xref: /freebsd/sys/amd64/linux32/linux32_sysvec.c (revision c5f49ece16a92380092191fad13cb1da72af5d37)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2004 Tim J. Robbins
5  * Copyright (c) 2003 Peter Wemm
6  * Copyright (c) 2002 Doug Rabson
7  * Copyright (c) 1998-1999 Andrew Gallatin
8  * Copyright (c) 1994-1996 Søren Schmidt
9  * All rights reserved.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer
16  *    in this position and unchanged.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. The name of the author may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 #ifndef COMPAT_FREEBSD32
37 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!"
38 #endif
39 
40 #define	__ELF_WORD_SIZE	32
41 
42 #include <sys/param.h>
43 #include <sys/exec.h>
44 #include <sys/fcntl.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/module.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/stddef.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysent.h>
57 
58 #include <vm/pmap.h>
59 #include <vm/vm.h>
60 #include <vm/vm_map.h>
61 #include <vm/vm_page.h>
62 
63 #include <machine/cpu.h>
64 #include <machine/md_var.h>
65 #include <machine/pcb.h>
66 #include <machine/specialreg.h>
67 #include <machine/trap.h>
68 
69 #include <x86/linux/linux_x86.h>
70 #include <amd64/linux32/linux.h>
71 #include <amd64/linux32/linux32_proto.h>
72 #include <compat/linux/linux_elf.h>
73 #include <compat/linux/linux_emul.h>
74 #include <compat/linux/linux_fork.h>
75 #include <compat/linux/linux_ioctl.h>
76 #include <compat/linux/linux_mib.h>
77 #include <compat/linux/linux_misc.h>
78 #include <compat/linux/linux_signal.h>
79 #include <compat/linux/linux_util.h>
80 #include <compat/linux/linux_vdso.h>
81 
82 #include <x86/linux/linux_x86_sigframe.h>
83 
84 MODULE_VERSION(linux, 1);
85 
86 #define	LINUX32_MAXUSER		((1ul << 32) - PAGE_SIZE)
87 #define	LINUX32_VDSOPAGE_SIZE	PAGE_SIZE * 2
88 #define	LINUX32_VDSOPAGE	(LINUX32_MAXUSER - LINUX32_VDSOPAGE_SIZE)
89 #define	LINUX32_SHAREDPAGE	(LINUX32_VDSOPAGE - PAGE_SIZE)
90 				/*
91 				 * PAGE_SIZE - the size
92 				 * of the native SHAREDPAGE
93 				 */
94 #define	LINUX32_USRSTACK	LINUX32_SHAREDPAGE
95 
96 static int linux_szsigcode;
97 static vm_object_t linux_vdso_obj;
98 static char *linux_vdso_mapping;
99 extern char _binary_linux32_vdso_so_o_start;
100 extern char _binary_linux32_vdso_so_o_end;
101 static vm_offset_t linux_vdso_base;
102 
103 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
104 extern const char *linux32_syscallnames[];
105 
106 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
107 
108 static int	linux_copyout_strings(struct image_params *imgp,
109 		    uintptr_t *stack_base);
110 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
111 static void	linux_exec_setregs(struct thread *td,
112 				   struct image_params *imgp, uintptr_t stack);
113 static void	linux_exec_sysvec_init(void *param);
114 static int	linux_on_exec_vmspace(struct proc *p,
115 		    struct image_params *imgp);
116 static void	linux32_fixlimit(struct rlimit *rl, int which);
117 static void	linux_vdso_install(const void *param);
118 static void	linux_vdso_deinstall(const void *param);
119 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
120 static void	linux32_set_fork_retval(struct thread *td);
121 static void	linux32_set_syscall_retval(struct thread *td, int error);
122 
123 struct linux32_ps_strings {
124 	u_int32_t ps_argvstr;	/* first of 0 or more argument strings */
125 	u_int ps_nargvstr;	/* the number of argument strings */
126 	u_int32_t ps_envstr;	/* first of 0 or more environment strings */
127 	u_int ps_nenvstr;	/* the number of environment strings */
128 };
129 #define	LINUX32_PS_STRINGS	(LINUX32_USRSTACK - \
130 				    sizeof(struct linux32_ps_strings))
131 
132 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
133 LINUX_VDSO_SYM_INTPTR(linux32_vdso_sigcode);
134 LINUX_VDSO_SYM_INTPTR(linux32_vdso_rt_sigcode);
135 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
136 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
137 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
138 LINUX_VDSO_SYM_CHAR(linux_platform);
139 
140 void
141 linux32_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos)
142 {
143 
144 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO, __kernel_vsyscall);
145 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
146 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature);
147 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2());
148 	AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform));
149 }
150 
151 static void
152 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
153 {
154 	struct thread *td = curthread;
155 	struct proc *p = td->td_proc;
156 	struct sigacts *psp;
157 	struct trapframe *regs;
158 	struct l_rt_sigframe *fp, frame;
159 	int oonstack;
160 	int sig;
161 	int code;
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_rsp);
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_rsp - 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 	/*
193 	 * Build the signal context to be used by sigreturn and libgcc unwind.
194 	 */
195 	frame.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
196 	frame.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
197 	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
198 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
199 	PROC_UNLOCK(p);
200 
201 	bsd_to_linux_sigset(mask, &frame.sf_uc.uc_sigmask);
202 
203 	frame.sf_uc.uc_mcontext.sc_mask   = frame.sf_uc.uc_sigmask.__mask;
204 	frame.sf_uc.uc_mcontext.sc_edi    = regs->tf_rdi;
205 	frame.sf_uc.uc_mcontext.sc_esi    = regs->tf_rsi;
206 	frame.sf_uc.uc_mcontext.sc_ebp    = regs->tf_rbp;
207 	frame.sf_uc.uc_mcontext.sc_ebx    = regs->tf_rbx;
208 	frame.sf_uc.uc_mcontext.sc_esp    = regs->tf_rsp;
209 	frame.sf_uc.uc_mcontext.sc_edx    = regs->tf_rdx;
210 	frame.sf_uc.uc_mcontext.sc_ecx    = regs->tf_rcx;
211 	frame.sf_uc.uc_mcontext.sc_eax    = regs->tf_rax;
212 	frame.sf_uc.uc_mcontext.sc_eip    = regs->tf_rip;
213 	frame.sf_uc.uc_mcontext.sc_cs     = regs->tf_cs;
214 	frame.sf_uc.uc_mcontext.sc_gs     = regs->tf_gs;
215 	frame.sf_uc.uc_mcontext.sc_fs     = regs->tf_fs;
216 	frame.sf_uc.uc_mcontext.sc_es     = regs->tf_es;
217 	frame.sf_uc.uc_mcontext.sc_ds     = regs->tf_ds;
218 	frame.sf_uc.uc_mcontext.sc_eflags = regs->tf_rflags;
219 	frame.sf_uc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
220 	frame.sf_uc.uc_mcontext.sc_ss     = regs->tf_ss;
221 	frame.sf_uc.uc_mcontext.sc_err    = regs->tf_err;
222 	frame.sf_uc.uc_mcontext.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
223 	frame.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
224 
225 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
226 		/*
227 		 * Process has trashed its stack; give it an illegal
228 		 * instruction to halt it in its tracks.
229 		 */
230 		PROC_LOCK(p);
231 		sigexit(td, SIGILL);
232 	}
233 
234 	/* Build context to run handler in. */
235 	regs->tf_rsp = PTROUT(fp);
236 	regs->tf_rip = linux32_vdso_rt_sigcode;
237 	regs->tf_rdi = PTROUT(catcher);
238 	regs->tf_rflags &= ~(PSL_T | PSL_D);
239 	regs->tf_cs = _ucode32sel;
240 	regs->tf_ss = _udatasel;
241 	regs->tf_ds = _udatasel;
242 	regs->tf_es = _udatasel;
243 	regs->tf_fs = _ufssel;
244 	regs->tf_gs = _ugssel;
245 	regs->tf_flags = TF_HASSEGS;
246 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
247 	PROC_LOCK(p);
248 	mtx_lock(&psp->ps_mtx);
249 }
250 
251 /*
252  * Send an interrupt to process.
253  *
254  * Stack is set up to allow sigcode stored
255  * in u. to call routine, followed by kcall
256  * to sigreturn routine below.  After sigreturn
257  * resets the signal mask, the stack, and the
258  * frame pointer, it returns to the user
259  * specified pc, psl.
260  */
261 static void
262 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
263 {
264 	struct thread *td = curthread;
265 	struct proc *p = td->td_proc;
266 	struct sigacts *psp;
267 	struct trapframe *regs;
268 	struct l_sigframe *fp, frame;
269 	l_sigset_t lmask;
270 	int oonstack;
271 	int sig, code;
272 
273 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
274 	code = ksi->ksi_code;
275 	PROC_LOCK_ASSERT(p, MA_OWNED);
276 	psp = p->p_sigacts;
277 	mtx_assert(&psp->ps_mtx, MA_OWNED);
278 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
279 		/* Signal handler installed with SA_SIGINFO. */
280 		linux_rt_sendsig(catcher, ksi, mask);
281 		return;
282 	}
283 
284 	regs = td->td_frame;
285 	oonstack = sigonstack(regs->tf_rsp);
286 
287 	/* Allocate space for the signal handler context. */
288 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
289 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
290 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
291 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
292 	} else
293 		fp = (struct l_sigframe *)regs->tf_rsp - 1;
294 	mtx_unlock(&psp->ps_mtx);
295 	PROC_UNLOCK(p);
296 
297 	/* Build the argument list for the signal handler. */
298 	sig = bsd_to_linux_signal(sig);
299 
300 	bzero(&frame, sizeof(frame));
301 
302 	frame.sf_sig = sig;
303 	frame.sf_sigmask = *mask;
304 	bsd_to_linux_sigset(mask, &lmask);
305 
306 	/* Build the signal context to be used by sigreturn. */
307 	frame.sf_sc.sc_mask   = lmask.__mask;
308 	frame.sf_sc.sc_gs     = regs->tf_gs;
309 	frame.sf_sc.sc_fs     = regs->tf_fs;
310 	frame.sf_sc.sc_es     = regs->tf_es;
311 	frame.sf_sc.sc_ds     = regs->tf_ds;
312 	frame.sf_sc.sc_edi    = regs->tf_rdi;
313 	frame.sf_sc.sc_esi    = regs->tf_rsi;
314 	frame.sf_sc.sc_ebp    = regs->tf_rbp;
315 	frame.sf_sc.sc_ebx    = regs->tf_rbx;
316 	frame.sf_sc.sc_esp    = regs->tf_rsp;
317 	frame.sf_sc.sc_edx    = regs->tf_rdx;
318 	frame.sf_sc.sc_ecx    = regs->tf_rcx;
319 	frame.sf_sc.sc_eax    = regs->tf_rax;
320 	frame.sf_sc.sc_eip    = regs->tf_rip;
321 	frame.sf_sc.sc_cs     = regs->tf_cs;
322 	frame.sf_sc.sc_eflags = regs->tf_rflags;
323 	frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
324 	frame.sf_sc.sc_ss     = regs->tf_ss;
325 	frame.sf_sc.sc_err    = regs->tf_err;
326 	frame.sf_sc.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
327 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
328 
329 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
330 		/*
331 		 * Process has trashed its stack; give it an illegal
332 		 * instruction to halt it in its tracks.
333 		 */
334 		PROC_LOCK(p);
335 		sigexit(td, SIGILL);
336 	}
337 
338 	/* Build context to run handler in. */
339 	regs->tf_rsp = PTROUT(fp);
340 	regs->tf_rip = linux32_vdso_sigcode;
341 	regs->tf_rdi = PTROUT(catcher);
342 	regs->tf_rflags &= ~(PSL_T | PSL_D);
343 	regs->tf_cs = _ucode32sel;
344 	regs->tf_ss = _udatasel;
345 	regs->tf_ds = _udatasel;
346 	regs->tf_es = _udatasel;
347 	regs->tf_fs = _ufssel;
348 	regs->tf_gs = _ugssel;
349 	regs->tf_flags = TF_HASSEGS;
350 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
351 	PROC_LOCK(p);
352 	mtx_lock(&psp->ps_mtx);
353 }
354 
355 /*
356  * System call to cleanup state after a signal
357  * has been taken.  Reset signal mask and
358  * stack state from context left by sendsig (above).
359  * Return to previous pc and psl as specified by
360  * context left by sendsig. Check carefully to
361  * make sure that the user has not modified the
362  * psl to gain improper privileges or to cause
363  * a machine fault.
364  */
365 int
366 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
367 {
368 	struct l_sigframe frame;
369 	struct trapframe *regs;
370 	int eflags;
371 	ksiginfo_t ksi;
372 
373 	regs = td->td_frame;
374 
375 	/*
376 	 * The trampoline code hands us the sigframe.
377 	 * It is unsafe to keep track of it ourselves, in the event that a
378 	 * program jumps out of a signal handler.
379 	 */
380 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
381 		return (EFAULT);
382 
383 	/* Check for security violations. */
384 	eflags = frame.sf_sc.sc_eflags;
385 	if (!EFL_SECURE(eflags, regs->tf_rflags))
386 		return(EINVAL);
387 
388 	/*
389 	 * Don't allow users to load a valid privileged %cs.  Let the
390 	 * hardware check for invalid selectors, excess privilege in
391 	 * other selectors, invalid %eip's and invalid %esp's.
392 	 */
393 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
394 		ksiginfo_init_trap(&ksi);
395 		ksi.ksi_signo = SIGBUS;
396 		ksi.ksi_code = BUS_OBJERR;
397 		ksi.ksi_trapno = T_PROTFLT;
398 		ksi.ksi_addr = (void *)regs->tf_rip;
399 		trapsignal(td, &ksi);
400 		return(EINVAL);
401 	}
402 
403 	kern_sigprocmask(td, SIG_SETMASK, &frame.sf_sigmask, NULL, 0);
404 
405 	/* Restore signal context. */
406 	regs->tf_rdi    = frame.sf_sc.sc_edi;
407 	regs->tf_rsi    = frame.sf_sc.sc_esi;
408 	regs->tf_rbp    = frame.sf_sc.sc_ebp;
409 	regs->tf_rbx    = frame.sf_sc.sc_ebx;
410 	regs->tf_rdx    = frame.sf_sc.sc_edx;
411 	regs->tf_rcx    = frame.sf_sc.sc_ecx;
412 	regs->tf_rax    = frame.sf_sc.sc_eax;
413 	regs->tf_rip    = frame.sf_sc.sc_eip;
414 	regs->tf_cs     = frame.sf_sc.sc_cs;
415 	regs->tf_ds     = frame.sf_sc.sc_ds;
416 	regs->tf_es     = frame.sf_sc.sc_es;
417 	regs->tf_fs     = frame.sf_sc.sc_fs;
418 	regs->tf_gs     = frame.sf_sc.sc_gs;
419 	regs->tf_rflags = eflags;
420 	regs->tf_rsp    = frame.sf_sc.sc_esp_at_signal;
421 	regs->tf_ss     = frame.sf_sc.sc_ss;
422 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
423 
424 	return (EJUSTRETURN);
425 }
426 
427 /*
428  * System call to cleanup state after a signal
429  * has been taken.  Reset signal mask and
430  * stack state from context left by rt_sendsig (above).
431  * Return to previous pc and psl as specified by
432  * context left by sendsig. Check carefully to
433  * make sure that the user has not modified the
434  * psl to gain improper privileges or to cause
435  * a machine fault.
436  */
437 int
438 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
439 {
440 	struct l_ucontext uc;
441 	struct l_sigcontext *context;
442 	sigset_t bmask;
443 	l_stack_t *lss;
444 	stack_t ss;
445 	struct trapframe *regs;
446 	int eflags;
447 	ksiginfo_t ksi;
448 
449 	regs = td->td_frame;
450 
451 	/*
452 	 * The trampoline code hands us the ucontext.
453 	 * It is unsafe to keep track of it ourselves, in the event that a
454 	 * program jumps out of a signal handler.
455 	 */
456 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
457 		return (EFAULT);
458 
459 	context = &uc.uc_mcontext;
460 
461 	/* Check for security violations. */
462 	eflags = context->sc_eflags;
463 	if (!EFL_SECURE(eflags, regs->tf_rflags))
464 		return(EINVAL);
465 
466 	/*
467 	 * Don't allow users to load a valid privileged %cs.  Let the
468 	 * hardware check for invalid selectors, excess privilege in
469 	 * other selectors, invalid %eip's and invalid %esp's.
470 	 */
471 	if (!CS_SECURE(context->sc_cs)) {
472 		ksiginfo_init_trap(&ksi);
473 		ksi.ksi_signo = SIGBUS;
474 		ksi.ksi_code = BUS_OBJERR;
475 		ksi.ksi_trapno = T_PROTFLT;
476 		ksi.ksi_addr = (void *)regs->tf_rip;
477 		trapsignal(td, &ksi);
478 		return(EINVAL);
479 	}
480 
481 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
482 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
483 
484 	/*
485 	 * Restore signal context
486 	 */
487 	regs->tf_gs	= context->sc_gs;
488 	regs->tf_fs	= context->sc_fs;
489 	regs->tf_es	= context->sc_es;
490 	regs->tf_ds	= context->sc_ds;
491 	regs->tf_rdi    = context->sc_edi;
492 	regs->tf_rsi    = context->sc_esi;
493 	regs->tf_rbp    = context->sc_ebp;
494 	regs->tf_rbx    = context->sc_ebx;
495 	regs->tf_rdx    = context->sc_edx;
496 	regs->tf_rcx    = context->sc_ecx;
497 	regs->tf_rax    = context->sc_eax;
498 	regs->tf_rip    = context->sc_eip;
499 	regs->tf_cs     = context->sc_cs;
500 	regs->tf_rflags = eflags;
501 	regs->tf_rsp    = context->sc_esp_at_signal;
502 	regs->tf_ss     = context->sc_ss;
503 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
504 
505 	/*
506 	 * call sigaltstack & ignore results..
507 	 */
508 	lss = &uc.uc_stack;
509 	ss.ss_sp = PTRIN(lss->ss_sp);
510 	ss.ss_size = lss->ss_size;
511 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
512 
513 	(void)kern_sigaltstack(td, &ss, NULL);
514 
515 	return (EJUSTRETURN);
516 }
517 
518 static int
519 linux32_fetch_syscall_args(struct thread *td)
520 {
521 	struct proc *p;
522 	struct trapframe *frame;
523 	struct syscall_args *sa;
524 
525 	p = td->td_proc;
526 	frame = td->td_frame;
527 	sa = &td->td_sa;
528 
529 	sa->args[0] = frame->tf_rbx;
530 	sa->args[1] = frame->tf_rcx;
531 	sa->args[2] = frame->tf_rdx;
532 	sa->args[3] = frame->tf_rsi;
533 	sa->args[4] = frame->tf_rdi;
534 	sa->args[5] = frame->tf_rbp;	/* Unconfirmed */
535 	sa->code = frame->tf_rax;
536 	sa->original_code = sa->code;
537 
538 	if (sa->code >= p->p_sysent->sv_size)
539 		/* nosys */
540 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
541 	else
542 		sa->callp = &p->p_sysent->sv_table[sa->code];
543 
544 	td->td_retval[0] = 0;
545 	td->td_retval[1] = frame->tf_rdx;
546 
547 	return (0);
548 }
549 
550 static void
551 linux32_set_syscall_retval(struct thread *td, int error)
552 {
553 	struct trapframe *frame = td->td_frame;
554 
555 	cpu_set_syscall_retval(td, error);
556 
557 	if (__predict_false(error != 0)) {
558 		if (error != ERESTART && error != EJUSTRETURN)
559 			frame->tf_rax = bsd_to_linux_errno(error);
560 	}
561 }
562 
563 static void
564 linux32_set_fork_retval(struct thread *td)
565 {
566 	struct trapframe *frame = td->td_frame;
567 
568 	frame->tf_rax = 0;
569 }
570 
571 /*
572  * Clear registers on exec
573  * XXX copied from ia32_signal.c.
574  */
575 static void
576 linux_exec_setregs(struct thread *td, struct image_params *imgp,
577     uintptr_t stack)
578 {
579 	struct trapframe *regs = td->td_frame;
580 	struct pcb *pcb = td->td_pcb;
581 	register_t saved_rflags;
582 
583 	regs = td->td_frame;
584 	pcb = td->td_pcb;
585 
586 	if (td->td_proc->p_md.md_ldt != NULL)
587 		user_ldt_free(td);
588 
589 	critical_enter();
590 	wrmsr(MSR_FSBASE, 0);
591 	wrmsr(MSR_KGSBASE, 0);	/* User value while we're in the kernel */
592 	pcb->pcb_fsbase = 0;
593 	pcb->pcb_gsbase = 0;
594 	critical_exit();
595 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
596 
597 	saved_rflags = regs->tf_rflags & PSL_T;
598 	bzero((char *)regs, sizeof(struct trapframe));
599 	regs->tf_rip = imgp->entry_addr;
600 	regs->tf_rsp = stack;
601 	regs->tf_rflags = PSL_USER | saved_rflags;
602 	regs->tf_gs = _ugssel;
603 	regs->tf_fs = _ufssel;
604 	regs->tf_es = _udatasel;
605 	regs->tf_ds = _udatasel;
606 	regs->tf_ss = _udatasel;
607 	regs->tf_flags = TF_HASSEGS;
608 	regs->tf_cs = _ucode32sel;
609 	regs->tf_rbx = (register_t)imgp->ps_strings;
610 
611 	x86_clear_dbregs(pcb);
612 
613 	fpstate_drop(td);
614 
615 	/* Do full restore on return so that we can change to a different %cs */
616 	set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
617 }
618 
619 /*
620  * XXX copied from ia32_sysvec.c.
621  */
622 static int
623 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
624 {
625 	int argc, envc, error;
626 	u_int32_t *vectp;
627 	char *stringp;
628 	uintptr_t destp, ustringp;
629 	struct linux32_ps_strings *arginfo;
630 	char canary[LINUX_AT_RANDOM_LEN];
631 	size_t execpath_len;
632 
633 	arginfo = (struct linux32_ps_strings *)PROC_PS_STRINGS(imgp->proc);
634 	destp = (uintptr_t)arginfo;
635 
636 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
637 		execpath_len = strlen(imgp->execpath) + 1;
638 		destp -= execpath_len;
639 		destp = rounddown2(destp, sizeof(uint32_t));
640 		imgp->execpathp = (void *)destp;
641 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
642 		if (error != 0)
643 			return (error);
644 	}
645 
646 	/* Prepare the canary for SSP. */
647 	arc4rand(canary, sizeof(canary), 0);
648 	destp -= roundup(sizeof(canary), sizeof(uint32_t));
649 	imgp->canary = (void *)destp;
650 	error = copyout(canary, imgp->canary, sizeof(canary));
651 	if (error != 0)
652 		return (error);
653 
654 	/* Allocate room for the argument and environment strings. */
655 	destp -= ARG_MAX - imgp->args->stringspace;
656 	destp = rounddown2(destp, sizeof(uint32_t));
657 	ustringp = destp;
658 
659 	if (imgp->auxargs) {
660 		/*
661 		 * Allocate room on the stack for the ELF auxargs
662 		 * array.  It has LINUX_AT_COUNT entries.
663 		 */
664 		destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
665 		destp = rounddown2(destp, sizeof(uint32_t));
666 	}
667 
668 	vectp = (uint32_t *)destp;
669 
670 	/*
671 	 * Allocate room for the argv[] and env vectors including the
672 	 * terminating NULL pointers.
673 	 */
674 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
675 
676 	/* vectp also becomes our initial stack base. */
677 	*stack_base = (uintptr_t)vectp;
678 
679 	stringp = imgp->args->begin_argv;
680 	argc = imgp->args->argc;
681 	envc = imgp->args->envc;
682 
683 	/* Copy out strings - arguments and environment. */
684 	error = copyout(stringp, (void *)ustringp,
685 	    ARG_MAX - imgp->args->stringspace);
686 	if (error != 0)
687 		return (error);
688 
689 	/* Fill in "ps_strings" struct for ps, w, etc. */
690 	if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 ||
691 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
692 		return (EFAULT);
693 
694 	/* Fill in argument portion of vector table. */
695 	for (; argc > 0; --argc) {
696 		if (suword32(vectp++, ustringp) != 0)
697 			return (EFAULT);
698 		while (*stringp++ != 0)
699 			ustringp++;
700 		ustringp++;
701 	}
702 
703 	/* A null vector table pointer separates the argp's from the envp's. */
704 	if (suword32(vectp++, 0) != 0)
705 		return (EFAULT);
706 
707 	if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 ||
708 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
709 		return (EFAULT);
710 
711 	/* Fill in environment portion of vector table. */
712 	for (; envc > 0; --envc) {
713 		if (suword32(vectp++, ustringp) != 0)
714 			return (EFAULT);
715 		while (*stringp++ != 0)
716 			ustringp++;
717 		ustringp++;
718 	}
719 
720 	/* The end of the vector table is a null pointer. */
721 	if (suword32(vectp, 0) != 0)
722 		return (EFAULT);
723 
724 	if (imgp->auxargs) {
725 		vectp++;
726 		error = imgp->sysent->sv_copyout_auxargs(imgp,
727 		    (uintptr_t)vectp);
728 		if (error != 0)
729 			return (error);
730 	}
731 
732 	return (0);
733 }
734 
735 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
736     "32-bit Linux emulation");
737 
738 static u_long	linux32_maxdsiz = LINUX32_MAXDSIZ;
739 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
740     &linux32_maxdsiz, 0, "");
741 static u_long	linux32_maxssiz = LINUX32_MAXSSIZ;
742 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
743     &linux32_maxssiz, 0, "");
744 static u_long	linux32_maxvmem = LINUX32_MAXVMEM;
745 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
746     &linux32_maxvmem, 0, "");
747 bool linux32_emulate_i386 = false;
748 SYSCTL_BOOL(_compat_linux32, OID_AUTO, emulate_i386, CTLFLAG_RWTUN,
749     &linux32_emulate_i386, 0, "Emulate the real i386");
750 
751 static void
752 linux32_fixlimit(struct rlimit *rl, int which)
753 {
754 
755 	switch (which) {
756 	case RLIMIT_DATA:
757 		if (linux32_maxdsiz != 0) {
758 			if (rl->rlim_cur > linux32_maxdsiz)
759 				rl->rlim_cur = linux32_maxdsiz;
760 			if (rl->rlim_max > linux32_maxdsiz)
761 				rl->rlim_max = linux32_maxdsiz;
762 		}
763 		break;
764 	case RLIMIT_STACK:
765 		if (linux32_maxssiz != 0) {
766 			if (rl->rlim_cur > linux32_maxssiz)
767 				rl->rlim_cur = linux32_maxssiz;
768 			if (rl->rlim_max > linux32_maxssiz)
769 				rl->rlim_max = linux32_maxssiz;
770 		}
771 		break;
772 	case RLIMIT_VMEM:
773 		if (linux32_maxvmem != 0) {
774 			if (rl->rlim_cur > linux32_maxvmem)
775 				rl->rlim_cur = linux32_maxvmem;
776 			if (rl->rlim_max > linux32_maxvmem)
777 				rl->rlim_max = linux32_maxvmem;
778 		}
779 		break;
780 	}
781 }
782 
783 struct sysentvec elf_linux_sysvec = {
784 	.sv_size	= LINUX32_SYS_MAXSYSCALL,
785 	.sv_table	= linux32_sysent,
786 	.sv_fixup	= elf32_freebsd_fixup,
787 	.sv_sendsig	= linux_sendsig,
788 	.sv_sigcode	= &_binary_linux32_vdso_so_o_start,
789 	.sv_szsigcode	= &linux_szsigcode,
790 	.sv_name	= "Linux ELF32",
791 	.sv_coredump	= elf32_coredump,
792 	.sv_elf_core_osabi = ELFOSABI_NONE,
793 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
794 	.sv_elf_core_prepare_notes = linux32_prepare_notes,
795 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
796 	.sv_minuser	= VM_MIN_ADDRESS,
797 	.sv_maxuser	= LINUX32_MAXUSER,
798 	.sv_usrstack	= LINUX32_USRSTACK,
799 	.sv_psstrings	= LINUX32_PS_STRINGS,
800 	.sv_psstringssz	= sizeof(struct linux32_ps_strings),
801 	.sv_stackprot	= VM_PROT_ALL,
802 	.sv_copyout_auxargs = __linuxN(copyout_auxargs),
803 	.sv_copyout_strings = linux_copyout_strings,
804 	.sv_setregs	= linux_exec_setregs,
805 	.sv_fixlimit	= linux32_fixlimit,
806 	.sv_maxssiz	= &linux32_maxssiz,
807 	.sv_flags	= SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP |
808 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
809 	.sv_set_syscall_retval = linux32_set_syscall_retval,
810 	.sv_fetch_syscall_args = linux32_fetch_syscall_args,
811 	.sv_syscallnames = linux32_syscallnames,
812 	.sv_shared_page_base = LINUX32_SHAREDPAGE,
813 	.sv_shared_page_len = PAGE_SIZE,
814 	.sv_schedtail	= linux_schedtail,
815 	.sv_thread_detach = linux_thread_detach,
816 	.sv_trap	= NULL,
817 	.sv_hwcap	= NULL,
818 	.sv_hwcap2	= NULL,
819 	.sv_onexec	= linux_on_exec_vmspace,
820 	.sv_onexit	= linux_on_exit,
821 	.sv_ontdexit	= linux_thread_dtor,
822 	.sv_setid_allowed = &linux_setid_allowed_query,
823 	.sv_set_fork_retval = linux32_set_fork_retval,
824 };
825 
826 static int
827 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
828 {
829 	int error;
830 
831 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
832 	    LINUX32_VDSOPAGE_SIZE, imgp);
833 	if (error == 0)
834 		error = linux_on_exec(p, imgp);
835 	return (error);
836 }
837 
838 /*
839  * linux_vdso_install() and linux_exec_sysvec_init() must be called
840  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
841  */
842 static void
843 linux_exec_sysvec_init(void *param)
844 {
845 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
846 	struct sysentvec *sv;
847 	ptrdiff_t tkoff;
848 
849 	sv = param;
850 	/* Fill timekeep_base */
851 	exec_sysvec_init(sv);
852 
853 	tkoff = kern_timekeep_base - linux_vdso_base;
854 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
855 	*ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
856 
857 	tkoff = kern_tsc_selector - linux_vdso_base;
858 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
859 	*ktsc_selector = linux_vdso_tsc_selector_idx();
860 	if (bootverbose)
861 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
862 
863 	tkoff = kern_cpu_selector - linux_vdso_base;
864 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
865 	*ktsc_selector = linux_vdso_cpu_selector_idx();
866 	if (bootverbose)
867 		printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector);
868 }
869 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
870     linux_exec_sysvec_init, &elf_linux_sysvec);
871 
872 static void
873 linux_vdso_install(const void *param)
874 {
875 	char *vdso_start = &_binary_linux32_vdso_so_o_start;
876 	char *vdso_end = &_binary_linux32_vdso_so_o_end;
877 
878 	linux_szsigcode = vdso_end - vdso_start;
879 	MPASS(linux_szsigcode <= LINUX32_VDSOPAGE_SIZE);
880 
881 	linux_vdso_base = LINUX32_VDSOPAGE;
882 
883 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
884 
885 	linux_vdso_obj = __elfN(linux_shared_page_init)
886 	    (&linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
887 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
888 
889 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
890 }
891 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
892     linux_vdso_install, NULL);
893 
894 static void
895 linux_vdso_deinstall(const void *param)
896 {
897 
898 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
899 	    linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
900 }
901 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
902     linux_vdso_deinstall, NULL);
903 
904 static void
905 linux_vdso_reloc(char *mapping, Elf_Addr offset)
906 {
907 	const Elf_Shdr *shdr;
908 	const Elf_Rel *rel;
909 	const Elf_Ehdr *ehdr;
910 	Elf32_Addr *where;
911 	Elf_Size rtype, symidx;
912 	Elf32_Addr addr, addend;
913 	int i, relcnt;
914 
915 	MPASS(offset != 0);
916 
917 	relcnt = 0;
918 	ehdr = (const Elf_Ehdr *)mapping;
919 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
920 	for (i = 0; i < ehdr->e_shnum; i++)
921 	{
922 		switch (shdr[i].sh_type) {
923 		case SHT_REL:
924 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
925 			relcnt = shdr[i].sh_size / sizeof(*rel);
926 			break;
927 		case SHT_RELA:
928 			printf("Linux i386 vDSO: unexpected Rela section\n");
929 			break;
930 		}
931 	}
932 
933 	for (i = 0; i < relcnt; i++, rel++) {
934 		where = (Elf32_Addr *)(mapping + rel->r_offset);
935 		addend = *where;
936 		rtype = ELF_R_TYPE(rel->r_info);
937 		symidx = ELF_R_SYM(rel->r_info);
938 
939 		switch (rtype) {
940 		case R_386_NONE:	/* none */
941 			break;
942 
943 		case R_386_RELATIVE:	/* B + A */
944 			addr = (Elf32_Addr)PTROUT(offset + addend);
945 			if (*where != addr)
946 				*where = addr;
947 			break;
948 
949 		case R_386_IRELATIVE:
950 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
951 			    "symbol index %ld\n", (intmax_t)symidx);
952 			break;
953 		default:
954 			printf("Linux i386 vDSO: unexpected relocation type %ld, "
955 			    "symbol index %ld\n", (intmax_t)rtype, (intmax_t)symidx);
956 		}
957 	}
958 }
959 
960 static Elf_Brandnote linux32_brandnote = {
961 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
962 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
963 	.hdr.n_type	= 1,
964 	.vendor		= GNU_ABI_VENDOR,
965 	.flags		= BN_TRANSLATE_OSREL,
966 	.trans_osrel	= linux_trans_osrel
967 };
968 
969 static Elf32_Brandinfo linux_brand = {
970 	.brand		= ELFOSABI_LINUX,
971 	.machine	= EM_386,
972 	.compat_3_brand	= "Linux",
973 	.interp_path	= "/lib/ld-linux.so.1",
974 	.sysvec		= &elf_linux_sysvec,
975 	.interp_newpath	= NULL,
976 	.brand_note	= &linux32_brandnote,
977 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
978 };
979 
980 static Elf32_Brandinfo linux_glibc2brand = {
981 	.brand		= ELFOSABI_LINUX,
982 	.machine	= EM_386,
983 	.compat_3_brand	= "Linux",
984 	.interp_path	= "/lib/ld-linux.so.2",
985 	.sysvec		= &elf_linux_sysvec,
986 	.interp_newpath	= NULL,
987 	.brand_note	= &linux32_brandnote,
988 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
989 };
990 
991 static Elf32_Brandinfo linux_muslbrand = {
992 	.brand		= ELFOSABI_LINUX,
993 	.machine	= EM_386,
994 	.compat_3_brand	= "Linux",
995 	.interp_path	= "/lib/ld-musl-i386.so.1",
996 	.sysvec		= &elf_linux_sysvec,
997 	.interp_newpath	= NULL,
998 	.brand_note	= &linux32_brandnote,
999 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
1000 			    LINUX_BI_FUTEX_REQUEUE
1001 };
1002 
1003 Elf32_Brandinfo *linux_brandlist[] = {
1004 	&linux_brand,
1005 	&linux_glibc2brand,
1006 	&linux_muslbrand,
1007 	NULL
1008 };
1009 
1010 static int
1011 linux_elf_modevent(module_t mod, int type, void *data)
1012 {
1013 	Elf32_Brandinfo **brandinfo;
1014 	int error;
1015 	struct linux_ioctl_handler **lihp;
1016 
1017 	error = 0;
1018 
1019 	switch(type) {
1020 	case MOD_LOAD:
1021 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1022 		     ++brandinfo)
1023 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1024 				error = EINVAL;
1025 		if (error == 0) {
1026 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1027 				linux32_ioctl_register_handler(*lihp);
1028 			stclohz = (stathz ? stathz : hz);
1029 			if (bootverbose)
1030 				printf("Linux i386 ELF exec handler installed\n");
1031 		} else
1032 			printf("cannot insert Linux i386 ELF brand handler\n");
1033 		break;
1034 	case MOD_UNLOAD:
1035 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1036 		     ++brandinfo)
1037 			if (elf32_brand_inuse(*brandinfo))
1038 				error = EBUSY;
1039 		if (error == 0) {
1040 			for (brandinfo = &linux_brandlist[0];
1041 			     *brandinfo != NULL; ++brandinfo)
1042 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1043 					error = EINVAL;
1044 		}
1045 		if (error == 0) {
1046 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1047 				linux32_ioctl_unregister_handler(*lihp);
1048 			if (bootverbose)
1049 				printf("Linux i386 ELF exec handler removed\n");
1050 		} else
1051 			printf("Could not deinstall Linux i386 ELF interpreter entry\n");
1052 		break;
1053 	default:
1054 		return (EOPNOTSUPP);
1055 	}
1056 	return (error);
1057 }
1058 
1059 static moduledata_t linux_elf_mod = {
1060 	"linuxelf",
1061 	linux_elf_modevent,
1062 	0
1063 };
1064 
1065 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1066 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1067 FEATURE(linux, "Linux 32bit support");
1068