Make CloudABI's way of doing TLS more friendly to userspace emulators.

We're currently seeing how hard it would be to run CloudABI binaries on
operating systems cannot be modified easily (Windows, Mac OS X). The
idea is that we want to just run them without any sandboxing. Now
that CloudABI executables are PIE, this is already a bit easier, but TLS
is still problematic:

- CloudABI executables want to write to the %fs, which typically
  requires extra system calls by the emulator every time it needs to
  switch between CloudABI's and its own TLS.

- If CloudABI executables overwrite the %fs base unconditionally, it
  also becomes harder for the emulator to store a backup of the old
  value of %fs. To solve this, let's no longer overwrite %fs, but just
  %fs:0.

As CloudABI's C library does not use a TCB, this space can now be used
by an emulator to keep track of its internal state. The executable can
now safely overwrite %fs:0, as long as it makes sure that the TCB is
copied over to the new TLS area.

Ensure that there is an initial TLS area set up when the process starts,
only containing a bogus TCB. We don't really care about its contents on
FreeBSD.

Reviewed by:	kib
Differential Revision:	https://reviews.freebsd.org/D5836

sys/amd64/cloudabi64/cloudabi64_sysvec.c

@@ -27,6 +27,7 @@
 __FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
+#include <sys/imgact.h>
 #include <sys/kernel.h>
 #include <sys/proc.h>
 #include <sys/sysent.h>
@@ -46,6 +47,45 @@ __FBSDID("$FreeBSD$");
 extern const char *cloudabi64_syscallnames[];
 extern struct sysent cloudabi64_sysent[];
 
+static int
+cloudabi64_fixup_tcb(register_t **stack_base, struct image_params *imgp)
+{
+	int error;
+	register_t tcbptr;
+
+	/* Place auxiliary vector and TCB on the stack. */
+	error = cloudabi64_fixup(stack_base, imgp);
+	if (error != 0)
+		return (error);
+	
+	/*
+	 * On x86-64, the TCB is referred to by %fs:0. Take some space
+	 * from the top of the stack to store a single element array,
+	 * containing a pointer to the TCB. %fs base will point to this.
+	 */
+	tcbptr = (register_t)*stack_base;
+	return (copyout(&tcbptr, --*stack_base, sizeof(tcbptr)));
+}
+
+static void
+cloudabi64_proc_setregs(struct thread *td, struct image_params *imgp,
+    unsigned long stack)
+{
+	struct trapframe *regs;
+
+	exec_setregs(td, imgp, stack);
+
+	/*
+	 * The stack now contains a pointer to the TCB, the TCB itself,
+	 * and the auxiliary vector. Let %rdx point to the auxiliary
+	 * vector, and set %fs base to the address of the TCB.
+	 */
+	regs = td->td_frame;
+	regs->tf_rdi = stack + sizeof(register_t) +
+	    roundup(sizeof(cloudabi64_tcb_t), sizeof(register_t));
+	(void)cpu_set_user_tls(td, (void *)stack);
+}
+
 static int
 cloudabi64_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 {
@@ -109,16 +149,29 @@ cloudabi64_schedtail(struct thread *td)
 	frame->tf_rdx = td->td_tid;
 }
 
-void
+int
 cloudabi64_thread_setregs(struct thread *td,
-    const cloudabi64_threadattr_t *attr)
+    const cloudabi64_threadattr_t *attr, uint64_t tcb)
 {
 	struct trapframe *frame;
 	stack_t stack;
+	uint64_t tcbptr;
+	int error;
+
+	/*
+	 * On x86-64, the TCB is referred to by %fs:0. Take some space
+	 * from the top of the stack to store a single element array,
+	 * containing a pointer to the TCB. %fs base will point to this.
+	 */
+	tcbptr = rounddown(attr->stack + attr->stack_size - sizeof(tcbptr),
+	    _Alignof(tcbptr));
+	error = copyout(&tcb, (void *)tcbptr, sizeof(tcb));
+	if (error != 0)
+		return (error);
 
 	/* Perform standard register initialization. */
 	stack.ss_sp = (void *)attr->stack;
-	stack.ss_size = attr->stack_size;
+	stack.ss_size = tcbptr - attr->stack;
 	cpu_set_upcall_kse(td, (void *)attr->entry_point, NULL, &stack);
 
 	/*
@@ -129,12 +182,14 @@ cloudabi64_thread_setregs(struct thread *td,
 	frame = td->td_frame;
 	frame->tf_rdi = td->td_tid;
 	frame->tf_rsi = attr->argument;
+
+	return (cpu_set_user_tls(td, (void *)tcbptr));
 }
 
 static struct sysentvec cloudabi64_elf_sysvec = {
 	.sv_size		= CLOUDABI64_SYS_MAXSYSCALL,
 	.sv_table		= cloudabi64_sysent,
-	.sv_fixup		= cloudabi64_fixup,
+	.sv_fixup		= cloudabi64_fixup_tcb,
 	.sv_name		= "CloudABI ELF64",
 	.sv_coredump		= elf64_coredump,
 	.sv_pagesize		= PAGE_SIZE,
@@ -143,6 +198,7 @@ static struct sysentvec cloudabi64_elf_sysvec = {
 	.sv_usrstack		= USRSTACK,
 	.sv_stackprot		= VM_PROT_READ | VM_PROT_WRITE,
 	.sv_copyout_strings	= cloudabi64_copyout_strings,
+	.sv_setregs		= cloudabi64_proc_setregs,
 	.sv_flags		= SV_ABI_CLOUDABI | SV_CAPSICUM | SV_LP64,
 	.sv_set_syscall_retval	= cloudabi64_set_syscall_retval,
 	.sv_fetch_syscall_args	= cloudabi64_fetch_syscall_args,

sys/arm64/cloudabi64/cloudabi64_sysvec.c

@@ -27,6 +27,7 @@
 __FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
+#include <sys/imgact.h>
 #include <sys/kernel.h>
 #include <sys/proc.h>
 #include <sys/sysent.h>
@@ -46,6 +47,25 @@ __FBSDID("$FreeBSD$");
 extern const char *cloudabi64_syscallnames[];
 extern struct sysent cloudabi64_sysent[];
 
+static void
+cloudabi64_proc_setregs(struct thread *td, struct image_params *imgp,
+    unsigned long stack)
+{
+	struct trapframe *regs;
+
+	exec_setregs(td, imgp, stack);
+
+	/*
+	 * The stack now contains a pointer to the TCB and the auxiliary
+	 * vector. Let x0 point to the auxiliary vector, and set
+	 * tpidr_el0 to the TCB.
+	 */
+	regs = td->td_frame;
+	regs->tf_x[0] = td->td_retval[0] =
+	    stack + roundup(sizeof(cloudabi64_tcb_t), sizeof(register_t));
+	(void)cpu_set_user_tls(td, (void *)stack);
+}
+
 static int
 cloudabi64_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 {
@@ -110,9 +130,9 @@ cloudabi64_schedtail(struct thread *td)
 	}
 }
 
-void
+int
 cloudabi64_thread_setregs(struct thread *td,
-    const cloudabi64_threadattr_t *attr)
+    const cloudabi64_threadattr_t *attr, uint64_t tcb)
 {
 	struct trapframe *frame;
 	stack_t stack;
@@ -130,6 +150,9 @@ cloudabi64_thread_setregs(struct thread *td,
 	frame = td->td_frame;
 	frame->tf_x[0] = td->td_tid;
 	frame->tf_x[1] = attr->argument;
+
+	/* Set up TLS. */
+	return (cpu_set_user_tls(td, (void *)tcb));
 }
 
 static struct sysentvec cloudabi64_elf_sysvec = {
@@ -144,6 +167,7 @@ static struct sysentvec cloudabi64_elf_sysvec = {
 	.sv_usrstack		= USRSTACK,
 	.sv_stackprot		= VM_PROT_READ | VM_PROT_WRITE,
 	.sv_copyout_strings	= cloudabi64_copyout_strings,
+	.sv_setregs		= cloudabi64_proc_setregs,
 	.sv_flags		= SV_ABI_CLOUDABI | SV_CAPSICUM | SV_LP64,
 	.sv_set_syscall_retval	= cloudabi64_set_syscall_retval,
 	.sv_fetch_syscall_args	= cloudabi64_fetch_syscall_args,

sys/compat/cloudabi64/cloudabi64_module.c

@@ -112,7 +112,13 @@ cloudabi64_fixup(register_t **stack_base, struct image_params *imgp)
 		{ .a_type = CLOUDABI_AT_NULL },
 	};
 	*stack_base -= howmany(sizeof(auxv), sizeof(register_t));
-	return (copyout(auxv, *stack_base, sizeof(auxv)));
+	error = copyout(auxv, *stack_base, sizeof(auxv));
+	if (error != 0)
+		return (error);
+
+	/* Reserve space for storing the TCB. */
+	*stack_base -= howmany(sizeof(cloudabi64_tcb_t), sizeof(register_t));
+	return (0);
 }
 
 static int

sys/compat/cloudabi64/cloudabi64_thread.c

@@ -37,6 +37,7 @@ __FBSDID("$FreeBSD$");
 
 struct thread_create_args {
 	cloudabi64_threadattr_t attr;
+	uint64_t tcb;
 	lwpid_t tid;
 };
 
@@ -49,8 +50,7 @@ initialize_thread(struct thread *td, void *thunk)
 	args->tid = td->td_tid;
 
 	/* Set up initial register contents. */
-	cloudabi64_thread_setregs(td, &args->attr);
-	return (0);
+	return (cloudabi64_thread_setregs(td, &args->attr, args->tcb));
 }
 
 int
@@ -63,6 +63,12 @@ cloudabi64_sys_thread_create(struct thread *td,
 	error = copyin(uap->attr, &args.attr, sizeof(args.attr));
 	if (error != 0)
 		return (error);
+
+	/* Remove some space on the top of the stack for the TCB. */
+	args.tcb = rounddown(args.attr.stack + args.attr.stack_size -
+	    sizeof(cloudabi64_tcb_t), _Alignof(cloudabi64_tcb_t));
+	args.attr.stack_size = args.tcb - args.attr.stack;
+
 	error = thread_create(td, NULL, initialize_thread, &args);
 	if (error != 0)
 		return (error);

sys/compat/cloudabi64/cloudabi64_util.h

@@ -42,7 +42,7 @@ extern Elf64_Brandinfo cloudabi64_brand;
 register_t *cloudabi64_copyout_strings(struct image_params *);
 int	cloudabi64_fixup(register_t **, struct image_params *);
 
-void	cloudabi64_thread_setregs(struct thread *,
-    const cloudabi64_threadattr_t *);
+int	cloudabi64_thread_setregs(struct thread *,
+    const cloudabi64_threadattr_t *, uint64_t);
 
 #endif

sys/contrib/cloudabi/cloudabi64_types.h

@@ -192,6 +192,13 @@ _Static_assert(offsetof(cloudabi64_subscription_t, proc_terminate.fd) == 16, "In
 _Static_assert(sizeof(cloudabi64_subscription_t) == 56, "Incorrect layout");
 _Static_assert(_Alignof(cloudabi64_subscription_t) == 8, "Incorrect layout");
 
+typedef struct {
+	_Alignas(8) uint64_t parent;
+} cloudabi64_tcb_t;
+_Static_assert(offsetof(cloudabi64_tcb_t, parent) == 0, "Incorrect layout");
+_Static_assert(sizeof(cloudabi64_tcb_t) == 8, "Incorrect layout");
+_Static_assert(_Alignof(cloudabi64_tcb_t) == 8, "Incorrect layout");
+
 typedef void cloudabi64_threadentry_t(cloudabi_tid_t tid, uint64_t aux);
 
 typedef struct {