xref: /netbsd-src/sys/compat/linux/common/linux_exec.c (revision 62a8debe1dc62962e18a1c918def78666141273b)
1 /*	$NetBSD: linux_exec.c,v 1.113 2009/10/25 01:14:03 rmind 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 1.113 2009/10/25 01:14:03 rmind 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 struct emul emul_linux = {
93 	.e_name =		"linux",
94 	.e_path =		"/emul/linux",
95 #ifndef __HAVE_MINIMAL_EMUL
96 	.e_flags =		0,
97 	.e_errno =		(const int *)native_to_linux_errno,
98 	.e_nosys =		LINUX_SYS_syscall,
99 	.e_nsysent =		LINUX_SYS_NSYSENT,
100 #endif
101 	.e_sysent =		linux_sysent,
102 	.e_syscallnames =	linux_syscallnames,
103 	.e_sendsig =		linux_sendsig,
104 	.e_trapsignal =		linux_trapsignal,
105 	.e_tracesig =		NULL,
106 	.e_sigcode =		linux_sigcode,
107 	.e_esigcode =		linux_esigcode,
108 	.e_sigobject =		&emul_linux_object,
109 	.e_setregs =		linux_setregs,
110 	.e_proc_exec =		linux_e_proc_exec,
111 	.e_proc_fork =		linux_e_proc_fork,
112 	.e_proc_exit =		linux_e_proc_exit,
113 	.e_lwp_fork =		NULL,
114 	.e_lwp_exit =		NULL,
115 #ifdef __HAVE_SYSCALL_INTERN
116 	.e_syscall_intern =	linux_syscall_intern,
117 #else
118 #error Implement __HAVE_SYSCALL_INTERN for this platform
119 #endif
120 	.e_sysctlovly =		NULL,
121 	.e_fault =		NULL,
122 	.e_vm_default_addr =	uvm_default_mapaddr,
123 	.e_usertrap =		linux_usertrap,
124 	.e_sa =			NULL,
125 	.e_ucsize =		0,
126 	.e_startlwp =		NULL
127 };
128 
129 static void
130 linux_e_proc_init(struct proc *p, struct proc *parent, int forkflags)
131 {
132 	struct linux_emuldata *e = p->p_emuldata;
133 	struct linux_emuldata_shared *s;
134 	struct linux_emuldata *ep = NULL;
135 
136 	if (!e) {
137 		/* allocate new Linux emuldata */
138 		e = malloc(sizeof(struct linux_emuldata),
139 			M_EMULDATA, M_WAITOK);
140 	} else  {
141 		mutex_enter(proc_lock);
142 		e->s->refs--;
143 		if (e->s->refs == 0)
144 			free(e->s, M_EMULDATA);
145 		mutex_exit(proc_lock);
146 	}
147 
148 	memset(e, '\0', sizeof(struct linux_emuldata));
149 
150 	e->proc = p;
151 	e->robust_futexes = NULL;
152 
153 	if (parent)
154 		ep = parent->p_emuldata;
155 
156 	if (forkflags & FORK_SHAREVM) {
157 		mutex_enter(proc_lock);
158 #ifdef DIAGNOSTIC
159 		if (ep == NULL) {
160 			killproc(p, "FORK_SHAREVM while emuldata is NULL\n");
161 			mutex_exit(proc_lock);
162 			free(e, M_EMULDATA);
163 			return;
164 		}
165 #endif
166 		s = ep->s;
167 		s->refs++;
168 	} else {
169 		struct vmspace *vm;
170 
171 		s = malloc(sizeof(struct linux_emuldata_shared),
172 			M_EMULDATA, M_WAITOK);
173 		s->refs = 1;
174 
175 		/*
176 		 * Set the process idea of the break to the real value.
177 		 * For fork, we use parent's vmspace since our's
178 		 * is not setup at the time of this call and is going
179 		 * to be copy of parent's anyway. For exec, just
180 		 * use our own vmspace.
181 		 */
182 		vm = (parent) ? parent->p_vmspace : p->p_vmspace;
183 		s->p_break = (char *)vm->vm_daddr + ctob(vm->vm_dsize);
184 
185 		/*
186 		 * Linux threads are emulated as NetBSD processes (not lwp)
187 		 * We use native PID for Linux TID. The Linux TID is the
188 		 * PID of the first process in the group. It is stored
189 		 * here
190 		 */
191 		s->group_pid = p->p_pid;
192 
193 		/*
194 		 * Initialize the list of threads in the group
195 		 */
196 		LIST_INIT(&s->threads);
197 
198 		s->xstat = 0;
199 		s->flags = 0;
200 		mutex_enter(proc_lock);
201 	}
202 
203 	e->s = s;
204 
205 	/*
206 	 * Add this thread in the group thread list
207 	 */
208 	LIST_INSERT_HEAD(&s->threads, e, threads);
209 
210 #ifdef LINUX_NPTL
211 	if (ep != NULL) {
212 		e->parent_tidptr = ep->parent_tidptr;
213 		e->child_tidptr = ep->child_tidptr;
214 		e->clone_flags = ep->clone_flags;
215 	}
216 #endif /* LINUX_NPTL */
217 
218 	mutex_exit(proc_lock);
219 	p->p_emuldata = e;
220 }
221 
222 /*
223  * Allocate new per-process structures. Called when executing Linux
224  * process. We can reuse the old emuldata - if it's not null,
225  * the executed process is of same emulation as original forked one.
226  */
227 static void
228 linux_e_proc_exec(struct proc *p, struct exec_package *epp)
229 {
230 
231 	/* exec, use our vmspace */
232 	linux_e_proc_init(p, NULL, 0);
233 }
234 
235 /*
236  * Emulation per-process exit hook.
237  */
238 static void
239 linux_e_proc_exit(struct proc *p)
240 {
241 	struct linux_emuldata *e = p->p_emuldata;
242 
243 #ifdef LINUX_NPTL
244 	linux_nptl_proc_exit(p);
245 	release_futexes(p);
246 #endif
247 
248 	/* Remove the thread for the group thread list */
249 	mutex_enter(proc_lock);
250 	LIST_REMOVE(e, threads);
251 
252 	/* free Linux emuldata and set the pointer to null */
253 	e->s->refs--;
254 	if (e->s->refs == 0)
255 		free(e->s, M_EMULDATA);
256 	p->p_emuldata = NULL;
257 	mutex_exit(proc_lock);
258 	free(e, M_EMULDATA);
259 }
260 
261 /*
262  * Emulation fork hook.
263  */
264 static void
265 linux_e_proc_fork(struct proc *p, struct proc *parent, int forkflags)
266 {
267 
268 	/*
269 	 * The new process might share some vmspace-related stuff
270 	 * with parent, depending on fork flags (CLONE_VM et.al).
271 	 * Force allocation of new base emuldata, and share the
272 	 * VM-related parts only if necessary.
273 	 */
274 	p->p_emuldata = NULL;
275 	linux_e_proc_init(p, parent, forkflags);
276 
277 #ifdef LINUX_NPTL
278 	linux_nptl_proc_fork(p, parent, linux_userret);
279 #endif
280 
281 	return;
282 }
283 
284 #ifdef LINUX_NPTL
285 void
286 linux_userret(void)
287 {
288 	struct lwp *l = curlwp;
289 	struct proc *p = l->l_proc;
290 	struct linux_emuldata *led = p->p_emuldata;
291 	int error;
292 
293 	/* LINUX_CLONE_CHILD_SETTID: copy child's TID to child's memory  */
294 	if (led->clone_flags & LINUX_CLONE_CHILD_SETTID) {
295 		if ((error = copyout(&l->l_proc->p_pid,
296 		    led->child_tidptr,  sizeof(l->l_proc->p_pid))) != 0)
297 			printf("linux_userret: LINUX_CLONE_CHILD_SETTID "
298 			    "failed (led->child_tidptr = %p, p->p_pid = %d)\n",
299 			    led->child_tidptr, p->p_pid);
300 	}
301 
302 	/* LINUX_CLONE_SETTLS: allocate a new TLS */
303 	if (led->clone_flags & LINUX_CLONE_SETTLS) {
304 		if (linux_set_newtls(l, linux_get_newtls(l)) != 0)
305 			printf("linux_userret: linux_set_tls failed");
306 	}
307 
308 	return;
309 }
310 
311 void
312 linux_nptl_proc_exit(struct proc *p)
313 {
314 	struct linux_emuldata *e = p->p_emuldata;
315 
316 	mutex_enter(proc_lock);
317 
318 	/*
319 	 * Check if we are a thread group leader victim of another
320 	 * thread doing exit_group(). If we are, change the exit code.
321 	 */
322 	if ((e->s->group_pid == p->p_pid) &&
323 	    (e->s->flags & LINUX_LES_INEXITGROUP)) {
324 		p->p_xstat = e->s->xstat;
325 	}
326 
327 	/*
328 	 * Members of the thread groups others than the leader should
329 	 * exit quietely: no zombie stage, no signal. We do that by
330 	 * reparenting to init. init will collect us and nobody will
331 	 * notice what happened.
332 	 */
333 #ifdef DEBUG_LINUX
334 	printf("%s:%d e->s->group_pid = %d, p->p_pid = %d, flags = 0x%x\n",
335 	    __func__, __LINE__, e->s->group_pid, p->p_pid, e->s->flags);
336 #endif
337 	if ((e->s->group_pid != p->p_pid) &&
338 	    (e->clone_flags & LINUX_CLONE_THREAD)) {
339 		proc_reparent(p, initproc);
340 		cv_broadcast(&initproc->p_waitcv);
341 	}
342 
343 	mutex_exit(proc_lock);
344 
345 	/* Emulate LINUX_CLONE_CHILD_CLEARTID */
346 	if (e->clear_tid != NULL) {
347 		int error;
348 		int null = 0;
349 		struct linux_sys_futex_args cup;
350 		register_t retval;
351 
352 		error = copyout(&null, e->clear_tid, sizeof(null));
353 #ifdef DEBUG_LINUX
354 		if (error != 0)
355 			printf("%s: cannot clear TID\n", __func__);
356 #endif
357 
358 		SCARG(&cup, uaddr) = e->clear_tid;
359 		SCARG(&cup, op) = LINUX_FUTEX_WAKE;
360 		SCARG(&cup, val) = 0x7fffffff; /* Awake everyone */
361 		SCARG(&cup, timeout) = NULL;
362 		SCARG(&cup, uaddr2) = NULL;
363 		SCARG(&cup, val3) = 0;
364 		if ((error = linux_sys_futex(curlwp, &cup, &retval)) != 0)
365 			printf("%s: linux_sys_futex failed\n", __func__);
366 	}
367 
368 	return;
369 }
370 
371 void
372 linux_nptl_proc_fork(struct proc *p, struct proc *parent, void (*luserret)(void))
373 {
374 	struct linux_emuldata *e = p->p_emuldata;
375 
376 	/* LINUX_CLONE_CHILD_CLEARTID: clear TID in child's memory on exit() */
377 	if (e->clone_flags & LINUX_CLONE_CHILD_CLEARTID)
378 		e->clear_tid = e->child_tidptr;
379 
380 	/* LINUX_CLONE_PARENT_SETTID: set child's TID in parent's memory */
381 	if (e->clone_flags & LINUX_CLONE_PARENT_SETTID) {
382 		if (copyout_proc(parent, &p->p_pid,
383 		    e->parent_tidptr,  sizeof(p->p_pid)) != 0)
384 			printf("%s: LINUX_CLONE_PARENT_SETTID "
385 			    "failed (e->parent_tidptr = %p, "
386 			    "parent->p_pid = %d, p->p_pid = %d)\n",
387 			    __func__, e->parent_tidptr,
388 			    parent->p_pid, p->p_pid);
389 	}
390 
391 	/*
392 	 * CLONE_CHILD_SETTID and LINUX_CLONE_SETTLS require child's VM
393 	 * setup to be completed, we postpone them until userret time.
394 	 */
395 	if (e->clone_flags &
396 	    (LINUX_CLONE_CHILD_CLEARTID | LINUX_CLONE_SETTLS))
397 		p->p_userret = luserret;
398 
399 	return;
400 }
401 
402 void
403 linux_nptl_proc_init(struct proc *p, struct proc *parent)
404 {
405 	struct linux_emuldata *e = p->p_emuldata;
406 	struct linux_emuldata *ep;
407 
408 	if ((parent != NULL) && (parent->p_emuldata != NULL)) {
409 		ep = parent->p_emuldata;
410 
411 		e->parent_tidptr = ep->parent_tidptr;
412 		e->child_tidptr = ep->child_tidptr;
413 		e->clone_flags = ep->clone_flags;
414 	}
415 
416 	return;
417 }
418 #endif /* LINUX_NPTL */
419