xref: /freebsd-13-stable/sys/i386/linux/linux_sysvec.c (revision dc2622bb27f78b1119eb9cbd9db0e0c2b16fa099)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 #define __ELF_WORD_SIZE	32
31 
32 #include <sys/param.h>
33 #include <sys/exec.h>
34 #include <sys/fcntl.h>
35 #include <sys/imgact.h>
36 #include <sys/imgact_aout.h>
37 #include <sys/imgact_elf.h>
38 #include <sys/kernel.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/mutex.h>
43 #include <sys/proc.h>
44 #include <sys/stddef.h>
45 #include <sys/syscallsubr.h>
46 #include <sys/sysctl.h>
47 #include <sys/sysent.h>
48 #include <sys/sysproto.h>
49 
50 #include <vm/pmap.h>
51 #include <vm/vm.h>
52 #include <vm/vm_map.h>
53 #include <vm/vm_page.h>
54 
55 #include <machine/cpu.h>
56 #include <machine/cputypes.h>
57 #include <machine/md_var.h>
58 #include <machine/pcb.h>
59 #include <machine/trap.h>
60 
61 #include <x86/linux/linux_x86.h>
62 #include <i386/linux/linux.h>
63 #include <i386/linux/linux_proto.h>
64 #include <compat/linux/linux_elf.h>
65 #include <compat/linux/linux_emul.h>
66 #include <compat/linux/linux_fork.h>
67 #include <compat/linux/linux_ioctl.h>
68 #include <compat/linux/linux_mib.h>
69 #include <compat/linux/linux_misc.h>
70 #include <compat/linux/linux_signal.h>
71 #include <compat/linux/linux_util.h>
72 #include <compat/linux/linux_vdso.h>
73 
74 #include <x86/linux/linux_x86_sigframe.h>
75 
76 MODULE_VERSION(linux, 1);
77 
78 #define	LINUX_VDSOPAGE_SIZE	PAGE_SIZE * 2
79 #define	LINUX_VDSOPAGE		(VM_MAXUSER_ADDRESS - LINUX_VDSOPAGE_SIZE)
80 #define	LINUX_SHAREDPAGE	(LINUX_VDSOPAGE - PAGE_SIZE)
81 				/*
82 				 * PAGE_SIZE - the size
83 				 * of the native SHAREDPAGE
84 				 */
85 #define	LINUX_USRSTACK		LINUX_SHAREDPAGE
86 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
87 
88 static int linux_szsigcode;
89 static vm_object_t linux_vdso_obj;
90 static char *linux_vdso_mapping;
91 extern char _binary_linux_vdso_so_o_start;
92 extern char _binary_linux_vdso_so_o_end;
93 static vm_offset_t linux_vdso_base;
94 
95 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
96 extern const char *linux_syscallnames[];
97 
98 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
99 
100 static int	linux_fixup(uintptr_t *stack_base,
101 		    struct image_params *iparams);
102 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
103 static void	linux_exec_setregs(struct thread *td,
104 		    struct image_params *imgp, uintptr_t stack);
105 static void	linux_exec_sysvec_init(void *param);
106 static int	linux_on_exec_vmspace(struct proc *p,
107 		    struct image_params *imgp);
108 static void	linux_set_fork_retval(struct thread *td);
109 static void	linux_vdso_install(const void *param);
110 static void	linux_vdso_deinstall(const void *param);
111 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
112 
113 LINUX_VDSO_SYM_CHAR(linux_platform);
114 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
115 LINUX_VDSO_SYM_INTPTR(linux_vdso_sigcode);
116 LINUX_VDSO_SYM_INTPTR(linux_vdso_rt_sigcode);
117 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
118 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
119 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
120 
121 static int
linux_fixup(uintptr_t * stack_base,struct image_params * imgp)122 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
123 {
124 	register_t *base, *argv, *envp;
125 
126 	base = (register_t *)*stack_base;
127 	argv = base;
128 	envp = base + (imgp->args->argc + 1);
129 	base--;
130 	if (suword(base, (intptr_t)envp) != 0)
131 		return (EFAULT);
132 	base--;
133 	if (suword(base, (intptr_t)argv) != 0)
134 		return (EFAULT);
135 	base--;
136 	if (suword(base, imgp->args->argc) != 0)
137 		return (EFAULT);
138 	*stack_base = (uintptr_t)base;
139 	return (0);
140 }
141 
142 void
linux32_arch_copyout_auxargs(struct image_params * imgp,Elf_Auxinfo ** pos)143 linux32_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos)
144 {
145 
146 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
147 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO, __kernel_vsyscall);
148 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature);
149 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2());
150 	AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform));
151 }
152 
153 static void
linux_rt_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)154 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
155 {
156 	struct thread *td = curthread;
157 	struct proc *p = td->td_proc;
158 	struct sigacts *psp;
159 	struct trapframe *regs;
160 	struct l_rt_sigframe *fp, frame;
161 	int sig, code;
162 	int oonstack;
163 
164 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
165 	code = ksi->ksi_code;
166 	PROC_LOCK_ASSERT(p, MA_OWNED);
167 	psp = p->p_sigacts;
168 	mtx_assert(&psp->ps_mtx, MA_OWNED);
169 	regs = td->td_frame;
170 	oonstack = sigonstack(regs->tf_esp);
171 
172 	/* Allocate space for the signal handler context. */
173 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
174 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
175 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
176 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
177 	} else
178 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
179 	mtx_unlock(&psp->ps_mtx);
180 
181 	/* Build the argument list for the signal handler. */
182 	sig = bsd_to_linux_signal(sig);
183 
184 	bzero(&frame, sizeof(frame));
185 
186 	frame.sf_sig = sig;
187 	frame.sf_siginfo = PTROUT(&fp->sf_si);
188 	frame.sf_ucontext = PTROUT(&fp->sf_uc);
189 
190 	/* Fill in POSIX parts. */
191 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
192 
193 	/* Build the signal context to be used by sigreturn. */
194 	frame.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
195 	frame.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
196 	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
197 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
198 	PROC_UNLOCK(p);
199 
200 	bsd_to_linux_sigset(mask, &frame.sf_uc.uc_sigmask);
201 
202 	frame.sf_uc.uc_mcontext.sc_mask   = frame.sf_uc.uc_sigmask.__mask;
203 	frame.sf_uc.uc_mcontext.sc_gs     = rgs();
204 	frame.sf_uc.uc_mcontext.sc_fs     = regs->tf_fs;
205 	frame.sf_uc.uc_mcontext.sc_es     = regs->tf_es;
206 	frame.sf_uc.uc_mcontext.sc_ds     = regs->tf_ds;
207 	frame.sf_uc.uc_mcontext.sc_edi    = regs->tf_edi;
208 	frame.sf_uc.uc_mcontext.sc_esi    = regs->tf_esi;
209 	frame.sf_uc.uc_mcontext.sc_ebp    = regs->tf_ebp;
210 	frame.sf_uc.uc_mcontext.sc_ebx    = regs->tf_ebx;
211 	frame.sf_uc.uc_mcontext.sc_esp    = regs->tf_esp;
212 	frame.sf_uc.uc_mcontext.sc_edx    = regs->tf_edx;
213 	frame.sf_uc.uc_mcontext.sc_ecx    = regs->tf_ecx;
214 	frame.sf_uc.uc_mcontext.sc_eax    = regs->tf_eax;
215 	frame.sf_uc.uc_mcontext.sc_eip    = regs->tf_eip;
216 	frame.sf_uc.uc_mcontext.sc_cs     = regs->tf_cs;
217 	frame.sf_uc.uc_mcontext.sc_eflags = regs->tf_eflags;
218 	frame.sf_uc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
219 	frame.sf_uc.uc_mcontext.sc_ss     = regs->tf_ss;
220 	frame.sf_uc.uc_mcontext.sc_err    = regs->tf_err;
221 	frame.sf_uc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
222 	frame.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
223 
224 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
225 		/*
226 		 * Process has trashed its stack; give it an illegal
227 		 * instruction to halt it in its tracks.
228 		 */
229 		PROC_LOCK(p);
230 		sigexit(td, SIGILL);
231 	}
232 
233 	/* Build context to run handler in. */
234 	regs->tf_esp = PTROUT(fp);
235 	regs->tf_eip = linux_vdso_rt_sigcode;
236 	regs->tf_edi = PTROUT(catcher);
237 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
238 	regs->tf_cs = _ucodesel;
239 	regs->tf_ds = _udatasel;
240 	regs->tf_es = _udatasel;
241 	regs->tf_fs = _udatasel;
242 	regs->tf_ss = _udatasel;
243 	PROC_LOCK(p);
244 	mtx_lock(&psp->ps_mtx);
245 }
246 
247 /*
248  * Send an interrupt to process.
249  *
250  * Stack is set up to allow sigcode stored
251  * in u. to call routine, followed by kcall
252  * to sigreturn routine below.  After sigreturn
253  * resets the signal mask, the stack, and the
254  * frame pointer, it returns to the user
255  * specified pc, psl.
256  */
257 static void
linux_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)258 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
259 {
260 	struct thread *td = curthread;
261 	struct proc *p = td->td_proc;
262 	struct sigacts *psp;
263 	struct trapframe *regs;
264 	struct l_sigframe *fp, frame;
265 	l_sigset_t lmask;
266 	int sig;
267 	int oonstack;
268 
269 	PROC_LOCK_ASSERT(p, MA_OWNED);
270 	psp = p->p_sigacts;
271 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
272 	mtx_assert(&psp->ps_mtx, MA_OWNED);
273 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
274 		/* Signal handler installed with SA_SIGINFO. */
275 		linux_rt_sendsig(catcher, ksi, mask);
276 		return;
277 	}
278 	regs = td->td_frame;
279 	oonstack = sigonstack(regs->tf_esp);
280 
281 	/* Allocate space for the signal handler context. */
282 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
283 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
284 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
285 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
286 	} else
287 		fp = (struct l_sigframe *)regs->tf_esp - 1;
288 	mtx_unlock(&psp->ps_mtx);
289 	PROC_UNLOCK(p);
290 
291 	/* Build the argument list for the signal handler. */
292 	sig = bsd_to_linux_signal(sig);
293 
294 	bzero(&frame, sizeof(frame));
295 
296 	frame.sf_sig = sig;
297 	frame.sf_sigmask = *mask;
298 	bsd_to_linux_sigset(mask, &lmask);
299 
300 	/* Build the signal context to be used by sigreturn. */
301 	frame.sf_sc.sc_mask   = lmask.__mask;
302 	frame.sf_sc.sc_gs     = rgs();
303 	frame.sf_sc.sc_fs     = regs->tf_fs;
304 	frame.sf_sc.sc_es     = regs->tf_es;
305 	frame.sf_sc.sc_ds     = regs->tf_ds;
306 	frame.sf_sc.sc_edi    = regs->tf_edi;
307 	frame.sf_sc.sc_esi    = regs->tf_esi;
308 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
309 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
310 	frame.sf_sc.sc_esp    = regs->tf_esp;
311 	frame.sf_sc.sc_edx    = regs->tf_edx;
312 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
313 	frame.sf_sc.sc_eax    = regs->tf_eax;
314 	frame.sf_sc.sc_eip    = regs->tf_eip;
315 	frame.sf_sc.sc_cs     = regs->tf_cs;
316 	frame.sf_sc.sc_eflags = regs->tf_eflags;
317 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
318 	frame.sf_sc.sc_ss     = regs->tf_ss;
319 	frame.sf_sc.sc_err    = regs->tf_err;
320 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
321 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
322 
323 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
324 		/*
325 		 * Process has trashed its stack; give it an illegal
326 		 * instruction to halt it in its tracks.
327 		 */
328 		PROC_LOCK(p);
329 		sigexit(td, SIGILL);
330 	}
331 
332 	/* Build context to run handler in. */
333 	regs->tf_esp = PTROUT(fp);
334 	regs->tf_eip = linux_vdso_sigcode;
335 	regs->tf_edi = PTROUT(catcher);
336 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
337 	regs->tf_cs = _ucodesel;
338 	regs->tf_ds = _udatasel;
339 	regs->tf_es = _udatasel;
340 	regs->tf_fs = _udatasel;
341 	regs->tf_ss = _udatasel;
342 	PROC_LOCK(p);
343 	mtx_lock(&psp->ps_mtx);
344 }
345 
346 /*
347  * System call to cleanup state after a signal
348  * has been taken.  Reset signal mask and
349  * stack state from context left by sendsig (above).
350  * Return to previous pc and psl as specified by
351  * context left by sendsig. Check carefully to
352  * make sure that the user has not modified the
353  * psl to gain improper privileges or to cause
354  * a machine fault.
355  */
356 int
linux_sigreturn(struct thread * td,struct linux_sigreturn_args * args)357 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
358 {
359 	struct l_sigframe frame;
360 	struct trapframe *regs;
361 	int eflags;
362 	ksiginfo_t ksi;
363 
364 	regs = td->td_frame;
365 
366 	/*
367 	 * The trampoline code hands us the sigframe.
368 	 * It is unsafe to keep track of it ourselves, in the event that a
369 	 * program jumps out of a signal handler.
370 	 */
371 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
372 		return (EFAULT);
373 
374 	/* Check for security violations. */
375 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
376 	eflags = frame.sf_sc.sc_eflags;
377 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
378 		return (EINVAL);
379 
380 	/*
381 	 * Don't allow users to load a valid privileged %cs.  Let the
382 	 * hardware check for invalid selectors, excess privilege in
383 	 * other selectors, invalid %eip's and invalid %esp's.
384 	 */
385 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
386 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
387 		ksiginfo_init_trap(&ksi);
388 		ksi.ksi_signo = SIGBUS;
389 		ksi.ksi_code = BUS_OBJERR;
390 		ksi.ksi_trapno = T_PROTFLT;
391 		ksi.ksi_addr = (void *)regs->tf_eip;
392 		trapsignal(td, &ksi);
393 		return (EINVAL);
394 	}
395 
396 	kern_sigprocmask(td, SIG_SETMASK, &frame.sf_sigmask, NULL, 0);
397 
398 	/* Restore signal context. */
399 	/* %gs was restored by the trampoline. */
400 	regs->tf_fs     = frame.sf_sc.sc_fs;
401 	regs->tf_es     = frame.sf_sc.sc_es;
402 	regs->tf_ds     = frame.sf_sc.sc_ds;
403 	regs->tf_edi    = frame.sf_sc.sc_edi;
404 	regs->tf_esi    = frame.sf_sc.sc_esi;
405 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
406 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
407 	regs->tf_edx    = frame.sf_sc.sc_edx;
408 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
409 	regs->tf_eax    = frame.sf_sc.sc_eax;
410 	regs->tf_eip    = frame.sf_sc.sc_eip;
411 	regs->tf_cs     = frame.sf_sc.sc_cs;
412 	regs->tf_eflags = eflags;
413 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
414 	regs->tf_ss     = frame.sf_sc.sc_ss;
415 
416 	return (EJUSTRETURN);
417 }
418 
419 /*
420  * System call to cleanup state after a signal
421  * has been taken.  Reset signal mask and
422  * stack state from context left by rt_sendsig (above).
423  * Return to previous pc and psl as specified by
424  * context left by sendsig. Check carefully to
425  * make sure that the user has not modified the
426  * psl to gain improper privileges or to cause
427  * a machine fault.
428  */
429 int
linux_rt_sigreturn(struct thread * td,struct linux_rt_sigreturn_args * args)430 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
431 {
432 	struct l_ucontext uc;
433 	struct l_sigcontext *context;
434 	sigset_t bmask;
435 	l_stack_t *lss;
436 	stack_t ss;
437 	struct trapframe *regs;
438 	int eflags;
439 	ksiginfo_t ksi;
440 
441 	regs = td->td_frame;
442 
443 	/*
444 	 * The trampoline code hands us the ucontext.
445 	 * It is unsafe to keep track of it ourselves, in the event that a
446 	 * program jumps out of a signal handler.
447 	 */
448 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
449 		return (EFAULT);
450 
451 	context = &uc.uc_mcontext;
452 
453 	/* Check for security violations. */
454 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
455 	eflags = context->sc_eflags;
456 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
457 		return (EINVAL);
458 
459 	/*
460 	 * Don't allow users to load a valid privileged %cs.  Let the
461 	 * hardware check for invalid selectors, excess privilege in
462 	 * other selectors, invalid %eip's and invalid %esp's.
463 	 */
464 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
465 	if (!CS_SECURE(context->sc_cs)) {
466 		ksiginfo_init_trap(&ksi);
467 		ksi.ksi_signo = SIGBUS;
468 		ksi.ksi_code = BUS_OBJERR;
469 		ksi.ksi_trapno = T_PROTFLT;
470 		ksi.ksi_addr = (void *)regs->tf_eip;
471 		trapsignal(td, &ksi);
472 		return (EINVAL);
473 	}
474 
475 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
476 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
477 
478 	/* Restore signal context. */
479 	/* %gs was restored by the trampoline. */
480 	regs->tf_fs     = context->sc_fs;
481 	regs->tf_es     = context->sc_es;
482 	regs->tf_ds     = context->sc_ds;
483 	regs->tf_edi    = context->sc_edi;
484 	regs->tf_esi    = context->sc_esi;
485 	regs->tf_ebp    = context->sc_ebp;
486 	regs->tf_ebx    = context->sc_ebx;
487 	regs->tf_edx    = context->sc_edx;
488 	regs->tf_ecx    = context->sc_ecx;
489 	regs->tf_eax    = context->sc_eax;
490 	regs->tf_eip    = context->sc_eip;
491 	regs->tf_cs     = context->sc_cs;
492 	regs->tf_eflags = eflags;
493 	regs->tf_esp    = context->sc_esp_at_signal;
494 	regs->tf_ss     = context->sc_ss;
495 
496 	/* Call sigaltstack & ignore results. */
497 	lss = &uc.uc_stack;
498 	ss.ss_sp = PTRIN(lss->ss_sp);
499 	ss.ss_size = lss->ss_size;
500 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
501 
502 	(void)kern_sigaltstack(td, &ss, NULL);
503 
504 	return (EJUSTRETURN);
505 }
506 
507 static int
linux_fetch_syscall_args(struct thread * td)508 linux_fetch_syscall_args(struct thread *td)
509 {
510 	struct proc *p;
511 	struct trapframe *frame;
512 	struct syscall_args *sa;
513 
514 	p = td->td_proc;
515 	frame = td->td_frame;
516 	sa = &td->td_sa;
517 
518 	sa->code = frame->tf_eax;
519 	sa->args[0] = frame->tf_ebx;
520 	sa->args[1] = frame->tf_ecx;
521 	sa->args[2] = frame->tf_edx;
522 	sa->args[3] = frame->tf_esi;
523 	sa->args[4] = frame->tf_edi;
524 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
525 
526 	if (sa->code >= p->p_sysent->sv_size)
527 		/* nosys */
528 		sa->callp = &nosys_sysent;
529 	else
530 		sa->callp = &p->p_sysent->sv_table[sa->code];
531 
532 	td->td_retval[0] = 0;
533 	td->td_retval[1] = frame->tf_edx;
534 
535 	return (0);
536 }
537 
538 static void
linux_set_syscall_retval(struct thread * td,int error)539 linux_set_syscall_retval(struct thread *td, int error)
540 {
541 	struct trapframe *frame = td->td_frame;
542 
543 	cpu_set_syscall_retval(td, error);
544 
545 	if (__predict_false(error != 0)) {
546 		if (error != ERESTART && error != EJUSTRETURN)
547 			frame->tf_eax = bsd_to_linux_errno(error);
548 	}
549 }
550 
551 static void
linux_set_fork_retval(struct thread * td)552 linux_set_fork_retval(struct thread *td)
553 {
554 	struct trapframe *frame = td->td_frame;
555 
556 	frame->tf_eax = 0;
557 }
558 
559 /*
560  * exec_setregs may initialize some registers differently than Linux
561  * does, thus potentially confusing Linux binaries. If necessary, we
562  * override the exec_setregs default(s) here.
563  */
564 static void
linux_exec_setregs(struct thread * td,struct image_params * imgp,uintptr_t stack)565 linux_exec_setregs(struct thread *td, struct image_params *imgp,
566     uintptr_t stack)
567 {
568 	struct pcb *pcb = td->td_pcb;
569 
570 	exec_setregs(td, imgp, stack);
571 
572 	/* Linux sets %gs to 0, we default to _udatasel. */
573 	pcb->pcb_gs = 0;
574 	load_gs(0);
575 
576 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
577 }
578 
579 struct sysentvec linux_sysvec = {
580 	.sv_size	= LINUX_SYS_MAXSYSCALL,
581 	.sv_table	= linux_sysent,
582 	.sv_fixup	= linux_fixup,
583 	.sv_sendsig	= linux_sendsig,
584 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
585 	.sv_szsigcode	= &linux_szsigcode,
586 	.sv_name	= "Linux a.out",
587 	.sv_coredump	= NULL,
588 	.sv_imgact_try	= linux_exec_imgact_try,
589 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
590 	.sv_minuser	= VM_MIN_ADDRESS,
591 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
592 	.sv_usrstack	= LINUX_USRSTACK,
593 	.sv_psstrings	= PS_STRINGS,
594 	.sv_psstringssz	= sizeof(struct ps_strings),
595 	.sv_stackprot	= VM_PROT_ALL,
596 	.sv_copyout_strings = exec_copyout_strings,
597 	.sv_setregs	= linux_exec_setregs,
598 	.sv_fixlimit	= NULL,
599 	.sv_maxssiz	= NULL,
600 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32 |
601 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
602 	.sv_set_syscall_retval = linux_set_syscall_retval,
603 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
604 	.sv_syscallnames = linux_syscallnames,
605 	.sv_schedtail	= linux_schedtail,
606 	.sv_thread_detach = linux_thread_detach,
607 	.sv_trap	= NULL,
608 	.sv_hwcap	= NULL,
609 	.sv_hwcap2	= NULL,
610 	.sv_onexec	= linux_on_exec_vmspace,
611 	.sv_onexit	= linux_on_exit,
612 	.sv_ontdexit	= linux_thread_dtor,
613 	.sv_setid_allowed = &linux_setid_allowed_query,
614 	.sv_set_fork_retval = linux_set_fork_retval,
615 };
616 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
617 
618 struct sysentvec elf_linux_sysvec = {
619 	.sv_size	= LINUX_SYS_MAXSYSCALL,
620 	.sv_table	= linux_sysent,
621 	.sv_fixup	= __elfN(freebsd_fixup),
622 	.sv_sendsig	= linux_sendsig,
623 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
624 	.sv_szsigcode	= &linux_szsigcode,
625 	.sv_name	= "Linux ELF32",
626 	.sv_coredump	= elf32_coredump,
627 	.sv_elf_core_osabi = ELFOSABI_NONE,
628 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
629 	.sv_elf_core_prepare_notes = __linuxN(prepare_notes),
630 	.sv_imgact_try	= linux_exec_imgact_try,
631 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
632 	.sv_minuser	= VM_MIN_ADDRESS,
633 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
634 	.sv_usrstack	= LINUX_USRSTACK,
635 	.sv_psstrings	= LINUX_PS_STRINGS,
636 	.sv_psstringssz	= sizeof(struct ps_strings),
637 	.sv_stackprot	= VM_PROT_ALL,
638 	.sv_copyout_auxargs = __linuxN(copyout_auxargs),
639 	.sv_copyout_strings = __linuxN(copyout_strings),
640 	.sv_setregs	= linux_exec_setregs,
641 	.sv_fixlimit	= NULL,
642 	.sv_maxssiz	= NULL,
643 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP |
644 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
645 	.sv_set_syscall_retval = linux_set_syscall_retval,
646 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
647 	.sv_syscallnames = NULL,
648 	.sv_shared_page_base = LINUX_SHAREDPAGE,
649 	.sv_shared_page_len = PAGE_SIZE,
650 	.sv_schedtail	= linux_schedtail,
651 	.sv_thread_detach = linux_thread_detach,
652 	.sv_trap	= NULL,
653 	.sv_hwcap	= NULL,
654 	.sv_hwcap2	= NULL,
655 	.sv_onexec	= linux_on_exec_vmspace,
656 	.sv_onexit	= linux_on_exit,
657 	.sv_ontdexit	= linux_thread_dtor,
658 	.sv_setid_allowed = &linux_setid_allowed_query,
659 	.sv_set_fork_retval = linux_set_fork_retval,
660 };
661 
662 static int
linux_on_exec_vmspace(struct proc * p,struct image_params * imgp)663 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
664 {
665 	int error = 0;
666 
667 	if (SV_PROC_FLAG(p, SV_SHP) != 0)
668 		error = linux_map_vdso(p, linux_vdso_obj,
669 		    linux_vdso_base, LINUX_VDSOPAGE_SIZE, imgp);
670 	if (error == 0)
671 		linux_on_exec(p, imgp);
672 	return (error);
673 }
674 
675 /*
676  * linux_vdso_install() and linux_exec_sysvec_init() must be called
677  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
678  */
679 static void
linux_exec_sysvec_init(void * param)680 linux_exec_sysvec_init(void *param)
681 {
682 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
683 	struct sysentvec *sv;
684 	ptrdiff_t tkoff;
685 
686 	sv = param;
687 	/* Fill timekeep_base */
688 	exec_sysvec_init(sv);
689 
690 	tkoff = kern_timekeep_base - linux_vdso_base;
691 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
692 	*ktimekeep_base = sv->sv_timekeep_base;
693 
694 	tkoff = kern_tsc_selector - linux_vdso_base;
695 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
696 	*ktsc_selector = linux_vdso_tsc_selector_idx();
697 	if (bootverbose)
698 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
699 
700 	tkoff = kern_cpu_selector - linux_vdso_base;
701 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
702 	*ktsc_selector = linux_vdso_cpu_selector_idx();
703 	if (bootverbose)
704 		printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector);
705 }
706 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
707     linux_exec_sysvec_init, &elf_linux_sysvec);
708 
709 static void
linux_vdso_install(const void * param)710 linux_vdso_install(const void *param)
711 {
712 	char *vdso_start = &_binary_linux_vdso_so_o_start;
713 	char *vdso_end = &_binary_linux_vdso_so_o_end;
714 
715 	linux_szsigcode = vdso_end - vdso_start;
716 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
717 
718 	linux_vdso_base = LINUX_VDSOPAGE;
719 
720 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
721 
722 	linux_vdso_obj = __elfN(linux_shared_page_init)
723 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
724 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
725 
726 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
727 }
728 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
729     linux_vdso_install, NULL);
730 
731 static void
linux_vdso_deinstall(const void * param)732 linux_vdso_deinstall(const void *param)
733 {
734 
735 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
736 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
737 }
738 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
739     linux_vdso_deinstall, NULL);
740 
741 static void
linux_vdso_reloc(char * mapping,Elf_Addr offset)742 linux_vdso_reloc(char *mapping, Elf_Addr offset)
743 {
744 	const Elf_Shdr *shdr;
745 	const Elf_Rel *rel;
746 	const Elf_Ehdr *ehdr;
747 	Elf_Addr *where;
748 	Elf_Size rtype, symidx;
749 	Elf_Addr addr, addend;
750 	int i, relcnt;
751 
752 	MPASS(offset != 0);
753 
754 	relcnt = 0;
755 	ehdr = (const Elf_Ehdr *)mapping;
756 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
757 	for (i = 0; i < ehdr->e_shnum; i++)
758 	{
759 		switch (shdr[i].sh_type) {
760 		case SHT_REL:
761 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
762 			relcnt = shdr[i].sh_size / sizeof(*rel);
763 			break;
764 		case SHT_RELA:
765 			printf("Linux i386 vDSO: unexpected Rela section\n");
766 			break;
767 		}
768 	}
769 
770 	for (i = 0; i < relcnt; i++, rel++) {
771 		where = (Elf_Addr *)(mapping + rel->r_offset);
772 		addend = *where;
773 		rtype = ELF_R_TYPE(rel->r_info);
774 		symidx = ELF_R_SYM(rel->r_info);
775 
776 		switch (rtype) {
777 		case R_386_NONE:	/* none */
778 			break;
779 
780 		case R_386_RELATIVE:	/* B + A */
781 			addr = (Elf_Addr)PTROUT(offset + addend);
782 			if (*where != addr)
783 				*where = addr;
784 			break;
785 
786 		case R_386_IRELATIVE:
787 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
788 			    "symbol index %d\n", symidx);
789 			break;
790 		default:
791 			printf("Linux i386 vDSO: unexpected relocation type %d, "
792 			    "symbol index %d\n", rtype, symidx);
793 		}
794 	}
795 }
796 
797 static Elf_Brandnote linux_brandnote = {
798 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
799 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
800 	.hdr.n_type	= 1,
801 	.vendor		= GNU_ABI_VENDOR,
802 	.flags		= BN_TRANSLATE_OSREL,
803 	.trans_osrel	= linux_trans_osrel
804 };
805 
806 static Elf32_Brandinfo linux_brand = {
807 	.brand		= ELFOSABI_LINUX,
808 	.machine	= EM_386,
809 	.compat_3_brand	= "Linux",
810 	.emul_path	= linux_emul_path,
811 	.interp_path	= "/lib/ld-linux.so.1",
812 	.sysvec		= &elf_linux_sysvec,
813 	.interp_newpath	= NULL,
814 	.brand_note	= &linux_brandnote,
815 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
816 };
817 
818 static Elf32_Brandinfo linux_glibc2brand = {
819 	.brand		= ELFOSABI_LINUX,
820 	.machine	= EM_386,
821 	.compat_3_brand	= "Linux",
822 	.emul_path	= linux_emul_path,
823 	.interp_path	= "/lib/ld-linux.so.2",
824 	.sysvec		= &elf_linux_sysvec,
825 	.interp_newpath	= NULL,
826 	.brand_note	= &linux_brandnote,
827 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
828 };
829 
830 static Elf32_Brandinfo linux_muslbrand = {
831 	.brand		= ELFOSABI_LINUX,
832 	.machine	= EM_386,
833 	.compat_3_brand	= "Linux",
834 	.emul_path	= linux_emul_path,
835 	.interp_path	= "/lib/ld-musl-i386.so.1",
836 	.sysvec		= &elf_linux_sysvec,
837 	.interp_newpath	= NULL,
838 	.brand_note	= &linux_brandnote,
839 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
840 			    LINUX_BI_FUTEX_REQUEUE
841 };
842 
843 Elf32_Brandinfo *linux_brandlist[] = {
844 	&linux_brand,
845 	&linux_glibc2brand,
846 	&linux_muslbrand,
847 	NULL
848 };
849 
850 static int
linux_elf_modevent(module_t mod,int type,void * data)851 linux_elf_modevent(module_t mod, int type, void *data)
852 {
853 	Elf32_Brandinfo **brandinfo;
854 	int error;
855 	struct linux_ioctl_handler **lihp;
856 
857 	error = 0;
858 
859 	switch(type) {
860 	case MOD_LOAD:
861 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
862 		     ++brandinfo)
863 			if (elf32_insert_brand_entry(*brandinfo) < 0)
864 				error = EINVAL;
865 		if (error == 0) {
866 			SET_FOREACH(lihp, linux_ioctl_handler_set)
867 				linux_ioctl_register_handler(*lihp);
868 			linux_dev_shm_create();
869 			linux_osd_jail_register();
870 			linux_netlink_register();
871 			stclohz = (stathz ? stathz : hz);
872 			if (bootverbose)
873 				printf("Linux ELF exec handler installed\n");
874 		} else
875 			printf("cannot insert Linux ELF brand handler\n");
876 		break;
877 	case MOD_UNLOAD:
878 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
879 		     ++brandinfo)
880 			if (elf32_brand_inuse(*brandinfo))
881 				error = EBUSY;
882 		if (error == 0) {
883 			for (brandinfo = &linux_brandlist[0];
884 			     *brandinfo != NULL; ++brandinfo)
885 				if (elf32_remove_brand_entry(*brandinfo) < 0)
886 					error = EINVAL;
887 		}
888 		if (error == 0) {
889 			SET_FOREACH(lihp, linux_ioctl_handler_set)
890 				linux_ioctl_unregister_handler(*lihp);
891 			linux_netlink_deregister();
892 			linux_dev_shm_destroy();
893 			linux_osd_jail_deregister();
894 			if (bootverbose)
895 				printf("Linux ELF exec handler removed\n");
896 		} else
897 			printf("Could not deinstall ELF interpreter entry\n");
898 		break;
899 	default:
900 		return (EOPNOTSUPP);
901 	}
902 	return (error);
903 }
904 
905 static moduledata_t linux_elf_mod = {
906 	"linuxelf",
907 	linux_elf_modevent,
908 	0
909 };
910 
911 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
912 MODULE_DEPEND(linuxelf, netlink, 1, 1, 1);
913 FEATURE(linux, "Linux 32bit support");
914