Merge branch 'patch-1' into 'master'

Clean up everything and add lots of new functions

I hope everything here is okay. I fixed some bugs in existing functions (mostly wrong prototypes or off-by-one errors) and formatted everything to a consistent style. If you'd like me to adjust the style, that's no problem. It wasn't very consistent to start with (within the code, and compared to the documentation).

*I added specific notes you may want to comment on to some of the commit messages.*

Major features:
* Pass correct `argv[0]` in loader
* Intercept `readlink("/proc/self/exe")` to allow re-exec
* Add almost all reasonable-to-implement functions in LSB 5.0.0 core generic libc. Remaining functions are:
  - Impossible-to-implement: sigreturn
  - Not useful: reentrant random (`*rand48_r`, etc.), argz, envz, pmap, rpc (clnt_*, svc*, xdr*), bindresvport
  - Got tired of it: checked wchar
* Add additional functions used by android and its NDK tools (clang, cmake, lldb, ninja, etc.).

At this point, I am able to run Android Studio with the bundled prebuilt JDK, and compile, install, and run an android application (including one with native libraries) on a real device, with only a few minor issues:
* must export `LD_LIBRARY_PATH=/opt/android-studio/jre/jre/lib/amd64/server` because musl and glibc interpret the variable differently with regards to `dlopen`. This is something that has to be patched in musl builds of openjdk, so it's not a gcompat issue.
* ld.bfd fails to parse the argument `--sysroot=/path`, but can parse `--sysroot /path`. So there's some difference with `getopt_long_only` (or getopt in general). May be a bug, may be just an API difference. May be it can be patched up.
* LLDB fails to connect to the android phone for native debugging -- I haven't tried it on glibc yet, so it may not be a gcompat issue at all (may be a phone or the app issue).

I'd be happy to send some documentation later.

See merge request !1

1 files changed, 12 insertions, 12 deletions

diff --git a/libgcompat/resource.c b/libgcompat/resource.c
index 029b31e..836d5e2 100644
--- a/libgcompat/resource.c
+++ b/libgcompat/resource.c
@@ -1,8 +1,8 @@
-#include <sys/resource.h>	/* setrlimit, struct rlimit */
-#include <assert.h>		/* assert */
-#include <dlfcn.h>		/* dlsym, RTLD_NEXT */
-#include <stdlib.h>		/* NULL */
-#include <string.h>		/* memcpy */
+#include <assert.h>       /* assert */
+#include <dlfcn.h>        /* dlsym, RTLD_NEXT */
+#include <stdlib.h>       /* NULL */
+#include <string.h>       /* memcpy */
+#include <sys/resource.h> /* setrlimit, struct rlimit */
 
 /* Sigh.
  * Valve compiled Steam against the glibc2.2 version of setrlimit.
@@ -11,24 +11,24 @@
  * So, what you have to do is: if you want to run steam with this gcompat,
  * ensure you compile *without* defining NO_BROKEN_SHADOW_SETRLIMIT.
  * If you do *not* want to run steam with this gcompat, define it.
- * 
+ *
  * The only problem with enabling this all the time is that if a binary
  * really does need a ulimit to be 0 for any reason (such as coredumps), it
  * very obviously won't work here.
  */
 #ifndef NO_BROKEN_SHADOW_SETRLIMIT
-int (*real_rlimit)(int, const struct rlimit *);
+static int (*real_rlimit)(int, const struct rlimit *);
 
 int setrlimit(int resource, const struct rlimit *rlim)
 {
 	struct rlimit my_rlim;
-	real_rlimit = dlsym(RTLD_NEXT, "setrlimit");
-	assert(real_rlimit != NULL);
 
+	if (real_rlimit == NULL) {
+		real_rlimit = dlsym(RTLD_NEXT, "setrlimit");
+		assert(real_rlimit);
+	}
 	memcpy(&my_rlim, rlim, sizeof(struct rlimit));
-
-	if(my_rlim.rlim_cur == 0)
-	{
+	if (my_rlim.rlim_cur == 0) {
 		my_rlim.rlim_cur = my_rlim.rlim_max;
 	}