attaching to a running process.

Makefile

@@ -186,6 +186,7 @@ LIBMCOUNT_UTILS_SRCS += $(srcdir)/utils/rbtree.c $(srcdir)/utils/filter.c
 LIBMCOUNT_UTILS_SRCS += $(srcdir)/utils/demangle.c $(srcdir)/utils/utils.c
 LIBMCOUNT_UTILS_SRCS += $(srcdir)/utils/script.c $(srcdir)/utils/script-python.c $(srcdir)/utils/script-luajit.c
 LIBMCOUNT_UTILS_SRCS += $(srcdir)/utils/auto-args.c $(srcdir)/utils/dwarf.c
+LIBMCOUNT_UTILS_SRCS += $(srcdir)/utils/env-file.c
 LIBMCOUNT_UTILS_SRCS += $(wildcard $(srcdir)/utils/symbol*.c)
 LIBMCOUNT_UTILS_OBJS := $(patsubst $(srcdir)/utils/%.c,$(objdir)/libmcount/%.op,$(LIBMCOUNT_UTILS_SRCS))
 

arch/aarch64/mcount-arch.h

@@ -1,6 +1,20 @@
 #ifndef MCOUNT_ARCH_H
 #define MCOUNT_ARCH_H
 
+#include <sys/user.h>
+#define ARCH_REGS       struct user_regs_struct
+#define ARCH_PC_TYPE    typeof(((ARCH_REGS *)0)->pc)
+
+inline ARCH_PC_TYPE get_pc(ARCH_REGS regs)
+{
+	return regs.pc;
+}
+
+inline void set_pc(ARCH_REGS *regs, ARCH_PC_TYPE pc)
+{
+	regs->pc = pc;
+}
+
 #define mcount_regs  mcount_regs
 
 struct mcount_regs {

arch/arm/mcount-arch.h

@@ -1,6 +1,20 @@
 #ifndef MCOUNT_ARCH_H
 #define MCOUNT_ARCH_H
 
+#include <sys/user.h>
+#define ARCH_REGS       struct user_regs
+#define ARCH_PC_TYPE    typeof(((ARCH_REGS *)0)->uregs[0])
+
+inline ARCH_PC_TYPE get_pc(ARCH_REGS regs)
+{
+	return regs.uregs[15];
+}
+
+inline void set_pc(ARCH_REGS *regs, ARCH_PC_TYPE pc)
+{
+	regs->uregs[15] = pc;
+}
+
 #define mcount_regs  mcount_regs
 
 struct mcount_regs {

arch/i386/mcount-arch.h

@@ -2,6 +2,20 @@
 #ifndef __MCOUNT_ARCH_H__
 #define __MCOUNT_ARCH_H__
 
+#include <sys/user.h>
+#define ARCH_REGS       struct user_regs_struct
+#define ARCH_PC_TYPE    typeof(((ARCH_REGS *)0)->eip)
+
+inline ARCH_PC_TYPE get_pc(ARCH_REGS regs)
+{
+	return regs.eip;
+}
+
+inline void set_pc(ARCH_REGS *regs, ARCH_PC_TYPE pc)
+{
+	regs->eip = pc;
+}
+
 #define mcount_regs  mcount_regs
 
 struct mcount_regs {

arch/x86_64/Makefile

@@ -8,17 +8,19 @@ include $(srcdir)/Makefile.include
 ARCH_ENTRY_SRC = $(wildcard $(sdir)/*.S)
 ARCH_MCOUNT_SRC = $(wildcard $(sdir)/mcount-*.c) $(sdir)/symbol.c
 ARCH_UFTRACE_SRC = $(sdir)/cpuinfo.c $(sdir)/symbol.c
+ARCH_ATTACH_ASM = $(wildcard $(sdir)/inject*.S)
 
 ARCH_MCOUNT_OBJS  = $(patsubst $(sdir)/%.S,$(odir)/%.op,$(ARCH_ENTRY_SRC))
 ARCH_MCOUNT_OBJS += $(patsubst $(sdir)/%.c,$(odir)/%.op,$(ARCH_MCOUNT_SRC))
 ARCH_UFTRACE_OBJS = $(patsubst $(sdir)/%.c,$(odir)/%.o,$(ARCH_UFTRACE_SRC))
+ARCH_ATTACH_OBJS  = $(patsubst $(sdir)/%.S,$(odir)/%.op,$(ARCH_ATTACH_ASM))
 
 all: $(odir)/entry.op
 
 $(odir)/mcount-entry.op: $(ARCH_MCOUNT_OBJS)
 	$(QUIET_LINK)$(LD) $(LINKFLAGS) -o $@ $^
 
-$(odir)/uftrace.o: $(ARCH_UFTRACE_OBJS)
+$(odir)/uftrace.o: $(ARCH_UFTRACE_OBJS) $(ARCH_ATTACH_OBJS)
 	$(QUIET_LINK)$(LD) $(LINKFLAGS) -o $@ $^
 
 $(odir)/%.op: $(sdir)/%.S

arch/x86_64/inject-content.S

@@ -0,0 +1,94 @@
+#include "utils/asm.h"
+
+	.file	"inject.c"
+	.section .data
+FUNC(inject_contents_start)
+END(inject_contents_start)
+
+FUNC(inject_so_loader)
+	/*
+	 * The PC of the target program is changed by uftrace to reach here
+	 * to make load the shared object module the libmcount. uftrace use
+	 * ptrace to do this. but sometime target program have not jump to PC
+	 * which assigned by uftrace. there is some different between assigned
+	 * address and real jump address. maybe it cause by ptrace inner work.
+	 *
+	 * [NOTICE]
+	 * nop instructions has been added to mitigate this problem.
+	 */
+	nop
+	nop
+	nop
+
+	/*
+	 * save registers what used in below codes
+	 */
+	pushq %rax
+	pushq %rsi
+	pushq %rdi
+	pushq %r9
+	/* create new stack frame to save registers */
+	pushq %rsp
+	pushq %rbp
+	movq %rsp, %rbp
+
+	/*
+	 * since recently dlopen use movabs instruction,
+	 * stack must be aligned by 16byte.
+	 */
+	andq $-16, %rsp
+
+	/*
+	 * 1st argument to dlopen():
+	 * absolute path of shared object to loaded.
+	 */
+	leaq inject_so_path(%rip), %rdi
+	/*
+	 * 2nd argument
+	 * 0x80000000 : __RTLD_DLOPEN
+	 * 0x00000001 : RTLD_NOW
+	 */
+	movabs $0x80000001, %rsi
+	movq inject_dlopen_addr(%rip), %r9
+
+	/*
+	 * call dlopen()
+	 */
+	callq *%r9
+
+	/*
+	 * rewind stack
+	 */
+	movq %rbp, %rsp
+	popq %rbp
+	popq %rsp
+	/*
+	 * restore used registers.
+	 */
+	popq %r9
+	popq %rdi
+	popq %rsi
+	popq %rax
+
+	/*
+	 * return to original control flow that
+	 * before to trapped by ptrace.
+	 */
+	jmpq *0(%rip)
+END(inject_so_loader)
+
+OBJECT(inject_so_loader_ret)
+	.zero	8
+END(inject_so_loader_ret)
+
+OBJECT(inject_dlopen_addr)
+	.zero	8
+END(inject_dlopen_addr)
+
+OBJECT(inject_so_path)
+	.string "absolute path of shared object"
+	.zero	97
+END(inject_so_path)
+
+FUNC(inject_contents_end)
+END(inject_contents_end)

arch/x86_64/mcount-arch.h

@@ -1,9 +1,23 @@
 #ifndef MCOUNT_ARCH_H
 #define MCOUNT_ARCH_H
 
-#include "utils/arch.h"
+#include <sys/user.h>
+#include <stdbool.h>
 #include "utils/list.h"
 
+#define ARCH_REGS	struct user_regs_struct
+#define ARCH_PC_TYPE	typeof(((ARCH_REGS *)0)->rip)
+
+inline ARCH_PC_TYPE get_pc(ARCH_REGS regs)
+{
+	return regs.rip;
+}
+
+inline void set_pc(ARCH_REGS *regs, ARCH_PC_TYPE pc)
+{
+	regs->rip = pc;
+}
+
 #define mcount_regs  mcount_regs
 
 struct mcount_regs {
@@ -37,7 +51,7 @@ struct mcount_arch_context {
 #define ARCH_CAN_RESTORE_PLTHOOK   1
 
 struct plthook_arch_context {
-	bool	has_plt_sec;
+	bool    has_plt_sec;
 };
 
 struct mcount_disasm_engine;

arch/x86_64/mcount-dynamic.c

@@ -11,7 +11,6 @@
 #include "utils/utils.h"
 #include "utils/symbol.h"
 
-#define PAGE_SIZE  4096
 #define XRAY_SECT  "xray_instr_map"
 #define MCOUNTLOC_SECT  "__mcount_loc"
 

arch/x86_64/mcount-event.c

@@ -11,7 +11,6 @@
 #include "libmcount/internal.h"
 
 #define INVALID_OPCODE  0xce
-#define PAGE_SIZE       4096
 #define PAGE_ADDR(a)    ((void *)((a) & ~(PAGE_SIZE - 1)))
 
 static void sdt_handler(int sig, siginfo_t *info, void *arg)