xref: /freebsd/sys/compat/linux/linux_emul.c (revision cc68614da8232d8baaca0ae0d0dd8f890f06623e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * Copyright (c) 2006 Roman Divacky
6  * All rights reserved.
7  * Copyright (c) 2013 Dmitry Chagin <dchagin@FreeBSD.org>
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 __diagused;
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 __diagused;
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, bool init_thread)
143 {
144 	struct linux_emuldata *em;
145 	struct linux_pemuldata *pem;
146 	struct proc *p;
147 
148 	if (newtd != NULL) {
149 		p = newtd->td_proc;
150 
151 		/* non-exec call */
152 		em = malloc(sizeof(*em), M_TEMP, M_WAITOK | M_ZERO);
153 		if (init_thread) {
154 			LINUX_CTR1(proc_init, "thread newtd(%d)",
155 			    newtd->td_tid);
156 
157 			em->em_tid = newtd->td_tid;
158 		} else {
159 			LINUX_CTR1(proc_init, "fork newtd(%d)", p->p_pid);
160 
161 			em->em_tid = p->p_pid;
162 
163 			pem = malloc(sizeof(*pem), M_LINUX, M_WAITOK | M_ZERO);
164 			sx_init(&pem->pem_sx, "lpemlk");
165 			p->p_emuldata = pem;
166 		}
167 		newtd->td_emuldata = em;
168 
169 		linux_set_default_openfiles(td, p);
170 		linux_set_default_stacksize(td, p);
171 	} else {
172 		p = td->td_proc;
173 
174 		/* exec */
175 		LINUX_CTR1(proc_init, "exec newtd(%d)", p->p_pid);
176 
177 		/* lookup the old one */
178 		em = em_find(td);
179 		KASSERT(em != NULL, ("proc_init: emuldata not found in exec case.\n"));
180 
181 		em->em_tid = p->p_pid;
182 		em->flags = 0;
183 		em->robust_futexes = NULL;
184 		em->child_clear_tid = NULL;
185 		em->child_set_tid = NULL;
186 
187 		pem = pem_find(p);
188 		KASSERT(pem != NULL, ("proc_exit: proc emuldata not found.\n"));
189 		pem->persona = 0;
190 	}
191 
192 }
193 
194 void
195 linux_on_exit(struct proc *p)
196 {
197 	struct linux_pemuldata *pem;
198 	struct thread *td = curthread;
199 
200 	MPASS(SV_CURPROC_ABI() == SV_ABI_LINUX);
201 
202 	LINUX_CTR3(proc_exit, "thread(%d) proc(%d) p %p",
203 	    td->td_tid, p->p_pid, p);
204 
205 	pem = pem_find(p);
206 	if (pem == NULL)
207 		return;
208 	(p->p_sysent->sv_thread_detach)(td);
209 
210 	p->p_emuldata = NULL;
211 
212 	sx_destroy(&pem->pem_sx);
213 	free(pem, M_LINUX);
214 }
215 
216 /*
217  * If a Linux binary is exec'ing something, try this image activator
218  * first.  We override standard shell script execution in order to
219  * be able to modify the interpreter path.  We only do this if a Linux
220  * binary is doing the exec, so we do not create an EXEC module for it.
221  */
222 int
223 linux_exec_imgact_try(struct image_params *imgp)
224 {
225 	const char *head = (const char *)imgp->image_header;
226 	char *rpath;
227 	int error = -1;
228 
229 	/*
230 	 * The interpreter for shell scripts run from a Linux binary needs
231 	 * to be located in /compat/linux if possible in order to recursively
232 	 * maintain Linux path emulation.
233 	 */
234 	if (((const short *)head)[0] == SHELLMAGIC) {
235 		/*
236 		 * Run our normal shell image activator.  If it succeeds attempt
237 		 * to use the alternate path for the interpreter.  If an
238 		 * alternate path is found, use our stringspace to store it.
239 		 */
240 		if ((error = exec_shell_imgact(imgp)) == 0) {
241 			linux_emul_convpath(imgp->interpreter_name, UIO_SYSSPACE,
242 			    &rpath, 0, AT_FDCWD);
243 			if (rpath != NULL)
244 				imgp->args->fname_buf =
245 				    imgp->interpreter_name = rpath;
246 		}
247 	}
248 	return (error);
249 }
250 
251 int
252 linux_common_execve(struct thread *td, struct image_args *eargs)
253 {
254 	struct linux_pemuldata *pem;
255 	struct vmspace *oldvmspace;
256 	struct linux_emuldata *em;
257 	struct proc *p;
258 	int error;
259 
260 	p = td->td_proc;
261 
262 	error = pre_execve(td, &oldvmspace);
263 	if (error != 0)
264 		return (error);
265 
266 	error = kern_execve(td, eargs, NULL, oldvmspace);
267 	post_execve(td, error, oldvmspace);
268 	if (error != EJUSTRETURN)
269 		return (error);
270 
271 	/*
272 	 * In a case of transition from Linux binary execing to
273 	 * FreeBSD binary we destroy Linux emuldata thread & proc entries.
274 	 */
275 	if (SV_CURPROC_ABI() != SV_ABI_LINUX) {
276 		PROC_LOCK(p);
277 		em = em_find(td);
278 		KASSERT(em != NULL, ("proc_exec: thread emuldata not found.\n"));
279 		td->td_emuldata = NULL;
280 
281 		pem = pem_find(p);
282 		KASSERT(pem != NULL, ("proc_exec: proc pemuldata not found.\n"));
283 		p->p_emuldata = NULL;
284 		PROC_UNLOCK(p);
285 
286 		free(em, M_TEMP);
287 		free(pem, M_LINUX);
288 	}
289 	return (EJUSTRETURN);
290 }
291 
292 void
293 linux_on_exec(struct proc *p, struct image_params *imgp)
294 {
295 	struct thread *td;
296 	struct thread *othertd;
297 #if defined(__amd64__)
298 	struct linux_pemuldata *pem;
299 #endif
300 
301 	td = curthread;
302 	MPASS((imgp->sysent->sv_flags & SV_ABI_MASK) == SV_ABI_LINUX);
303 
304 	/*
305 	 * When execing to Linux binary, we create Linux emuldata
306 	 * thread entry.
307 	 */
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, false);
315 	} else {
316 		/*
317 		 * We are switching the process to Linux emulator.
318 		 */
319 		linux_proc_init(td, td, false);
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 				continue;
330 			linux_proc_init(td, othertd, true);
331 		}
332 	}
333 #if defined(__amd64__)
334 	/*
335 	 * An IA32 executable which has executable stack will have the
336 	 * READ_IMPLIES_EXEC personality flag set automatically.
337 	 */
338 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
339 	    imgp->stack_prot & VM_PROT_EXECUTE) {
340 		pem = pem_find(p);
341 		pem->persona |= LINUX_READ_IMPLIES_EXEC;
342 	}
343 #endif
344 }
345 
346 void
347 linux_thread_dtor(struct thread *td)
348 {
349 	struct linux_emuldata *em;
350 
351 	em = em_find(td);
352 	if (em == NULL)
353 		return;
354 	td->td_emuldata = NULL;
355 
356 	LINUX_CTR1(thread_dtor, "thread(%d)", em->em_tid);
357 
358 	free(em, M_TEMP);
359 }
360 
361 void
362 linux_schedtail(struct thread *td)
363 {
364 	struct linux_emuldata *em;
365 #ifdef KTR
366 	int error;
367 #else
368 	int error __unused;
369 #endif
370 	int *child_set_tid;
371 
372 	em = em_find(td);
373 	KASSERT(em != NULL, ("linux_schedtail: thread emuldata not found.\n"));
374 	child_set_tid = em->child_set_tid;
375 
376 	if (child_set_tid != NULL) {
377 		error = copyout(&em->em_tid, child_set_tid,
378 		    sizeof(em->em_tid));
379 		LINUX_CTR4(schedtail, "thread(%d) %p stored %d error %d",
380 		    td->td_tid, child_set_tid, em->em_tid, error);
381 	} else
382 		LINUX_CTR1(schedtail, "thread(%d)", em->em_tid);
383 }
384