Clean up the TLS code

libc/runtime/enable_tls.c

@@ -21,6 +21,8 @@
 #include "libc/calls/syscall-sysv.internal.h"
 #include "libc/dce.h"
 #include "libc/errno.h"
+#include "libc/intrin/asan.internal.h"
+#include "libc/intrin/asancodes.h"
 #include "libc/intrin/bits.h"
 #include "libc/intrin/weaken.h"
 #include "libc/log/libfatal.internal.h"
@@ -57,7 +59,7 @@ __msabi extern typeof(TlsAlloc) *const __imp_TlsAlloc;
 
 extern unsigned char __tls_mov_nt_rax[];
 extern unsigned char __tls_add_nt_rax[];
-_Alignas(long) static char __static_tls[5008];
+_Alignas(TLS_ALIGNMENT) static char __static_tls[5008];
 
 /**
  * Enables thread local storage for main process.
@@ -103,7 +105,6 @@ privileged void __enable_tls(void) {
     // if tls requirement is small then use the static tls block
     // which helps avoid a system call for appes with little tls
     // this is crucial to keeping life.com 16 kilobytes in size!
-    _Static_assert(alignof(__static_tls) >= alignof(struct CosmoTib));
     mem = __static_tls;
   } else {
     // if this binary needs a hefty tls block then we'll bank on
@@ -115,6 +116,12 @@ privileged void __enable_tls(void) {
     mem = weaken(_mapanon)(siz);
     assert(mem);
   }
+  if (IsAsan()) {
+    // poison the space between .tdata and .tbss
+    __asan_poison(mem + (intptr_t)_tdata_size,
+                  (intptr_t)_tbss_offset - (intptr_t)_tdata_size,
+                  kAsanProtected);
+  }
   tib = (struct CosmoTib *)(mem + siz - _TIBZ);
   tls = mem + siz - _TIBZ - _TLSZ;
   tib->tib_self = tib;

libc/runtime/internal.h

@@ -23,6 +23,8 @@ extern unsigned char _tdata_end[];
 extern unsigned char _tdata_size[];
 extern unsigned char _tbss_start[];
 extern unsigned char _tbss_end[];
+extern unsigned char _tbss_size[];
+extern unsigned char _tbss_offset[];
 extern unsigned char _tls_size[];
 extern unsigned char _tls_content[];
 

libc/stubs/ld.S

@@ -50,6 +50,8 @@
 	_tdata_size = 0
 	_tbss_start = 0
 	_tbss_end = 0
+	_tbss_offset = 0
+	_tbss_size = 0
 	_tls_size = 0
 	_tls_content = 0
 
@@ -73,6 +75,8 @@
 	.globl	_tdata_size
 	.globl	_tbss_start
 	.globl	_tbss_end
+	.globl	_tbss_size
+	.globl	_tbss_offset
 	.globl	_tls_size
 	.globl	_tls_content
 	.globl	__data_start
@@ -100,8 +104,10 @@
 	.weak	_tdata_size
 	.weak	_tbss_start
 	.weak	_tbss_end
+	.weak	_tbss_size
 	.weak	_tls_size
 	.weak	_tls_content
+	.weak	_tbss_offset
 	.weak	__data_start
 	.weak	__data_end
 	.weak	__bss_start

libc/thread/mktls.c

@@ -16,6 +16,9 @@
 │ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
+#include "libc/dce.h"
+#include "libc/intrin/asan.internal.h"
+#include "libc/intrin/asancodes.h"
 #include "libc/macros.internal.h"
 #include "libc/mem/mem.h"
 #include "libc/runtime/internal.h"
@@ -24,10 +27,9 @@
 #include "libc/thread/spawn.h"
 #include "libc/thread/tls.h"
 
-#define _TLSZ ((intptr_t)_tls_size)
-#define _TLDZ ((intptr_t)_tdata_size)
-#define _TIBZ sizeof(struct CosmoTib)
-#define _MEMZ ROUNDUP(_TLSZ + _TIBZ, _Alignof(struct CosmoTib))
+#define I(x) ((intptr_t)x)
+
+void Bzero(void *, size_t) asm("bzero");  // gcc bug
 
 /**
  * Allocates thread-local storage memory for new thread.
@@ -37,17 +39,25 @@ char *_mktls(char **out_tib) {
   char *tls;
   struct CosmoTib *tib;
 
-  // Allocate enough TLS memory for all the GNU Linuker (_tls_size)
-  // organized _Thread_local data, as well as Cosmpolitan Libc (64)
-  if (!(tls = calloc(1, _MEMZ))) return 0;
+  // allocate memory for tdata, tbss, and tib
+  tls = memalign(TLS_ALIGNMENT, I(_tls_size) + sizeof(struct CosmoTib));
+  if (!tls) return 0;
+
+  // poison memory between tdata and tbss
+  if (IsAsan()) {
+    __asan_poison(tls + I(_tdata_size), I(_tbss_offset) - I(_tdata_size),
+                  kAsanProtected);
+  }
+
+  // initialize tdata and clear tbss
+  memmove(tls, _tdata_start, I(_tdata_size));
+  Bzero(tls + I(_tbss_offset), I(_tbss_size) + sizeof(struct CosmoTib));
 
   // set up thread information block
-  tib = (struct CosmoTib *)(tls + _MEMZ - _TIBZ);
+  tib = (struct CosmoTib *)(tls + I(_tls_size));
   tib->tib_self = tib;
   tib->tib_self2 = tib;
-  tib->tib_errno = 0;
   tib->tib_tid = -1;
-  memmove(tls, _tdata_start, _TLDZ);
 
   if (out_tib) {
     *out_tib = (char *)tib;

libc/thread/tls.h

@@ -1,5 +1,8 @@
 #ifndef COSMOPOLITAN_LIBC_THREAD_TLS_H_
 #define COSMOPOLITAN_LIBC_THREAD_TLS_H_
+
+#define TLS_ALIGNMENT 64
+
 #if !(__ASSEMBLER__ + __LINKER__ + 0)
 COSMOPOLITAN_C_START_