xref: /freebsd/sys/compat/linux/linux_emul.c (revision eda14cbc264d6969b02f2b1994cef11148e914f1)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * Copyright (c) 2006 Roman Divacky
6  * Copyright (c) 2013 Dmitry Chagin
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/fcntl.h>
37 #include <sys/imgact.h>
38 #include <sys/kernel.h>
39 #include <sys/ktr.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mutex.h>
43 #include <sys/sx.h>
44 #include <sys/proc.h>
45 #include <sys/resourcevar.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysent.h>
48 
49 #include <compat/linux/linux_emul.h>
50 #include <compat/linux/linux_mib.h>
51 #include <compat/linux/linux_misc.h>
52 #include <compat/linux/linux_persona.h>
53 #include <compat/linux/linux_util.h>
54 
55 #if BYTE_ORDER == LITTLE_ENDIAN
56 #define SHELLMAGIC	0x2123 /* #! */
57 #else
58 #define SHELLMAGIC	0x2321
59 #endif
60 
61 /*
62  * This returns reference to the thread emuldata entry (if found)
63  *
64  * Hold PROC_LOCK when referencing emuldata from other threads.
65  */
66 struct linux_emuldata *
67 em_find(struct thread *td)
68 {
69 	struct linux_emuldata *em;
70 
71 	em = td->td_emuldata;
72 
73 	return (em);
74 }
75 
76 /*
77  * This returns reference to the proc pemuldata entry (if found)
78  *
79  * Hold PROC_LOCK when referencing proc pemuldata from other threads.
80  * Hold LINUX_PEM_LOCK wher referencing pemuldata members.
81  */
82 struct linux_pemuldata *
83 pem_find(struct proc *p)
84 {
85 	struct linux_pemuldata *pem;
86 
87 	pem = p->p_emuldata;
88 
89 	return (pem);
90 }
91 
92 /*
93  * Linux apps generally expect the soft open file limit to be set
94  * to 1024, often iterating over all the file descriptors up to that
95  * limit instead of using closefrom(2).  Give them what they want,
96  * unless there already is a resource limit in place.
97  */
98 static void
99 linux_set_default_openfiles(struct thread *td, struct proc *p)
100 {
101 	struct rlimit rlim;
102 	int error;
103 
104 	if (linux_default_openfiles < 0)
105 		return;
106 
107 	PROC_LOCK(p);
108 	lim_rlimit_proc(p, RLIMIT_NOFILE, &rlim);
109 	PROC_UNLOCK(p);
110 	if (rlim.rlim_cur != rlim.rlim_max ||
111 	    rlim.rlim_cur <= linux_default_openfiles)
112 		return;
113 	rlim.rlim_cur = linux_default_openfiles;
114 	error = kern_proc_setrlimit(td, p, RLIMIT_NOFILE, &rlim);
115 	KASSERT(error == 0, ("kern_proc_setrlimit failed"));
116 }
117 
118 void
119 linux_proc_init(struct thread *td, struct thread *newtd, int flags)
120 {
121 	struct linux_emuldata *em;
122 	struct linux_pemuldata *pem;
123 	struct epoll_emuldata *emd;
124 	struct proc *p;
125 
126 	if (newtd != NULL) {
127 		p = newtd->td_proc;
128 
129 		/* non-exec call */
130 		em = malloc(sizeof(*em), M_TEMP, M_WAITOK | M_ZERO);
131 		if (flags & LINUX_CLONE_THREAD) {
132 			LINUX_CTR1(proc_init, "thread newtd(%d)",
133 			    newtd->td_tid);
134 
135 			em->em_tid = newtd->td_tid;
136 		} else {
137 			LINUX_CTR1(proc_init, "fork newtd(%d)", p->p_pid);
138 
139 			em->em_tid = p->p_pid;
140 
141 			pem = malloc(sizeof(*pem), M_LINUX, M_WAITOK | M_ZERO);
142 			sx_init(&pem->pem_sx, "lpemlk");
143 			p->p_emuldata = pem;
144 		}
145 		newtd->td_emuldata = em;
146 
147 		linux_set_default_openfiles(td, p);
148 	} else {
149 		p = td->td_proc;
150 
151 		/* exec */
152 		LINUX_CTR1(proc_init, "exec newtd(%d)", p->p_pid);
153 
154 		/* lookup the old one */
155 		em = em_find(td);
156 		KASSERT(em != NULL, ("proc_init: emuldata not found in exec case.\n"));
157 
158 		em->em_tid = p->p_pid;
159 		em->flags = 0;
160 		em->robust_futexes = NULL;
161 		em->child_clear_tid = NULL;
162 		em->child_set_tid = NULL;
163 
164 		 /* epoll should be destroyed in a case of exec. */
165 		pem = pem_find(p);
166 		KASSERT(pem != NULL, ("proc_exit: proc emuldata not found.\n"));
167 		pem->persona = 0;
168 		if (pem->epoll != NULL) {
169 			emd = pem->epoll;
170 			pem->epoll = NULL;
171 			free(emd, M_EPOLL);
172 		}
173 	}
174 
175 }
176 
177 void
178 linux_proc_exit(void *arg __unused, struct proc *p)
179 {
180 	struct linux_pemuldata *pem;
181 	struct epoll_emuldata *emd;
182 	struct thread *td = curthread;
183 
184 	if (__predict_false(SV_CURPROC_ABI() != SV_ABI_LINUX))
185 		return;
186 
187 	LINUX_CTR3(proc_exit, "thread(%d) proc(%d) p %p",
188 	    td->td_tid, p->p_pid, p);
189 
190 	pem = pem_find(p);
191 	if (pem == NULL)
192 		return;
193 	(p->p_sysent->sv_thread_detach)(td);
194 
195 	p->p_emuldata = NULL;
196 
197 	if (pem->epoll != NULL) {
198 		emd = pem->epoll;
199 		pem->epoll = NULL;
200 		free(emd, M_EPOLL);
201 	}
202 
203 	sx_destroy(&pem->pem_sx);
204 	free(pem, M_LINUX);
205 }
206 
207 /*
208  * If a Linux binary is exec'ing something, try this image activator
209  * first.  We override standard shell script execution in order to
210  * be able to modify the interpreter path.  We only do this if a Linux
211  * binary is doing the exec, so we do not create an EXEC module for it.
212  */
213 int
214 linux_exec_imgact_try(struct image_params *imgp)
215 {
216 	const char *head = (const char *)imgp->image_header;
217 	char *rpath;
218 	int error = -1;
219 
220 	/*
221 	 * The interpreter for shell scripts run from a Linux binary needs
222 	 * to be located in /compat/linux if possible in order to recursively
223 	 * maintain Linux path emulation.
224 	 */
225 	if (((const short *)head)[0] == SHELLMAGIC) {
226 		/*
227 		 * Run our normal shell image activator.  If it succeeds attempt
228 		 * to use the alternate path for the interpreter.  If an
229 		 * alternate path is found, use our stringspace to store it.
230 		 */
231 		if ((error = exec_shell_imgact(imgp)) == 0) {
232 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
233 			    imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
234 			    AT_FDCWD);
235 			if (rpath != NULL)
236 				imgp->args->fname_buf =
237 				    imgp->interpreter_name = rpath;
238 		}
239 	}
240 	return (error);
241 }
242 
243 int
244 linux_common_execve(struct thread *td, struct image_args *eargs)
245 {
246 	struct linux_pemuldata *pem;
247 	struct epoll_emuldata *emd;
248 	struct vmspace *oldvmspace;
249 	struct linux_emuldata *em;
250 	struct proc *p;
251 	int error;
252 
253 	p = td->td_proc;
254 
255 	error = pre_execve(td, &oldvmspace);
256 	if (error != 0)
257 		return (error);
258 
259 	error = kern_execve(td, eargs, NULL);
260 	post_execve(td, error, oldvmspace);
261 	if (error != EJUSTRETURN)
262 		return (error);
263 
264 	/*
265 	 * In a case of transition from Linux binary execing to
266 	 * FreeBSD binary we destroy Linux emuldata thread & proc entries.
267 	 */
268 	if (SV_CURPROC_ABI() != SV_ABI_LINUX) {
269 		PROC_LOCK(p);
270 		em = em_find(td);
271 		KASSERT(em != NULL, ("proc_exec: thread emuldata not found.\n"));
272 		td->td_emuldata = NULL;
273 
274 		pem = pem_find(p);
275 		KASSERT(pem != NULL, ("proc_exec: proc pemuldata not found.\n"));
276 		p->p_emuldata = NULL;
277 		PROC_UNLOCK(p);
278 
279 		if (pem->epoll != NULL) {
280 			emd = pem->epoll;
281 			pem->epoll = NULL;
282 			free(emd, M_EPOLL);
283 		}
284 
285 		free(em, M_TEMP);
286 		free(pem, M_LINUX);
287 	}
288 	return (EJUSTRETURN);
289 }
290 
291 void
292 linux_proc_exec(void *arg __unused, struct proc *p, struct image_params *imgp)
293 {
294 	struct thread *td;
295 	struct thread *othertd;
296 #if defined(__amd64__)
297 	struct linux_pemuldata *pem;
298 #endif
299 
300 	td = curthread;
301 
302 	/*
303 	 * In a case of execing to Linux binary we create Linux
304 	 * emuldata thread entry.
305 	 */
306 	if (__predict_false((imgp->sysent->sv_flags & SV_ABI_MASK) ==
307 	    SV_ABI_LINUX)) {
308 		if (SV_PROC_ABI(p) == SV_ABI_LINUX) {
309 			/*
310 			 * Process already was under Linuxolator
311 			 * before exec.  Update emuldata to reflect
312 			 * single-threaded cleaned state after exec.
313 			 */
314 			linux_proc_init(td, NULL, 0);
315 		} else {
316 			/*
317 			 * We are switching the process to Linux emulator.
318 			 */
319 			linux_proc_init(td, td, 0);
320 
321 			/*
322 			 * Create a transient td_emuldata for all suspended
323 			 * threads, so that p->p_sysent->sv_thread_detach() ==
324 			 * linux_thread_detach() can find expected but unused
325 			 * emuldata.
326 			 */
327 			FOREACH_THREAD_IN_PROC(td->td_proc, othertd) {
328 				if (othertd != td) {
329 					linux_proc_init(td, othertd,
330 					    LINUX_CLONE_THREAD);
331 				}
332 			}
333 		}
334 #if defined(__amd64__)
335 		/*
336 		 * An IA32 executable which has executable stack will have the
337 		 * READ_IMPLIES_EXEC personality flag set automatically.
338 		 */
339 		if (SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
340 		    imgp->stack_prot & VM_PROT_EXECUTE) {
341 			pem = pem_find(p);
342 			pem->persona |= LINUX_READ_IMPLIES_EXEC;
343 		}
344 #endif
345 	}
346 }
347 
348 void
349 linux_thread_dtor(void *arg __unused, struct thread *td)
350 {
351 	struct linux_emuldata *em;
352 
353 	em = em_find(td);
354 	if (em == NULL)
355 		return;
356 	td->td_emuldata = NULL;
357 
358 	LINUX_CTR1(thread_dtor, "thread(%d)", em->em_tid);
359 
360 	free(em, M_TEMP);
361 }
362 
363 void
364 linux_schedtail(struct thread *td)
365 {
366 	struct linux_emuldata *em;
367 	struct proc *p;
368 	int error = 0;
369 	int *child_set_tid;
370 
371 	p = td->td_proc;
372 
373 	em = em_find(td);
374 	KASSERT(em != NULL, ("linux_schedtail: thread emuldata not found.\n"));
375 	child_set_tid = em->child_set_tid;
376 
377 	if (child_set_tid != NULL) {
378 		error = copyout(&em->em_tid, child_set_tid,
379 		    sizeof(em->em_tid));
380 		LINUX_CTR4(schedtail, "thread(%d) %p stored %d error %d",
381 		    td->td_tid, child_set_tid, em->em_tid, error);
382 	} else
383 		LINUX_CTR1(schedtail, "thread(%d)", em->em_tid);
384 }
385