xref: /freebsd/sys/compat/linux/linux_emul.c (revision e2eeea75eb8b6dd50c1298067a0655880d186734)
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 /*
119  * The default stack size limit in Linux is 8MB.
120  */
121 static void
122 linux_set_default_stacksize(struct thread *td, struct proc *p)
123 {
124 	struct rlimit rlim;
125 	int error;
126 
127 	if (linux_default_stacksize < 0)
128 		return;
129 
130 	PROC_LOCK(p);
131 	lim_rlimit_proc(p, RLIMIT_STACK, &rlim);
132 	PROC_UNLOCK(p);
133 	if (rlim.rlim_cur != rlim.rlim_max ||
134 	    rlim.rlim_cur <= linux_default_stacksize)
135 		return;
136 	rlim.rlim_cur = linux_default_stacksize;
137 	error = kern_proc_setrlimit(td, p, RLIMIT_STACK, &rlim);
138 	KASSERT(error == 0, ("kern_proc_setrlimit failed"));
139 }
140 
141 void
142 linux_proc_init(struct thread *td, struct thread *newtd, int flags)
143 {
144 	struct linux_emuldata *em;
145 	struct linux_pemuldata *pem;
146 	struct epoll_emuldata *emd;
147 	struct proc *p;
148 
149 	if (newtd != NULL) {
150 		p = newtd->td_proc;
151 
152 		/* non-exec call */
153 		em = malloc(sizeof(*em), M_TEMP, M_WAITOK | M_ZERO);
154 		if (flags & LINUX_CLONE_THREAD) {
155 			LINUX_CTR1(proc_init, "thread newtd(%d)",
156 			    newtd->td_tid);
157 
158 			em->em_tid = newtd->td_tid;
159 		} else {
160 			LINUX_CTR1(proc_init, "fork newtd(%d)", p->p_pid);
161 
162 			em->em_tid = p->p_pid;
163 
164 			pem = malloc(sizeof(*pem), M_LINUX, M_WAITOK | M_ZERO);
165 			sx_init(&pem->pem_sx, "lpemlk");
166 			p->p_emuldata = pem;
167 		}
168 		newtd->td_emuldata = em;
169 
170 		linux_set_default_openfiles(td, p);
171 		linux_set_default_stacksize(td, p);
172 	} else {
173 		p = td->td_proc;
174 
175 		/* exec */
176 		LINUX_CTR1(proc_init, "exec newtd(%d)", p->p_pid);
177 
178 		/* lookup the old one */
179 		em = em_find(td);
180 		KASSERT(em != NULL, ("proc_init: emuldata not found in exec case.\n"));
181 
182 		em->em_tid = p->p_pid;
183 		em->flags = 0;
184 		em->robust_futexes = NULL;
185 		em->child_clear_tid = NULL;
186 		em->child_set_tid = NULL;
187 
188 		 /* epoll should be destroyed in a case of exec. */
189 		pem = pem_find(p);
190 		KASSERT(pem != NULL, ("proc_exit: proc emuldata not found.\n"));
191 		pem->persona = 0;
192 		if (pem->epoll != NULL) {
193 			emd = pem->epoll;
194 			pem->epoll = NULL;
195 			free(emd, M_EPOLL);
196 		}
197 	}
198 
199 }
200 
201 void
202 linux_proc_exit(void *arg __unused, struct proc *p)
203 {
204 	struct linux_pemuldata *pem;
205 	struct epoll_emuldata *emd;
206 	struct thread *td = curthread;
207 
208 	if (__predict_false(SV_CURPROC_ABI() != SV_ABI_LINUX))
209 		return;
210 
211 	LINUX_CTR3(proc_exit, "thread(%d) proc(%d) p %p",
212 	    td->td_tid, p->p_pid, p);
213 
214 	pem = pem_find(p);
215 	if (pem == NULL)
216 		return;
217 	(p->p_sysent->sv_thread_detach)(td);
218 
219 	p->p_emuldata = NULL;
220 
221 	if (pem->epoll != NULL) {
222 		emd = pem->epoll;
223 		pem->epoll = NULL;
224 		free(emd, M_EPOLL);
225 	}
226 
227 	sx_destroy(&pem->pem_sx);
228 	free(pem, M_LINUX);
229 }
230 
231 /*
232  * If a Linux binary is exec'ing something, try this image activator
233  * first.  We override standard shell script execution in order to
234  * be able to modify the interpreter path.  We only do this if a Linux
235  * binary is doing the exec, so we do not create an EXEC module for it.
236  */
237 int
238 linux_exec_imgact_try(struct image_params *imgp)
239 {
240 	const char *head = (const char *)imgp->image_header;
241 	char *rpath;
242 	int error = -1;
243 
244 	/*
245 	 * The interpreter for shell scripts run from a Linux binary needs
246 	 * to be located in /compat/linux if possible in order to recursively
247 	 * maintain Linux path emulation.
248 	 */
249 	if (((const short *)head)[0] == SHELLMAGIC) {
250 		/*
251 		 * Run our normal shell image activator.  If it succeeds attempt
252 		 * to use the alternate path for the interpreter.  If an
253 		 * alternate path is found, use our stringspace to store it.
254 		 */
255 		if ((error = exec_shell_imgact(imgp)) == 0) {
256 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
257 			    imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
258 			    AT_FDCWD);
259 			if (rpath != NULL)
260 				imgp->args->fname_buf =
261 				    imgp->interpreter_name = rpath;
262 		}
263 	}
264 	return (error);
265 }
266 
267 int
268 linux_common_execve(struct thread *td, struct image_args *eargs)
269 {
270 	struct linux_pemuldata *pem;
271 	struct epoll_emuldata *emd;
272 	struct vmspace *oldvmspace;
273 	struct linux_emuldata *em;
274 	struct proc *p;
275 	int error;
276 
277 	p = td->td_proc;
278 
279 	error = pre_execve(td, &oldvmspace);
280 	if (error != 0)
281 		return (error);
282 
283 	error = kern_execve(td, eargs, NULL, oldvmspace);
284 	post_execve(td, error, oldvmspace);
285 	if (error != EJUSTRETURN)
286 		return (error);
287 
288 	/*
289 	 * In a case of transition from Linux binary execing to
290 	 * FreeBSD binary we destroy Linux emuldata thread & proc entries.
291 	 */
292 	if (SV_CURPROC_ABI() != SV_ABI_LINUX) {
293 		PROC_LOCK(p);
294 		em = em_find(td);
295 		KASSERT(em != NULL, ("proc_exec: thread emuldata not found.\n"));
296 		td->td_emuldata = NULL;
297 
298 		pem = pem_find(p);
299 		KASSERT(pem != NULL, ("proc_exec: proc pemuldata not found.\n"));
300 		p->p_emuldata = NULL;
301 		PROC_UNLOCK(p);
302 
303 		if (pem->epoll != NULL) {
304 			emd = pem->epoll;
305 			pem->epoll = NULL;
306 			free(emd, M_EPOLL);
307 		}
308 
309 		free(em, M_TEMP);
310 		free(pem, M_LINUX);
311 	}
312 	return (EJUSTRETURN);
313 }
314 
315 void
316 linux_proc_exec(void *arg __unused, struct proc *p, struct image_params *imgp)
317 {
318 	struct thread *td;
319 	struct thread *othertd;
320 #if defined(__amd64__)
321 	struct linux_pemuldata *pem;
322 #endif
323 
324 	td = curthread;
325 
326 	/*
327 	 * In a case of execing to Linux binary we create Linux
328 	 * emuldata thread entry.
329 	 */
330 	if (__predict_false((imgp->sysent->sv_flags & SV_ABI_MASK) ==
331 	    SV_ABI_LINUX)) {
332 		if (SV_PROC_ABI(p) == SV_ABI_LINUX) {
333 			/*
334 			 * Process already was under Linuxolator
335 			 * before exec.  Update emuldata to reflect
336 			 * single-threaded cleaned state after exec.
337 			 */
338 			linux_proc_init(td, NULL, 0);
339 		} else {
340 			/*
341 			 * We are switching the process to Linux emulator.
342 			 */
343 			linux_proc_init(td, td, 0);
344 
345 			/*
346 			 * Create a transient td_emuldata for all suspended
347 			 * threads, so that p->p_sysent->sv_thread_detach() ==
348 			 * linux_thread_detach() can find expected but unused
349 			 * emuldata.
350 			 */
351 			FOREACH_THREAD_IN_PROC(td->td_proc, othertd) {
352 				if (othertd != td) {
353 					linux_proc_init(td, othertd,
354 					    LINUX_CLONE_THREAD);
355 				}
356 			}
357 		}
358 #if defined(__amd64__)
359 		/*
360 		 * An IA32 executable which has executable stack will have the
361 		 * READ_IMPLIES_EXEC personality flag set automatically.
362 		 */
363 		if (SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
364 		    imgp->stack_prot & VM_PROT_EXECUTE) {
365 			pem = pem_find(p);
366 			pem->persona |= LINUX_READ_IMPLIES_EXEC;
367 		}
368 #endif
369 	}
370 }
371 
372 void
373 linux_thread_dtor(void *arg __unused, struct thread *td)
374 {
375 	struct linux_emuldata *em;
376 
377 	em = em_find(td);
378 	if (em == NULL)
379 		return;
380 	td->td_emuldata = NULL;
381 
382 	LINUX_CTR1(thread_dtor, "thread(%d)", em->em_tid);
383 
384 	free(em, M_TEMP);
385 }
386 
387 void
388 linux_schedtail(struct thread *td)
389 {
390 	struct linux_emuldata *em;
391 	struct proc *p;
392 	int error = 0;
393 	int *child_set_tid;
394 
395 	p = td->td_proc;
396 
397 	em = em_find(td);
398 	KASSERT(em != NULL, ("linux_schedtail: thread emuldata not found.\n"));
399 	child_set_tid = em->child_set_tid;
400 
401 	if (child_set_tid != NULL) {
402 		error = copyout(&em->em_tid, child_set_tid,
403 		    sizeof(em->em_tid));
404 		LINUX_CTR4(schedtail, "thread(%d) %p stored %d error %d",
405 		    td->td_tid, child_set_tid, em->em_tid, error);
406 	} else
407 		LINUX_CTR1(schedtail, "thread(%d)", em->em_tid);
408 }
409