1 /* $NetBSD: linux_exec.c,v 1.105 2008/04/28 20:23:43 martin Exp $ */ 2 3 /*- 4 * Copyright (c) 1994, 1995, 1998, 2000, 2007, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Christos Zoulas, Frank van der Linden, Eric Haszlakiewicz and 9 * Thor Lancelot Simon. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 #include <sys/cdefs.h> 34 __KERNEL_RCSID(0, "$NetBSD: linux_exec.c,v 1.105 2008/04/28 20:23:43 martin Exp $"); 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/kernel.h> 39 #include <sys/proc.h> 40 #include <sys/malloc.h> 41 #include <sys/namei.h> 42 #include <sys/vnode.h> 43 #include <sys/mount.h> 44 #include <sys/exec.h> 45 #include <sys/exec_elf.h> 46 47 #include <sys/mman.h> 48 #include <sys/syscallargs.h> 49 50 #include <sys/ptrace.h> /* For proc_reparent() */ 51 52 #include <uvm/uvm_extern.h> 53 54 #include <sys/cpu.h> 55 #include <machine/reg.h> 56 57 #include <compat/linux/common/linux_types.h> 58 #include <compat/linux/common/linux_signal.h> 59 #include <compat/linux/common/linux_util.h> 60 #include <compat/linux/common/linux_sched.h> 61 #include <compat/linux/common/linux_machdep.h> 62 #include <compat/linux/common/linux_exec.h> 63 #include <compat/linux/common/linux_futex.h> 64 #include <compat/linux/common/linux_ipc.h> 65 #include <compat/linux/common/linux_sem.h> 66 67 #include <compat/linux/linux_syscallargs.h> 68 #include <compat/linux/linux_syscall.h> 69 #include <compat/linux/common/linux_misc.h> 70 #include <compat/linux/common/linux_errno.h> 71 #include <compat/linux/common/linux_emuldata.h> 72 73 extern struct sysent linux_sysent[]; 74 extern const char * const linux_syscallnames[]; 75 extern char linux_sigcode[], linux_esigcode[]; 76 77 static void linux_e_proc_exec(struct proc *, struct exec_package *); 78 static void linux_e_proc_fork(struct proc *, struct proc *, int); 79 static void linux_e_proc_exit(struct proc *); 80 static void linux_e_proc_init(struct proc *, struct proc *, int); 81 82 #ifdef LINUX_NPTL 83 void linux_userret(void); 84 #endif 85 86 /* 87 * Emulation switch. 88 */ 89 90 struct uvm_object *emul_linux_object; 91 92 const struct emul emul_linux = { 93 "linux", 94 "/emul/linux", 95 #ifndef __HAVE_MINIMAL_EMUL 96 0, 97 (const int *)native_to_linux_errno, 98 LINUX_SYS_syscall, 99 LINUX_SYS_NSYSENT, 100 #endif 101 linux_sysent, 102 linux_syscallnames, 103 linux_sendsig, 104 linux_trapsignal, 105 NULL, 106 linux_sigcode, 107 linux_esigcode, 108 &emul_linux_object, 109 linux_setregs, 110 linux_e_proc_exec, 111 linux_e_proc_fork, 112 linux_e_proc_exit, 113 NULL, 114 NULL, 115 #ifdef __HAVE_SYSCALL_INTERN 116 linux_syscall_intern, 117 #else 118 #error Implement __HAVE_SYSCALL_INTERN for this platform 119 #endif 120 NULL, 121 NULL, 122 123 uvm_default_mapaddr, 124 125 linux_usertrap, 126 0, 127 NULL, /* e_startlwp */ 128 }; 129 130 static void 131 linux_e_proc_init(p, parent, forkflags) 132 struct proc *p, *parent; 133 int forkflags; 134 { 135 struct linux_emuldata *e = p->p_emuldata; 136 struct linux_emuldata_shared *s; 137 struct linux_emuldata *ep = NULL; 138 139 if (!e) { 140 /* allocate new Linux emuldata */ 141 MALLOC(e, void *, sizeof(struct linux_emuldata), 142 M_EMULDATA, M_WAITOK); 143 } else { 144 mutex_enter(proc_lock); 145 e->s->refs--; 146 if (e->s->refs == 0) 147 FREE(e->s, M_EMULDATA); 148 mutex_exit(proc_lock); 149 } 150 151 memset(e, '\0', sizeof(struct linux_emuldata)); 152 153 e->proc = p; 154 155 if (parent) 156 ep = parent->p_emuldata; 157 158 if (forkflags & FORK_SHAREVM) { 159 mutex_enter(proc_lock); 160 #ifdef DIAGNOSTIC 161 if (ep == NULL) { 162 killproc(p, "FORK_SHAREVM while emuldata is NULL\n"); 163 mutex_exit(proc_lock); 164 FREE(e, M_EMULDATA); 165 return; 166 } 167 #endif 168 s = ep->s; 169 s->refs++; 170 } else { 171 struct vmspace *vm; 172 173 MALLOC(s, void *, sizeof(struct linux_emuldata_shared), 174 M_EMULDATA, M_WAITOK); 175 s->refs = 1; 176 177 /* 178 * Set the process idea of the break to the real value. 179 * For fork, we use parent's vmspace since our's 180 * is not setup at the time of this call and is going 181 * to be copy of parent's anyway. For exec, just 182 * use our own vmspace. 183 */ 184 vm = (parent) ? parent->p_vmspace : p->p_vmspace; 185 s->p_break = (char *)vm->vm_daddr + ctob(vm->vm_dsize); 186 187 /* 188 * Linux threads are emulated as NetBSD processes (not lwp) 189 * We use native PID for Linux TID. The Linux TID is the 190 * PID of the first process in the group. It is stored 191 * here 192 */ 193 s->group_pid = p->p_pid; 194 195 /* 196 * Initialize the list of threads in the group 197 */ 198 LIST_INIT(&s->threads); 199 200 s->xstat = 0; 201 s->flags = 0; 202 mutex_enter(proc_lock); 203 } 204 205 e->s = s; 206 207 /* 208 * Add this thread in the group thread list 209 */ 210 LIST_INSERT_HEAD(&s->threads, e, threads); 211 212 #ifdef LINUX_NPTL 213 if (ep != NULL) { 214 e->parent_tidptr = ep->parent_tidptr; 215 e->child_tidptr = ep->child_tidptr; 216 e->clone_flags = ep->clone_flags; 217 } 218 #endif /* LINUX_NPTL */ 219 220 mutex_exit(proc_lock); 221 p->p_emuldata = e; 222 } 223 224 /* 225 * Allocate new per-process structures. Called when executing Linux 226 * process. We can reuse the old emuldata - if it's not null, 227 * the executed process is of same emulation as original forked one. 228 */ 229 static void 230 linux_e_proc_exec(struct proc *p, struct exec_package *epp) 231 { 232 233 /* exec, use our vmspace */ 234 linux_e_proc_init(p, NULL, 0); 235 } 236 237 /* 238 * Emulation per-process exit hook. 239 */ 240 static void 241 linux_e_proc_exit(struct proc *p) 242 { 243 struct linux_emuldata *e = p->p_emuldata; 244 245 #ifdef LINUX_NPTL 246 linux_nptl_proc_exit(p); 247 #endif 248 249 /* Remove the thread for the group thread list */ 250 mutex_enter(proc_lock); 251 LIST_REMOVE(e, threads); 252 253 /* free Linux emuldata and set the pointer to null */ 254 e->s->refs--; 255 if (e->s->refs == 0) 256 FREE(e->s, M_EMULDATA); 257 p->p_emuldata = NULL; 258 mutex_exit(proc_lock); 259 FREE(e, M_EMULDATA); 260 } 261 262 /* 263 * Emulation fork hook. 264 */ 265 static void 266 linux_e_proc_fork(p, parent, forkflags) 267 struct proc *p, *parent; 268 int forkflags; 269 { 270 271 /* 272 * The new process might share some vmspace-related stuff 273 * with parent, depending on fork flags (CLONE_VM et.al). 274 * Force allocation of new base emuldata, and share the 275 * VM-related parts only if necessary. 276 */ 277 p->p_emuldata = NULL; 278 linux_e_proc_init(p, parent, forkflags); 279 280 #ifdef LINUX_NPTL 281 linux_nptl_proc_fork(p, parent, linux_userret); 282 #endif 283 284 return; 285 } 286 287 #ifdef LINUX_NPTL 288 void 289 linux_userret(void) 290 { 291 struct lwp *l = curlwp; 292 struct proc *p = l->l_proc; 293 struct linux_emuldata *led = p->p_emuldata; 294 int error; 295 296 /* LINUX_CLONE_CHILD_SETTID: copy child's TID to child's memory */ 297 if (led->clone_flags & LINUX_CLONE_CHILD_SETTID) { 298 if ((error = copyout(&l->l_proc->p_pid, 299 led->child_tidptr, sizeof(l->l_proc->p_pid))) != 0) 300 printf("linux_userret: LINUX_CLONE_CHILD_SETTID " 301 "failed (led->child_tidptr = %p, p->p_pid = %d)\n", 302 led->child_tidptr, p->p_pid); 303 } 304 305 /* LINUX_CLONE_SETTLS: allocate a new TLS */ 306 if (led->clone_flags & LINUX_CLONE_SETTLS) { 307 if (linux_set_newtls(l, linux_get_newtls(l)) != 0) 308 printf("linux_userret: linux_set_tls failed"); 309 } 310 311 return; 312 } 313 314 void 315 linux_nptl_proc_exit(struct proc *p) 316 { 317 struct linux_emuldata *e = p->p_emuldata; 318 319 mutex_enter(proc_lock); 320 321 /* 322 * Check if we are a thread group leader victim of another 323 * thread doing exit_group(). If we are, change the exit code. 324 */ 325 if ((e->s->group_pid == p->p_pid) && 326 (e->s->flags & LINUX_LES_INEXITGROUP)) { 327 p->p_xstat = e->s->xstat; 328 } 329 330 /* 331 * Members of the thread groups others than the leader should 332 * exit quietely: no zombie stage, no signal. We do that by 333 * reparenting to init. init will collect us and nobody will 334 * notice what happened. 335 */ 336 #ifdef DEBUG_LINUX 337 printf("%s:%d e->s->group_pid = %d, p->p_pid = %d, flags = 0x%x\n", 338 __func__, __LINE__, e->s->group_pid, p->p_pid, e->s->flags); 339 #endif 340 if ((e->s->group_pid != p->p_pid) && 341 (e->clone_flags & LINUX_CLONE_THREAD)) { 342 proc_reparent(p, initproc); 343 cv_broadcast(&initproc->p_waitcv); 344 } 345 346 mutex_exit(proc_lock); 347 348 /* Emulate LINUX_CLONE_CHILD_CLEARTID */ 349 if (e->clear_tid != NULL) { 350 int error; 351 int null = 0; 352 struct linux_sys_futex_args cup; 353 register_t retval; 354 355 error = copyout(&null, e->clear_tid, sizeof(null)); 356 #ifdef DEBUG_LINUX 357 if (error != 0) 358 printf("%s: cannot clear TID\n", __func__); 359 #endif 360 361 SCARG(&cup, uaddr) = e->clear_tid; 362 SCARG(&cup, op) = LINUX_FUTEX_WAKE; 363 SCARG(&cup, val) = 0x7fffffff; /* Awake everyone */ 364 SCARG(&cup, timeout) = NULL; 365 SCARG(&cup, uaddr2) = NULL; 366 SCARG(&cup, val3) = 0; 367 if ((error = linux_sys_futex(curlwp, &cup, &retval)) != 0) 368 printf("%s: linux_sys_futex failed\n", __func__); 369 } 370 371 return; 372 } 373 374 void 375 linux_nptl_proc_fork(p, parent, luserret) 376 struct proc *p; 377 struct proc *parent; 378 void (*luserret)(void); 379 { 380 struct linux_emuldata *e = p->p_emuldata; 381 382 /* LINUX_CLONE_CHILD_CLEARTID: clear TID in child's memory on exit() */ 383 if (e->clone_flags & LINUX_CLONE_CHILD_CLEARTID) 384 e->clear_tid = e->child_tidptr; 385 386 /* LINUX_CLONE_PARENT_SETTID: set child's TID in parent's memory */ 387 if (e->clone_flags & LINUX_CLONE_PARENT_SETTID) { 388 if (copyout_proc(parent, &p->p_pid, 389 e->parent_tidptr, sizeof(p->p_pid)) != 0) 390 printf("%s: LINUX_CLONE_PARENT_SETTID " 391 "failed (e->parent_tidptr = %p, " 392 "parent->p_pid = %d, p->p_pid = %d)\n", 393 __func__, e->parent_tidptr, 394 parent->p_pid, p->p_pid); 395 } 396 397 /* 398 * CLONE_CHILD_SETTID and LINUX_CLONE_SETTLS require child's VM 399 * setup to be completed, we postpone them until userret time. 400 */ 401 if (e->clone_flags & 402 (LINUX_CLONE_CHILD_CLEARTID | LINUX_CLONE_SETTLS)) 403 p->p_userret = luserret; 404 405 return; 406 } 407 408 void 409 linux_nptl_proc_init(struct proc *p, struct proc *parent) 410 { 411 struct linux_emuldata *e = p->p_emuldata; 412 struct linux_emuldata *ep; 413 414 if ((parent != NULL) && (parent->p_emuldata != NULL)) { 415 ep = parent->p_emuldata; 416 417 e->parent_tidptr = ep->parent_tidptr; 418 e->child_tidptr = ep->child_tidptr; 419 e->clone_flags = ep->clone_flags; 420 } 421 422 return; 423 } 424 #endif /* LINUX_NPTL */ 425