xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision 71625ec9ad2a9bc8c09784fbd23b759830e0ee5f)
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 __FBSDID("$FreeBSD$");
31 
32 #define __ELF_WORD_SIZE	32
33 
34 #include <sys/param.h>
35 #include <sys/exec.h>
36 #include <sys/fcntl.h>
37 #include <sys/imgact.h>
38 #include <sys/imgact_aout.h>
39 #include <sys/imgact_elf.h>
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/module.h>
44 #include <sys/mutex.h>
45 #include <sys/proc.h>
46 #include <sys/stddef.h>
47 #include <sys/syscallsubr.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
50 #include <sys/sysproto.h>
51 
52 #include <vm/pmap.h>
53 #include <vm/vm.h>
54 #include <vm/vm_map.h>
55 #include <vm/vm_page.h>
56 
57 #include <machine/cpu.h>
58 #include <machine/cputypes.h>
59 #include <machine/md_var.h>
60 #include <machine/pcb.h>
61 #include <machine/trap.h>
62 
63 #include <x86/linux/linux_x86.h>
64 #include <i386/linux/linux.h>
65 #include <i386/linux/linux_proto.h>
66 #include <compat/linux/linux_elf.h>
67 #include <compat/linux/linux_emul.h>
68 #include <compat/linux/linux_fork.h>
69 #include <compat/linux/linux_ioctl.h>
70 #include <compat/linux/linux_mib.h>
71 #include <compat/linux/linux_misc.h>
72 #include <compat/linux/linux_signal.h>
73 #include <compat/linux/linux_util.h>
74 #include <compat/linux/linux_vdso.h>
75 
76 #include <x86/linux/linux_x86_sigframe.h>
77 
78 MODULE_VERSION(linux, 1);
79 
80 #define	LINUX_VDSOPAGE_SIZE	PAGE_SIZE * 2
81 #define	LINUX_VDSOPAGE		(VM_MAXUSER_ADDRESS - LINUX_VDSOPAGE_SIZE)
82 #define	LINUX_SHAREDPAGE	(LINUX_VDSOPAGE - PAGE_SIZE)
83 				/*
84 				 * PAGE_SIZE - the size
85 				 * of the native SHAREDPAGE
86 				 */
87 #define	LINUX_USRSTACK		LINUX_SHAREDPAGE
88 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
89 
90 static int linux_szsigcode;
91 static vm_object_t linux_vdso_obj;
92 static char *linux_vdso_mapping;
93 extern char _binary_linux_vdso_so_o_start;
94 extern char _binary_linux_vdso_so_o_end;
95 static vm_offset_t linux_vdso_base;
96 
97 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
98 extern const char *linux_syscallnames[];
99 
100 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
101 
102 static int	linux_fixup(uintptr_t *stack_base,
103 		    struct image_params *iparams);
104 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
105 static void	linux_exec_setregs(struct thread *td,
106 		    struct image_params *imgp, uintptr_t stack);
107 static void	linux_exec_sysvec_init(void *param);
108 static int	linux_on_exec_vmspace(struct proc *p,
109 		    struct image_params *imgp);
110 static void	linux_set_fork_retval(struct thread *td);
111 static void	linux_vdso_install(const void *param);
112 static void	linux_vdso_deinstall(const void *param);
113 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
114 
115 LINUX_VDSO_SYM_CHAR(linux_platform);
116 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
117 LINUX_VDSO_SYM_INTPTR(linux_vdso_sigcode);
118 LINUX_VDSO_SYM_INTPTR(linux_vdso_rt_sigcode);
119 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
120 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
121 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
122 
123 static int
124 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
125 {
126 	register_t *base, *argv, *envp;
127 
128 	base = (register_t *)*stack_base;
129 	argv = base;
130 	envp = base + (imgp->args->argc + 1);
131 	base--;
132 	suword(base, (intptr_t)envp);
133 	base--;
134 	suword(base, (intptr_t)argv);
135 	base--;
136 	suword(base, imgp->args->argc);
137 	*stack_base = (uintptr_t)base;
138 	return (0);
139 }
140 
141 void
142 linux32_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos)
143 {
144 
145 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
146 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO, __kernel_vsyscall);
147 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature);
148 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2());
149 	AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform));
150 }
151 
152 static void
153 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
154 {
155 	struct thread *td = curthread;
156 	struct proc *p = td->td_proc;
157 	struct sigacts *psp;
158 	struct trapframe *regs;
159 	struct l_rt_sigframe *fp, frame;
160 	int sig, code;
161 	int oonstack;
162 
163 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
164 	code = ksi->ksi_code;
165 	PROC_LOCK_ASSERT(p, MA_OWNED);
166 	psp = p->p_sigacts;
167 	mtx_assert(&psp->ps_mtx, MA_OWNED);
168 	regs = td->td_frame;
169 	oonstack = sigonstack(regs->tf_esp);
170 
171 	/* Allocate space for the signal handler context. */
172 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
173 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
174 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
175 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
176 	} else
177 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
178 	mtx_unlock(&psp->ps_mtx);
179 
180 	/* Build the argument list for the signal handler. */
181 	sig = bsd_to_linux_signal(sig);
182 
183 	bzero(&frame, sizeof(frame));
184 
185 	frame.sf_sig = sig;
186 	frame.sf_siginfo = PTROUT(&fp->sf_si);
187 	frame.sf_ucontext = PTROUT(&fp->sf_uc);
188 
189 	/* Fill in POSIX parts. */
190 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
191 
192 	/* Build the signal context to be used by sigreturn. */
193 	frame.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
194 	frame.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
195 	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
196 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
197 	PROC_UNLOCK(p);
198 
199 	bsd_to_linux_sigset(mask, &frame.sf_uc.uc_sigmask);
200 
201 	frame.sf_uc.uc_mcontext.sc_mask   = frame.sf_uc.uc_sigmask.__mask;
202 	frame.sf_uc.uc_mcontext.sc_gs     = rgs();
203 	frame.sf_uc.uc_mcontext.sc_fs     = regs->tf_fs;
204 	frame.sf_uc.uc_mcontext.sc_es     = regs->tf_es;
205 	frame.sf_uc.uc_mcontext.sc_ds     = regs->tf_ds;
206 	frame.sf_uc.uc_mcontext.sc_edi    = regs->tf_edi;
207 	frame.sf_uc.uc_mcontext.sc_esi    = regs->tf_esi;
208 	frame.sf_uc.uc_mcontext.sc_ebp    = regs->tf_ebp;
209 	frame.sf_uc.uc_mcontext.sc_ebx    = regs->tf_ebx;
210 	frame.sf_uc.uc_mcontext.sc_esp    = regs->tf_esp;
211 	frame.sf_uc.uc_mcontext.sc_edx    = regs->tf_edx;
212 	frame.sf_uc.uc_mcontext.sc_ecx    = regs->tf_ecx;
213 	frame.sf_uc.uc_mcontext.sc_eax    = regs->tf_eax;
214 	frame.sf_uc.uc_mcontext.sc_eip    = regs->tf_eip;
215 	frame.sf_uc.uc_mcontext.sc_cs     = regs->tf_cs;
216 	frame.sf_uc.uc_mcontext.sc_eflags = regs->tf_eflags;
217 	frame.sf_uc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
218 	frame.sf_uc.uc_mcontext.sc_ss     = regs->tf_ss;
219 	frame.sf_uc.uc_mcontext.sc_err    = regs->tf_err;
220 	frame.sf_uc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
221 	frame.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
222 
223 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
224 		/*
225 		 * Process has trashed its stack; give it an illegal
226 		 * instruction to halt it in its tracks.
227 		 */
228 		PROC_LOCK(p);
229 		sigexit(td, SIGILL);
230 	}
231 
232 	/* Build context to run handler in. */
233 	regs->tf_esp = PTROUT(fp);
234 	regs->tf_eip = linux_vdso_rt_sigcode;
235 	regs->tf_edi = PTROUT(catcher);
236 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
237 	regs->tf_cs = _ucodesel;
238 	regs->tf_ds = _udatasel;
239 	regs->tf_es = _udatasel;
240 	regs->tf_fs = _udatasel;
241 	regs->tf_ss = _udatasel;
242 	PROC_LOCK(p);
243 	mtx_lock(&psp->ps_mtx);
244 }
245 
246 /*
247  * Send an interrupt to process.
248  *
249  * Stack is set up to allow sigcode stored
250  * in u. to call routine, followed by kcall
251  * to sigreturn routine below.  After sigreturn
252  * resets the signal mask, the stack, and the
253  * frame pointer, it returns to the user
254  * specified pc, psl.
255  */
256 static void
257 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
258 {
259 	struct thread *td = curthread;
260 	struct proc *p = td->td_proc;
261 	struct sigacts *psp;
262 	struct trapframe *regs;
263 	struct l_sigframe *fp, frame;
264 	l_sigset_t lmask;
265 	int sig;
266 	int oonstack;
267 
268 	PROC_LOCK_ASSERT(p, MA_OWNED);
269 	psp = p->p_sigacts;
270 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
271 	mtx_assert(&psp->ps_mtx, MA_OWNED);
272 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
273 		/* Signal handler installed with SA_SIGINFO. */
274 		linux_rt_sendsig(catcher, ksi, mask);
275 		return;
276 	}
277 	regs = td->td_frame;
278 	oonstack = sigonstack(regs->tf_esp);
279 
280 	/* Allocate space for the signal handler context. */
281 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
282 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
283 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
284 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
285 	} else
286 		fp = (struct l_sigframe *)regs->tf_esp - 1;
287 	mtx_unlock(&psp->ps_mtx);
288 	PROC_UNLOCK(p);
289 
290 	/* Build the argument list for the signal handler. */
291 	sig = bsd_to_linux_signal(sig);
292 
293 	bzero(&frame, sizeof(frame));
294 
295 	frame.sf_sig = sig;
296 	frame.sf_sigmask = *mask;
297 	bsd_to_linux_sigset(mask, &lmask);
298 
299 	/* Build the signal context to be used by sigreturn. */
300 	frame.sf_sc.sc_mask   = lmask.__mask;
301 	frame.sf_sc.sc_gs     = rgs();
302 	frame.sf_sc.sc_fs     = regs->tf_fs;
303 	frame.sf_sc.sc_es     = regs->tf_es;
304 	frame.sf_sc.sc_ds     = regs->tf_ds;
305 	frame.sf_sc.sc_edi    = regs->tf_edi;
306 	frame.sf_sc.sc_esi    = regs->tf_esi;
307 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
308 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
309 	frame.sf_sc.sc_esp    = regs->tf_esp;
310 	frame.sf_sc.sc_edx    = regs->tf_edx;
311 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
312 	frame.sf_sc.sc_eax    = regs->tf_eax;
313 	frame.sf_sc.sc_eip    = regs->tf_eip;
314 	frame.sf_sc.sc_cs     = regs->tf_cs;
315 	frame.sf_sc.sc_eflags = regs->tf_eflags;
316 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
317 	frame.sf_sc.sc_ss     = regs->tf_ss;
318 	frame.sf_sc.sc_err    = regs->tf_err;
319 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
320 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
321 
322 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
323 		/*
324 		 * Process has trashed its stack; give it an illegal
325 		 * instruction to halt it in its tracks.
326 		 */
327 		PROC_LOCK(p);
328 		sigexit(td, SIGILL);
329 	}
330 
331 	/* Build context to run handler in. */
332 	regs->tf_esp = PTROUT(fp);
333 	regs->tf_eip = linux_vdso_sigcode;
334 	regs->tf_edi = PTROUT(catcher);
335 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
336 	regs->tf_cs = _ucodesel;
337 	regs->tf_ds = _udatasel;
338 	regs->tf_es = _udatasel;
339 	regs->tf_fs = _udatasel;
340 	regs->tf_ss = _udatasel;
341 	PROC_LOCK(p);
342 	mtx_lock(&psp->ps_mtx);
343 }
344 
345 /*
346  * System call to cleanup state after a signal
347  * has been taken.  Reset signal mask and
348  * stack state from context left by sendsig (above).
349  * Return to previous pc and psl as specified by
350  * context left by sendsig. Check carefully to
351  * make sure that the user has not modified the
352  * psl to gain improper privileges or to cause
353  * a machine fault.
354  */
355 int
356 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
357 {
358 	struct l_sigframe frame;
359 	struct trapframe *regs;
360 	int eflags;
361 	ksiginfo_t ksi;
362 
363 	regs = td->td_frame;
364 
365 	/*
366 	 * The trampoline code hands us the sigframe.
367 	 * It is unsafe to keep track of it ourselves, in the event that a
368 	 * program jumps out of a signal handler.
369 	 */
370 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
371 		return (EFAULT);
372 
373 	/* Check for security violations. */
374 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
375 	eflags = frame.sf_sc.sc_eflags;
376 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
377 		return (EINVAL);
378 
379 	/*
380 	 * Don't allow users to load a valid privileged %cs.  Let the
381 	 * hardware check for invalid selectors, excess privilege in
382 	 * other selectors, invalid %eip's and invalid %esp's.
383 	 */
384 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
385 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
386 		ksiginfo_init_trap(&ksi);
387 		ksi.ksi_signo = SIGBUS;
388 		ksi.ksi_code = BUS_OBJERR;
389 		ksi.ksi_trapno = T_PROTFLT;
390 		ksi.ksi_addr = (void *)regs->tf_eip;
391 		trapsignal(td, &ksi);
392 		return (EINVAL);
393 	}
394 
395 	kern_sigprocmask(td, SIG_SETMASK, &frame.sf_sigmask, NULL, 0);
396 
397 	/* Restore signal context. */
398 	/* %gs was restored by the trampoline. */
399 	regs->tf_fs     = frame.sf_sc.sc_fs;
400 	regs->tf_es     = frame.sf_sc.sc_es;
401 	regs->tf_ds     = frame.sf_sc.sc_ds;
402 	regs->tf_edi    = frame.sf_sc.sc_edi;
403 	regs->tf_esi    = frame.sf_sc.sc_esi;
404 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
405 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
406 	regs->tf_edx    = frame.sf_sc.sc_edx;
407 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
408 	regs->tf_eax    = frame.sf_sc.sc_eax;
409 	regs->tf_eip    = frame.sf_sc.sc_eip;
410 	regs->tf_cs     = frame.sf_sc.sc_cs;
411 	regs->tf_eflags = eflags;
412 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
413 	regs->tf_ss     = frame.sf_sc.sc_ss;
414 
415 	return (EJUSTRETURN);
416 }
417 
418 /*
419  * System call to cleanup state after a signal
420  * has been taken.  Reset signal mask and
421  * stack state from context left by rt_sendsig (above).
422  * Return to previous pc and psl as specified by
423  * context left by sendsig. Check carefully to
424  * make sure that the user has not modified the
425  * psl to gain improper privileges or to cause
426  * a machine fault.
427  */
428 int
429 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
430 {
431 	struct l_ucontext uc;
432 	struct l_sigcontext *context;
433 	sigset_t bmask;
434 	l_stack_t *lss;
435 	stack_t ss;
436 	struct trapframe *regs;
437 	int eflags;
438 	ksiginfo_t ksi;
439 
440 	regs = td->td_frame;
441 
442 	/*
443 	 * The trampoline code hands us the ucontext.
444 	 * It is unsafe to keep track of it ourselves, in the event that a
445 	 * program jumps out of a signal handler.
446 	 */
447 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
448 		return (EFAULT);
449 
450 	context = &uc.uc_mcontext;
451 
452 	/* Check for security violations. */
453 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
454 	eflags = context->sc_eflags;
455 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
456 		return (EINVAL);
457 
458 	/*
459 	 * Don't allow users to load a valid privileged %cs.  Let the
460 	 * hardware check for invalid selectors, excess privilege in
461 	 * other selectors, invalid %eip's and invalid %esp's.
462 	 */
463 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
464 	if (!CS_SECURE(context->sc_cs)) {
465 		ksiginfo_init_trap(&ksi);
466 		ksi.ksi_signo = SIGBUS;
467 		ksi.ksi_code = BUS_OBJERR;
468 		ksi.ksi_trapno = T_PROTFLT;
469 		ksi.ksi_addr = (void *)regs->tf_eip;
470 		trapsignal(td, &ksi);
471 		return (EINVAL);
472 	}
473 
474 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
475 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
476 
477 	/* Restore signal context. */
478 	/* %gs was restored by the trampoline. */
479 	regs->tf_fs     = context->sc_fs;
480 	regs->tf_es     = context->sc_es;
481 	regs->tf_ds     = context->sc_ds;
482 	regs->tf_edi    = context->sc_edi;
483 	regs->tf_esi    = context->sc_esi;
484 	regs->tf_ebp    = context->sc_ebp;
485 	regs->tf_ebx    = context->sc_ebx;
486 	regs->tf_edx    = context->sc_edx;
487 	regs->tf_ecx    = context->sc_ecx;
488 	regs->tf_eax    = context->sc_eax;
489 	regs->tf_eip    = context->sc_eip;
490 	regs->tf_cs     = context->sc_cs;
491 	regs->tf_eflags = eflags;
492 	regs->tf_esp    = context->sc_esp_at_signal;
493 	regs->tf_ss     = context->sc_ss;
494 
495 	/* Call sigaltstack & ignore results. */
496 	lss = &uc.uc_stack;
497 	ss.ss_sp = PTRIN(lss->ss_sp);
498 	ss.ss_size = lss->ss_size;
499 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
500 
501 	(void)kern_sigaltstack(td, &ss, NULL);
502 
503 	return (EJUSTRETURN);
504 }
505 
506 static int
507 linux_fetch_syscall_args(struct thread *td)
508 {
509 	struct proc *p;
510 	struct trapframe *frame;
511 	struct syscall_args *sa;
512 
513 	p = td->td_proc;
514 	frame = td->td_frame;
515 	sa = &td->td_sa;
516 
517 	sa->code = frame->tf_eax;
518 	sa->original_code = sa->code;
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 = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
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
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
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
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_minsigstksz	= LINUX_MINSIGSTKSZ,
589 	.sv_minuser	= VM_MIN_ADDRESS,
590 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
591 	.sv_usrstack	= LINUX_USRSTACK,
592 	.sv_psstrings	= PS_STRINGS,
593 	.sv_psstringssz	= sizeof(struct ps_strings),
594 	.sv_stackprot	= VM_PROT_ALL,
595 	.sv_copyout_strings = exec_copyout_strings,
596 	.sv_setregs	= linux_exec_setregs,
597 	.sv_fixlimit	= NULL,
598 	.sv_maxssiz	= NULL,
599 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32 |
600 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
601 	.sv_set_syscall_retval = linux_set_syscall_retval,
602 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
603 	.sv_syscallnames = linux_syscallnames,
604 	.sv_schedtail	= linux_schedtail,
605 	.sv_thread_detach = linux_thread_detach,
606 	.sv_trap	= NULL,
607 	.sv_hwcap	= NULL,
608 	.sv_hwcap2	= NULL,
609 	.sv_onexec	= linux_on_exec_vmspace,
610 	.sv_onexit	= linux_on_exit,
611 	.sv_ontdexit	= linux_thread_dtor,
612 	.sv_setid_allowed = &linux_setid_allowed_query,
613 	.sv_set_fork_retval = linux_set_fork_retval,
614 };
615 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
616 
617 struct sysentvec elf_linux_sysvec = {
618 	.sv_size	= LINUX_SYS_MAXSYSCALL,
619 	.sv_table	= linux_sysent,
620 	.sv_fixup	= __elfN(freebsd_fixup),
621 	.sv_sendsig	= linux_sendsig,
622 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
623 	.sv_szsigcode	= &linux_szsigcode,
624 	.sv_name	= "Linux ELF32",
625 	.sv_coredump	= elf32_coredump,
626 	.sv_elf_core_osabi = ELFOSABI_NONE,
627 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
628 	.sv_elf_core_prepare_notes = __linuxN(prepare_notes),
629 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
630 	.sv_minuser	= VM_MIN_ADDRESS,
631 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
632 	.sv_usrstack	= LINUX_USRSTACK,
633 	.sv_psstrings	= LINUX_PS_STRINGS,
634 	.sv_psstringssz	= sizeof(struct ps_strings),
635 	.sv_stackprot	= VM_PROT_ALL,
636 	.sv_copyout_auxargs = __linuxN(copyout_auxargs),
637 	.sv_copyout_strings = __linuxN(copyout_strings),
638 	.sv_setregs	= linux_exec_setregs,
639 	.sv_fixlimit	= NULL,
640 	.sv_maxssiz	= NULL,
641 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP |
642 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
643 	.sv_set_syscall_retval = linux_set_syscall_retval,
644 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
645 	.sv_syscallnames = NULL,
646 	.sv_shared_page_base = LINUX_SHAREDPAGE,
647 	.sv_shared_page_len = PAGE_SIZE,
648 	.sv_schedtail	= linux_schedtail,
649 	.sv_thread_detach = linux_thread_detach,
650 	.sv_trap	= NULL,
651 	.sv_hwcap	= NULL,
652 	.sv_hwcap2	= NULL,
653 	.sv_onexec	= linux_on_exec_vmspace,
654 	.sv_onexit	= linux_on_exit,
655 	.sv_ontdexit	= linux_thread_dtor,
656 	.sv_setid_allowed = &linux_setid_allowed_query,
657 	.sv_set_fork_retval = linux_set_fork_retval,
658 };
659 
660 static int
661 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
662 {
663 	int error = 0;
664 
665 	if (SV_PROC_FLAG(p, SV_SHP) != 0)
666 		error = linux_map_vdso(p, linux_vdso_obj,
667 		    linux_vdso_base, LINUX_VDSOPAGE_SIZE, imgp);
668 	if (error == 0)
669 		error = linux_on_exec(p, imgp);
670 	return (error);
671 }
672 
673 /*
674  * linux_vdso_install() and linux_exec_sysvec_init() must be called
675  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
676  */
677 static void
678 linux_exec_sysvec_init(void *param)
679 {
680 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
681 	struct sysentvec *sv;
682 	ptrdiff_t tkoff;
683 
684 	sv = param;
685 	/* Fill timekeep_base */
686 	exec_sysvec_init(sv);
687 
688 	tkoff = kern_timekeep_base - linux_vdso_base;
689 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
690 	*ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
691 
692 	tkoff = kern_tsc_selector - linux_vdso_base;
693 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
694 	*ktsc_selector = linux_vdso_tsc_selector_idx();
695 	if (bootverbose)
696 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
697 
698 	tkoff = kern_cpu_selector - linux_vdso_base;
699 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
700 	*ktsc_selector = linux_vdso_cpu_selector_idx();
701 	if (bootverbose)
702 		printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector);
703 }
704 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
705     linux_exec_sysvec_init, &elf_linux_sysvec);
706 
707 static void
708 linux_vdso_install(const void *param)
709 {
710 	char *vdso_start = &_binary_linux_vdso_so_o_start;
711 	char *vdso_end = &_binary_linux_vdso_so_o_end;
712 
713 	linux_szsigcode = vdso_end - vdso_start;
714 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
715 
716 	linux_vdso_base = LINUX_VDSOPAGE;
717 
718 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
719 
720 	linux_vdso_obj = __elfN(linux_shared_page_init)
721 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
722 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
723 
724 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
725 }
726 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
727     linux_vdso_install, NULL);
728 
729 static void
730 linux_vdso_deinstall(const void *param)
731 {
732 
733 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
734 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
735 }
736 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
737     linux_vdso_deinstall, NULL);
738 
739 static void
740 linux_vdso_reloc(char *mapping, Elf_Addr offset)
741 {
742 	const Elf_Shdr *shdr;
743 	const Elf_Rel *rel;
744 	const Elf_Ehdr *ehdr;
745 	Elf_Addr *where;
746 	Elf_Size rtype, symidx;
747 	Elf_Addr addr, addend;
748 	int i, relcnt;
749 
750 	MPASS(offset != 0);
751 
752 	relcnt = 0;
753 	ehdr = (const Elf_Ehdr *)mapping;
754 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
755 	for (i = 0; i < ehdr->e_shnum; i++)
756 	{
757 		switch (shdr[i].sh_type) {
758 		case SHT_REL:
759 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
760 			relcnt = shdr[i].sh_size / sizeof(*rel);
761 			break;
762 		case SHT_RELA:
763 			printf("Linux i386 vDSO: unexpected Rela section\n");
764 			break;
765 		}
766 	}
767 
768 	for (i = 0; i < relcnt; i++, rel++) {
769 		where = (Elf_Addr *)(mapping + rel->r_offset);
770 		addend = *where;
771 		rtype = ELF_R_TYPE(rel->r_info);
772 		symidx = ELF_R_SYM(rel->r_info);
773 
774 		switch (rtype) {
775 		case R_386_NONE:	/* none */
776 			break;
777 
778 		case R_386_RELATIVE:	/* B + A */
779 			addr = (Elf_Addr)PTROUT(offset + addend);
780 			if (*where != addr)
781 				*where = addr;
782 			break;
783 
784 		case R_386_IRELATIVE:
785 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
786 			    "symbol index %d\n", symidx);
787 			break;
788 		default:
789 			printf("Linux i386 vDSO: unexpected relocation type %d, "
790 			    "symbol index %d\n", rtype, symidx);
791 		}
792 	}
793 }
794 
795 static Elf_Brandnote linux_brandnote = {
796 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
797 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
798 	.hdr.n_type	= 1,
799 	.vendor		= GNU_ABI_VENDOR,
800 	.flags		= BN_TRANSLATE_OSREL,
801 	.trans_osrel	= linux_trans_osrel
802 };
803 
804 static Elf32_Brandinfo linux_brand = {
805 	.brand		= ELFOSABI_LINUX,
806 	.machine	= EM_386,
807 	.compat_3_brand	= "Linux",
808 	.interp_path	= "/lib/ld-linux.so.1",
809 	.sysvec		= &elf_linux_sysvec,
810 	.interp_newpath	= NULL,
811 	.brand_note	= &linux_brandnote,
812 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
813 };
814 
815 static Elf32_Brandinfo linux_glibc2brand = {
816 	.brand		= ELFOSABI_LINUX,
817 	.machine	= EM_386,
818 	.compat_3_brand	= "Linux",
819 	.interp_path	= "/lib/ld-linux.so.2",
820 	.sysvec		= &elf_linux_sysvec,
821 	.interp_newpath	= NULL,
822 	.brand_note	= &linux_brandnote,
823 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
824 };
825 
826 static Elf32_Brandinfo linux_muslbrand = {
827 	.brand		= ELFOSABI_LINUX,
828 	.machine	= EM_386,
829 	.compat_3_brand	= "Linux",
830 	.interp_path	= "/lib/ld-musl-i386.so.1",
831 	.sysvec		= &elf_linux_sysvec,
832 	.interp_newpath	= NULL,
833 	.brand_note	= &linux_brandnote,
834 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
835 			    LINUX_BI_FUTEX_REQUEUE
836 };
837 
838 Elf32_Brandinfo *linux_brandlist[] = {
839 	&linux_brand,
840 	&linux_glibc2brand,
841 	&linux_muslbrand,
842 	NULL
843 };
844 
845 static int
846 linux_elf_modevent(module_t mod, int type, void *data)
847 {
848 	Elf32_Brandinfo **brandinfo;
849 	int error;
850 	struct linux_ioctl_handler **lihp;
851 
852 	error = 0;
853 
854 	switch(type) {
855 	case MOD_LOAD:
856 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
857 		     ++brandinfo)
858 			if (elf32_insert_brand_entry(*brandinfo) < 0)
859 				error = EINVAL;
860 		if (error == 0) {
861 			SET_FOREACH(lihp, linux_ioctl_handler_set)
862 				linux_ioctl_register_handler(*lihp);
863 			linux_dev_shm_create();
864 			linux_osd_jail_register();
865 			linux_netlink_register();
866 			stclohz = (stathz ? stathz : hz);
867 			if (bootverbose)
868 				printf("Linux ELF exec handler installed\n");
869 		} else
870 			printf("cannot insert Linux ELF brand handler\n");
871 		break;
872 	case MOD_UNLOAD:
873 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
874 		     ++brandinfo)
875 			if (elf32_brand_inuse(*brandinfo))
876 				error = EBUSY;
877 		if (error == 0) {
878 			for (brandinfo = &linux_brandlist[0];
879 			     *brandinfo != NULL; ++brandinfo)
880 				if (elf32_remove_brand_entry(*brandinfo) < 0)
881 					error = EINVAL;
882 		}
883 		if (error == 0) {
884 			SET_FOREACH(lihp, linux_ioctl_handler_set)
885 				linux_ioctl_unregister_handler(*lihp);
886 			linux_netlink_deregister();
887 			linux_dev_shm_destroy();
888 			linux_osd_jail_deregister();
889 			if (bootverbose)
890 				printf("Linux ELF exec handler removed\n");
891 		} else
892 			printf("Could not deinstall ELF interpreter entry\n");
893 		break;
894 	default:
895 		return (EOPNOTSUPP);
896 	}
897 	return (error);
898 }
899 
900 static moduledata_t linux_elf_mod = {
901 	"linuxelf",
902 	linux_elf_modevent,
903 	0
904 };
905 
906 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
907 MODULE_DEPEND(linuxelf, netlink, 1, 1, 1);
908 FEATURE(linux, "Linux 32bit support");
909