syscall.c

linux进程跟踪的工具和源代码

		if (tcp->scno >= 0 && tcp->scno < nsyscalls && sysent[tcp->scno].nargs != -1)
			tcp->u_nargs = sysent[tcp->scno].nargs;
		else 
     	        	tcp->u_nargs = MAX_ARGS;
		for (i = 0; i < tcp->u_nargs; i++)
			tcp->u_arg[i] = *((&regs.r_o0) + i);
	}
#elif defined (HPPA)
	{
		int i;

		if (tcp->scno >= 0 && tcp->scno < nsyscalls && sysent[tcp->scno].nargs != -1)
			tcp->u_nargs = sysent[tcp->scno].nargs;
		else 
     	        	tcp->u_nargs = MAX_ARGS;
		for (i = 0; i < tcp->u_nargs; i++) {
			if (upeek(pid, PT_GR26-4*i, &tcp->u_arg[i]) < 0)
				return -1;
		}
	}
#else /* Other architecture (like i386) (32bits specific) */
	{
		int i;
		if (tcp->scno >= 0 && tcp->scno < nsyscalls && sysent[tcp->scno].nargs != -1)
			tcp->u_nargs = sysent[tcp->scno].nargs;
		else 
     	        	tcp->u_nargs = MAX_ARGS;
		for (i = 0; i < tcp->u_nargs; i++) {
			if (upeek(pid, i*4, &tcp->u_arg[i]) < 0)
				return -1;
		}
	}
#endif 
#endif /* LINUX */
#ifdef SUNOS4
	{
		int i;
		if (tcp->scno >= 0 && tcp->scno < nsyscalls && sysent[tcp->scno].nargs != -1)
			tcp->u_nargs = sysent[tcp->scno].nargs;
		else 
     	        	tcp->u_nargs = MAX_ARGS;
		for (i = 0; i < tcp->u_nargs; i++) {
			struct user *u;

			if (upeek(pid, uoff(u_arg[0]) +
			    (i*sizeof(u->u_arg[0])), &tcp->u_arg[i]) < 0)
				return -1;
		}
	}
#endif /* SUNOS4 */
#ifdef SVR4
#ifdef MIPS
	/*
	 * SGI is broken: even though it has pr_sysarg, it doesn't
	 * set them on system call entry.  Get a clue.
	 */
	if (tcp->scno >= 0 && tcp->scno < nsyscalls && sysent[tcp->scno].nargs != -1)
		tcp->u_nargs = sysent[tcp->scno].nargs;
	else
		tcp->u_nargs = tcp->status.pr_nsysarg;
	if (tcp->u_nargs > 4) {
		memcpy(tcp->u_arg, &tcp->status.pr_reg[CTX_A0],
			4*sizeof(tcp->u_arg[0]));
		umoven(tcp, tcp->status.pr_reg[CTX_SP] + 16,
			(tcp->u_nargs - 4)*sizeof(tcp->u_arg[0]), (char *) (tcp->u_arg + 4));
	}
	else {
		memcpy(tcp->u_arg, &tcp->status.pr_reg[CTX_A0],
			tcp->u_nargs*sizeof(tcp->u_arg[0]));
	}
#elif UNIXWARE >= 2
	/*
	 * Like SGI, UnixWare doesn't set pr_sysarg until system call exit
	 */
	if (tcp->scno >= 0 && tcp->scno < nsyscalls && sysent[tcp->scno].nargs != -1)
		tcp->u_nargs = sysent[tcp->scno].nargs;
	else
		tcp->u_nargs = tcp->status.pr_lwp.pr_nsysarg;
	umoven(tcp, tcp->status.PR_REG[UESP] + 4,
		tcp->u_nargs*sizeof(tcp->u_arg[0]), (char *) tcp->u_arg);
#elif defined (HAVE_PR_SYSCALL)
	if (tcp->scno >= 0 && tcp->scno < nsyscalls && sysent[tcp->scno].nargs != -1)
		tcp->u_nargs = sysent[tcp->scno].nargs;
	else
		tcp->u_nargs = tcp->status.pr_nsysarg;
	{
		int i;
		for (i = 0; i < tcp->u_nargs; i++)
			tcp->u_arg[i] = tcp->status.pr_sysarg[i];
	}
#elif defined (I386)
	if (tcp->scno >= 0 && tcp->scno < nsyscalls && sysent[tcp->scno].nargs != -1)
		tcp->u_nargs = sysent[tcp->scno].nargs;
	else
		tcp->u_nargs = 5;
	umoven(tcp, tcp->status.PR_REG[UESP] + 4,
		tcp->u_nargs*sizeof(tcp->u_arg[0]), (char *) tcp->u_arg);
#else
	I DONT KNOW WHAT TO DO
#endif /* !HAVE_PR_SYSCALL */
#endif /* SVR4 */
#ifdef FREEBSD
	if (tcp->scno >= 0 && tcp->scno < nsyscalls &&
	    sysent[tcp->scno].nargs > tcp->status.val)
		tcp->u_nargs = sysent[tcp->scno].nargs;
	else 
	  	tcp->u_nargs = tcp->status.val;
	if (tcp->u_nargs < 0)
		tcp->u_nargs = 0;
	if (tcp->u_nargs > MAX_ARGS)
		tcp->u_nargs = MAX_ARGS;
	switch(regs.r_eax) {
	case SYS___syscall:
		pread(tcp->pfd, &tcp->u_arg, tcp->u_nargs * sizeof(unsigned long),
		      regs.r_esp + sizeof(int) + sizeof(quad_t));
	  break;
        case SYS_syscall:
		pread(tcp->pfd, &tcp->u_arg, tcp->u_nargs * sizeof(unsigned long),
		      regs.r_esp + 2 * sizeof(int));
	  break;
        default:
		pread(tcp->pfd, &tcp->u_arg, tcp->u_nargs * sizeof(unsigned long),
		      regs.r_esp + sizeof(int));
	  break;
	}
#endif /* FREEBSD */
	return 1;
}

int
trace_syscall(tcp)
struct tcb *tcp;
{
	int sys_res;
	struct timeval tv;
	int res;

	/* Measure the exit time as early as possible to avoid errors. */
	if (dtime && (tcp->flags & TCB_INSYSCALL))
		gettimeofday(&tv, NULL);

	res = get_scno(tcp);
	if (res != 1)
		return res;

	res = syscall_fixup(tcp);
	if (res != 1)
		return res;

	if (tcp->flags & TCB_INSYSCALL) {
		long u_error;
		res = get_error(tcp);
		if (res != 1)
			return res;
		u_error = tcp->u_error;


		internal_syscall(tcp);
		if (tcp->scno >= 0 && tcp->scno < nsyscalls &&
		    !(qual_flags[tcp->scno] & QUAL_TRACE)) {
			tcp->flags &= ~TCB_INSYSCALL;
			return 0;
		}

		if (tcp->flags & TCB_REPRINT) {
			printleader(tcp);
			tprintf("<... ");
			if (tcp->scno >= nsyscalls || tcp->scno < 0)
				tprintf("syscall_%lu", tcp->scno);
			else
				tprintf("%s", sysent[tcp->scno].sys_name);
			tprintf(" resumed> ");
		}

		if (cflag && tcp->scno < nsyscalls && tcp->scno >= 0) {
			call_count[tcp->scno]++;
			if (tcp->u_error)
				error_count[tcp->scno]++;
			tv_sub(&tv, &tv, &tcp->etime);
#ifdef LINUX
			if (tv_cmp(&tv, &tcp->dtime) > 0) {
				static struct timeval one_tick =
					{ 0, 1000000 / HZ };

				if (tv_nz(&tcp->dtime))
					tv = tcp->dtime;
				else if (tv_cmp(&tv, &one_tick) > 0) {
					if (tv_cmp(&shortest, &one_tick) < 0)
						tv = shortest;
					else
						tv = one_tick;
				}
			}
#endif /* LINUX */
			if (tv_cmp(&tv, &shortest) < 0)
				shortest = tv;
			tv_add(&tv_count[tcp->scno],
				&tv_count[tcp->scno], &tv);
			tcp->flags &= ~TCB_INSYSCALL;
			return 0;
		}

		if (tcp->scno >= nsyscalls || tcp->scno < 0
		    || (qual_flags[tcp->scno] & QUAL_RAW))
			sys_res = printargs(tcp);
		else
			sys_res = (*sysent[tcp->scno].sys_func)(tcp);
		u_error = tcp->u_error;
		tprintf(") ");
		tabto(acolumn);
		if (tcp->scno >= nsyscalls || tcp->scno < 0 ||
		    qual_flags[tcp->scno] & QUAL_RAW) {
			if (u_error)
				tprintf("= -1 (errno %ld)", u_error);
			else
				tprintf("= %#lx", tcp->u_rval);
		}
		else if (!(sys_res & RVAL_NONE) && u_error) {
			switch (u_error) {
#ifdef LINUX
			case ERESTARTSYS:
				tprintf("= ? ERESTARTSYS (To be restarted)");
				break;
			case ERESTARTNOINTR:
				tprintf("= ? ERESTARTNOINTR (To be restarted)");
				break;
			case ERESTARTNOHAND:
				tprintf("= ? ERESTARTNOHAND (To be restarted)");
				break;
#endif /* LINUX */
			default:
				tprintf("= -1 ");
				if (u_error < nerrnos && u_error < sys_nerr)
					tprintf("%s (%s)", errnoent[u_error],
						sys_errlist[u_error]);
				else if (u_error < nerrnos)
					tprintf("%s (errno %ld)",
						errnoent[u_error], u_error);
				else if (u_error < sys_nerr)
					tprintf("ERRNO_%ld (%s)", u_error,
						sys_errlist[u_error]);
				else
					tprintf("E??? (errno %ld)", u_error);
				break;
			}
		}
		else {
			if (sys_res & RVAL_NONE)
				tprintf("= ?");
			else {
				switch (sys_res & RVAL_MASK) {
				case RVAL_HEX:
					tprintf("= %#lx", tcp->u_rval);
					break;
				case RVAL_OCTAL:
					tprintf("= %#lo", tcp->u_rval);
					break;
				case RVAL_UDECIMAL:
					tprintf("= %lu", tcp->u_rval);
					break;
				case RVAL_DECIMAL:
					tprintf("= %ld", tcp->u_rval);
					break;
#ifdef HAVE_LONG_LONG
				case RVAL_LHEX:
					tprintf("= %#llx", tcp->u_lrval);
					break;
				case RVAL_LOCTAL:
					tprintf("= %#llo", tcp->u_lrval);
					break;
				case RVAL_LUDECIMAL:
					tprintf("= %llu", tcp->u_lrval);
					break;
				case RVAL_LDECIMAL:
					tprintf("= %lld", tcp->u_lrval);
					break;
#endif
				default:
					fprintf(stderr,
						"invalid rval format\n");
					break;
				}
			}
			if ((sys_res & RVAL_STR) && tcp->auxstr)
				tprintf(" (%s)", tcp->auxstr);
		}
		if (dtime) {
			tv_sub(&tv, &tv, &tcp->etime);
			tprintf(" <%ld.%06ld>",
				(long) tv.tv_sec, (long) tv.tv_usec);
		}
		printtrailer(tcp);

		dumpio(tcp);
		if (fflush(tcp->outf) == EOF)
			return -1;
		tcp->flags &= ~TCB_INSYSCALL;
		return 0;
	}

	/* Entering system call */
	res = syscall_enter(tcp);
	if (res != 1)
		return res;

	switch (tcp->scno + NR_SYSCALL_BASE) {
#ifdef LINUX
#if !defined (ALPHA) && !defined(IA64) && !defined(SPARC) && !defined(MIPS) && !defined(HPPA)
	case SYS_socketcall:
		decode_subcall(tcp, SYS_socket_subcall,
			SYS_socket_nsubcalls, deref_style);
		break;
	case SYS_ipc:
		decode_subcall(tcp, SYS_ipc_subcall,
			SYS_ipc_nsubcalls, shift_style);
		break;
#endif /* !ALPHA && !IA64 && !MIPS && !SPARC */
#ifdef SPARC
	case SYS_socketcall:
		sparc_socket_decode (tcp);
		break;
#endif
#endif /* LINUX */
#ifdef SVR4
#ifdef SYS_pgrpsys_subcall
	case SYS_pgrpsys:
		decode_subcall(tcp, SYS_pgrpsys_subcall,
			SYS_pgrpsys_nsubcalls, shift_style);
		break;
#endif /* SYS_pgrpsys_subcall */
#ifdef SYS_sigcall_subcall
	case SYS_sigcall:
		decode_subcall(tcp, SYS_sigcall_subcall,
			SYS_sigcall_nsubcalls, mask_style);
		break;
#endif /* SYS_sigcall_subcall */
	case SYS_msgsys:
		decode_subcall(tcp, SYS_msgsys_subcall,
			SYS_msgsys_nsubcalls, shift_style);
		break;
	case SYS_shmsys:
		decode_subcall(tcp, SYS_shmsys_subcall,
			SYS_shmsys_nsubcalls, shift_style);
		break;
	case SYS_semsys:
		decode_subcall(tcp, SYS_semsys_subcall,
			SYS_semsys_nsubcalls, shift_style);
		break;
#if 0 /* broken */
	case SYS_utssys:
		decode_subcall(tcp, SYS_utssys_subcall,
			SYS_utssys_nsubcalls, shift_style);
		break;
#endif
	case SYS_sysfs:
		decode_subcall(tcp, SYS_sysfs_subcall,
			SYS_sysfs_nsubcalls, shift_style);
		break;
	case SYS_spcall:
		decode_subcall(tcp, SYS_spcall_subcall,
			SYS_spcall_nsubcalls, shift_style);
		break;
#ifdef SYS_context_subcall
	case SYS_context:
		decode_subcall(tcp, SYS_context_subcall,
			SYS_context_nsubcalls, shift_style);
		break;
#endif /* SYS_context_subcall */
#ifdef SYS_door_subcall
	case SYS_door:
		decode_subcall(tcp, SYS_door_subcall,
			SYS_door_nsubcalls, door_style);
		break;
#endif /* SYS_door_subcall */
#ifdef SYS_kaio_subcall
	case SYS_kaio:
		decode_subcall(tcp, SYS_kaio_subcall,
			SYS_kaio_nsubcalls, shift_style);
		break;
#endif	
#endif /* SVR4 */
#ifdef FREEBSD
	case SYS_msgsys:
	case SYS_shmsys:
	case SYS_semsys:
		decode_subcall(tcp, 0, 0, table_style);
		break;
#endif		
#ifdef SUNOS4
	case SYS_semsys:
		decode_subcall(tcp, SYS_semsys_subcall,
			SYS_semsys_nsubcalls, shift_style);
		break;
	case SYS_msgsys:
		decode_subcall(tcp, SYS_msgsys_subcall,
			SYS_msgsys_nsubcalls, shift_style);
		break;
	case SYS_shmsys:
		decode_subcall(tcp, SYS_shmsys_subcall,
			SYS_shmsys_nsubcalls, shift_style);
		break;
#endif
	}

	internal_syscall(tcp);
	if (tcp->scno >=0 && tcp->scno < nsyscalls && !(qual_flags[tcp->scno] & QUAL_TRACE)) {
		tcp->flags |= TCB_INSYSCALL;
		return 0;
	}

	if (cflag) {
		gettimeofday(&tcp->etime, NULL);
		tcp->flags |= TCB_INSYSCALL;
		return 0;
	}

	printleader(tcp);
	tcp->flags &= ~TCB_REPRINT;
	tcp_last = tcp;
	if (tcp->scno >= nsyscalls || tcp->scno < 0)
		tprintf("syscall_%lu(", tcp->scno);
	else
		tprintf("%s(", sysent[tcp->scno].sys_name);
	if (tcp->scno >= nsyscalls || tcp->scno < 0 || 
	    ((qual_flags[tcp->scno] & QUAL_RAW) && tcp->scno != SYS_exit))
		sys_res = printargs(tcp);
	else
		sys_res = (*sysent[tcp->scno].sys_func)(tcp);
	if (fflush(tcp->outf) == EOF)
		return -1;
	tcp->flags |= TCB_INSYSCALL;
	/* Measure the entrance time as late as possible to avoid errors. */
	if (dtime)
		gettimeofday(&tcp->etime, NULL);
	return sys_res;
}

int
printargs(tcp)
struct tcb *tcp;
{
	if (entering(tcp)) {
		int i;

		for (i = 0; i < tcp->u_nargs; i++)
			tprintf("%s%#lx", i ? ", " : "", tcp->u_arg[i]);
	}
	return 0;
}

long
getrval2(tcp)
struct tcb *tcp;
{
	long val = -1;

#ifdef LINUX
#ifdef SPARC
	struct regs regs;
	if (ptrace(PTRACE_GETREGS,tcp->pid,(char *)&regs,0) < 0)
		return -1;
	val = regs.r_o1;
#endif /* SPARC */
#endif /* LINUX */

#ifdef SUNOS4
	if (upeek(tcp->pid, uoff(u_rval2), &val) < 0)
		return -1;
#endif /* SUNOS4 */

#ifdef SVR4
#ifdef SPARC
	val = tcp->status.PR_REG[R_O1];
#endif /* SPARC */
#ifdef I386
	val = tcp->status.PR_REG[EDX];
#endif /* I386 */
#ifdef MIPS
	val = tcp->status.PR_REG[CTX_V1];
#endif /* MIPS */
#endif /* SVR4 */
#ifdef FREEBSD
	struct reg regs;
	pread(tcp->pfd_reg, &regs, sizeof(regs), 0);
	val = regs.r_edx;
#endif	
	return val;
}

/*
 * Apparently, indirect system calls have already be converted by ptrace(2),
 * so if you see "indir" this program has gone astray.
 */
int
sys_indir(tcp)
struct tcb *tcp;
{
	int i, scno, nargs;

	if (entering(tcp)) {
		if ((scno = tcp->u_arg[0]) > nsyscalls) {
			fprintf(stderr, "Bogus syscall: %u\n", scno);
			return 0;
		}
		nargs = sysent[scno].nargs;
		tprintf("%s", sysent[scno].sys_name);
		for (i = 0; i < nargs; i++)
			tprintf(", %#lx", tcp->u_arg[i+1]);
	}
	return 0;
}

static int
time_cmp(a, b)
void *a;
void *b;
{
	return -tv_cmp(&tv_count[*((int *) a)], &tv_count[*((int *) b)]);
}

static int
syscall_cmp(a, b)
void *a;
void *b;
{
	return strcmp(sysent[*((int *) a)].sys_name,
		sysent[*((int *) b)].sys_name);
}

static int
count_cmp(a, b)
void *a;
void *b;
{
	int m = call_count[*((int *) a)], n = call_count[*((int *) b)];

	return (m < n) ? 1 : (m > n) ? -1 : 0;
}

static int (*sortfun)();
static struct timeval overhead = { -1, -1 };

void
set_sortby(sortby)
char *sortby;
{
	if (strcmp(sortby, "time") == 0)
		sortfun = time_cmp;
	else if (strcmp(sortby, "calls") == 0)
		sortfun = count_cmp;
	else if (strcmp(sortby, "name") == 0)
		sortfun = syscall_cmp;
	else if (strcmp(sortby, "nothing") == 0)
		sortfun = NULL;
	else {
		fprintf(stderr, "invalid sortby: `%s'\n", sortby);
		exit(1);
	}
}

void set_overhead(n)
int n;
{
	overhead.tv_sec = n / 1000000;
	overhead.tv_usec = n % 1000000;
}

void
call_summary(outf)
FILE *outf;
{
	int i, j;
	int call_cum, error_cum;
	struct timeval tv_cum, dtv;
	double percent;
	char *dashes = "-------------------------";
	char error_str[16];

	call_cum = error_cum = tv_cum.tv_sec = tv_cum.tv_usec = 0;
	if (overhead.tv_sec == -1) {
		tv_mul(&overhead, &shortest, 8);
		tv_div(&overhead, &overhead, 10);
	}
	for (i = 0; i < nsyscalls; i++) {
		sorted_count[i] = i;
		if (call_count[i] == 0)
			continue;
		tv_mul(&dtv, &overhead, call_count[i]);
		tv_sub(&tv_count[i], &tv_count[i], &dtv);
		call_cum += call_count[i];
		error_cum += error_count[i];
		tv_add(&tv_cum, &tv_cum, &tv_count[i]);
	}
	if (sortfun)
		qsort((void *) sorted_count, nsyscalls, sizeof(int), sortfun);
	fprintf(outf, "%6.6s %11.11s %11.11s %9.9s %9.9s %s\n",
		"% time", "seconds", "usecs/call",
		"calls", "errors", "syscall");
	fprintf(outf, "%6.6s %11.11s %11.11s %9.9s %9.9s %-16.16s\n",
		dashes, dashes, dashes, dashes, dashes, dashes);
	for (i = 0; i < nsyscalls; i++) {
		j = sorted_count[i];
		if (call_count[j] == 0)
			continue;
		tv_div(&dtv, &tv_count[j], call_count[j]);
		if (error_count[j])
			sprintf(error_str, "%d", error_count[j]);
		else
			error_str[0] = '\0';
		percent = 100.0*tv_float(&tv_count[j])/tv_float(&tv_cum);
		fprintf(outf, "%6.2f %4ld.%06ld %11ld %9d %9.9s %s\n",
			percent, (long) tv_count[j].tv_sec,
			(long) tv_count[j].tv_usec,
			(long) 1000000 * dtv.tv_sec + dtv.tv_usec,
			call_count[j], error_str, sysent[j].sys_name);
	}
	fprintf(outf, "%6.6s %11.11s %11.11s %9.9s %9.9s %-16.16s\n",
		dashes, dashes, dashes, dashes, dashes, dashes);
	if (error_cum)
		sprintf(error_str, "%d", error_cum);
	else
		error_str[0] = '\0';
	fprintf(outf, "%6.6s %4ld.%06ld %11.11s %9d %9.9s %s\n",
		"100.00", (long) tv_cum.tv_sec, (long) tv_cum.tv_usec, "",
		call_cum, error_str, "total");
}