xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
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 #define __ELF_WORD_SIZE	32
30 
31 #include <sys/param.h>
32 #include <sys/exec.h>
33 #include <sys/fcntl.h>
34 #include <sys/imgact.h>
35 #include <sys/imgact_aout.h>
36 #include <sys/imgact_elf.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/module.h>
41 #include <sys/mutex.h>
42 #include <sys/proc.h>
43 #include <sys/stddef.h>
44 #include <sys/syscallsubr.h>
45 #include <sys/sysctl.h>
46 #include <sys/sysent.h>
47 #include <sys/sysproto.h>
48 
49 #include <vm/pmap.h>
50 #include <vm/vm.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_page.h>
53 
54 #include <machine/cpu.h>
55 #include <machine/cputypes.h>
56 #include <machine/md_var.h>
57 #include <machine/pcb.h>
58 #include <machine/trap.h>
59 
60 #include <x86/linux/linux_x86.h>
61 #include <i386/linux/linux.h>
62 #include <i386/linux/linux_proto.h>
63 #include <compat/linux/linux_elf.h>
64 #include <compat/linux/linux_emul.h>
65 #include <compat/linux/linux_fork.h>
66 #include <compat/linux/linux_ioctl.h>
67 #include <compat/linux/linux_mib.h>
68 #include <compat/linux/linux_misc.h>
69 #include <compat/linux/linux_signal.h>
70 #include <compat/linux/linux_util.h>
71 #include <compat/linux/linux_vdso.h>
72 
73 #include <x86/linux/linux_x86_sigframe.h>
74 
75 MODULE_VERSION(linux, 1);
76 
77 #define	LINUX_VDSOPAGE_SIZE	PAGE_SIZE * 2
78 #define	LINUX_VDSOPAGE		(VM_MAXUSER_ADDRESS - LINUX_VDSOPAGE_SIZE)
79 #define	LINUX_SHAREDPAGE	(LINUX_VDSOPAGE - PAGE_SIZE)
80 				/*
81 				 * PAGE_SIZE - the size
82 				 * of the native SHAREDPAGE
83 				 */
84 #define	LINUX_USRSTACK		LINUX_SHAREDPAGE
85 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
86 
87 static int linux_szsigcode;
88 static vm_object_t linux_vdso_obj;
89 static char *linux_vdso_mapping;
90 extern char _binary_linux_vdso_so_o_start;
91 extern char _binary_linux_vdso_so_o_end;
92 static vm_offset_t linux_vdso_base;
93 
94 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
95 extern const char *linux_syscallnames[];
96 
97 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
98 
99 static int	linux_fixup(uintptr_t *stack_base,
100 		    struct image_params *iparams);
101 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
102 static void	linux_exec_setregs(struct thread *td,
103 		    struct image_params *imgp, uintptr_t stack);
104 static void	linux_exec_sysvec_init(void *param);
105 static int	linux_on_exec_vmspace(struct proc *p,
106 		    struct image_params *imgp);
107 static void	linux_set_fork_retval(struct thread *td);
108 static void	linux_vdso_install(const void *param);
109 static void	linux_vdso_deinstall(const void *param);
110 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
111 
112 LINUX_VDSO_SYM_CHAR(linux_platform);
113 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
114 LINUX_VDSO_SYM_INTPTR(linux_vdso_sigcode);
115 LINUX_VDSO_SYM_INTPTR(linux_vdso_rt_sigcode);
116 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
117 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
118 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
119 
120 static int
121 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
122 {
123 	register_t *base, *argv, *envp;
124 
125 	base = (register_t *)*stack_base;
126 	argv = base;
127 	envp = base + (imgp->args->argc + 1);
128 	base--;
129 	suword(base, (intptr_t)envp);
130 	base--;
131 	suword(base, (intptr_t)argv);
132 	base--;
133 	suword(base, imgp->args->argc);
134 	*stack_base = (uintptr_t)base;
135 	return (0);
136 }
137 
138 void
139 linux32_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos)
140 {
141 
142 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
143 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO, __kernel_vsyscall);
144 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature);
145 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2());
146 	AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform));
147 }
148 
149 static void
150 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
151 {
152 	struct thread *td = curthread;
153 	struct proc *p = td->td_proc;
154 	struct sigacts *psp;
155 	struct trapframe *regs;
156 	struct l_rt_sigframe *fp, frame;
157 	int sig, code;
158 	int oonstack;
159 
160 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
161 	code = ksi->ksi_code;
162 	PROC_LOCK_ASSERT(p, MA_OWNED);
163 	psp = p->p_sigacts;
164 	mtx_assert(&psp->ps_mtx, MA_OWNED);
165 	regs = td->td_frame;
166 	oonstack = sigonstack(regs->tf_esp);
167 
168 	/* Allocate space for the signal handler context. */
169 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
170 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
171 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
172 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
173 	} else
174 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
175 	mtx_unlock(&psp->ps_mtx);
176 
177 	/* Build the argument list for the signal handler. */
178 	sig = bsd_to_linux_signal(sig);
179 
180 	bzero(&frame, sizeof(frame));
181 
182 	frame.sf_sig = sig;
183 	frame.sf_siginfo = PTROUT(&fp->sf_si);
184 	frame.sf_ucontext = PTROUT(&fp->sf_uc);
185 
186 	/* Fill in POSIX parts. */
187 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
188 
189 	/* Build the signal context to be used by sigreturn. */
190 	frame.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
191 	frame.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
192 	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
193 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
194 	PROC_UNLOCK(p);
195 
196 	bsd_to_linux_sigset(mask, &frame.sf_uc.uc_sigmask);
197 
198 	frame.sf_uc.uc_mcontext.sc_mask   = frame.sf_uc.uc_sigmask.__mask;
199 	frame.sf_uc.uc_mcontext.sc_gs     = rgs();
200 	frame.sf_uc.uc_mcontext.sc_fs     = regs->tf_fs;
201 	frame.sf_uc.uc_mcontext.sc_es     = regs->tf_es;
202 	frame.sf_uc.uc_mcontext.sc_ds     = regs->tf_ds;
203 	frame.sf_uc.uc_mcontext.sc_edi    = regs->tf_edi;
204 	frame.sf_uc.uc_mcontext.sc_esi    = regs->tf_esi;
205 	frame.sf_uc.uc_mcontext.sc_ebp    = regs->tf_ebp;
206 	frame.sf_uc.uc_mcontext.sc_ebx    = regs->tf_ebx;
207 	frame.sf_uc.uc_mcontext.sc_esp    = regs->tf_esp;
208 	frame.sf_uc.uc_mcontext.sc_edx    = regs->tf_edx;
209 	frame.sf_uc.uc_mcontext.sc_ecx    = regs->tf_ecx;
210 	frame.sf_uc.uc_mcontext.sc_eax    = regs->tf_eax;
211 	frame.sf_uc.uc_mcontext.sc_eip    = regs->tf_eip;
212 	frame.sf_uc.uc_mcontext.sc_cs     = regs->tf_cs;
213 	frame.sf_uc.uc_mcontext.sc_eflags = regs->tf_eflags;
214 	frame.sf_uc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
215 	frame.sf_uc.uc_mcontext.sc_ss     = regs->tf_ss;
216 	frame.sf_uc.uc_mcontext.sc_err    = regs->tf_err;
217 	frame.sf_uc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
218 	frame.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
219 
220 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
221 		/*
222 		 * Process has trashed its stack; give it an illegal
223 		 * instruction to halt it in its tracks.
224 		 */
225 		PROC_LOCK(p);
226 		sigexit(td, SIGILL);
227 	}
228 
229 	/* Build context to run handler in. */
230 	regs->tf_esp = PTROUT(fp);
231 	regs->tf_eip = linux_vdso_rt_sigcode;
232 	regs->tf_edi = PTROUT(catcher);
233 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
234 	regs->tf_cs = _ucodesel;
235 	regs->tf_ds = _udatasel;
236 	regs->tf_es = _udatasel;
237 	regs->tf_fs = _udatasel;
238 	regs->tf_ss = _udatasel;
239 	PROC_LOCK(p);
240 	mtx_lock(&psp->ps_mtx);
241 }
242 
243 /*
244  * Send an interrupt to process.
245  *
246  * Stack is set up to allow sigcode stored
247  * in u. to call routine, followed by kcall
248  * to sigreturn routine below.  After sigreturn
249  * resets the signal mask, the stack, and the
250  * frame pointer, it returns to the user
251  * specified pc, psl.
252  */
253 static void
254 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
255 {
256 	struct thread *td = curthread;
257 	struct proc *p = td->td_proc;
258 	struct sigacts *psp;
259 	struct trapframe *regs;
260 	struct l_sigframe *fp, frame;
261 	l_sigset_t lmask;
262 	int sig;
263 	int oonstack;
264 
265 	PROC_LOCK_ASSERT(p, MA_OWNED);
266 	psp = p->p_sigacts;
267 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
268 	mtx_assert(&psp->ps_mtx, MA_OWNED);
269 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
270 		/* Signal handler installed with SA_SIGINFO. */
271 		linux_rt_sendsig(catcher, ksi, mask);
272 		return;
273 	}
274 	regs = td->td_frame;
275 	oonstack = sigonstack(regs->tf_esp);
276 
277 	/* Allocate space for the signal handler context. */
278 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
279 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
280 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
281 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
282 	} else
283 		fp = (struct l_sigframe *)regs->tf_esp - 1;
284 	mtx_unlock(&psp->ps_mtx);
285 	PROC_UNLOCK(p);
286 
287 	/* Build the argument list for the signal handler. */
288 	sig = bsd_to_linux_signal(sig);
289 
290 	bzero(&frame, sizeof(frame));
291 
292 	frame.sf_sig = sig;
293 	frame.sf_sigmask = *mask;
294 	bsd_to_linux_sigset(mask, &lmask);
295 
296 	/* Build the signal context to be used by sigreturn. */
297 	frame.sf_sc.sc_mask   = lmask.__mask;
298 	frame.sf_sc.sc_gs     = rgs();
299 	frame.sf_sc.sc_fs     = regs->tf_fs;
300 	frame.sf_sc.sc_es     = regs->tf_es;
301 	frame.sf_sc.sc_ds     = regs->tf_ds;
302 	frame.sf_sc.sc_edi    = regs->tf_edi;
303 	frame.sf_sc.sc_esi    = regs->tf_esi;
304 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
305 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
306 	frame.sf_sc.sc_esp    = regs->tf_esp;
307 	frame.sf_sc.sc_edx    = regs->tf_edx;
308 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
309 	frame.sf_sc.sc_eax    = regs->tf_eax;
310 	frame.sf_sc.sc_eip    = regs->tf_eip;
311 	frame.sf_sc.sc_cs     = regs->tf_cs;
312 	frame.sf_sc.sc_eflags = regs->tf_eflags;
313 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
314 	frame.sf_sc.sc_ss     = regs->tf_ss;
315 	frame.sf_sc.sc_err    = regs->tf_err;
316 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
317 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
318 
319 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
320 		/*
321 		 * Process has trashed its stack; give it an illegal
322 		 * instruction to halt it in its tracks.
323 		 */
324 		PROC_LOCK(p);
325 		sigexit(td, SIGILL);
326 	}
327 
328 	/* Build context to run handler in. */
329 	regs->tf_esp = PTROUT(fp);
330 	regs->tf_eip = linux_vdso_sigcode;
331 	regs->tf_edi = PTROUT(catcher);
332 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
333 	regs->tf_cs = _ucodesel;
334 	regs->tf_ds = _udatasel;
335 	regs->tf_es = _udatasel;
336 	regs->tf_fs = _udatasel;
337 	regs->tf_ss = _udatasel;
338 	PROC_LOCK(p);
339 	mtx_lock(&psp->ps_mtx);
340 }
341 
342 /*
343  * System call to cleanup state after a signal
344  * has been taken.  Reset signal mask and
345  * stack state from context left by sendsig (above).
346  * Return to previous pc and psl as specified by
347  * context left by sendsig. Check carefully to
348  * make sure that the user has not modified the
349  * psl to gain improper privileges or to cause
350  * a machine fault.
351  */
352 int
353 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
354 {
355 	struct l_sigframe frame;
356 	struct trapframe *regs;
357 	int eflags;
358 	ksiginfo_t ksi;
359 
360 	regs = td->td_frame;
361 
362 	/*
363 	 * The trampoline code hands us the sigframe.
364 	 * It is unsafe to keep track of it ourselves, in the event that a
365 	 * program jumps out of a signal handler.
366 	 */
367 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
368 		return (EFAULT);
369 
370 	/* Check for security violations. */
371 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
372 	eflags = frame.sf_sc.sc_eflags;
373 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
374 		return (EINVAL);
375 
376 	/*
377 	 * Don't allow users to load a valid privileged %cs.  Let the
378 	 * hardware check for invalid selectors, excess privilege in
379 	 * other selectors, invalid %eip's and invalid %esp's.
380 	 */
381 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
382 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
383 		ksiginfo_init_trap(&ksi);
384 		ksi.ksi_signo = SIGBUS;
385 		ksi.ksi_code = BUS_OBJERR;
386 		ksi.ksi_trapno = T_PROTFLT;
387 		ksi.ksi_addr = (void *)regs->tf_eip;
388 		trapsignal(td, &ksi);
389 		return (EINVAL);
390 	}
391 
392 	kern_sigprocmask(td, SIG_SETMASK, &frame.sf_sigmask, NULL, 0);
393 
394 	/* Restore signal context. */
395 	/* %gs was restored by the trampoline. */
396 	regs->tf_fs     = frame.sf_sc.sc_fs;
397 	regs->tf_es     = frame.sf_sc.sc_es;
398 	regs->tf_ds     = frame.sf_sc.sc_ds;
399 	regs->tf_edi    = frame.sf_sc.sc_edi;
400 	regs->tf_esi    = frame.sf_sc.sc_esi;
401 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
402 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
403 	regs->tf_edx    = frame.sf_sc.sc_edx;
404 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
405 	regs->tf_eax    = frame.sf_sc.sc_eax;
406 	regs->tf_eip    = frame.sf_sc.sc_eip;
407 	regs->tf_cs     = frame.sf_sc.sc_cs;
408 	regs->tf_eflags = eflags;
409 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
410 	regs->tf_ss     = frame.sf_sc.sc_ss;
411 
412 	return (EJUSTRETURN);
413 }
414 
415 /*
416  * System call to cleanup state after a signal
417  * has been taken.  Reset signal mask and
418  * stack state from context left by rt_sendsig (above).
419  * Return to previous pc and psl as specified by
420  * context left by sendsig. Check carefully to
421  * make sure that the user has not modified the
422  * psl to gain improper privileges or to cause
423  * a machine fault.
424  */
425 int
426 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
427 {
428 	struct l_ucontext uc;
429 	struct l_sigcontext *context;
430 	sigset_t bmask;
431 	l_stack_t *lss;
432 	stack_t ss;
433 	struct trapframe *regs;
434 	int eflags;
435 	ksiginfo_t ksi;
436 
437 	regs = td->td_frame;
438 
439 	/*
440 	 * The trampoline code hands us the ucontext.
441 	 * It is unsafe to keep track of it ourselves, in the event that a
442 	 * program jumps out of a signal handler.
443 	 */
444 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
445 		return (EFAULT);
446 
447 	context = &uc.uc_mcontext;
448 
449 	/* Check for security violations. */
450 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
451 	eflags = context->sc_eflags;
452 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
453 		return (EINVAL);
454 
455 	/*
456 	 * Don't allow users to load a valid privileged %cs.  Let the
457 	 * hardware check for invalid selectors, excess privilege in
458 	 * other selectors, invalid %eip's and invalid %esp's.
459 	 */
460 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
461 	if (!CS_SECURE(context->sc_cs)) {
462 		ksiginfo_init_trap(&ksi);
463 		ksi.ksi_signo = SIGBUS;
464 		ksi.ksi_code = BUS_OBJERR;
465 		ksi.ksi_trapno = T_PROTFLT;
466 		ksi.ksi_addr = (void *)regs->tf_eip;
467 		trapsignal(td, &ksi);
468 		return (EINVAL);
469 	}
470 
471 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
472 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
473 
474 	/* Restore signal context. */
475 	/* %gs was restored by the trampoline. */
476 	regs->tf_fs     = context->sc_fs;
477 	regs->tf_es     = context->sc_es;
478 	regs->tf_ds     = context->sc_ds;
479 	regs->tf_edi    = context->sc_edi;
480 	regs->tf_esi    = context->sc_esi;
481 	regs->tf_ebp    = context->sc_ebp;
482 	regs->tf_ebx    = context->sc_ebx;
483 	regs->tf_edx    = context->sc_edx;
484 	regs->tf_ecx    = context->sc_ecx;
485 	regs->tf_eax    = context->sc_eax;
486 	regs->tf_eip    = context->sc_eip;
487 	regs->tf_cs     = context->sc_cs;
488 	regs->tf_eflags = eflags;
489 	regs->tf_esp    = context->sc_esp_at_signal;
490 	regs->tf_ss     = context->sc_ss;
491 
492 	/* Call sigaltstack & ignore results. */
493 	lss = &uc.uc_stack;
494 	ss.ss_sp = PTRIN(lss->ss_sp);
495 	ss.ss_size = lss->ss_size;
496 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
497 
498 	(void)kern_sigaltstack(td, &ss, NULL);
499 
500 	return (EJUSTRETURN);
501 }
502 
503 static int
504 linux_fetch_syscall_args(struct thread *td)
505 {
506 	struct proc *p;
507 	struct trapframe *frame;
508 	struct syscall_args *sa;
509 
510 	p = td->td_proc;
511 	frame = td->td_frame;
512 	sa = &td->td_sa;
513 
514 	sa->code = frame->tf_eax;
515 	sa->original_code = sa->code;
516 	sa->args[0] = frame->tf_ebx;
517 	sa->args[1] = frame->tf_ecx;
518 	sa->args[2] = frame->tf_edx;
519 	sa->args[3] = frame->tf_esi;
520 	sa->args[4] = frame->tf_edi;
521 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
522 
523 	if (sa->code >= p->p_sysent->sv_size)
524 		/* nosys */
525 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
526 	else
527 		sa->callp = &p->p_sysent->sv_table[sa->code];
528 
529 	td->td_retval[0] = 0;
530 	td->td_retval[1] = frame->tf_edx;
531 
532 	return (0);
533 }
534 
535 static void
536 linux_set_syscall_retval(struct thread *td, int error)
537 {
538 	struct trapframe *frame = td->td_frame;
539 
540 	cpu_set_syscall_retval(td, error);
541 
542 	if (__predict_false(error != 0)) {
543 		if (error != ERESTART && error != EJUSTRETURN)
544 			frame->tf_eax = bsd_to_linux_errno(error);
545 	}
546 }
547 
548 static void
549 linux_set_fork_retval(struct thread *td)
550 {
551 	struct trapframe *frame = td->td_frame;
552 
553 	frame->tf_eax = 0;
554 }
555 
556 /*
557  * exec_setregs may initialize some registers differently than Linux
558  * does, thus potentially confusing Linux binaries. If necessary, we
559  * override the exec_setregs default(s) here.
560  */
561 static void
562 linux_exec_setregs(struct thread *td, struct image_params *imgp,
563     uintptr_t stack)
564 {
565 	struct pcb *pcb = td->td_pcb;
566 
567 	exec_setregs(td, imgp, stack);
568 
569 	/* Linux sets %gs to 0, we default to _udatasel. */
570 	pcb->pcb_gs = 0;
571 	load_gs(0);
572 
573 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
574 }
575 
576 struct sysentvec linux_sysvec = {
577 	.sv_size	= LINUX_SYS_MAXSYSCALL,
578 	.sv_table	= linux_sysent,
579 	.sv_fixup	= linux_fixup,
580 	.sv_sendsig	= linux_sendsig,
581 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
582 	.sv_szsigcode	= &linux_szsigcode,
583 	.sv_name	= "Linux a.out",
584 	.sv_coredump	= NULL,
585 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
586 	.sv_minuser	= VM_MIN_ADDRESS,
587 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
588 	.sv_usrstack	= LINUX_USRSTACK,
589 	.sv_psstrings	= PS_STRINGS,
590 	.sv_psstringssz	= sizeof(struct ps_strings),
591 	.sv_stackprot	= VM_PROT_ALL,
592 	.sv_copyout_strings = exec_copyout_strings,
593 	.sv_setregs	= linux_exec_setregs,
594 	.sv_fixlimit	= NULL,
595 	.sv_maxssiz	= NULL,
596 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32 |
597 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
598 	.sv_set_syscall_retval = linux_set_syscall_retval,
599 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
600 	.sv_syscallnames = linux_syscallnames,
601 	.sv_schedtail	= linux_schedtail,
602 	.sv_thread_detach = linux_thread_detach,
603 	.sv_trap	= NULL,
604 	.sv_hwcap	= NULL,
605 	.sv_hwcap2	= NULL,
606 	.sv_onexec	= linux_on_exec_vmspace,
607 	.sv_onexit	= linux_on_exit,
608 	.sv_ontdexit	= linux_thread_dtor,
609 	.sv_setid_allowed = &linux_setid_allowed_query,
610 	.sv_set_fork_retval = linux_set_fork_retval,
611 };
612 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
613 
614 struct sysentvec elf_linux_sysvec = {
615 	.sv_size	= LINUX_SYS_MAXSYSCALL,
616 	.sv_table	= linux_sysent,
617 	.sv_fixup	= __elfN(freebsd_fixup),
618 	.sv_sendsig	= linux_sendsig,
619 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
620 	.sv_szsigcode	= &linux_szsigcode,
621 	.sv_name	= "Linux ELF32",
622 	.sv_coredump	= elf32_coredump,
623 	.sv_elf_core_osabi = ELFOSABI_NONE,
624 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
625 	.sv_elf_core_prepare_notes = __linuxN(prepare_notes),
626 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
627 	.sv_minuser	= VM_MIN_ADDRESS,
628 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
629 	.sv_usrstack	= LINUX_USRSTACK,
630 	.sv_psstrings	= LINUX_PS_STRINGS,
631 	.sv_psstringssz	= sizeof(struct ps_strings),
632 	.sv_stackprot	= VM_PROT_ALL,
633 	.sv_copyout_auxargs = __linuxN(copyout_auxargs),
634 	.sv_copyout_strings = __linuxN(copyout_strings),
635 	.sv_setregs	= linux_exec_setregs,
636 	.sv_fixlimit	= NULL,
637 	.sv_maxssiz	= NULL,
638 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP |
639 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
640 	.sv_set_syscall_retval = linux_set_syscall_retval,
641 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
642 	.sv_syscallnames = NULL,
643 	.sv_shared_page_base = LINUX_SHAREDPAGE,
644 	.sv_shared_page_len = PAGE_SIZE,
645 	.sv_schedtail	= linux_schedtail,
646 	.sv_thread_detach = linux_thread_detach,
647 	.sv_trap	= NULL,
648 	.sv_hwcap	= NULL,
649 	.sv_hwcap2	= NULL,
650 	.sv_onexec	= linux_on_exec_vmspace,
651 	.sv_onexit	= linux_on_exit,
652 	.sv_ontdexit	= linux_thread_dtor,
653 	.sv_setid_allowed = &linux_setid_allowed_query,
654 	.sv_set_fork_retval = linux_set_fork_retval,
655 };
656 
657 static int
658 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
659 {
660 	int error = 0;
661 
662 	if (SV_PROC_FLAG(p, SV_SHP) != 0)
663 		error = linux_map_vdso(p, linux_vdso_obj,
664 		    linux_vdso_base, LINUX_VDSOPAGE_SIZE, imgp);
665 	if (error == 0)
666 		error = linux_on_exec(p, imgp);
667 	return (error);
668 }
669 
670 /*
671  * linux_vdso_install() and linux_exec_sysvec_init() must be called
672  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
673  */
674 static void
675 linux_exec_sysvec_init(void *param)
676 {
677 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
678 	struct sysentvec *sv;
679 	ptrdiff_t tkoff;
680 
681 	sv = param;
682 	/* Fill timekeep_base */
683 	exec_sysvec_init(sv);
684 
685 	tkoff = kern_timekeep_base - linux_vdso_base;
686 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
687 	*ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
688 
689 	tkoff = kern_tsc_selector - linux_vdso_base;
690 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
691 	*ktsc_selector = linux_vdso_tsc_selector_idx();
692 	if (bootverbose)
693 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
694 
695 	tkoff = kern_cpu_selector - linux_vdso_base;
696 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
697 	*ktsc_selector = linux_vdso_cpu_selector_idx();
698 	if (bootverbose)
699 		printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector);
700 }
701 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
702     linux_exec_sysvec_init, &elf_linux_sysvec);
703 
704 static void
705 linux_vdso_install(const void *param)
706 {
707 	char *vdso_start = &_binary_linux_vdso_so_o_start;
708 	char *vdso_end = &_binary_linux_vdso_so_o_end;
709 
710 	linux_szsigcode = vdso_end - vdso_start;
711 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
712 
713 	linux_vdso_base = LINUX_VDSOPAGE;
714 
715 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
716 
717 	linux_vdso_obj = __elfN(linux_shared_page_init)
718 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
719 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
720 
721 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
722 }
723 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
724     linux_vdso_install, NULL);
725 
726 static void
727 linux_vdso_deinstall(const void *param)
728 {
729 
730 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
731 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
732 }
733 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
734     linux_vdso_deinstall, NULL);
735 
736 static void
737 linux_vdso_reloc(char *mapping, Elf_Addr offset)
738 {
739 	const Elf_Shdr *shdr;
740 	const Elf_Rel *rel;
741 	const Elf_Ehdr *ehdr;
742 	Elf_Addr *where;
743 	Elf_Size rtype, symidx;
744 	Elf_Addr addr, addend;
745 	int i, relcnt;
746 
747 	MPASS(offset != 0);
748 
749 	relcnt = 0;
750 	ehdr = (const Elf_Ehdr *)mapping;
751 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
752 	for (i = 0; i < ehdr->e_shnum; i++)
753 	{
754 		switch (shdr[i].sh_type) {
755 		case SHT_REL:
756 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
757 			relcnt = shdr[i].sh_size / sizeof(*rel);
758 			break;
759 		case SHT_RELA:
760 			printf("Linux i386 vDSO: unexpected Rela section\n");
761 			break;
762 		}
763 	}
764 
765 	for (i = 0; i < relcnt; i++, rel++) {
766 		where = (Elf_Addr *)(mapping + rel->r_offset);
767 		addend = *where;
768 		rtype = ELF_R_TYPE(rel->r_info);
769 		symidx = ELF_R_SYM(rel->r_info);
770 
771 		switch (rtype) {
772 		case R_386_NONE:	/* none */
773 			break;
774 
775 		case R_386_RELATIVE:	/* B + A */
776 			addr = (Elf_Addr)PTROUT(offset + addend);
777 			if (*where != addr)
778 				*where = addr;
779 			break;
780 
781 		case R_386_IRELATIVE:
782 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
783 			    "symbol index %d\n", symidx);
784 			break;
785 		default:
786 			printf("Linux i386 vDSO: unexpected relocation type %d, "
787 			    "symbol index %d\n", rtype, symidx);
788 		}
789 	}
790 }
791 
792 static Elf_Brandnote linux_brandnote = {
793 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
794 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
795 	.hdr.n_type	= 1,
796 	.vendor		= GNU_ABI_VENDOR,
797 	.flags		= BN_TRANSLATE_OSREL,
798 	.trans_osrel	= linux_trans_osrel
799 };
800 
801 static Elf32_Brandinfo linux_brand = {
802 	.brand		= ELFOSABI_LINUX,
803 	.machine	= EM_386,
804 	.compat_3_brand	= "Linux",
805 	.interp_path	= "/lib/ld-linux.so.1",
806 	.sysvec		= &elf_linux_sysvec,
807 	.interp_newpath	= NULL,
808 	.brand_note	= &linux_brandnote,
809 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
810 };
811 
812 static Elf32_Brandinfo linux_glibc2brand = {
813 	.brand		= ELFOSABI_LINUX,
814 	.machine	= EM_386,
815 	.compat_3_brand	= "Linux",
816 	.interp_path	= "/lib/ld-linux.so.2",
817 	.sysvec		= &elf_linux_sysvec,
818 	.interp_newpath	= NULL,
819 	.brand_note	= &linux_brandnote,
820 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
821 };
822 
823 static Elf32_Brandinfo linux_muslbrand = {
824 	.brand		= ELFOSABI_LINUX,
825 	.machine	= EM_386,
826 	.compat_3_brand	= "Linux",
827 	.interp_path	= "/lib/ld-musl-i386.so.1",
828 	.sysvec		= &elf_linux_sysvec,
829 	.interp_newpath	= NULL,
830 	.brand_note	= &linux_brandnote,
831 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
832 			    LINUX_BI_FUTEX_REQUEUE
833 };
834 
835 Elf32_Brandinfo *linux_brandlist[] = {
836 	&linux_brand,
837 	&linux_glibc2brand,
838 	&linux_muslbrand,
839 	NULL
840 };
841 
842 static int
843 linux_elf_modevent(module_t mod, int type, void *data)
844 {
845 	Elf32_Brandinfo **brandinfo;
846 	int error;
847 	struct linux_ioctl_handler **lihp;
848 
849 	error = 0;
850 
851 	switch(type) {
852 	case MOD_LOAD:
853 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
854 		     ++brandinfo)
855 			if (elf32_insert_brand_entry(*brandinfo) < 0)
856 				error = EINVAL;
857 		if (error == 0) {
858 			SET_FOREACH(lihp, linux_ioctl_handler_set)
859 				linux_ioctl_register_handler(*lihp);
860 			linux_dev_shm_create();
861 			linux_osd_jail_register();
862 			linux_netlink_register();
863 			stclohz = (stathz ? stathz : hz);
864 			if (bootverbose)
865 				printf("Linux ELF exec handler installed\n");
866 		} else
867 			printf("cannot insert Linux ELF brand handler\n");
868 		break;
869 	case MOD_UNLOAD:
870 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
871 		     ++brandinfo)
872 			if (elf32_brand_inuse(*brandinfo))
873 				error = EBUSY;
874 		if (error == 0) {
875 			for (brandinfo = &linux_brandlist[0];
876 			     *brandinfo != NULL; ++brandinfo)
877 				if (elf32_remove_brand_entry(*brandinfo) < 0)
878 					error = EINVAL;
879 		}
880 		if (error == 0) {
881 			SET_FOREACH(lihp, linux_ioctl_handler_set)
882 				linux_ioctl_unregister_handler(*lihp);
883 			linux_netlink_deregister();
884 			linux_dev_shm_destroy();
885 			linux_osd_jail_deregister();
886 			if (bootverbose)
887 				printf("Linux ELF exec handler removed\n");
888 		} else
889 			printf("Could not deinstall ELF interpreter entry\n");
890 		break;
891 	default:
892 		return (EOPNOTSUPP);
893 	}
894 	return (error);
895 }
896 
897 static moduledata_t linux_elf_mod = {
898 	"linuxelf",
899 	linux_elf_modevent,
900 	0
901 };
902 
903 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
904 MODULE_DEPEND(linuxelf, netlink, 1, 1, 1);
905 FEATURE(linux, "Linux 32bit support");
906