xref: /freebsd/sys/amd64/linux/linux_machdep.c (revision 4db78cacdee1b6f3b7880eb8c5560e8edaf91698)
1 /*-
2  * Copyright (c) 2013 Dmitry Chagin
3  * Copyright (c) 2004 Tim J. Robbins
4  * Copyright (c) 2002 Doug Rabson
5  * Copyright (c) 2000 Marcel Moolenaar
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer
13  *    in this position and unchanged.
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  * 3. The name of the author may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/capsicum.h>
37 #include <sys/clock.h>
38 #include <sys/dirent.h>
39 #include <sys/fcntl.h>
40 #include <sys/file.h>
41 #include <sys/filedesc.h>
42 #include <sys/imgact.h>
43 #include <sys/kernel.h>
44 #include <sys/ktr.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mman.h>
49 #include <sys/mutex.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/resource.h>
53 #include <sys/resourcevar.h>
54 #include <sys/sched.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysproto.h>
57 #include <sys/systm.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/wait.h>
61 
62 #include <security/mac/mac_framework.h>
63 
64 #include <ufs/ufs/extattr.h>
65 #include <ufs/ufs/quota.h>
66 #include <ufs/ufs/ufsmount.h>
67 
68 #include <machine/frame.h>
69 #include <machine/md_var.h>
70 #include <machine/pcb.h>
71 #include <machine/psl.h>
72 #include <machine/segments.h>
73 #include <machine/specialreg.h>
74 
75 #include <vm/pmap.h>
76 #include <vm/vm.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_map.h>
80 
81 #include <x86/ifunc.h>
82 #include <x86/sysarch.h>
83 
84 #include <amd64/linux/linux.h>
85 #include <amd64/linux/linux_proto.h>
86 #include <compat/linux/linux_emul.h>
87 #include <compat/linux/linux_file.h>
88 #include <compat/linux/linux_ipc.h>
89 #include <compat/linux/linux_misc.h>
90 #include <compat/linux/linux_mmap.h>
91 #include <compat/linux/linux_signal.h>
92 #include <compat/linux/linux_util.h>
93 
94 int
95 linux_execve(struct thread *td, struct linux_execve_args *args)
96 {
97 	struct image_args eargs;
98 	char *path;
99 	int error;
100 
101 	LCONVPATHEXIST(td, args->path, &path);
102 
103 	LINUX_CTR(execve);
104 
105 	error = exec_copyin_args(&eargs, path, UIO_SYSSPACE, args->argp,
106 	    args->envp);
107 	free(path, M_TEMP);
108 	if (error == 0)
109 		error = linux_common_execve(td, &eargs);
110 	return (error);
111 }
112 
113 int
114 linux_set_upcall_kse(struct thread *td, register_t stack)
115 {
116 
117 	if (stack)
118 		td->td_frame->tf_rsp = stack;
119 
120 	/*
121 	 * The newly created Linux thread returns
122 	 * to the user space by the same path that a parent do.
123 	 */
124 	td->td_frame->tf_rax = 0;
125 	return (0);
126 }
127 
128 int
129 linux_mmap2(struct thread *td, struct linux_mmap2_args *args)
130 {
131 
132 	return (linux_mmap_common(td, PTROUT(args->addr), args->len, args->prot,
133 		args->flags, args->fd, args->pgoff));
134 }
135 
136 int
137 linux_mprotect(struct thread *td, struct linux_mprotect_args *uap)
138 {
139 
140 	return (linux_mprotect_common(td, PTROUT(uap->addr), uap->len, uap->prot));
141 }
142 
143 int
144 linux_iopl(struct thread *td, struct linux_iopl_args *args)
145 {
146 	int error;
147 
148 	LINUX_CTR(iopl);
149 
150 	if (args->level > 3)
151 		return (EINVAL);
152 	if ((error = priv_check(td, PRIV_IO)) != 0)
153 		return (error);
154 	if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
155 		return (error);
156 	td->td_frame->tf_rflags = (td->td_frame->tf_rflags & ~PSL_IOPL) |
157 	    (args->level * (PSL_IOPL / 3));
158 
159 	return (0);
160 }
161 
162 int
163 linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap)
164 {
165 	l_sigset_t lmask;
166 	sigset_t sigmask;
167 	int error;
168 
169 	LINUX_CTR2(rt_sigsuspend, "%p, %ld",
170 	    uap->newset, uap->sigsetsize);
171 
172 	if (uap->sigsetsize != sizeof(l_sigset_t))
173 		return (EINVAL);
174 
175 	error = copyin(uap->newset, &lmask, sizeof(l_sigset_t));
176 	if (error)
177 		return (error);
178 
179 	linux_to_bsd_sigset(&lmask, &sigmask);
180 	return (kern_sigsuspend(td, sigmask));
181 }
182 
183 int
184 linux_pause(struct thread *td, struct linux_pause_args *args)
185 {
186 	struct proc *p = td->td_proc;
187 	sigset_t sigmask;
188 
189 	LINUX_CTR(pause);
190 
191 	PROC_LOCK(p);
192 	sigmask = td->td_sigmask;
193 	PROC_UNLOCK(p);
194 	return (kern_sigsuspend(td, sigmask));
195 }
196 
197 int
198 linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap)
199 {
200 	stack_t ss, oss;
201 	l_stack_t lss;
202 	int error;
203 
204 	LINUX_CTR2(sigaltstack, "%p, %p", uap->uss, uap->uoss);
205 
206 	if (uap->uss != NULL) {
207 		error = copyin(uap->uss, &lss, sizeof(l_stack_t));
208 		if (error)
209 			return (error);
210 
211 		ss.ss_sp = PTRIN(lss.ss_sp);
212 		ss.ss_size = lss.ss_size;
213 		ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags);
214 	}
215 	error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL,
216 	    (uap->uoss != NULL) ? &oss : NULL);
217 	if (!error && uap->uoss != NULL) {
218 		lss.ss_sp = PTROUT(oss.ss_sp);
219 		lss.ss_size = oss.ss_size;
220 		lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags);
221 		error = copyout(&lss, uap->uoss, sizeof(l_stack_t));
222 	}
223 
224 	return (error);
225 }
226 
227 int
228 linux_arch_prctl(struct thread *td, struct linux_arch_prctl_args *args)
229 {
230 	int error;
231 	struct sysarch_args bsd_args;
232 
233 	LINUX_CTR2(arch_prctl, "0x%x, %p", args->code, args->addr);
234 
235 	switch (args->code) {
236 	case LINUX_ARCH_SET_GS:
237 		bsd_args.op = AMD64_SET_GSBASE;
238 		bsd_args.parms = (void *)args->addr;
239 		error = sysarch(td, &bsd_args);
240 		if (error == EINVAL)
241 			error = EPERM;
242 		break;
243 	case LINUX_ARCH_SET_FS:
244 		bsd_args.op = AMD64_SET_FSBASE;
245 		bsd_args.parms = (void *)args->addr;
246 		error = sysarch(td, &bsd_args);
247 		if (error == EINVAL)
248 			error = EPERM;
249 		break;
250 	case LINUX_ARCH_GET_FS:
251 		bsd_args.op = AMD64_GET_FSBASE;
252 		bsd_args.parms = (void *)args->addr;
253 		error = sysarch(td, &bsd_args);
254 		break;
255 	case LINUX_ARCH_GET_GS:
256 		bsd_args.op = AMD64_GET_GSBASE;
257 		bsd_args.parms = (void *)args->addr;
258 		error = sysarch(td, &bsd_args);
259 		break;
260 	default:
261 		error = EINVAL;
262 	}
263 	return (error);
264 }
265 
266 int
267 linux_set_cloned_tls(struct thread *td, void *desc)
268 {
269 	struct pcb *pcb;
270 
271 	if ((uint64_t)desc >= VM_MAXUSER_ADDRESS)
272 		return (EPERM);
273 
274 	pcb = td->td_pcb;
275 	pcb->pcb_fsbase = (register_t)desc;
276 	td->td_frame->tf_fs = _ufssel;
277 
278 	return (0);
279 }
280 
281 int futex_xchgl_nosmap(int oparg, uint32_t *uaddr, int *oldval);
282 int futex_xchgl_smap(int oparg, uint32_t *uaddr, int *oldval);
283 DEFINE_IFUNC(, int, futex_xchgl, (int, uint32_t *, int *), static)
284 {
285 
286 	return ((cpu_stdext_feature & CPUID_STDEXT_SMAP) != 0 ?
287 	    futex_xchgl_smap : futex_xchgl_nosmap);
288 }
289 
290 int futex_addl_nosmap(int oparg, uint32_t *uaddr, int *oldval);
291 int futex_addl_smap(int oparg, uint32_t *uaddr, int *oldval);
292 DEFINE_IFUNC(, int, futex_addl, (int, uint32_t *, int *), static)
293 {
294 
295 	return ((cpu_stdext_feature & CPUID_STDEXT_SMAP) != 0 ?
296 	    futex_addl_smap : futex_addl_nosmap);
297 }
298 
299 int futex_orl_nosmap(int oparg, uint32_t *uaddr, int *oldval);
300 int futex_orl_smap(int oparg, uint32_t *uaddr, int *oldval);
301 DEFINE_IFUNC(, int, futex_orl, (int, uint32_t *, int *), static)
302 {
303 
304 	return ((cpu_stdext_feature & CPUID_STDEXT_SMAP) != 0 ?
305 	    futex_orl_smap : futex_orl_nosmap);
306 }
307 
308 int futex_andl_nosmap(int oparg, uint32_t *uaddr, int *oldval);
309 int futex_andl_smap(int oparg, uint32_t *uaddr, int *oldval);
310 DEFINE_IFUNC(, int, futex_andl, (int, uint32_t *, int *), static)
311 {
312 
313 	return ((cpu_stdext_feature & CPUID_STDEXT_SMAP) != 0 ?
314 	    futex_andl_smap : futex_andl_nosmap);
315 }
316 
317 int futex_xorl_nosmap(int oparg, uint32_t *uaddr, int *oldval);
318 int futex_xorl_smap(int oparg, uint32_t *uaddr, int *oldval);
319 DEFINE_IFUNC(, int, futex_xorl, (int, uint32_t *, int *), static)
320 {
321 
322 	return ((cpu_stdext_feature & CPUID_STDEXT_SMAP) != 0 ?
323 	    futex_xorl_smap : futex_xorl_nosmap);
324 }
325