they were used in the program or not!
I am using automake 1.11.1 and gcc 4.5.1 on Fedora 14. I am also using Libtool 2.2.10 to build plugin.so (but not the convenience library.)
I didn't forget to put -Wl,--export-dynamic in myprogram_LDFLAGS, nor did I forget to put the source that contains func() in libconvenience_a_SOURCES (some Googling suggests that these are common causes of this problem.)
Can somebody help me understand what is going on here?
I managed to solve it. It was this note from John Calcote's excellent Autotools book that pointed me in the right direction:
Linkers add to the binary product every object file specified explicitly on the command line, but they only extract from archives those object files that are actually referenced in the code being linked.
To counteract this behavior, one can use the --whole-archive flag to libtool. However, this causes all the symbols from all the system libraries to be pulled in also, causing lots of double symbol definition errors. So --whole-archive needs to be right before libconvenience.a on the linker command line, and it needs to be followed by --no-whole-archive so that the other libraries aren't treated that way. This is a bit difficult since automake and libtool don't really guarantee keeping your flags in the same order on the command line, but this line in Makefile.am did the trick:
myprogram_LDFLAGS = -Wl,--export-dynamic \
-Wl,--whole-archive,libconvenience/libconvenience.a,--no-whole-archive
If you need func to be in plugin.so, you should try and locate it there if possible. Convenience libraries are meant to be just that -- a convenience to link to an executable or lib as an intermediate step.

Resources