首页
论坛
课程
招聘
[原创]Linux系统调用机制浅析
2021-10-27 17:08 5646

[原创]Linux系统调用机制浅析

erfze 活跃值
12
2021-10-27 17:08
5646

0x00 前言

本文不会介绍CPU特权级别,中断,MSR,段机制及页机制等相关前置知识,如果读者此前未接触过这些,建议阅读Intel SDM对应篇章或者参阅链接<SUP>1</SUP>之后再继续下面篇幅。本文基于如下环境:

  • CPU:Intel
  • Kernel Version:4.15.0
  • Debugging Env:Ubuntu 20.04.02 x64(Kernel Version—5.11.0)

0x01 INT $0x80

a. 源码分析

首先从源码角度分析传统系统调用,即int 0x80。IDT(Interrupt Descriptor Table)建立位于arch/x86/kernel/traps.c中:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
void __init trap_init(void)
{
    /* Init cpu_entry_area before IST entries are set up */
    setup_cpu_entry_areas();
 
    idt_setup_traps();
 
    /*
     * Set the IDT descriptor to a fixed read-only location, so that the
     * "sidt" instruction will not leak the location of the kernel, and
     * to defend the IDT against arbitrary memory write vulnerabilities.
     * It will be reloaded in cpu_init() */
    cea_set_pte(CPU_ENTRY_AREA_RO_IDT_VADDR, __pa_symbol(idt_table),
            PAGE_KERNEL_RO);
    idt_descr.address = CPU_ENTRY_AREA_RO_IDT;
 
    /*
     * Should be a barrier for any external CPU state:
     */
    cpu_init();
 
    idt_setup_ist_traps();
 
    x86_init.irqs.trap_init();
 
    idt_setup_debugidt_traps();
}

idt_setup_traps()函数定义在arch/x86/kernel/idt.c中:

1
2
3
4
5
6
7
/**
 * idt_setup_traps - Initialize the idt table with default traps
 */
void __init idt_setup_traps(void)
{
    idt_setup_from_table(idt_table, def_idts, ARRAY_SIZE(def_idts), true);
}

其调用idt_setup_from_table函数同样位于该文件:

1
2
3
4
5
6
7
8
9
10
11
12
static void
idt_setup_from_table(gate_desc *idt, const struct idt_data *t, int size, bool sys)
{
    gate_desc desc;
 
    for (; size > 0; t++, size--) {
        idt_init_desc(&desc, t);
        write_idt_entry(idt, t->vector, &desc);
        if (sys)
            set_bit(t->vector, system_vectors);
    }
}

def_idts存储了IDT各项默认值,其定义如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
/*
 * The default IDT entries which are set up in trap_init() before
 * cpu_init() is invoked. Interrupt stacks cannot be used at that point and
 * the traps which use them are reinitialized with IST after cpu_init() has
 * set up TSS.
 */
static const __initconst struct idt_data def_idts[] = {
    INTG(X86_TRAP_DE,        divide_error),
    INTG(X86_TRAP_NMI,        nmi),
    INTG(X86_TRAP_BR,        bounds),
    INTG(X86_TRAP_UD,        invalid_op),
    INTG(X86_TRAP_NM,        device_not_available),
    INTG(X86_TRAP_OLD_MF,        coprocessor_segment_overrun),
    INTG(X86_TRAP_TS,        invalid_TSS),
    INTG(X86_TRAP_NP,        segment_not_present),
    INTG(X86_TRAP_SS,        stack_segment),
    INTG(X86_TRAP_GP,        general_protection),
    INTG(X86_TRAP_SPURIOUS,        spurious_interrupt_bug),
    INTG(X86_TRAP_MF,        coprocessor_error),
    INTG(X86_TRAP_AC,        alignment_check),
    INTG(X86_TRAP_XF,        simd_coprocessor_error),
 
#ifdef CONFIG_X86_32
    TSKG(X86_TRAP_DF,        GDT_ENTRY_DOUBLEFAULT_TSS),
#else
    INTG(X86_TRAP_DF,        double_fault),
#endif
    INTG(X86_TRAP_DB,        debug),
 
#ifdef CONFIG_X86_MCE
    INTG(X86_TRAP_MC,        &machine_check),
#endif
 
    SYSG(X86_TRAP_OF,        overflow),
#if defined(CONFIG_IA32_EMULATION)
    SYSG(IA32_SYSCALL_VECTOR,    entry_INT80_compat),
#elif defined(CONFIG_X86_32)
    SYSG(IA32_SYSCALL_VECTOR,    entry_INT80_32),
#endif
};

根据配置选项不同,IA32_SYSCALL_VECTOR项值不同——若启用CONFIG_IA32_EMULATION,则以64位兼容模式运行32位程序;否则是32位。IA32_SYSCALL_VECTOR定义如下:

1
#define IA32_SYSCALL_VECTOR        0x80

INTGSYSG定义不同之处在于DPL:

1
2
3
4
5
6
7
/* Interrupt gate */
#define INTG(_vector, _addr)                \
    G(_vector, _addr, DEFAULT_STACK, GATE_INTERRUPT, DPL0, __KERNEL_CS)
 
/* System interrupt gate */
#define SYSG(_vector, _addr)                \
    G(_vector, _addr, DEFAULT_STACK, GATE_INTERRUPT, DPL3, __KERNEL_CS)

相关定义如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
#define DPL0        0x0
#define DPL3        0x3
 
#define DEFAULT_STACK    0
 
#define G(_vector, _addr, _ist, _type, _dpl, _segment)    \
    {                        \
        .vector        = _vector,        \
        .bits.ist    = _ist,            \
        .bits.type    = _type,        \
        .bits.dpl    = _dpl,            \
        .bits.p        = 1,            \
        .addr        = _addr,        \
        .segment    = _segment,        \
    }

门描述符及类型定义如下(位于/arch/x86/include/asm/desc_defs.h):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
struct gate_struct {
    u16        offset_low;
    u16        segment;
    struct idt_bits    bits;
    u16        offset_middle;
#ifdef CONFIG_X86_64
    u32        offset_high;
    u32        reserved;
#endif
} __attribute__((packed));
 
enum {
    GATE_INTERRUPT = 0xE,
    GATE_TRAP = 0xF,
    GATE_CALL = 0xC,
    GATE_TASK = 0x5,
};

对应于Intel SDM中:

 

 

idt_init_desc函数定义如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
static inline void idt_init_desc(gate_desc *gate, const struct idt_data *d)
{
    unsigned long addr = (unsigned long) d->addr;
 
    gate->offset_low    = (u16) addr;
    gate->segment        = (u16) d->segment;
    gate->bits        = d->bits;
    gate->offset_middle    = (u16) (addr >> 16);
#ifdef CONFIG_X86_64
    gate->offset_high    = (u32) (addr >> 32);
    gate->reserved        = 0;
#endif
}

write_idt_entrymemcpy函数的简单包装:

1
2
3
4
5
6
#define write_idt_entry(dt, entry, g)        native_write_idt_entry(dt, entry, g)
......
static inline void native_write_idt_entry(gate_desc *idt, int entry, const gate_desc *gate)
{
    memcpy(&idt[entry], gate, sizeof(*gate));
}

如此一来,便在IDT 0x80项写入了系统调用函数地址。上述函数调用关系为:

 


 

entry_INT80_32定义位于arch/x86/entry/entry_32.S文件中:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
ENTRY(entry_INT80_32)
    ASM_CLAC
    pushl    %eax            /* pt_regs->orig_ax */
    SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 
    /*
     * User mode is traced as though IRQs are on, and the interrupt gate
     * turned them off.
     */
    TRACE_IRQS_OFF
 
    movl    %esp, %eax
    call    do_int80_syscall_32
.Lsyscall_32_done:
 
restore_all:
    TRACE_IRQS_IRET
.Lrestore_all_notrace:
#ifdef CONFIG_X86_ESPFIX32
    ALTERNATIVE    "jmp .Lrestore_nocheck", "", X86_BUG_ESPFIX
 
    movl    PT_EFLAGS(%esp), %eax        # mix EFLAGS, SS and CS
    /*
     * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
     * are returning to the kernel.
     * See comments in process.c:copy_thread() for details.
     */
    movb    PT_OLDSS(%esp), %ah
    movb    PT_CS(%esp), %al
    andl    $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
    cmpl    $((SEGMENT_LDT << 8) | USER_RPL), %eax
    je .Lldt_ss                # returning to user-space with LDT SS
#endif
.Lrestore_nocheck:
    RESTORE_REGS 4                # skip orig_eax/error_code
.Lirq_return:
    INTERRUPT_RETURN
 
.section .fixup, "ax"
ENTRY(iret_exc    )
    pushl    $0                # no error code
    pushl    $do_iret_error
    jmp    common_exception
.previous
    _ASM_EXTABLE(.Lirq_return, iret_exc)
 
#ifdef CONFIG_X86_ESPFIX32
.Lldt_ss:
/*
 * Setup and switch to ESPFIX stack
 *
 * We're returning to userspace with a 16 bit stack. The CPU will not
 * restore the high word of ESP for us on executing iret... This is an
 * "official" bug of all the x86-compatible CPUs, which we can work
 * around to make dosemu and wine happy. We do this by preloading the
 * high word of ESP with the high word of the userspace ESP while
 * compensating for the offset by changing to the ESPFIX segment with
 * a base address that matches for the difference.
 */
#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
    mov    %esp, %edx            /* load kernel esp */
    mov    PT_OLDESP(%esp), %eax        /* load userspace esp */
    mov    %dx, %ax            /* eax: new kernel esp */
    sub    %eax, %edx            /* offset (low word is 0) */
    shr    $16, %edx
    mov    %dl, GDT_ESPFIX_SS + 4        /* bits 16..23 */
    mov    %dh, GDT_ESPFIX_SS + 7        /* bits 24..31 */
    pushl    $__ESPFIX_SS
    pushl    %eax                /* new kernel esp */
    /*
     * Disable interrupts, but do not irqtrace this section: we
     * will soon execute iret and the tracer was already set to
     * the irqstate after the IRET:
     */
    DISABLE_INTERRUPTS(CLBR_ANY)
    lss    (%esp), %esp            /* switch to espfix segment */
    jmp    .Lrestore_nocheck
#endif
ENDPROC(entry_INT80_32)

执行系统调用的主要代码位于do_int80_syscall_32(arch/x86/entry/common.c):

1
2
3
4
5
6
7
/* Handles int $0x80 */
__visible void do_int80_syscall_32(struct pt_regs *regs)
{
    enter_from_user_mode();
    local_irq_enable();
    do_syscall_32_irqs_on(regs);
}

do_syscall_32_irqs_on定义如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
/*
 * Does a 32-bit syscall.  Called with IRQs on in CONTEXT_KERNEL.  Does
 * all entry and exit work and returns with IRQs off.  This function is
 * extremely hot in workloads that use it, and it's usually called from
 * do_fast_syscall_32, so forcibly inline it to improve performance.
 */
static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
{
    struct thread_info *ti = current_thread_info();
    unsigned int nr = (unsigned int)regs->orig_ax;
 
#ifdef CONFIG_IA32_EMULATION
    current->thread.status |= TS_COMPAT;
#endif
 
    if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
        /*
         * Subtlety here: if ptrace pokes something larger than
         * 2^32-1 into orig_ax, this truncates it.  This may or
         * may not be necessary, but it matches the old asm
         * behavior.
         */
        nr = syscall_trace_enter(regs);
    }
 
    if (likely(nr < IA32_NR_syscalls)) {
        /*
         * It's possible that a 32-bit syscall implementation
         * takes a 64-bit parameter but nonetheless assumes that
         * the high bits are zero.  Make sure we zero-extend all
         * of the args.
         */
        regs->ax = ia32_sys_call_table[nr](
            (unsigned int)regs->bx, (unsigned int)regs->cx,
            (unsigned int)regs->dx, (unsigned int)regs->si,
            (unsigned int)regs->di, (unsigned int)regs->bp);
    }
 
    syscall_return_slowpath(regs);
}

上述函数调用关系为:

 

 

ia32_sys_call_table定义位于同目录的syscall_32.c文件中:

1
2
3
4
5
6
7
8
9
10
extern asmlinkage long sys_ni_syscall(unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long);
 
__visible const sys_call_ptr_t ia32_sys_call_table[__NR_syscall_compat_max+1] = {
    /*
     * Smells like a compiler bug -- it doesn't work
     * when the & below is removed.
     */
    [0 ... __NR_syscall_compat_max] = &sys_ni_syscall,
#include <asm/syscalls_32.h>
};

sys_ni_syscall(kernel/sys_ni.c)定义如下,对应于未实现的系统调用:

1
2
3
4
5
6
7
/*
 * Non-implemented system calls get redirected here.
 */
asmlinkage long sys_ni_syscall(void)
{
    return -ENOSYS;
}

asm/syscalls_32.h文件内容由syscalltbl.sh脚本根据syscall_32.tbl生成,具体定义在arch/x86/entry/syscalls/Makefile中:

1
2
3
4
5
6
7
8
9
10
syscall32 := $(srctree)/$(src)/syscall_32.tbl
syscall64 := $(srctree)/$(src)/syscall_64.tbl
 
syshdr := $(srctree)/$(src)/syscallhdr.sh
systbl := $(srctree)/$(src)/syscalltbl.sh
......
$(out)/syscalls_32.h: $(syscall32) $(systbl)
    $(call if_changed,systbl)
$(out)/syscalls_64.h: $(syscall64) $(systbl)
    $(call if_changed,systbl)

syscall_32.tbl中存储了系统调用名称,调用号及入口等内容:

 

 

syscall_32.c文件中有如下宏定义:

1
2
3
4
5
#define __SYSCALL_I386(nr, sym, qual) extern asmlinkage long sym(unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long) ;
#include <asm/syscalls_32.h>
#undef __SYSCALL_I386
 
#define __SYSCALL_I386(nr, sym, qual) [nr] = sym,

那么ia32_sys_call_table数组内容会成为如下形式:

1
2
3
4
[0 ... __NR_syscall_compat_max] = &sys_ni_syscall,
[0] = sys_restart_syscall,
[1] = sys_exit,
......

#define __SYSCALL_I386(nr, sym, qual) [nr] = sym,宏定义了ia32_sys_call_table数组项——以系统调用号为索引;#define __SYSCALL_I386(nr, sym, qual) extern asmlinkage long sym(unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long);定义了每项中系统调用函数Entry Point。

 

如此一来,ia32_sys_call_table[nr]((unsigned int)regs->bx, (unsigned int)regs->cx,(unsigned int)regs->dx, (unsigned int)regs->si,(unsigned int)regs->di, (unsigned int)regs->bp);便会调用真正实现功能函数。以sys_restart_syscall为例,其定义位于kernel/signal.c中:

1
2
3
4
5
6
7
8
/**
 *  sys_restart_syscall - restart a system call
 */
SYSCALL_DEFINE0(restart_syscall)
{
    struct restart_block *restart = &current->restart_block;
    return restart->fn(restart);
}

SYSCALL_DEFINE相关宏定义位于include/linux/syscalls.h中:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
#define SYSCALL_METADATA(sname, nb, ...)
 
static inline int is_syscall_trace_event(struct trace_event_call *tp_event)
{
    return 0;
}
#endif
 
#define SYSCALL_DEFINE0(sname)                    \
    SYSCALL_METADATA(_##sname, 0);                \
    asmlinkage long sys_##sname(void)
 
#define SYSCALL_DEFINE1(name, ...) SYSCALL_DEFINEx(1, _##name, __VA_ARGS__)
#define SYSCALL_DEFINE2(name, ...) SYSCALL_DEFINEx(2, _##name, __VA_ARGS__)
#define SYSCALL_DEFINE3(name, ...) SYSCALL_DEFINEx(3, _##name, __VA_ARGS__)
#define SYSCALL_DEFINE4(name, ...) SYSCALL_DEFINEx(4, _##name, __VA_ARGS__)
#define SYSCALL_DEFINE5(name, ...) SYSCALL_DEFINEx(5, _##name, __VA_ARGS__)
#define SYSCALL_DEFINE6(name, ...) SYSCALL_DEFINEx(6, _##name, __VA_ARGS__)
 
#define SYSCALL_DEFINE_MAXARGS    6
 
#define SYSCALL_DEFINEx(x, sname, ...)                \
    SYSCALL_METADATA(sname, x, __VA_ARGS__)            \
    __SYSCALL_DEFINEx(x, sname, __VA_ARGS__)
 
#define __PROTECT(...) asmlinkage_protect(__VA_ARGS__)
#define __SYSCALL_DEFINEx(x, name, ...)                    \
    asmlinkage long sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))    \
        __attribute__((alias(__stringify(SyS##name))));        \
    static inline long SYSC##name(__MAP(x,__SC_DECL,__VA_ARGS__));    \
    asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__));    \
    asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__))    \
    {                                \
        long ret = SYSC##name(__MAP(x,__SC_CAST,__VA_ARGS__));    \
        __MAP(x,__SC_TEST,__VA_ARGS__);                \
        __PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__));    \
        return ret;                        \
    }                                \
    static inline long SYSC##name(__MAP(x,__SC_DECL,__VA_ARGS__))

系统调用返回是通过IRET语句:

 

 

其弹出寄存器值在发生中断时已经保存在栈中:

 

b. 动态调试

下面通过动态调试(调试环境使用Qemu+GDB+Busybox搭建)来剖析传统系统调用过程。于entry_INT80_32设置断点后,键入clear命令,成功断下:

 

 

查看栈中各寄存器值:

 

 

确为INT $0x80传统系统调用:

 

 

保存系统调用号及相关寄存器值:

 

 

传递regs参数给do_int80_syscall_32及引用其成员值:

 

 

对应源码为:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
unsigned int nr = (unsigned int)regs->orig_ax;
......
if (likely(nr < IA32_NR_syscalls)) {
        /*
         * It's possible that a 32-bit syscall implementation
         * takes a 64-bit parameter but nonetheless assumes that
         * the high bits are zero.  Make sure we zero-extend all
         * of the args.
         */
        regs->ax = ia32_sys_call_table[nr](
            (unsigned int)regs->bx, (unsigned int)regs->cx,
            (unsigned int)regs->dx, (unsigned int)regs->si,
            (unsigned int)regs->di, (unsigned int)regs->bp);
    }

pt_regs结构定义如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
struct pt_regs {
    /*
     * NB: 32-bit x86 CPUs are inconsistent as what happens in the
     * following cases (where %seg represents a segment register):
     *
     * - pushl %seg: some do a 16-bit write and leave the high
     *   bits alone
     * - movl %seg, [mem]: some do a 16-bit write despite the movl
     * - IDT entry: some (e.g. 486) will leave the high bits of CS
     *   and (if applicable) SS undefined.
     *
     * Fortunately, x86-32 doesn't read the high bits on POP or IRET,
     * so we can just treat all of the segment registers as 16-bit
     * values.
     */
    unsigned long bx;
    unsigned long cx;
    unsigned long dx;
    unsigned long si;
    unsigned long di;
    unsigned long bp;
    unsigned long ax;
    unsigned short ds;
    unsigned short __dsh;
    unsigned short es;
    unsigned short __esh;
    unsigned short fs;
    unsigned short __fsh;
    unsigned short gs;
    unsigned short __gsh;
    unsigned long orig_ax;
    unsigned long ip;
    unsigned short cs;
    unsigned short __csh;
    unsigned long flags;
    unsigned long sp;
    unsigned short ss;
    unsigned short __ssh;
};

之后便是根据系统调用号进入真正实现功能函数:

 

 

 

检查EFLAGS中VM位,SS中TI位是否设置为1以及CS中RPL:

 

 

 

若TI位未设置,则使用GDT进行索引。之后恢复SAVE_ALL所保存的寄存器值(出栈及入栈顺序与pt_regs中所定义顺序一致)并执行IRET指令返回调用程序:

 

 

返回值则在之前由do_syscall_32_irqs_on函数保存在了栈中:

 

 

RESTORE_REGS恢复寄存器值时将其弹出到EAX以传递给调用程序。

0x02 SYSENTER

a. 源码分析

根据Intel SDM中描述,使用SYSENTER命令需要事先设置如下三个MSR寄存器值;

 

 

执行到SYSENTER命令时操作如下:

 

 

Linux源码中设置三个MSR寄存器值操作位于syscall_init函数(arch/x86/kernel/cpu/common.c)中:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
#ifdef CONFIG_IA32_EMULATION
    wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
    /*
     * This only works on Intel CPUs.
     * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
     * This does not cause SYSENTER to jump to the wrong location, because
     * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
     */
    wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
    wrmsrl_safe(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1));
    wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
#else
    wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
    wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
    wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
    wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
#endif

编译时需要启用CONFIG_IA32_EMULATION选项。entry_SYSENTER_compat定义位于arch/x86/entry/entry_64_compat.S中:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
/*
 * 32-bit SYSENTER entry.
 *
 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
 * on 64-bit kernels running on Intel CPUs.
 *
 * The SYSENTER instruction, in principle, should *only* occur in the
 * vDSO.  In practice, a small number of Android devices were shipped
 * with a copy of Bionic that inlined a SYSENTER instruction.  This
 * never happened in any of Google's Bionic versions -- it only happened
 * in a narrow range of Intel-provided versions.
 *
 * SYSENTER loads SS, RSP, CS, and RIP from previously programmed MSRs.
 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
 * SYSENTER does not save anything on the stack,
 * and does not save old RIP (!!!), RSP, or RFLAGS.
 *
 * Arguments:
 * eax  system call number
 * ebx  arg1
 * ecx  arg2
 * edx  arg3
 * esi  arg4
 * edi  arg5
 * ebp  user stack
 * 0(%ebp) arg6
 */
ENTRY(entry_SYSENTER_compat)
    /* Interrupts are off on entry. */
    SWAPGS
 
    /* We are about to clobber %rsp anyway, clobbering here is OK */
    SWITCH_TO_KERNEL_CR3 scratch_reg=%rsp
 
    movq    PER_CPU_VAR(cpu_current_top_of_stack), %rsp
 
    /*
     * User tracing code (ptrace or signal handlers) might assume that
     * the saved RAX contains a 32-bit number when we're invoking a 32-bit
     * syscall.  Just in case the high bits are nonzero, zero-extend
     * the syscall number.  (This could almost certainly be deleted
     * with no ill effects.)
     */
    movl    %eax, %eax
 
    /* Construct struct pt_regs on stack */
    pushq    $__USER32_DS        /* pt_regs->ss */
    pushq    %rbp            /* pt_regs->sp (stashed in bp) */
 
    /*
     * Push flags.  This is nasty.  First, interrupts are currently
     * off, but we need pt_regs->flags to have IF set.  Second, even
     * if TF was set when SYSENTER started, it's clear by now.  We fix
     * that later using TIF_SINGLESTEP.
     */
    pushfq                /* pt_regs->flags (except IF = 0) */
    orl    $X86_EFLAGS_IF, (%rsp)    /* Fix saved flags */
    pushq    $__USER32_CS        /* pt_regs->cs */
    pushq    $0            /* pt_regs->ip = 0 (placeholder) */
    pushq    %rax            /* pt_regs->orig_ax */
    pushq    %rdi            /* pt_regs->di */
    pushq    %rsi            /* pt_regs->si */
    pushq    %rdx            /* pt_regs->dx */
    pushq    %rcx            /* pt_regs->cx */
    pushq    $-ENOSYS        /* pt_regs->ax */
    pushq   $0            /* pt_regs->r8  = 0 */
    pushq   $0            /* pt_regs->r9  = 0 */
    pushq   $0            /* pt_regs->r10 = 0 */
    pushq   $0            /* pt_regs->r11 = 0 */
    pushq   %rbx                    /* pt_regs->rbx */
    pushq   %rbp                    /* pt_regs->rbp (will be overwritten) */
    pushq   $0            /* pt_regs->r12 = 0 */
    pushq   $0            /* pt_regs->r13 = 0 */
    pushq   $0            /* pt_regs->r14 = 0 */
    pushq   $0            /* pt_regs->r15 = 0 */
    cld
 
    /*
     * SYSENTER doesn't filter flags, so we need to clear NT and AC
     * ourselves.  To save a few cycles, we can check whether
     * either was set instead of doing an unconditional popfq.
     * This needs to happen before enabling interrupts so that
     * we don't get preempted with NT set.
     *
     * If TF is set, we will single-step all the way to here -- do_debug
     * will ignore all the traps.  (Yes, this is slow, but so is
     * single-stepping in general.  This allows us to avoid having
     * a more complicated code to handle the case where a user program
     * forces us to single-step through the SYSENTER entry code.)
     *
     * NB.: .Lsysenter_fix_flags is a label with the code under it moved
     * out-of-line as an optimization: NT is unlikely to be set in the
     * majority of the cases and instead of polluting the I$ unnecessarily,
     * we're keeping that code behind a branch which will predict as
     * not-taken and therefore its instructions won't be fetched.
     */
    testl    $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, EFLAGS(%rsp)
    jnz    .Lsysenter_fix_flags
.Lsysenter_flags_fixed:
 
    /*
     * User mode is traced as though IRQs are on, and SYSENTER
     * turned them off.
     */
    TRACE_IRQS_OFF
 
    movq    %rsp, %rdi
    call    do_fast_syscall_32
    /* XEN PV guests always use IRET path */
    ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
            "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
    jmp    sysret32_from_system_call
 
.Lsysenter_fix_flags:
    pushq    $X86_EFLAGS_FIXED
    popfq
    jmp    .Lsysenter_flags_fixed
GLOBAL(__end_entry_SYSENTER_compat)
ENDPROC(entry_SYSENTER_compat)

关于SWAPGS可阅读参阅链接<SUP>7</SUP>:

 

 

do_fast_syscall_32函数会调用do_syscall_32_irqs_on

1
2
3
4
5
6
7
8
/* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
__visible long do_fast_syscall_32(struct pt_regs *regs)
{
......
    /* Now this is just like a normal syscall. */
    do_syscall_32_irqs_on(regs);
......
}

该函数其余代码部分见后文描述。

b. 动态调试

使用如下代码作为示例(不建议这样去执行系统调用,下面的代码仅仅是作为展示):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
int
main(int argc, char *argv[])
{
  unsigned long syscall_nr = 1;
  long exit_status = 44;
 
  asm ("movl %0, %%eax\n"
       "movl %1, %%ebx\n"
       "sysenter"
    : /* output parameters, we aren't outputting anything, no none */
      /* (none) */
    : /* input parameters mapped to %0 and %1, repsectively */
      "m" (syscall_nr), "m" (exit_status)
    : /* registers that we are "clobbering", unneeded since we are calling exit */
      "eax", "ebx");
}

entry_SYSENTER_compat成功断下:

 

 

regs传递给do_fast_syscall_32

 

 

 

可以看到其orig_ax成员偏移与之前相比发生了变化, 这是因为regs对应结构定义为:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
struct pt_regs {
/*
 * C ABI says these regs are callee-preserved. They aren't saved on kernel entry
 * unless syscall needs a complete, fully filled "struct pt_regs".
 */
    unsigned long r15;
    unsigned long r14;
    unsigned long r13;
    unsigned long r12;
    unsigned long bp;
    unsigned long bx;
/* These regs are callee-clobbered. Always saved on kernel entry. */
    unsigned long r11;
    unsigned long r10;
    unsigned long r9;
    unsigned long r8;
    unsigned long ax;
    unsigned long cx;
    unsigned long dx;
    unsigned long si;
    unsigned long di;
/*
 * On syscall entry, this is syscall#. On CPU exception, this is error code.
 * On hw interrupt, it's IRQ number:
 */
    unsigned long orig_ax;
/* Return frame for iretq */
    unsigned long ip;
    unsigned long cs;
    unsigned long flags;
    unsigned long sp;
    unsigned long ss;
/* top of stack page */
}

通过sysret指令返回调用程序:

 

 

Intel SDM中对此命令描述如下:

 

c. __kernel_vsyscall

严格意义上来说,上一小节中给出示例不符合系统调用规范,笔者在实际测试时发现手动执行SYSENTER会出现错误。本小节示例如下:

1
2
3
4
5
6
7
8
9
10
11
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
 
int main(int argc, char *argv[]){
    char buffer[80] = "/tmp/test";
    int fd = open(buffer, O_RDONLY);
    int size = read(fd, buffer, sizeof(buffer));
    close(fd);
}

采用静态编译方式,目标平台32位。跟踪open函数调用如下:

 

 

对应源码位于arch/x86/entry/vdso/vdso32/system_call.S文件中:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
.text
    .globl __kernel_vsyscall
    .type __kernel_vsyscall,@function
    ALIGN
__kernel_vsyscall:
    CFI_STARTPROC
    pushl    %ecx
    CFI_ADJUST_CFA_OFFSET    4
    CFI_REL_OFFSET        ecx, 0
    pushl    %edx
    CFI_ADJUST_CFA_OFFSET    4
    CFI_REL_OFFSET        edx, 0
    pushl    %ebp
    CFI_ADJUST_CFA_OFFSET    4
    CFI_REL_OFFSET        ebp, 0
 
    #define SYSENTER_SEQUENCE    "movl %esp, %ebp; sysenter"
    #define SYSCALL_SEQUENCE    "movl %ecx, %ebp; syscall"
 
#ifdef CONFIG_X86_64
    /* If SYSENTER (Intel) or SYSCALL32 (AMD) is available, use it. */
    ALTERNATIVE_2 "", SYSENTER_SEQUENCE, X86_FEATURE_SYSENTER32, \
                      SYSCALL_SEQUENCE,  X86_FEATURE_SYSCALL32
#else
    ALTERNATIVE "", SYSENTER_SEQUENCE, X86_FEATURE_SEP
#endif
 
    /* Enter using int $0x80 */
    int    $0x80
GLOBAL(int80_landing_pad)
 
    /*
     * Restore EDX and ECX in case they were clobbered.  EBP is not
     * clobbered (the kernel restores it), but it's cleaner and
     * probably faster to pop it than to adjust ESP using addl.
     */
    popl    %ebp
    CFI_RESTORE        ebp
    CFI_ADJUST_CFA_OFFSET    -4
    popl    %edx
    CFI_RESTORE        edx
    CFI_ADJUST_CFA_OFFSET    -4
    popl    %ecx
    CFI_RESTORE        ecx
    CFI_ADJUST_CFA_OFFSET    -4
    ret
    CFI_ENDPROC
 
    .size __kernel_vsyscall,.-__kernel_vsyscall
    .previous

关于系统调用指令,根据平台选择是SYSENTER或是SYSCALL,若均不支持则执行传统系统调用int $0x80。

0x03 SYSCALL

Intel SDM:

 

 

同样是位于syscall_init函数中:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
void syscall_init(void)
{
    extern char _entry_trampoline[];
    extern char entry_SYSCALL_64_trampoline[];
 
    int cpu = smp_processor_id();
    unsigned long SYSCALL64_entry_trampoline =
        (unsigned long)get_cpu_entry_area(cpu)->entry_trampoline +
        (entry_SYSCALL_64_trampoline - _entry_trampoline);
 
    wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
    if (static_cpu_has(X86_FEATURE_PTI))
        wrmsrl(MSR_LSTAR, SYSCALL64_entry_trampoline);
    else
        wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
......
    /* Flags to clear on syscall */
    wrmsrl(MSR_SYSCALL_MASK,
           X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
           X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
}

entry_SYSCALL_64中执行系统调用是采用如下方式:

1
2
3
4
5
#ifdef CONFIG_RETPOLINE
    movq    sys_call_table(, %rax, 8), %rax
    call    __x86_indirect_thunk_rax
#else
    call    *sys_call_table(, %rax, 8)

调用约定是:

1
2
3
4
5
6
7
8
9
10
11
* Registers on entry:
* rax  system call number
* rcx  return address
* r11  saved rflags (note: r11 is callee-clobbered register in C ABI)
* rdi  arg0
* rsi  arg1
* rdx  arg2
* r10  arg3 (needs to be moved to rcx to conform to C ABI)
* r8   arg4
* r9   arg5
* (note: r12-r15, rbp, rbx are callee-preserved in C ABI)

返回依然是采用SYSRET指令:

1
2
3
4
5
6
#define USERGS_SYSRET64                \
    swapgs;                    \
    sysretq;
#define USERGS_SYSRET32                \
    swapgs;                    \
    sysretl

0x04 VDSO

VDSO全称是Virtual Dynamic Shared Object,它映射到用户地址空间中,可以被用户程序直接调用,但没有对应文件,是由内核直接映射:

 

 

其导出函数见arch/x86/entry/vdso/vdso.lds.S文件:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
VERSION {
    LINUX_2.6 {
    global:
        clock_gettime;
        __vdso_clock_gettime;
        gettimeofday;
        __vdso_gettimeofday;
        getcpu;
        __vdso_getcpu;
        time;
        __vdso_time;
    local: *;
    };
}

gettimeofday为例, 其定义位于同目录下vclock_gettime.c文件中:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
extern int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz);
......
notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
{
    if (likely(tv != NULL)) {
        if (unlikely(do_realtime((struct timespec *)tv) == VCLOCK_NONE))
            return vdso_fallback_gtod(tv, tz);
        tv->tv_usec /= 1000;
    }
    if (unlikely(tz != NULL)) {
        tz->tz_minuteswest = gtod->tz_minuteswest;
        tz->tz_dsttime = gtod->tz_dsttime;
    }
 
    return 0;
}
int gettimeofday(struct timeval *, struct timezone *)
    __attribute__((weak, alias("__vdso_gettimeofday")));

用户调用gettimeofday时,实际执行的是__vdso_gettimeofday。示例代码如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
#include <time.h>
#include <sys/time.h>
#include <stdio.h>
 
int main(int argc, char **argv)
{
    char buffer[40];
    struct timeval time;
 
    gettimeofday(&time, NULL);
 
    strftime(buffer, 40, "Current date/time: %m-%d-%Y/%T", localtime(&time.tv_sec));
    printf("%s\n",buffer);
 
    return 0;
}

编译之后跟踪gettimeofday函数调用:

 

 

查看内存空间映射情况:

 

 

可以看到执行指令确实映射在vdso区域内。

0x05 参阅链接

  1. The Definitive Guide to Linux System Calls——Prerequisite information
  2. Linux Kernel 实践(二):劫持系统调用
  3. 代码解析Linux系统调用
  4. 谈结构体struct 初始化多出的点号“.”,数组[]初始化多出的逗号“,
  5. x86 架构下 Linux 的系统调用与 vsyscall, vDSO
  6. Setup: Ubuntu host, QEMU vm, x86-64 kernel
  7. x86-64 Spec addition - SwapGS instruction
  8. Linux系统调用过程分析
  9. Timers and time management in the Linux kernel. Part 7

【看雪培训】《Adroid高级研修班》2022年夏季班招生中!

收藏
点赞0
打赏
分享
最新回复 (1)
雪    币: 7501
活跃值: 活跃值 (8034)
能力值: ( LV12,RANK:650 )
在线值:
发帖
回帖
粉丝
erfze 活跃值 12 2021-10-27 18:12
2
0
游客
登录 | 注册 方可回帖
返回