i386-linux.c revision 168569
1/*
2 * Copryight 1997 Sean Eric Fagan
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 * 1. Redistributions of source code must retain the above copyright
8 *    notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 *    notice, this list of conditions and the following disclaimer in the
11 *    documentation and/or other materials provided with the distribution.
12 * 3. All advertising materials mentioning features or use of this software
13 *    must display the following acknowledgement:
14 *	This product includes software developed by Sean Eric Fagan
15 * 4. Neither the name of the author may be used to endorse or promote
16 *    products derived from this software without specific prior written
17 *    permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32#ifndef lint
33static const char rcsid[] =
34  "$FreeBSD: head/usr.bin/truss/i386-linux.c 168569 2007-04-10 04:03:34Z delphij $";
35#endif /* not lint */
36
37/*
38 * Linux/i386-specific system call handling.  Given how much of this code
39 * is taken from the freebsd equivalent, I can probably put even more of
40 * it in support routines that can be used by any personality support.
41 */
42
43#include <sys/types.h>
44#include <sys/ptrace.h>
45
46#include <machine/reg.h>
47#include <machine/psl.h>
48
49#include <errno.h>
50#include <fcntl.h>
51#include <signal.h>
52#include <stdio.h>
53#include <stdlib.h>
54#include <string.h>
55#include <time.h>
56#include <unistd.h>
57
58#include "truss.h"
59#include "syscall.h"
60#include "extern.h"
61
62static int cpid = -1;
63
64#include "linux_syscalls.h"
65
66static int nsyscalls =
67	sizeof(linux_syscallnames) / sizeof(linux_syscallnames[0]);
68
69/*
70 * This is what this particular file uses to keep track of a system call.
71 * It is probably not quite sufficient -- I can probably use the same
72 * structure for the various syscall personalities, and I also probably
73 * need to nest system calls (for signal handlers).
74 *
75 * 'struct syscall' describes the system call; it may be NULL, however,
76 * if we don't know about this particular system call yet.
77 */
78static struct linux_syscall {
79	struct syscall *sc;
80	const char *name;
81	int number;
82	unsigned long args[5];
83	int nargs;	/* number of arguments -- *not* number of words! */
84	char **s_args;	/* the printable arguments */
85} fsc;
86
87/* Clear up and free parts of the fsc structure. */
88static __inline void
89clear_fsc(void) {
90  if (fsc.s_args) {
91    int i;
92    for (i = 0; i < fsc.nargs; i++)
93      if (fsc.s_args[i])
94	free(fsc.s_args[i]);
95    free(fsc.s_args);
96  }
97  memset(&fsc, 0, sizeof(fsc));
98}
99
100/*
101 * Called when a process has entered a system call.  nargs is the
102 * number of words, not number of arguments (a necessary distinction
103 * in some cases).  Note that if the STOPEVENT() code in i386/i386/trap.c
104 * is ever changed these functions need to keep up.
105 */
106
107void
108i386_linux_syscall_entry(struct trussinfo *trussinfo, int nargs) {
109  struct reg regs;
110  int syscall_num;
111  int i;
112  struct syscall *sc;
113
114  cpid = trussinfo->curthread->tid;
115
116  clear_fsc();
117
118  if (ptrace(PT_GETREGS, cpid, (caddr_t)&regs, 0) < 0)
119  {
120    fprintf(trussinfo->outfile, "-- CANNOT READ REGISTERS --\n");
121    return;
122  }
123  syscall_num = regs.r_eax;
124
125  fsc.number = syscall_num;
126  fsc.name =
127    (syscall_num < 0 || syscall_num > nsyscalls) ? NULL : linux_syscallnames[syscall_num];
128  if (!fsc.name) {
129    fprintf(trussinfo->outfile, "-- UNKNOWN SYSCALL %d --\n", syscall_num);
130  }
131
132  if (fsc.name && (trussinfo->flags & FOLLOWFORKS)
133   && ((!strcmp(fsc.name, "linux_fork")
134    || !strcmp(fsc.name, "linux_vfork"))))
135  {
136    trussinfo->curthread->in_fork = 1;
137  }
138
139  if (nargs == 0)
140    return;
141
142  /*
143   * Linux passes syscall arguments in registers, not
144   * on the stack.  Fortunately, we've got access to the
145   * register set.  Note that we don't bother checking the
146   * number of arguments.  And what does linux do for syscalls
147   * that have more than five arguments?
148   */
149
150  fsc.args[0] = regs.r_ebx;
151  fsc.args[1] = regs.r_ecx;
152  fsc.args[2] = regs.r_edx;
153  fsc.args[3] = regs.r_esi;
154  fsc.args[4] = regs.r_edi;
155
156  sc = get_syscall(fsc.name);
157  if (sc) {
158    fsc.nargs = sc->nargs;
159  } else {
160#if DEBUG
161    fprintf(trussinfo->outfile, "unknown syscall %s -- setting args to %d\n",
162	   fsc.name, nargs);
163#endif
164    fsc.nargs = nargs;
165  }
166
167  fsc.s_args = malloc((1+fsc.nargs) * sizeof(char*));
168  memset(fsc.s_args, 0, fsc.nargs * sizeof(char*));
169  fsc.sc = sc;
170
171  /*
172   * At this point, we set up the system call arguments.
173   * We ignore any OUT ones, however -- those are arguments that
174   * are set by the system call, and so are probably meaningless
175   * now.  This doesn't currently support arguments that are
176   * passed in *and* out, however.
177   */
178
179  if (fsc.name) {
180
181#if DEBUG
182    fprintf(stderr, "syscall %s(", fsc.name);
183#endif
184    for (i = 0; i < fsc.nargs; i++) {
185#if DEBUG
186      fprintf(stderr, "0x%x%s",
187	      sc
188	      ? fsc.args[sc->args[i].offset]
189	      : fsc.args[i],
190	      i < (fsc.nargs - 1) ? "," : "");
191#endif
192      if (sc && !(sc->args[i].type & OUT)) {
193	fsc.s_args[i] = print_arg(&sc->args[i], fsc.args, 0, trussinfo);
194      }
195    }
196#if DEBUG
197    fprintf(stderr, ")\n");
198#endif
199  }
200
201#if DEBUG
202  fprintf(trussinfo->outfile, "\n");
203#endif
204
205  /*
206   * Some system calls should be printed out before they are done --
207   * execve() and exit(), for example, never return.  Possibly change
208   * this to work for any system call that doesn't have an OUT
209   * parameter?
210   */
211
212  if (fsc.name != NULL &&
213      (!strcmp(fsc.name, "execve") || !strcmp(fsc.name, "exit"))) {
214
215    /* XXX
216     * This could be done in a more general
217     * manner but it still wouldn't be very pretty.
218     */
219    if (!strcmp(fsc.name, "linux_execve")) {
220        if ((trussinfo->flags & EXECVEARGS) == 0)
221          if (fsc.s_args[1]) {
222            free(fsc.s_args[1]);
223            fsc.s_args[1] = NULL;
224          }
225        if ((trussinfo->flags & EXECVEENVS) == 0)
226          if (fsc.s_args[2]) {
227            free(fsc.s_args[2]);
228            fsc.s_args[2] = NULL;
229          }
230    }
231
232    print_syscall(trussinfo, fsc.name, fsc.nargs, fsc.s_args);
233    fprintf(trussinfo->outfile, "\n");
234  }
235
236  return;
237}
238
239/*
240 * Linux syscalls return negative errno's, we do positive and map them
241 */
242const int bsd_to_linux_errno[] = {
243  	-0,  -1,  -2,  -3,  -4,  -5,  -6,  -7,  -8,  -9,
244 	-10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
245 	-20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
246 	-30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
247 	-90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
248	-100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
249	-110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
250	-116, -66,  -6,  -6,  -6,  -6,  -6, -37, -38,  -9,
251  	-6,
252};
253
254long
255i386_linux_syscall_exit(struct trussinfo *trussinfo, int syscall_num __unused)
256{
257  struct reg regs;
258  long retval;
259  int i;
260  int errorp;
261  struct syscall *sc;
262
263  cpid = trussinfo->curthread->tid;
264  if (ptrace(PT_GETREGS, cpid, (caddr_t)&regs, 0) < 0)
265  {
266    fprintf(trussinfo->outfile, "-- CANNOT READ REGISTERS --\n");
267    return (-1);
268  }
269
270  retval = regs.r_eax;
271  errorp = !!(regs.r_eflags & PSL_C);
272
273  /*
274   * This code, while simpler than the initial versions I used, could
275   * stand some significant cleaning.
276   */
277
278  sc = fsc.sc;
279  if (!sc) {
280    for (i = 0; i < fsc.nargs; i++)
281      asprintf(&fsc.s_args[i], "0x%lx", fsc.args[i]);
282  } else {
283    /*
284     * Here, we only look for arguments that have OUT masked in --
285     * otherwise, they were handled in the syscall_entry function.
286     */
287    for (i = 0; i < sc->nargs; i++) {
288      char *temp;
289      if (sc->args[i].type & OUT) {
290	/*
291	 * If an error occurred, than don't bothe getting the data;
292	 * it may not be valid.
293	 */
294	if (errorp)
295	  asprintf(&temp, "0x%lx", fsc.args[sc->args[i].offset]);
296	else
297	  temp = print_arg(&sc->args[i], fsc.args, retval, trussinfo);
298	fsc.s_args[i] = temp;
299      }
300    }
301  }
302
303  /*
304   * It would probably be a good idea to merge the error handling,
305   * but that complicates things considerably.
306   */
307  if (errorp) {
308    for (i = 0; (size_t)i < sizeof(bsd_to_linux_errno) / sizeof(int); i++)
309      if (retval == bsd_to_linux_errno[i])
310      break;
311  }
312  print_syscall_ret(trussinfo, fsc.name, fsc.nargs, fsc.s_args, errorp,
313                    errorp ? i : retval);
314  clear_fsc();
315
316  return (retval);
317}
318