xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision 8c2dd68caa963f1900a8228b0732b04f5d530ffa)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/imgact_aout.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/signalvar.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysctl.h>
48 #include <sys/sysent.h>
49 #include <sys/sysproto.h>
50 #include <sys/vnode.h>
51 
52 #include <vm/vm.h>
53 #include <vm/pmap.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_param.h>
59 
60 #include <machine/cpu.h>
61 #include <machine/cputypes.h>
62 #include <machine/md_var.h>
63 #include <machine/pcb.h>
64 #include <machine/trap.h>
65 
66 #include <i386/linux/linux.h>
67 #include <i386/linux/linux_proto.h>
68 #include <compat/linux/linux_emul.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 MODULE_VERSION(linux, 1);
77 
78 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
79 
80 static int linux_szsigcode;
81 static vm_object_t linux_shared_page_obj;
82 static char *linux_shared_page_mapping;
83 extern char _binary_linux_locore_o_start;
84 extern char _binary_linux_locore_o_end;
85 
86 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
87 
88 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
89 
90 static int	linux_fixup(uintptr_t *stack_base,
91 		    struct image_params *iparams);
92 static int	linux_fixup_elf(uintptr_t *stack_base,
93 		    struct image_params *iparams);
94 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
95 static void	linux_exec_setregs(struct thread *td,
96 		    struct image_params *imgp, uintptr_t stack);
97 static int	linux_copyout_strings(struct image_params *imgp,
98 		    uintptr_t *stack_base);
99 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static void	linux_vdso_install(void *param);
101 static void	linux_vdso_deinstall(void *param);
102 
103 #define LINUX_T_UNKNOWN  255
104 static int _bsd_to_linux_trapcode[] = {
105 	LINUX_T_UNKNOWN,	/* 0 */
106 	6,			/* 1  T_PRIVINFLT */
107 	LINUX_T_UNKNOWN,	/* 2 */
108 	3,			/* 3  T_BPTFLT */
109 	LINUX_T_UNKNOWN,	/* 4 */
110 	LINUX_T_UNKNOWN,	/* 5 */
111 	16,			/* 6  T_ARITHTRAP */
112 	254,			/* 7  T_ASTFLT */
113 	LINUX_T_UNKNOWN,	/* 8 */
114 	13,			/* 9  T_PROTFLT */
115 	1,			/* 10 T_TRCTRAP */
116 	LINUX_T_UNKNOWN,	/* 11 */
117 	14,			/* 12 T_PAGEFLT */
118 	LINUX_T_UNKNOWN,	/* 13 */
119 	17,			/* 14 T_ALIGNFLT */
120 	LINUX_T_UNKNOWN,	/* 15 */
121 	LINUX_T_UNKNOWN,	/* 16 */
122 	LINUX_T_UNKNOWN,	/* 17 */
123 	0,			/* 18 T_DIVIDE */
124 	2,			/* 19 T_NMI */
125 	4,			/* 20 T_OFLOW */
126 	5,			/* 21 T_BOUND */
127 	7,			/* 22 T_DNA */
128 	8,			/* 23 T_DOUBLEFLT */
129 	9,			/* 24 T_FPOPFLT */
130 	10,			/* 25 T_TSSFLT */
131 	11,			/* 26 T_SEGNPFLT */
132 	12,			/* 27 T_STKFLT */
133 	18,			/* 28 T_MCHK */
134 	19,			/* 29 T_XMMFLT */
135 	15			/* 30 T_RESERVED */
136 };
137 #define bsd_to_linux_trapcode(code) \
138     ((code)<nitems(_bsd_to_linux_trapcode)? \
139      _bsd_to_linux_trapcode[(code)]: \
140      LINUX_T_UNKNOWN)
141 
142 LINUX_VDSO_SYM_CHAR(linux_platform);
143 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
144 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
145 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
146 
147 /*
148  * If FreeBSD & Linux have a difference of opinion about what a trap
149  * means, deal with it here.
150  *
151  * MPSAFE
152  */
153 static int
154 linux_translate_traps(int signal, int trap_code)
155 {
156 	if (signal != SIGBUS)
157 		return (signal);
158 	switch (trap_code) {
159 	case T_PROTFLT:
160 	case T_TSSFLT:
161 	case T_DOUBLEFLT:
162 	case T_PAGEFLT:
163 		return (SIGSEGV);
164 	default:
165 		return (signal);
166 	}
167 }
168 
169 static int
170 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
171 {
172 	register_t *base, *argv, *envp;
173 
174 	base = (register_t *)*stack_base;
175 	argv = base;
176 	envp = base + (imgp->args->argc + 1);
177 	base--;
178 	suword(base, (intptr_t)envp);
179 	base--;
180 	suword(base, (intptr_t)argv);
181 	base--;
182 	suword(base, imgp->args->argc);
183 	*stack_base = (uintptr_t)base;
184 	return (0);
185 }
186 
187 static int
188 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
189 {
190 	struct proc *p;
191 	Elf32_Auxargs *args;
192 	Elf32_Auxinfo *argarray, *pos;
193 	struct ps_strings *arginfo;
194 	int error, issetugid;
195 
196 	p = imgp->proc;
197 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
198 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
199 	args = (Elf32_Auxargs *)imgp->auxargs;
200 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
201 	    M_WAITOK | M_ZERO);
202 
203 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
204 	    imgp->proc->p_sysent->sv_shared_page_base);
205 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
206 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
207 
208 	/*
209 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
210 	 * as it has appeared in the 2.4.0-rc7 first time.
211 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
212 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
213 	 * is not present.
214 	 * Also see linux_times() implementation.
215 	 */
216 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
217 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
218 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
219 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
220 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
221 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
222 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
223 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
224 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
225 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
226 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
227 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
228 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
229 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
230 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
231 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
232 	if (imgp->execpathp != 0)
233 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
234 	if (args->execfd != -1)
235 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
236 	AUXARGS_ENTRY(pos, AT_NULL, 0);
237 
238 	free(imgp->auxargs, M_TEMP);
239 	imgp->auxargs = NULL;
240 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
241 
242 	error = copyout(argarray, (void *)base,
243 	    sizeof(*argarray) * LINUX_AT_COUNT);
244 	free(argarray, M_TEMP);
245 	return (error);
246 }
247 
248 static int
249 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
250 {
251 	register_t *base;
252 
253 	base = (register_t *)*stack_base;
254 	base--;
255 	if (suword(base, (register_t)imgp->args->argc) == -1)
256 		return (EFAULT);
257 	*stack_base = (uintptr_t)base;
258 	return (0);
259 }
260 
261 /*
262  * Copied from kern/kern_exec.c
263  */
264 static int
265 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
266 {
267 	int argc, envc, error;
268 	char **vectp;
269 	char *stringp;
270 	uintptr_t destp, ustringp;
271 	struct ps_strings *arginfo;
272 	char canary[LINUX_AT_RANDOM_LEN];
273 	size_t execpath_len;
274 	struct proc *p;
275 
276 	/* Calculate string base and vector table pointers. */
277 	p = imgp->proc;
278 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
279 		execpath_len = strlen(imgp->execpath) + 1;
280 	else
281 		execpath_len = 0;
282 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
283 	destp = (uintptr_t)arginfo;
284 
285 	if (execpath_len != 0) {
286 		destp -= execpath_len;
287 		destp = rounddown2(destp, sizeof(void *));
288 		imgp->execpathp = (void *)destp;
289 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
290 		if (error != 0)
291 			return (error);
292 	}
293 
294 	/* Prepare the canary for SSP. */
295 	arc4rand(canary, sizeof(canary), 0);
296 	destp -= roundup(sizeof(canary), sizeof(void *));
297 	imgp->canary = (void *)destp;
298 	error = copyout(canary, imgp->canary, sizeof(canary));
299 	if (error != 0)
300 		return (error);
301 
302 	/* Allocate room for the argument and environment strings. */
303 	destp -= ARG_MAX - imgp->args->stringspace;
304 	destp = rounddown2(destp, sizeof(void *));
305 	ustringp = destp;
306 
307 	if (imgp->auxargs) {
308 		/*
309 		 * Allocate room on the stack for the ELF auxargs
310 		 * array.  It has LINUX_AT_COUNT entries.
311 		 */
312 		destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
313 		destp = rounddown2(destp, sizeof(void *));
314 	}
315 
316 	vectp = (char **)destp;
317 
318 	/*
319 	 * Allocate room for the argv[] and env vectors including the
320 	 * terminating NULL pointers.
321 	 */
322 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
323 
324 	/* vectp also becomes our initial stack base. */
325 	*stack_base = (uintptr_t)vectp;
326 
327 	stringp = imgp->args->begin_argv;
328 	argc = imgp->args->argc;
329 	envc = imgp->args->envc;
330 
331 	/* Copy out strings - arguments and environment. */
332 	error = copyout(stringp, (void *)ustringp,
333 	    ARG_MAX - imgp->args->stringspace);
334 	if (error != 0)
335 		return (error);
336 
337 	/* Fill in "ps_strings" struct for ps, w, etc. */
338 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
339 	    suword(&arginfo->ps_nargvstr, argc) != 0)
340 		return (EFAULT);
341 
342 	/* Fill in argument portion of vector table. */
343 	for (; argc > 0; --argc) {
344 		if (suword(vectp++, ustringp) != 0)
345 			return (EFAULT);
346 		while (*stringp++ != 0)
347 			ustringp++;
348 		ustringp++;
349 	}
350 
351 	/* A null vector table pointer separates the argp's from the envp's. */
352 	if (suword(vectp++, 0) != 0)
353 		return (EFAULT);
354 
355 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
356 	    suword(&arginfo->ps_nenvstr, envc) != 0)
357 		return (EFAULT);
358 
359 	/* Fill in environment portion of vector table. */
360 	for (; envc > 0; --envc) {
361 		if (suword(vectp++, ustringp) != 0)
362 			return (EFAULT);
363 		while (*stringp++ != 0)
364 			ustringp++;
365 		ustringp++;
366 	}
367 
368 	/* The end of the vector table is a null pointer. */
369 	if (suword(vectp, 0) != 0)
370 		return (EFAULT);
371 
372 	if (imgp->auxargs) {
373 		vectp++;
374 		error = imgp->sysent->sv_copyout_auxargs(imgp,
375 		    (uintptr_t)vectp);
376 		if (error != 0)
377 			return (error);
378 	}
379 
380 	return (0);
381 }
382 
383 static void
384 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
385 {
386 	struct thread *td = curthread;
387 	struct proc *p = td->td_proc;
388 	struct sigacts *psp;
389 	struct trapframe *regs;
390 	struct l_rt_sigframe *fp, frame;
391 	int sig, code;
392 	int oonstack;
393 
394 	sig = ksi->ksi_signo;
395 	code = ksi->ksi_code;
396 	PROC_LOCK_ASSERT(p, MA_OWNED);
397 	psp = p->p_sigacts;
398 	mtx_assert(&psp->ps_mtx, MA_OWNED);
399 	regs = td->td_frame;
400 	oonstack = sigonstack(regs->tf_esp);
401 
402 	/* Allocate space for the signal handler context. */
403 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
404 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
405 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
406 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
407 	} else
408 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
409 	mtx_unlock(&psp->ps_mtx);
410 
411 	/* Build the argument list for the signal handler. */
412 	sig = bsd_to_linux_signal(sig);
413 
414 	bzero(&frame, sizeof(frame));
415 
416 	frame.sf_handler = catcher;
417 	frame.sf_sig = sig;
418 	frame.sf_siginfo = &fp->sf_si;
419 	frame.sf_ucontext = &fp->sf_sc;
420 
421 	/* Fill in POSIX parts. */
422 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
423 
424 	/* Build the signal context to be used by sigreturn. */
425 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
426 	frame.sf_sc.uc_link = NULL;		/* XXX ??? */
427 
428 	frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
429 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
430 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
431 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
432 	PROC_UNLOCK(p);
433 
434 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
435 
436 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
437 	frame.sf_sc.uc_mcontext.sc_gs     = rgs();
438 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
439 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
440 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
441 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_edi;
442 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_esi;
443 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_ebp;
444 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_ebx;
445 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_esp;
446 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_edx;
447 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_ecx;
448 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_eax;
449 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_eip;
450 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
451 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
452 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
453 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
454 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
455 	frame.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
456 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
457 
458 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
459 		/*
460 		 * Process has trashed its stack; give it an illegal
461 		 * instruction to halt it in its tracks.
462 		 */
463 		PROC_LOCK(p);
464 		sigexit(td, SIGILL);
465 	}
466 
467 	/* Build context to run handler in. */
468 	regs->tf_esp = (int)fp;
469 	regs->tf_eip = linux_rt_sigcode;
470 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
471 	regs->tf_cs = _ucodesel;
472 	regs->tf_ds = _udatasel;
473 	regs->tf_es = _udatasel;
474 	regs->tf_fs = _udatasel;
475 	regs->tf_ss = _udatasel;
476 	PROC_LOCK(p);
477 	mtx_lock(&psp->ps_mtx);
478 }
479 
480 /*
481  * Send an interrupt to process.
482  *
483  * Stack is set up to allow sigcode stored
484  * in u. to call routine, followed by kcall
485  * to sigreturn routine below.  After sigreturn
486  * resets the signal mask, the stack, and the
487  * frame pointer, it returns to the user
488  * specified pc, psl.
489  */
490 static void
491 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
492 {
493 	struct thread *td = curthread;
494 	struct proc *p = td->td_proc;
495 	struct sigacts *psp;
496 	struct trapframe *regs;
497 	struct l_sigframe *fp, frame;
498 	l_sigset_t lmask;
499 	int sig, code;
500 	int oonstack;
501 
502 	PROC_LOCK_ASSERT(p, MA_OWNED);
503 	psp = p->p_sigacts;
504 	sig = ksi->ksi_signo;
505 	code = ksi->ksi_code;
506 	mtx_assert(&psp->ps_mtx, MA_OWNED);
507 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
508 		/* Signal handler installed with SA_SIGINFO. */
509 		linux_rt_sendsig(catcher, ksi, mask);
510 		return;
511 	}
512 	regs = td->td_frame;
513 	oonstack = sigonstack(regs->tf_esp);
514 
515 	/* Allocate space for the signal handler context. */
516 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
517 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
518 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
519 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
520 	} else
521 		fp = (struct l_sigframe *)regs->tf_esp - 1;
522 	mtx_unlock(&psp->ps_mtx);
523 	PROC_UNLOCK(p);
524 
525 	/* Build the argument list for the signal handler. */
526 	sig = bsd_to_linux_signal(sig);
527 
528 	bzero(&frame, sizeof(frame));
529 
530 	frame.sf_handler = catcher;
531 	frame.sf_sig = sig;
532 
533 	bsd_to_linux_sigset(mask, &lmask);
534 
535 	/* Build the signal context to be used by sigreturn. */
536 	frame.sf_sc.sc_mask   = lmask.__mask;
537 	frame.sf_sc.sc_gs     = rgs();
538 	frame.sf_sc.sc_fs     = regs->tf_fs;
539 	frame.sf_sc.sc_es     = regs->tf_es;
540 	frame.sf_sc.sc_ds     = regs->tf_ds;
541 	frame.sf_sc.sc_edi    = regs->tf_edi;
542 	frame.sf_sc.sc_esi    = regs->tf_esi;
543 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
544 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
545 	frame.sf_sc.sc_esp    = regs->tf_esp;
546 	frame.sf_sc.sc_edx    = regs->tf_edx;
547 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
548 	frame.sf_sc.sc_eax    = regs->tf_eax;
549 	frame.sf_sc.sc_eip    = regs->tf_eip;
550 	frame.sf_sc.sc_cs     = regs->tf_cs;
551 	frame.sf_sc.sc_eflags = regs->tf_eflags;
552 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
553 	frame.sf_sc.sc_ss     = regs->tf_ss;
554 	frame.sf_sc.sc_err    = regs->tf_err;
555 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
556 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
557 
558 	frame.sf_extramask[0] = lmask.__mask;
559 
560 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
561 		/*
562 		 * Process has trashed its stack; give it an illegal
563 		 * instruction to halt it in its tracks.
564 		 */
565 		PROC_LOCK(p);
566 		sigexit(td, SIGILL);
567 	}
568 
569 	/* Build context to run handler in. */
570 	regs->tf_esp = (int)fp;
571 	regs->tf_eip = linux_sigcode;
572 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
573 	regs->tf_cs = _ucodesel;
574 	regs->tf_ds = _udatasel;
575 	regs->tf_es = _udatasel;
576 	regs->tf_fs = _udatasel;
577 	regs->tf_ss = _udatasel;
578 	PROC_LOCK(p);
579 	mtx_lock(&psp->ps_mtx);
580 }
581 
582 /*
583  * System call to cleanup state after a signal
584  * has been taken.  Reset signal mask and
585  * stack state from context left by sendsig (above).
586  * Return to previous pc and psl as specified by
587  * context left by sendsig. Check carefully to
588  * make sure that the user has not modified the
589  * psl to gain improper privileges or to cause
590  * a machine fault.
591  */
592 int
593 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
594 {
595 	struct l_sigframe frame;
596 	struct trapframe *regs;
597 	l_sigset_t lmask;
598 	sigset_t bmask;
599 	int eflags;
600 	ksiginfo_t ksi;
601 
602 	regs = td->td_frame;
603 
604 	/*
605 	 * The trampoline code hands us the sigframe.
606 	 * It is unsafe to keep track of it ourselves, in the event that a
607 	 * program jumps out of a signal handler.
608 	 */
609 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
610 		return (EFAULT);
611 
612 	/* Check for security violations. */
613 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
614 	eflags = frame.sf_sc.sc_eflags;
615 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
616 		return (EINVAL);
617 
618 	/*
619 	 * Don't allow users to load a valid privileged %cs.  Let the
620 	 * hardware check for invalid selectors, excess privilege in
621 	 * other selectors, invalid %eip's and invalid %esp's.
622 	 */
623 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
624 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
625 		ksiginfo_init_trap(&ksi);
626 		ksi.ksi_signo = SIGBUS;
627 		ksi.ksi_code = BUS_OBJERR;
628 		ksi.ksi_trapno = T_PROTFLT;
629 		ksi.ksi_addr = (void *)regs->tf_eip;
630 		trapsignal(td, &ksi);
631 		return (EINVAL);
632 	}
633 
634 	lmask.__mask = frame.sf_sc.sc_mask;
635 	linux_to_bsd_sigset(&lmask, &bmask);
636 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
637 
638 	/* Restore signal context. */
639 	/* %gs was restored by the trampoline. */
640 	regs->tf_fs     = frame.sf_sc.sc_fs;
641 	regs->tf_es     = frame.sf_sc.sc_es;
642 	regs->tf_ds     = frame.sf_sc.sc_ds;
643 	regs->tf_edi    = frame.sf_sc.sc_edi;
644 	regs->tf_esi    = frame.sf_sc.sc_esi;
645 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
646 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
647 	regs->tf_edx    = frame.sf_sc.sc_edx;
648 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
649 	regs->tf_eax    = frame.sf_sc.sc_eax;
650 	regs->tf_eip    = frame.sf_sc.sc_eip;
651 	regs->tf_cs     = frame.sf_sc.sc_cs;
652 	regs->tf_eflags = eflags;
653 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
654 	regs->tf_ss     = frame.sf_sc.sc_ss;
655 
656 	return (EJUSTRETURN);
657 }
658 
659 /*
660  * System call to cleanup state after a signal
661  * has been taken.  Reset signal mask and
662  * stack state from context left by rt_sendsig (above).
663  * Return to previous pc and psl as specified by
664  * context left by sendsig. Check carefully to
665  * make sure that the user has not modified the
666  * psl to gain improper privileges or to cause
667  * a machine fault.
668  */
669 int
670 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
671 {
672 	struct l_ucontext uc;
673 	struct l_sigcontext *context;
674 	sigset_t bmask;
675 	l_stack_t *lss;
676 	stack_t ss;
677 	struct trapframe *regs;
678 	int eflags;
679 	ksiginfo_t ksi;
680 
681 	regs = td->td_frame;
682 
683 	/*
684 	 * The trampoline code hands us the ucontext.
685 	 * It is unsafe to keep track of it ourselves, in the event that a
686 	 * program jumps out of a signal handler.
687 	 */
688 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
689 		return (EFAULT);
690 
691 	context = &uc.uc_mcontext;
692 
693 	/* Check for security violations. */
694 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
695 	eflags = context->sc_eflags;
696 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
697 		return (EINVAL);
698 
699 	/*
700 	 * Don't allow users to load a valid privileged %cs.  Let the
701 	 * hardware check for invalid selectors, excess privilege in
702 	 * other selectors, invalid %eip's and invalid %esp's.
703 	 */
704 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
705 	if (!CS_SECURE(context->sc_cs)) {
706 		ksiginfo_init_trap(&ksi);
707 		ksi.ksi_signo = SIGBUS;
708 		ksi.ksi_code = BUS_OBJERR;
709 		ksi.ksi_trapno = T_PROTFLT;
710 		ksi.ksi_addr = (void *)regs->tf_eip;
711 		trapsignal(td, &ksi);
712 		return (EINVAL);
713 	}
714 
715 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
716 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
717 
718 	/* Restore signal context. */
719 	/* %gs was restored by the trampoline. */
720 	regs->tf_fs     = context->sc_fs;
721 	regs->tf_es     = context->sc_es;
722 	regs->tf_ds     = context->sc_ds;
723 	regs->tf_edi    = context->sc_edi;
724 	regs->tf_esi    = context->sc_esi;
725 	regs->tf_ebp    = context->sc_ebp;
726 	regs->tf_ebx    = context->sc_ebx;
727 	regs->tf_edx    = context->sc_edx;
728 	regs->tf_ecx    = context->sc_ecx;
729 	regs->tf_eax    = context->sc_eax;
730 	regs->tf_eip    = context->sc_eip;
731 	regs->tf_cs     = context->sc_cs;
732 	regs->tf_eflags = eflags;
733 	regs->tf_esp    = context->sc_esp_at_signal;
734 	regs->tf_ss     = context->sc_ss;
735 
736 	/* Call sigaltstack & ignore results. */
737 	lss = &uc.uc_stack;
738 	ss.ss_sp = lss->ss_sp;
739 	ss.ss_size = lss->ss_size;
740 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
741 
742 	(void)kern_sigaltstack(td, &ss, NULL);
743 
744 	return (EJUSTRETURN);
745 }
746 
747 static int
748 linux_fetch_syscall_args(struct thread *td)
749 {
750 	struct proc *p;
751 	struct trapframe *frame;
752 	struct syscall_args *sa;
753 
754 	p = td->td_proc;
755 	frame = td->td_frame;
756 	sa = &td->td_sa;
757 
758 	sa->code = frame->tf_eax;
759 	sa->args[0] = frame->tf_ebx;
760 	sa->args[1] = frame->tf_ecx;
761 	sa->args[2] = frame->tf_edx;
762 	sa->args[3] = frame->tf_esi;
763 	sa->args[4] = frame->tf_edi;
764 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
765 
766 	if (sa->code >= p->p_sysent->sv_size)
767 		/* nosys */
768 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
769 	else
770 		sa->callp = &p->p_sysent->sv_table[sa->code];
771 
772 	td->td_retval[0] = 0;
773 	td->td_retval[1] = frame->tf_edx;
774 
775 	return (0);
776 }
777 
778 static void
779 linux_set_syscall_retval(struct thread *td, int error)
780 {
781 	struct trapframe *frame = td->td_frame;
782 
783 	cpu_set_syscall_retval(td, error);
784 
785 	if (__predict_false(error != 0)) {
786 		if (error != ERESTART && error != EJUSTRETURN)
787 			frame->tf_eax = bsd_to_linux_errno(error);
788 	}
789 }
790 
791 /*
792  * exec_setregs may initialize some registers differently than Linux
793  * does, thus potentially confusing Linux binaries. If necessary, we
794  * override the exec_setregs default(s) here.
795  */
796 static void
797 linux_exec_setregs(struct thread *td, struct image_params *imgp,
798     uintptr_t stack)
799 {
800 	struct pcb *pcb = td->td_pcb;
801 
802 	exec_setregs(td, imgp, stack);
803 
804 	/* Linux sets %gs to 0, we default to _udatasel. */
805 	pcb->pcb_gs = 0;
806 	load_gs(0);
807 
808 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
809 }
810 
811 struct sysentvec linux_sysvec = {
812 	.sv_size	= LINUX_SYS_MAXSYSCALL,
813 	.sv_table	= linux_sysent,
814 	.sv_transtrap	= linux_translate_traps,
815 	.sv_fixup	= linux_fixup,
816 	.sv_sendsig	= linux_sendsig,
817 	.sv_sigcode	= &_binary_linux_locore_o_start,
818 	.sv_szsigcode	= &linux_szsigcode,
819 	.sv_name	= "Linux a.out",
820 	.sv_coredump	= NULL,
821 	.sv_imgact_try	= linux_exec_imgact_try,
822 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
823 	.sv_minuser	= VM_MIN_ADDRESS,
824 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
825 	.sv_usrstack	= LINUX_USRSTACK,
826 	.sv_psstrings	= PS_STRINGS,
827 	.sv_stackprot	= VM_PROT_ALL,
828 	.sv_copyout_strings = exec_copyout_strings,
829 	.sv_setregs	= linux_exec_setregs,
830 	.sv_fixlimit	= NULL,
831 	.sv_maxssiz	= NULL,
832 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32 |
833 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
834 	.sv_set_syscall_retval = linux_set_syscall_retval,
835 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
836 	.sv_syscallnames = NULL,
837 	.sv_schedtail	= linux_schedtail,
838 	.sv_thread_detach = linux_thread_detach,
839 	.sv_trap	= NULL,
840 	.sv_onexec	= linux_on_exec,
841 	.sv_onexit	= linux_on_exit,
842 	.sv_ontdexit	= linux_thread_dtor,
843 	.sv_setid_allowed = &linux_setid_allowed_query,
844 };
845 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
846 
847 struct sysentvec elf_linux_sysvec = {
848 	.sv_size	= LINUX_SYS_MAXSYSCALL,
849 	.sv_table	= linux_sysent,
850 	.sv_transtrap	= linux_translate_traps,
851 	.sv_fixup	= linux_fixup_elf,
852 	.sv_sendsig	= linux_sendsig,
853 	.sv_sigcode	= &_binary_linux_locore_o_start,
854 	.sv_szsigcode	= &linux_szsigcode,
855 	.sv_name	= "Linux ELF32",
856 	.sv_coredump	= elf32_coredump,
857 	.sv_elf_core_osabi = ELFOSABI_FREEBSD,
858 	.sv_elf_core_abi_vendor = FREEBSD_ABI_VENDOR,
859 	.sv_elf_core_prepare_notes = elf32_prepare_notes,
860 	.sv_imgact_try	= linux_exec_imgact_try,
861 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
862 	.sv_minuser	= VM_MIN_ADDRESS,
863 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
864 	.sv_usrstack	= LINUX_USRSTACK,
865 	.sv_psstrings	= LINUX_PS_STRINGS,
866 	.sv_stackprot	= VM_PROT_ALL,
867 	.sv_copyout_auxargs = linux_copyout_auxargs,
868 	.sv_copyout_strings = linux_copyout_strings,
869 	.sv_setregs	= linux_exec_setregs,
870 	.sv_fixlimit	= NULL,
871 	.sv_maxssiz	= NULL,
872 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP |
873 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
874 	.sv_set_syscall_retval = linux_set_syscall_retval,
875 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
876 	.sv_syscallnames = NULL,
877 	.sv_shared_page_base = LINUX_SHAREDPAGE,
878 	.sv_shared_page_len = PAGE_SIZE,
879 	.sv_schedtail	= linux_schedtail,
880 	.sv_thread_detach = linux_thread_detach,
881 	.sv_trap	= NULL,
882 	.sv_onexec	= linux_on_exec,
883 	.sv_onexit	= linux_on_exit,
884 	.sv_ontdexit	= linux_thread_dtor,
885 	.sv_setid_allowed = &linux_setid_allowed_query,
886 };
887 
888 static void
889 linux_vdso_install(void *param)
890 {
891 
892 	linux_szsigcode = (&_binary_linux_locore_o_end -
893 	    &_binary_linux_locore_o_start);
894 
895 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
896 		panic("Linux invalid vdso size\n");
897 
898 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
899 
900 	linux_shared_page_obj = __elfN(linux_shared_page_init)
901 	    (&linux_shared_page_mapping);
902 
903 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
904 
905 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
906 	    linux_szsigcode);
907 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
908 }
909 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
910     linux_vdso_install, NULL);
911 
912 static void
913 linux_vdso_deinstall(void *param)
914 {
915 
916 	__elfN(linux_shared_page_fini)(linux_shared_page_obj,
917 	    linux_shared_page_mapping);
918 }
919 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
920     linux_vdso_deinstall, NULL);
921 
922 static char GNU_ABI_VENDOR[] = "GNU";
923 static int GNULINUX_ABI_DESC = 0;
924 
925 static bool
926 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
927 {
928 	const Elf32_Word *desc;
929 	uintptr_t p;
930 
931 	p = (uintptr_t)(note + 1);
932 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
933 
934 	desc = (const Elf32_Word *)p;
935 	if (desc[0] != GNULINUX_ABI_DESC)
936 		return (false);
937 
938 	/*
939 	 * For Linux we encode osrel using the Linux convention of
940 	 * 	(version << 16) | (major << 8) | (minor)
941 	 * See macro in linux_mib.h
942 	 */
943 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
944 
945 	return (true);
946 }
947 
948 static Elf_Brandnote linux_brandnote = {
949 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
950 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
951 	.hdr.n_type	= 1,
952 	.vendor		= GNU_ABI_VENDOR,
953 	.flags		= BN_TRANSLATE_OSREL,
954 	.trans_osrel	= linux_trans_osrel
955 };
956 
957 static Elf32_Brandinfo linux_brand = {
958 	.brand		= ELFOSABI_LINUX,
959 	.machine	= EM_386,
960 	.compat_3_brand	= "Linux",
961 	.emul_path	= linux_emul_path,
962 	.interp_path	= "/lib/ld-linux.so.1",
963 	.sysvec		= &elf_linux_sysvec,
964 	.interp_newpath	= NULL,
965 	.brand_note	= &linux_brandnote,
966 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
967 };
968 
969 static Elf32_Brandinfo linux_glibc2brand = {
970 	.brand		= ELFOSABI_LINUX,
971 	.machine	= EM_386,
972 	.compat_3_brand	= "Linux",
973 	.emul_path	= linux_emul_path,
974 	.interp_path	= "/lib/ld-linux.so.2",
975 	.sysvec		= &elf_linux_sysvec,
976 	.interp_newpath	= NULL,
977 	.brand_note	= &linux_brandnote,
978 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
979 };
980 
981 static Elf32_Brandinfo linux_muslbrand = {
982 	.brand		= ELFOSABI_LINUX,
983 	.machine	= EM_386,
984 	.compat_3_brand	= "Linux",
985 	.emul_path	= linux_emul_path,
986 	.interp_path	= "/lib/ld-musl-i386.so.1",
987 	.sysvec		= &elf_linux_sysvec,
988 	.interp_newpath	= NULL,
989 	.brand_note	= &linux_brandnote,
990 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
991 };
992 
993 Elf32_Brandinfo *linux_brandlist[] = {
994 	&linux_brand,
995 	&linux_glibc2brand,
996 	&linux_muslbrand,
997 	NULL
998 };
999 
1000 static int
1001 linux_elf_modevent(module_t mod, int type, void *data)
1002 {
1003 	Elf32_Brandinfo **brandinfo;
1004 	int error;
1005 	struct linux_ioctl_handler **lihp;
1006 
1007 	error = 0;
1008 
1009 	switch(type) {
1010 	case MOD_LOAD:
1011 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1012 		     ++brandinfo)
1013 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1014 				error = EINVAL;
1015 		if (error == 0) {
1016 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1017 				linux_ioctl_register_handler(*lihp);
1018 			LIST_INIT(&futex_list);
1019 			mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1020 			linux_dev_shm_create();
1021 			linux_osd_jail_register();
1022 			stclohz = (stathz ? stathz : hz);
1023 			if (bootverbose)
1024 				printf("Linux ELF exec handler installed\n");
1025 		} else
1026 			printf("cannot insert Linux ELF brand handler\n");
1027 		break;
1028 	case MOD_UNLOAD:
1029 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1030 		     ++brandinfo)
1031 			if (elf32_brand_inuse(*brandinfo))
1032 				error = EBUSY;
1033 		if (error == 0) {
1034 			for (brandinfo = &linux_brandlist[0];
1035 			     *brandinfo != NULL; ++brandinfo)
1036 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1037 					error = EINVAL;
1038 		}
1039 		if (error == 0) {
1040 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1041 				linux_ioctl_unregister_handler(*lihp);
1042 			mtx_destroy(&futex_mtx);
1043 			linux_dev_shm_destroy();
1044 			linux_osd_jail_deregister();
1045 			if (bootverbose)
1046 				printf("Linux ELF exec handler removed\n");
1047 		} else
1048 			printf("Could not deinstall ELF interpreter entry\n");
1049 		break;
1050 	default:
1051 		return (EOPNOTSUPP);
1052 	}
1053 	return (error);
1054 }
1055 
1056 static moduledata_t linux_elf_mod = {
1057 	"linuxelf",
1058 	linux_elf_modevent,
1059 	0
1060 };
1061 
1062 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1063 FEATURE(linux, "Linux 32bit support");
1064