fork: Allow ABI to specify fork return values for child.

At least Linux x86 ABI's does not use carry bit and expects that the dx register
is preserved. For this add a new sv_set_fork_retval hook and call it from cpu_fork().

Add a short comment about touching dx in x86_set_fork_retval(), for more details
see phab comments from kib@ and imp@.

Reviewed by:		kib
Differential revision:	https://reviews.freebsd.org/D31472
MFC after:		2 weeks

sys/amd64/amd64/elf_machdep.c

@@ -87,6 +87,7 @@ struct sysentvec elf64_freebsd_sysvec_la48 = {
 	.sv_stackgap	= elf64_stackgap,
 	.sv_onexec_old	= exec_onexec_old,
 	.sv_onexit	= exit_onexit,
+	.sv_set_fork_retval = x86_set_fork_retval,
 };
 
 struct sysentvec elf64_freebsd_sysvec_la57 = {
@@ -127,6 +128,7 @@ struct sysentvec elf64_freebsd_sysvec_la57 = {
 	.sv_stackgap	= elf64_stackgap,
 	.sv_onexec_old	= exec_onexec_old,
 	.sv_onexit	= exit_onexit,
+	.sv_set_fork_retval=  x86_set_fork_retval,
 };
 
 static void

sys/amd64/amd64/vm_machdep.c

@@ -245,9 +245,8 @@ cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
 	td2->td_frame = (struct trapframe *)td2->td_md.md_stack_base - 1;
 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
 
-	td2->td_frame->tf_rax = 0;		/* Child returns zero */
-	td2->td_frame->tf_rflags &= ~PSL_C;	/* success */
-	td2->td_frame->tf_rdx = 1;
+	/* Set child return values. */
+	p2->p_sysent->sv_set_fork_retval(td2);
 
 	/*
 	 * If the parent process has the trap bit set (i.e. a debugger
@@ -300,6 +299,16 @@ cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
 	 */
 }
 
+void
+x86_set_fork_retval(struct thread *td)
+{
+	struct trapframe *frame = td->td_frame;
+
+	frame->tf_rax = 0;		/* Child returns zero */
+	frame->tf_rflags &= ~PSL_C;	/* success */
+	frame->tf_rdx = 1;		/* System V emulation */
+}
+
 /*
  * Intercept the return address from a freshly forked process that has NOT
  * been scheduled yet.

sys/amd64/cloudabi32/cloudabi32_sysvec.c

@@ -36,6 +36,7 @@ __FBSDID("$FreeBSD$");
 #include <vm/pmap.h>
 
 #include <machine/frame.h>
+#include <machine/md_var.h>
 #include <machine/pcb.h>
 #include <machine/vmparam.h>
 
@@ -225,6 +226,7 @@ static struct sysentvec cloudabi32_elf_sysvec = {
 	.sv_fetch_syscall_args	= cloudabi32_fetch_syscall_args,
 	.sv_syscallnames	= cloudabi32_syscallnames,
 	.sv_schedtail		= cloudabi32_schedtail,
+	.sv_set_fork_retval	= x86_set_fork_retval,
 };
 
 INIT_SYSENTVEC(elf_sysvec, &cloudabi32_elf_sysvec);

sys/amd64/cloudabi64/cloudabi64_sysvec.c

@@ -36,6 +36,7 @@ __FBSDID("$FreeBSD$");
 #include <vm/pmap.h>
 
 #include <machine/frame.h>
+#include <machine/md_var.h>
 #include <machine/pcb.h>
 #include <machine/vmparam.h>
 
@@ -212,6 +213,7 @@ static struct sysentvec cloudabi64_elf_sysvec = {
 	.sv_fetch_syscall_args	= cloudabi64_fetch_syscall_args,
 	.sv_syscallnames	= cloudabi64_syscallnames,
 	.sv_schedtail		= cloudabi64_schedtail,
+	.sv_set_fork_retval	= x86_set_fork_retval,
 };
 
 INIT_SYSENTVEC(elf_sysvec, &cloudabi64_elf_sysvec);

sys/amd64/linux/linux_sysvec.c

@@ -125,6 +125,7 @@ static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
 static void	linux_exec_sysvec_init(void *param);
 static int	linux_on_exec_vmspace(struct proc *p,
 		    struct image_params *imgp);
+static void	linux_set_fork_retval(struct thread *td);
 static int	linux_vsyscall(struct thread *td);
 
 #define LINUX_T_UNKNOWN  255
@@ -269,6 +270,14 @@ linux_set_syscall_retval(struct thread *td, int error)
 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
 }
 
+static void
+linux_set_fork_retval(struct thread *td)
+{
+	struct trapframe *frame = td->td_frame;
+
+	frame->tf_rax = 0;
+}
+
 static int
 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
 {
@@ -790,6 +799,7 @@ struct sysentvec elf_linux_sysvec = {
 	.sv_onexit	= linux_on_exit,
 	.sv_ontdexit	= linux_thread_dtor,
 	.sv_setid_allowed = &linux_setid_allowed_query,
+	.sv_set_fork_retval = linux_set_fork_retval,
 };
 
 static int

sys/amd64/linux32/linux32_sysvec.c

@@ -128,6 +128,7 @@ static bool	linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
 static void	linux_vdso_install(const void *param);
 static void	linux_vdso_deinstall(const void *param);
 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
+static void	linux32_set_fork_retval(struct thread *td);
 static void	linux32_set_syscall_retval(struct thread *td, int error);
 
 #define LINUX_T_UNKNOWN  255
@@ -703,6 +704,14 @@ linux32_set_syscall_retval(struct thread *td, int error)
 	}
 }
 
+static void
+linux32_set_fork_retval(struct thread *td)
+{
+	struct trapframe *frame = td->td_frame;
+
+	frame->tf_rax = 0;
+}
+
 /*
  * Clear registers on exec
  * XXX copied from ia32_signal.c.
@@ -956,6 +965,7 @@ struct sysentvec elf_linux_sysvec = {
 	.sv_onexit	= linux_on_exit,
 	.sv_ontdexit	= linux_thread_dtor,
 	.sv_setid_allowed = &linux_setid_allowed_query,
+	.sv_set_fork_retval = linux32_set_fork_retval,
 };
 
 static int

sys/compat/ia32/ia32_sysvec.c

@@ -133,6 +133,7 @@ struct sysentvec ia32_freebsd_sysvec = {
 	.sv_stackgap	= elf32_stackgap,
 	.sv_onexec_old	= exec_onexec_old,
 	.sv_onexit	= exit_onexit,
+	.sv_set_fork_retval = x86_set_fork_retval,
 };
 INIT_SYSENTVEC(elf_ia32_sysvec, &ia32_freebsd_sysvec);
 

sys/i386/cloudabi32/cloudabi32_sysvec.c

@@ -36,6 +36,7 @@ __FBSDID("$FreeBSD$");
 #include <vm/pmap.h>
 
 #include <machine/frame.h>
+#include <machine/md_var.h>
 #include <machine/pcb.h>
 #include <machine/vmparam.h>
 
@@ -195,6 +196,7 @@ static struct sysentvec cloudabi32_elf_sysvec = {
 	.sv_fetch_syscall_args	= cloudabi32_fetch_syscall_args,
 	.sv_syscallnames	= cloudabi32_syscallnames,
 	.sv_schedtail		= cloudabi32_schedtail,
+	.sv_set_fork_retval	= x86_set_fork_retval,
 };
 
 INIT_SYSENTVEC(elf_sysvec, &cloudabi32_elf_sysvec);

sys/i386/i386/elf_machdep.c

@@ -88,6 +88,7 @@ struct sysentvec elf32_freebsd_sysvec = {
 	.sv_trap	= NULL,
 	.sv_onexec_old	= exec_onexec_old,
 	.sv_onexit	= exit_onexit,
+	.sv_set_fork_retval = x86_set_fork_retval,
 };
 INIT_SYSENTVEC(elf32_sysvec, &elf32_freebsd_sysvec);
 

sys/i386/i386/vm_machdep.c

@@ -258,9 +258,8 @@ cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
 	    VM86_STACK_SPACE) - 1;
 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
 
-	td2->td_frame->tf_eax = 0;		/* Child returns zero */
-	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
-	td2->td_frame->tf_edx = 1;
+	/* Set child return values. */
+	p2->p_sysent->sv_set_fork_retval(td2);
 
 	/*
 	 * If the parent process has the trap bit set (i.e. a debugger
@@ -302,6 +301,16 @@ cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
 	 */
 }
 
+void
+x86_set_fork_retval(struct thread *td)
+{
+	struct trapframe * frame = td->td_frame;
+
+	frame->tf_eax = 0;		/* Child returns zero */
+	frame->tf_eflags &= ~PSL_C;	/* success */
+	frame->tf_edx = 1;		/* System V emulation */
+}
+
 /*
  * Intercept the return address from a freshly forked process that has NOT
  * been scheduled yet.

sys/i386/linux/linux_sysvec.c

@@ -108,6 +108,7 @@ static int	linux_on_exec_vmspace(struct proc *p,
 		    struct image_params *imgp);
 static int	linux_copyout_strings(struct image_params *imgp,
 		    uintptr_t *stack_base);
+static void	linux_set_fork_retval(struct thread *td);
 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
 static void	linux_vdso_install(const void *param);
 static void	linux_vdso_deinstall(const void *param);
@@ -803,6 +804,14 @@ linux_set_syscall_retval(struct thread *td, int error)
 	}
 }
 
+static void
+linux_set_fork_retval(struct thread *td)
+{
+	struct trapframe *frame = td->td_frame;
+
+	frame->tf_eax = 0;
+}
+
 /*
  * exec_setregs may initialize some registers differently than Linux
  * does, thus potentially confusing Linux binaries. If necessary, we
@@ -856,6 +865,7 @@ struct sysentvec linux_sysvec = {
 	.sv_onexit	= linux_on_exit,
 	.sv_ontdexit	= linux_thread_dtor,
 	.sv_setid_allowed = &linux_setid_allowed_query,
+	.sv_set_fork_retval = linux_set_fork_retval,
 };
 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
 
@@ -898,6 +908,7 @@ struct sysentvec elf_linux_sysvec = {
 	.sv_onexit	= linux_on_exit,
 	.sv_ontdexit	= linux_thread_dtor,
 	.sv_setid_allowed = &linux_setid_allowed_query,
+	.sv_set_fork_retval = linux_set_fork_retval,
 };
 
 static int

sys/kern/imgact_aout.c

@@ -103,6 +103,7 @@ struct sysentvec aout_sysvec = {
 	.sv_trap	= NULL,
 	.sv_onexec_old = exec_onexec_old,
 	.sv_onexit =  exit_onexit,
+	.sv_set_fork_retval = x86_set_fork_retval,
 };
 
 #elif defined(__amd64__)
@@ -141,6 +142,7 @@ struct sysentvec aout_sysvec = {
 	.sv_syscallnames = freebsd32_syscallnames,
 	.sv_onexec_old	= exec_onexec_old,
 	.sv_onexit	= exit_onexit,
+	.sv_set_fork_retval = x86_set_fork_retval,
 };
 #else
 #error "Port me"

sys/kern/init_main.c

@@ -402,6 +402,12 @@ null_set_syscall_retval(struct thread *td __unused, int error __unused)
 	panic("null_set_syscall_retval");
 }
 
+static void
+null_set_fork_retval(struct thread *td __unused)
+{
+
+}
+
 struct sysentvec null_sysvec = {
 	.sv_size	= 0,
 	.sv_table	= NULL,
@@ -430,6 +436,7 @@ struct sysentvec null_sysvec = {
 	.sv_schedtail	= NULL,
 	.sv_thread_detach = NULL,
 	.sv_trap	= NULL,
+	.sv_set_fork_retval = null_set_fork_retval,
 };
 
 /*

sys/sys/sysent.h

@@ -156,6 +156,8 @@ struct sysentvec {
 	void		(*sv_ontdexit)(struct thread *td);
 	int		(*sv_setid_allowed)(struct thread *td,
 			    struct image_params *imgp);
+	void		(*sv_set_fork_retval)(struct thread *);
+					/* Only used on x86 */
 };
 
 #define	SV_ILP32	0x000100	/* 32-bit executable. */

sys/x86/include/x86_var.h

@@ -153,6 +153,7 @@ int	pti_get_default(void);
 int	user_dbreg_trap(register_t dr6);
 int	minidumpsys(struct dumperinfo *);
 struct pcb *get_pcb_td(struct thread *td);
+void	x86_set_fork_retval(struct thread *td);
 
 /*
  * MSR ops for x86_msr_op()