Home | History | Annotate | Line # | Download | only in common
linux_exec.c revision 1.91.2.2
      1 /*	$NetBSD: linux_exec.c,v 1.91.2.2 2007/03/12 05:52:26 rmind Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1994, 1995, 1998, 2000, 2007 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  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *        This product includes software developed by the NetBSD
     22  *        Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 #include <sys/cdefs.h>
     41 __KERNEL_RCSID(0, "$NetBSD: linux_exec.c,v 1.91.2.2 2007/03/12 05:52:26 rmind Exp $");
     42 
     43 #include <sys/param.h>
     44 #include <sys/systm.h>
     45 #include <sys/kernel.h>
     46 #include <sys/proc.h>
     47 #include <sys/malloc.h>
     48 #include <sys/namei.h>
     49 #include <sys/vnode.h>
     50 #include <sys/mount.h>
     51 #include <sys/exec.h>
     52 #include <sys/exec_elf.h>
     53 
     54 #include <sys/mman.h>
     55 #include <sys/syscallargs.h>
     56 
     57 #include <sys/ptrace.h>	/* For proc_reparent() */
     58 
     59 #include <uvm/uvm_extern.h>
     60 
     61 #include <machine/cpu.h>
     62 #include <machine/reg.h>
     63 
     64 #include <compat/linux/common/linux_types.h>
     65 #include <compat/linux/common/linux_signal.h>
     66 #include <compat/linux/common/linux_util.h>
     67 #include <compat/linux/common/linux_sched.h>
     68 #include <compat/linux/common/linux_machdep.h>
     69 #include <compat/linux/common/linux_exec.h>
     70 #include <compat/linux/common/linux_futex.h>
     71 
     72 #include <compat/linux/linux_syscallargs.h>
     73 #include <compat/linux/linux_syscall.h>
     74 #include <compat/linux/common/linux_misc.h>
     75 #include <compat/linux/common/linux_errno.h>
     76 #include <compat/linux/common/linux_emuldata.h>
     77 
     78 extern struct sysent linux_sysent[];
     79 extern const char * const linux_syscallnames[];
     80 extern char linux_sigcode[], linux_esigcode[];
     81 
     82 static void linux_e_proc_exec __P((struct proc *, struct exec_package *));
     83 static void linux_e_proc_fork __P((struct proc *, struct proc *, int));
     84 static void linux_e_proc_exit __P((struct proc *));
     85 static void linux_e_proc_init __P((struct proc *, struct proc *, int));
     86 
     87 #ifdef LINUX_NPTL
     88 void linux_userret(void);
     89 #endif
     90 
     91 /*
     92  * Execve(2). Just check the alternate emulation path, and pass it on
     93  * to the NetBSD execve().
     94  */
     95 int
     96 linux_sys_execve(l, v, retval)
     97 	struct lwp *l;
     98 	void *v;
     99 	register_t *retval;
    100 {
    101 	struct linux_sys_execve_args /* {
    102 		syscallarg(const char *) path;
    103 		syscallarg(char **) argv;
    104 		syscallarg(char **) envp;
    105 	} */ *uap = v;
    106 	struct proc *p = l->l_proc;
    107 	struct sys_execve_args ap;
    108 	void *sg;
    109 
    110 	sg = stackgap_init(p, 0);
    111 	CHECK_ALT_EXIST(l, &sg, SCARG(uap, path));
    112 
    113 	SCARG(&ap, path) = SCARG(uap, path);
    114 	SCARG(&ap, argp) = SCARG(uap, argp);
    115 	SCARG(&ap, envp) = SCARG(uap, envp);
    116 
    117 	return sys_execve(l, &ap, retval);
    118 }
    119 
    120 /*
    121  * Emulation switch.
    122  */
    123 
    124 struct uvm_object *emul_linux_object;
    125 
    126 const struct emul emul_linux = {
    127 	"linux",
    128 	"/emul/linux",
    129 #ifndef __HAVE_MINIMAL_EMUL
    130 	0,
    131 	(const int *)native_to_linux_errno,
    132 	LINUX_SYS_syscall,
    133 	LINUX_SYS_NSYSENT,
    134 #endif
    135 	linux_sysent,
    136 	linux_syscallnames,
    137 	linux_sendsig,
    138 	linux_trapsignal,
    139 	NULL,
    140 	linux_sigcode,
    141 	linux_esigcode,
    142 	&emul_linux_object,
    143 	linux_setregs,
    144 	linux_e_proc_exec,
    145 	linux_e_proc_fork,
    146 	linux_e_proc_exit,
    147 	NULL,
    148 	NULL,
    149 #ifdef __HAVE_SYSCALL_INTERN
    150 	linux_syscall_intern,
    151 #else
    152 #error Implement __HAVE_SYSCALL_INTERN for this platform
    153 #endif
    154 	NULL,
    155 	NULL,
    156 
    157 	uvm_default_mapaddr,
    158 
    159 	linux_usertrap,
    160 	0,
    161 	NULL,		/* e_startlwp */
    162 };
    163 
    164 static void
    165 linux_e_proc_init(p, parent, forkflags)
    166 	struct proc *p, *parent;
    167 	int forkflags;
    168 {
    169 	struct linux_emuldata *e = p->p_emuldata;
    170 	struct linux_emuldata_shared *s;
    171 	struct linux_emuldata *ep = NULL;
    172 
    173 	if (!e) {
    174 		/* allocate new Linux emuldata */
    175 		MALLOC(e, void *, sizeof(struct linux_emuldata),
    176 			M_EMULDATA, M_WAITOK);
    177 	} else  {
    178 		e->s->refs--;
    179 		if (e->s->refs == 0)
    180 			FREE(e->s, M_EMULDATA);
    181 	}
    182 
    183 	memset(e, '\0', sizeof(struct linux_emuldata));
    184 
    185 	e->proc = p;
    186 
    187 	if (parent)
    188 		ep = parent->p_emuldata;
    189 
    190 	if (forkflags & FORK_SHAREVM) {
    191 #ifdef DIAGNOSTIC
    192 		if (ep == NULL) {
    193 			killproc(p, "FORK_SHAREVM while emuldata is NULL\n");
    194 			FREE(e, M_EMULDATA);
    195 			return;
    196 		}
    197 #endif
    198 		s = ep->s;
    199 		s->refs++;
    200 	} else {
    201 		struct vmspace *vm;
    202 
    203 		MALLOC(s, void *, sizeof(struct linux_emuldata_shared),
    204 			M_EMULDATA, M_WAITOK);
    205 		s->refs = 1;
    206 
    207 		/*
    208 		 * Set the process idea of the break to the real value.
    209 		 * For fork, we use parent's vmspace since our's
    210 		 * is not setup at the time of this call and is going
    211 		 * to be copy of parent's anyway. For exec, just
    212 		 * use our own vmspace.
    213 		 */
    214 		vm = (parent) ? parent->p_vmspace : p->p_vmspace;
    215 		s->p_break = (char *)vm->vm_daddr + ctob(vm->vm_dsize);
    216 
    217 		/*
    218 		 * Linux threads are emulated as NetBSD processes (not lwp)
    219 		 * We use native PID for Linux TID. The Linux TID is the
    220 		 * PID of the first process in the group. It is stored
    221 		 * here
    222 		 */
    223 		s->group_pid = p->p_pid;
    224 
    225 		/*
    226 		 * Initialize the list of threads in the group
    227 		 */
    228 		LIST_INIT(&s->threads);
    229 
    230 		s->xstat = 0;
    231 		s->flags = 0;
    232 	}
    233 
    234 	e->s = s;
    235 
    236 	/*
    237 	 * Add this thread in the group thread list
    238 	 */
    239 	LIST_INSERT_HEAD(&s->threads, e, threads);
    240 
    241 #ifdef LINUX_NPTL
    242 	if (ep != NULL) {
    243 		e->parent_tidptr = ep->parent_tidptr;
    244 		e->child_tidptr = ep->child_tidptr;
    245 		e->clone_flags = ep->clone_flags;
    246 	}
    247 #endif /* LINUX_NPTL */
    248 
    249 	p->p_emuldata = e;
    250 }
    251 
    252 /*
    253  * Allocate new per-process structures. Called when executing Linux
    254  * process. We can reuse the old emuldata - if it's not null,
    255  * the executed process is of same emulation as original forked one.
    256  */
    257 static void
    258 linux_e_proc_exec(struct proc *p, struct exec_package *epp)
    259 {
    260 	/* exec, use our vmspace */
    261 	linux_e_proc_init(p, NULL, 0);
    262 }
    263 
    264 /*
    265  * Emulation per-process exit hook.
    266  */
    267 static void
    268 linux_e_proc_exit(p)
    269 	struct proc *p;
    270 {
    271 	struct linux_emuldata *e = p->p_emuldata;
    272 
    273 #ifdef LINUX_NPTL
    274 	linux_nptl_proc_exit(p);
    275 #endif
    276 	/* Remove the thread for the group thread list */
    277 	LIST_REMOVE(e, threads);
    278 
    279 	/* free Linux emuldata and set the pointer to null */
    280 	e->s->refs--;
    281 	if (e->s->refs == 0)
    282 		FREE(e->s, M_EMULDATA);
    283 	FREE(e, M_EMULDATA);
    284 	p->p_emuldata = NULL;
    285 }
    286 
    287 /*
    288  * Emulation fork hook.
    289  */
    290 static void
    291 linux_e_proc_fork(p, parent, forkflags)
    292 	struct proc *p, *parent;
    293 	int forkflags;
    294 {
    295 	/*
    296 	 * The new process might share some vmspace-related stuff
    297 	 * with parent, depending on fork flags (CLONE_VM et.al).
    298 	 * Force allocation of new base emuldata, and share the
    299 	 * VM-related parts only if necessary.
    300 	 */
    301 	p->p_emuldata = NULL;
    302 	linux_e_proc_init(p, parent, forkflags);
    303 
    304 #ifdef LINUX_NPTL
    305 	linux_nptl_proc_fork(p, parent, linux_userret);
    306 #endif
    307 
    308 	return;
    309 }
    310 
    311 #ifdef LINUX_NPTL
    312 void
    313 linux_userret(void)
    314 {
    315 	struct lwp *l = curlwp;
    316 	struct proc *p = l->l_proc;
    317 	struct linux_emuldata *led = p->p_emuldata;
    318 	int error;
    319 
    320 	/* LINUX_CLONE_CHILD_SETTID: copy child's TID to child's memory  */
    321 	if (led->clone_flags & LINUX_CLONE_CHILD_SETTID) {
    322 		if ((error = copyout(&l->l_proc->p_pid,
    323 		    led->child_tidptr,  sizeof(l->l_proc->p_pid))) != 0)
    324 			printf("linux_userret: LINUX_CLONE_CHILD_SETTID "
    325 			    "failed (led->child_tidptr = %p, p->p_pid = %d)\n",
    326 			    led->child_tidptr, p->p_pid);
    327 	}
    328 
    329 	/* LINUX_CLONE_SETTLS: allocate a new TLS */
    330 	if (led->clone_flags & LINUX_CLONE_SETTLS) {
    331 		if (linux_set_newtls(l, linux_get_newtls(l)) != 0)
    332 			printf("linux_userret: linux_set_tls failed");
    333 	}
    334 
    335 	return;
    336 }
    337 
    338 void
    339 linux_nptl_proc_exit(p)
    340 	struct proc *p;
    341 {
    342 	struct linux_emuldata *e = p->p_emuldata;
    343 
    344 	mutex_enter(&proclist_lock);
    345 
    346 	/*
    347 	 * Check if we are a thread group leader victim of another
    348 	 * thread doing exit_group(). If we are, change the exit code.
    349 	 */
    350 	if ((e->s->group_pid == p->p_pid) &&
    351 	    (e->s->flags & LINUX_LES_INEXITGROUP)) {
    352 		p->p_xstat = e->s->xstat;
    353 	}
    354 
    355 	/*
    356 	 * Members of the thread groups others than the leader should
    357 	 * exit quietely: no zombie stage, no signal. We do that by
    358 	 * reparenting to init. init will collect us and nobody will
    359 	 * notice what happened.
    360 	 */
    361 #ifdef DEBUG_LINUX
    362 	printf("%s:%d e->s->group_pid = %d, p->p_pid = %d, flags = 0x%x\n",
    363 	    __func__, __LINE__, e->s->group_pid, p->p_pid, e->s->flags);
    364 #endif
    365 	if (e->s->group_pid != p->p_pid) {
    366 		proc_reparent(p, initproc);
    367 		cv_broadcast(&initproc->p_waitcv);
    368 	}
    369 
    370 	mutex_exit(&proclist_lock);
    371 
    372 	/* Emulate LINUX_CLONE_CHILD_CLEARTID */
    373 	if (e->clear_tid != NULL) {
    374 		int error;
    375 		int null = 0;
    376 		struct linux_sys_futex_args cup;
    377 		register_t retval;
    378 
    379 		error = copyout(&null, e->clear_tid, sizeof(null));
    380 #ifdef DEBUG_LINUX
    381 		if (error != 0)
    382 			printf("%s: cannot clear TID\n", __func__);
    383 #endif
    384 
    385 		SCARG(&cup, uaddr) = e->clear_tid;
    386 		SCARG(&cup, op) = LINUX_FUTEX_WAKE;
    387 		SCARG(&cup, val) = 0x7fffffff; /* Awake everyone */
    388 		SCARG(&cup, timeout) = NULL;
    389 		SCARG(&cup, uaddr2) = NULL;
    390 		SCARG(&cup, val3) = 0;
    391 		if ((error = linux_sys_futex(curlwp, &cup, &retval)) != 0)
    392 			printf("%s: linux_sys_futex failed\n", __func__);
    393 	}
    394 
    395 	return;
    396 }
    397 
    398 void
    399 linux_nptl_proc_fork(p, parent, luserret)
    400 	struct proc *p;
    401 	struct proc *parent;
    402 	void (*luserret)(void);
    403 {
    404 #ifdef LINUX_NPTL
    405 	struct linux_emuldata *e;
    406 #endif
    407 
    408 	e = p->p_emuldata;
    409 
    410 	/* LINUX_CLONE_CHILD_CLEARTID: clear TID in child's memory on exit() */
    411 	if (e->clone_flags & LINUX_CLONE_CHILD_CLEARTID)
    412 		e->clear_tid = e->child_tidptr;
    413 
    414 	/* LINUX_CLONE_PARENT_SETTID: set child's TID in parent's memory */
    415 	if (e->clone_flags & LINUX_CLONE_PARENT_SETTID) {
    416 		if (copyout_proc(parent, &p->p_pid,
    417 		    e->parent_tidptr,  sizeof(p->p_pid)) != 0)
    418 			printf("%s: LINUX_CLONE_PARENT_SETTID "
    419 			    "failed (e->parent_tidptr = %p, "
    420 			    "parent->p_pid = %d, p->p_pid = %d)\n",
    421 			    __func__, e->parent_tidptr,
    422 			    parent->p_pid, p->p_pid);
    423 	}
    424 
    425 	/*
    426 	 * CLONE_CHILD_SETTID and LINUX_CLONE_SETTLS require child's VM
    427 	 * setup to be completed, we postpone them until userret time.
    428 	 */
    429 	if (e->clone_flags &
    430 	    (LINUX_CLONE_CHILD_CLEARTID | LINUX_CLONE_SETTLS))
    431 		p->p_userret = luserret;
    432 
    433 	return;
    434 }
    435 
    436 void
    437 linux_nptl_proc_init(p, parent)
    438 	struct proc *p;
    439 	struct proc *parent;
    440 {
    441 	struct linux_emuldata *e = p->p_emuldata;
    442 	struct linux_emuldata *ep;
    443 
    444 	if ((parent != NULL) && (parent->p_emuldata != NULL)) {
    445 		ep = parent->p_emuldata;
    446 
    447 		e->parent_tidptr = ep->parent_tidptr;
    448 		e->child_tidptr = ep->child_tidptr;
    449 		e->clone_flags = ep->clone_flags;
    450 	}
    451 
    452 	return;
    453 }
    454 
    455 
    456 #endif /* LINUX_NPTL */
    457