i386-linux.c revision 158626
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 158626 2006-05-15 21:03:02Z pav $"; 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/ioctl.h> 45#include <sys/pioctl.h> 46 47#include <machine/reg.h> 48#include <machine/psl.h> 49 50#include <errno.h> 51#include <fcntl.h> 52#include <signal.h> 53#include <stdio.h> 54#include <stdlib.h> 55#include <string.h> 56#include <time.h> 57#include <unistd.h> 58 59#include "truss.h" 60#include "syscall.h" 61#include "extern.h" 62 63static int fd = -1; 64static int cpid = -1; 65 66#include "linux_syscalls.h" 67 68static int nsyscalls = 69 sizeof(linux_syscallnames) / sizeof(linux_syscallnames[0]); 70 71/* 72 * This is what this particular file uses to keep track of a system call. 73 * It is probably not quite sufficient -- I can probably use the same 74 * structure for the various syscall personalities, and I also probably 75 * need to nest system calls (for signal handlers). 76 * 77 * 'struct syscall' describes the system call; it may be NULL, however, 78 * if we don't know about this particular system call yet. 79 */ 80static struct linux_syscall { 81 struct syscall *sc; 82 const char *name; 83 int number; 84 unsigned long args[5]; 85 int nargs; /* number of arguments -- *not* number of words! */ 86 char **s_args; /* the printable arguments */ 87} fsc; 88 89/* Clear up and free parts of the fsc structure. */ 90static __inline void 91clear_fsc(void) { 92 if (fsc.s_args) { 93 int i; 94 for (i = 0; i < fsc.nargs; i++) 95 if (fsc.s_args[i]) 96 free(fsc.s_args[i]); 97 free(fsc.s_args); 98 } 99 memset(&fsc, 0, sizeof(fsc)); 100} 101 102/* 103 * Called when a process has entered a system call. nargs is the 104 * number of words, not number of arguments (a necessary distinction 105 * in some cases). Note that if the STOPEVENT() code in i386/i386/trap.c 106 * is ever changed these functions need to keep up. 107 */ 108 109void 110i386_linux_syscall_entry(struct trussinfo *trussinfo, int nargs) { 111 char buf[32]; 112 struct reg regs; 113 int syscall_num; 114 int i; 115 struct syscall *sc; 116 117 if (fd == -1 || trussinfo->pid != cpid) { 118 sprintf(buf, "/proc/%d/regs", trussinfo->pid); 119 fd = open(buf, O_RDWR); 120 if (fd == -1) { 121 fprintf(trussinfo->outfile, "-- CANNOT OPEN REGISTERS --\n"); 122 return; 123 } 124 cpid = trussinfo->pid; 125 } 126 127 clear_fsc(); 128 lseek(fd, 0L, 0); 129 if (read(fd, ®s, sizeof(regs)) != sizeof(regs)) { 130 fprintf(trussinfo->outfile, "-- CANNOT READ REGISTERS --\n"); 131 return; 132 } 133 syscall_num = regs.r_eax; 134 135 fsc.number = syscall_num; 136 fsc.name = 137 (syscall_num < 0 || syscall_num > nsyscalls) ? NULL : linux_syscallnames[syscall_num]; 138 if (!fsc.name) { 139 fprintf(trussinfo->outfile, "-- UNKNOWN SYSCALL %d --\n", syscall_num); 140 } 141 142 if (fsc.name && (trussinfo->flags & FOLLOWFORKS) 143 && ((!strcmp(fsc.name, "linux_fork") 144 || !strcmp(fsc.name, "linux_vfork")))) 145 { 146 trussinfo->in_fork = 1; 147 } 148 149 if (nargs == 0) 150 return; 151 152 /* 153 * Linux passes syscall arguments in registers, not 154 * on the stack. Fortunately, we've got access to the 155 * register set. Note that we don't bother checking the 156 * number of arguments. And what does linux do for syscalls 157 * that have more than five arguments? 158 */ 159 160 fsc.args[0] = regs.r_ebx; 161 fsc.args[1] = regs.r_ecx; 162 fsc.args[2] = regs.r_edx; 163 fsc.args[3] = regs.r_esi; 164 fsc.args[4] = regs.r_edi; 165 166 sc = get_syscall(fsc.name); 167 if (sc) { 168 fsc.nargs = sc->nargs; 169 } else { 170#if DEBUG 171 fprintf(trussinfo->outfile, "unknown syscall %s -- setting args to %d\n", 172 fsc.name, nargs); 173#endif 174 fsc.nargs = nargs; 175 } 176 177 fsc.s_args = malloc((1+fsc.nargs) * sizeof(char*)); 178 memset(fsc.s_args, 0, fsc.nargs * sizeof(char*)); 179 fsc.sc = sc; 180 181 /* 182 * At this point, we set up the system call arguments. 183 * We ignore any OUT ones, however -- those are arguments that 184 * are set by the system call, and so are probably meaningless 185 * now. This doesn't currently support arguments that are 186 * passed in *and* out, however. 187 */ 188 189 if (fsc.name) { 190 191#if DEBUG 192 fprintf(stderr, "syscall %s(", fsc.name); 193#endif 194 for (i = 0; i < fsc.nargs; i++) { 195#if DEBUG 196 fprintf(stderr, "0x%x%s", 197 sc 198 ? fsc.args[sc->args[i].offset] 199 : fsc.args[i], 200 i < (fsc.nargs - 1) ? "," : ""); 201#endif 202 if (sc && !(sc->args[i].type & OUT)) { 203 fsc.s_args[i] = print_arg(Procfd, &sc->args[i], fsc.args, 0, trussinfo); 204 } 205 } 206#if DEBUG 207 fprintf(stderr, ")\n"); 208#endif 209 } 210 211#if DEBUG 212 fprintf(trussinfo->outfile, "\n"); 213#endif 214 215 /* 216 * Some system calls should be printed out before they are done -- 217 * execve() and exit(), for example, never return. Possibly change 218 * this to work for any system call that doesn't have an OUT 219 * parameter? 220 */ 221 222 if (fsc.name != NULL && 223 (!strcmp(fsc.name, "execve") || !strcmp(fsc.name, "exit"))) { 224 225 /* XXX 226 * This could be done in a more general 227 * manner but it still wouldn't be very pretty. 228 */ 229 if (!strcmp(fsc.name, "linux_execve")) { 230 if ((trussinfo->flags & EXECVEARGS) == 0) 231 if (fsc.s_args[1]) { 232 free(fsc.s_args[1]); 233 fsc.s_args[1] = NULL; 234 } 235 if ((trussinfo->flags & EXECVEENVS) == 0) 236 if (fsc.s_args[2]) { 237 free(fsc.s_args[2]); 238 fsc.s_args[2] = NULL; 239 } 240 } 241 242 print_syscall(trussinfo, fsc.name, fsc.nargs, fsc.s_args); 243 fprintf(trussinfo->outfile, "\n"); 244 } 245 246 return; 247} 248 249/* 250 * Linux syscalls return negative errno's, we do positive and map them 251 */ 252const int bsd_to_linux_errno[] = { 253 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9, 254 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19, 255 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29, 256 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89, 257 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99, 258 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109, 259 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122, 260 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9, 261 -6, 262}; 263 264long 265i386_linux_syscall_exit(struct trussinfo *trussinfo, int syscall_num __unused) 266{ 267 char buf[32]; 268 struct reg regs; 269 long retval; 270 int i; 271 int errorp; 272 struct syscall *sc; 273 274 if (fd == -1 || trussinfo->pid != cpid) { 275 sprintf(buf, "/proc/%d/regs", trussinfo->pid); 276 fd = open(buf, O_RDONLY); 277 if (fd == -1) { 278 fprintf(trussinfo->outfile, "-- CANNOT OPEN REGISTERS --\n"); 279 return (-1); 280 } 281 cpid = trussinfo->pid; 282 } 283 284 lseek(fd, 0L, 0); 285 if (read(fd, ®s, sizeof(regs)) != sizeof(regs)) { 286 fprintf(trussinfo->outfile, "\n"); 287 return (-1); 288 } 289 retval = regs.r_eax; 290 errorp = !!(regs.r_eflags & PSL_C); 291 292 /* 293 * This code, while simpler than the initial versions I used, could 294 * stand some significant cleaning. 295 */ 296 297 sc = fsc.sc; 298 if (!sc) { 299 for (i = 0; i < fsc.nargs; i++) 300 asprintf(&fsc.s_args[i], "0x%lx", fsc.args[i]); 301 } else { 302 /* 303 * Here, we only look for arguments that have OUT masked in -- 304 * otherwise, they were handled in the syscall_entry function. 305 */ 306 for (i = 0; i < sc->nargs; i++) { 307 char *temp; 308 if (sc->args[i].type & OUT) { 309 /* 310 * If an error occurred, than don't bothe getting the data; 311 * it may not be valid. 312 */ 313 if (errorp) 314 asprintf(&temp, "0x%lx", fsc.args[sc->args[i].offset]); 315 else 316 temp = print_arg(Procfd, &sc->args[i], fsc.args, retval, trussinfo); 317 fsc.s_args[i] = temp; 318 } 319 } 320 } 321 322 /* 323 * It would probably be a good idea to merge the error handling, 324 * but that complicates things considerably. 325 */ 326 if (errorp) { 327 for (i = 0; (size_t)i < sizeof(bsd_to_linux_errno) / sizeof(int); i++) 328 if (retval == bsd_to_linux_errno[i]) 329 break; 330 } 331 print_syscall_ret(trussinfo, fsc.name, fsc.nargs, fsc.s_args, errorp, 332 errorp ? i : retval); 333 clear_fsc(); 334 335 return (retval); 336} 337