xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision 324cdd9320f58837c2fbaa7f6ceb9ea5c33d5b2a)
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 #include <sys/eventhandler.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_extern.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_param.h>
60 
61 #include <machine/cpu.h>
62 #include <machine/cputypes.h>
63 #include <machine/md_var.h>
64 #include <machine/pcb.h>
65 #include <machine/trap.h>
66 
67 #include <i386/linux/linux.h>
68 #include <i386/linux/linux_proto.h>
69 #include <compat/linux/linux_emul.h>
70 #include <compat/linux/linux_ioctl.h>
71 #include <compat/linux/linux_mib.h>
72 #include <compat/linux/linux_misc.h>
73 #include <compat/linux/linux_signal.h>
74 #include <compat/linux/linux_util.h>
75 #include <compat/linux/linux_vdso.h>
76 
77 MODULE_VERSION(linux, 1);
78 
79 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
80 
81 static int linux_szsigcode;
82 static vm_object_t linux_shared_page_obj;
83 static char *linux_shared_page_mapping;
84 extern char _binary_linux_locore_o_start;
85 extern char _binary_linux_locore_o_end;
86 
87 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
88 
89 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
90 
91 static int	linux_fixup(uintptr_t *stack_base,
92 		    struct image_params *iparams);
93 static int	linux_fixup_elf(uintptr_t *stack_base,
94 		    struct image_params *iparams);
95 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
96 static void	linux_exec_setregs(struct thread *td,
97 		    struct image_params *imgp, uintptr_t stack);
98 static int	linux_copyout_strings(struct image_params *imgp,
99 		    uintptr_t *stack_base);
100 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
101 static void	linux_vdso_install(void *param);
102 static void	linux_vdso_deinstall(void *param);
103 
104 static int linux_szplatform;
105 const char *linux_kplatform;
106 
107 static eventhandler_tag linux_exit_tag;
108 static eventhandler_tag linux_exec_tag;
109 static eventhandler_tag linux_thread_dtor_tag;
110 
111 #define LINUX_T_UNKNOWN  255
112 static int _bsd_to_linux_trapcode[] = {
113 	LINUX_T_UNKNOWN,	/* 0 */
114 	6,			/* 1  T_PRIVINFLT */
115 	LINUX_T_UNKNOWN,	/* 2 */
116 	3,			/* 3  T_BPTFLT */
117 	LINUX_T_UNKNOWN,	/* 4 */
118 	LINUX_T_UNKNOWN,	/* 5 */
119 	16,			/* 6  T_ARITHTRAP */
120 	254,			/* 7  T_ASTFLT */
121 	LINUX_T_UNKNOWN,	/* 8 */
122 	13,			/* 9  T_PROTFLT */
123 	1,			/* 10 T_TRCTRAP */
124 	LINUX_T_UNKNOWN,	/* 11 */
125 	14,			/* 12 T_PAGEFLT */
126 	LINUX_T_UNKNOWN,	/* 13 */
127 	17,			/* 14 T_ALIGNFLT */
128 	LINUX_T_UNKNOWN,	/* 15 */
129 	LINUX_T_UNKNOWN,	/* 16 */
130 	LINUX_T_UNKNOWN,	/* 17 */
131 	0,			/* 18 T_DIVIDE */
132 	2,			/* 19 T_NMI */
133 	4,			/* 20 T_OFLOW */
134 	5,			/* 21 T_BOUND */
135 	7,			/* 22 T_DNA */
136 	8,			/* 23 T_DOUBLEFLT */
137 	9,			/* 24 T_FPOPFLT */
138 	10,			/* 25 T_TSSFLT */
139 	11,			/* 26 T_SEGNPFLT */
140 	12,			/* 27 T_STKFLT */
141 	18,			/* 28 T_MCHK */
142 	19,			/* 29 T_XMMFLT */
143 	15			/* 30 T_RESERVED */
144 };
145 #define bsd_to_linux_trapcode(code) \
146     ((code)<nitems(_bsd_to_linux_trapcode)? \
147      _bsd_to_linux_trapcode[(code)]: \
148      LINUX_T_UNKNOWN)
149 
150 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
151 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
152 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
153 
154 /*
155  * If FreeBSD & Linux have a difference of opinion about what a trap
156  * means, deal with it here.
157  *
158  * MPSAFE
159  */
160 static int
161 linux_translate_traps(int signal, int trap_code)
162 {
163 	if (signal != SIGBUS)
164 		return (signal);
165 	switch (trap_code) {
166 	case T_PROTFLT:
167 	case T_TSSFLT:
168 	case T_DOUBLEFLT:
169 	case T_PAGEFLT:
170 		return (SIGSEGV);
171 	default:
172 		return (signal);
173 	}
174 }
175 
176 static int
177 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
178 {
179 	register_t *base, *argv, *envp;
180 
181 	base = (register_t *)*stack_base;
182 	argv = base;
183 	envp = base + (imgp->args->argc + 1);
184 	base--;
185 	suword(base, (intptr_t)envp);
186 	base--;
187 	suword(base, (intptr_t)argv);
188 	base--;
189 	suword(base, imgp->args->argc);
190 	*stack_base = (uintptr_t)base;
191 	return (0);
192 }
193 
194 static int
195 linux_copyout_auxargs(struct image_params *imgp, uintptr_t *base)
196 {
197 	struct proc *p;
198 	Elf32_Auxargs *args;
199 	Elf32_Auxinfo *argarray, *pos;
200 	Elf32_Addr *uplatform;
201 	struct ps_strings *arginfo;
202 	u_long auxlen;
203 	int error, issetugid;
204 
205 	p = imgp->proc;
206 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
207 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
208 	uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
209 	args = (Elf32_Auxargs *)imgp->auxargs;
210 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
211 	    M_WAITOK | M_ZERO);
212 
213 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
214 	    imgp->proc->p_sysent->sv_shared_page_base);
215 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
216 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
217 
218 	/*
219 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
220 	 * as it has appeared in the 2.4.0-rc7 first time.
221 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
222 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
223 	 * is not present.
224 	 * Also see linux_times() implementation.
225 	 */
226 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
227 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
228 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
229 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
230 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
231 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
232 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
233 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
234 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
235 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
236 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
237 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
238 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
239 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
240 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
241 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
242 	if (imgp->execpathp != 0)
243 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
244 	if (args->execfd != -1)
245 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
246 	AUXARGS_ENTRY(pos, AT_NULL, 0);
247 
248 	free(imgp->auxargs, M_TEMP);
249 	imgp->auxargs = NULL;
250 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
251 
252 	auxlen = sizeof(*argarray) * (pos - argarray);
253 	*base -= auxlen;
254 	error = copyout(argarray, (void *)*base, auxlen);
255 	free(argarray, M_TEMP);
256 	return (error);
257 }
258 
259 static int
260 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
261 {
262 	register_t *base;
263 
264 	base = (register_t *)*stack_base;
265 	base--;
266 	if (suword(base, (register_t)imgp->args->argc) == -1)
267 		return (EFAULT);
268 	*stack_base = (uintptr_t)base;
269 	return (0);
270 }
271 
272 /*
273  * Copied from kern/kern_exec.c
274  */
275 static int
276 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
277 {
278 	int argc, envc, error;
279 	char **vectp;
280 	char *stringp;
281 	uintptr_t destp, ustringp;
282 	struct ps_strings *arginfo;
283 	char canary[LINUX_AT_RANDOM_LEN];
284 	size_t execpath_len;
285 	struct proc *p;
286 
287 	/* Calculate string base and vector table pointers. */
288 	p = imgp->proc;
289 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
290 		execpath_len = strlen(imgp->execpath) + 1;
291 	else
292 		execpath_len = 0;
293 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
294 	destp = (uintptr_t)arginfo;
295 
296 	/* Install LINUX_PLATFORM. */
297 	destp -= linux_szplatform;
298 	destp = rounddown2(destp, sizeof(void *));
299 	error = copyout(linux_kplatform, (void *)destp, linux_szplatform);
300 	if (error != 0)
301 		return (error);
302 
303 	if (execpath_len != 0) {
304 		destp -= execpath_len;
305 		destp = rounddown2(destp, sizeof(void *));
306 		imgp->execpathp = destp;
307 		error = copyout(imgp->execpath, (void *)destp, execpath_len);
308 		if (error != 0)
309 			return (error);
310 	}
311 
312 	/* Prepare the canary for SSP. */
313 	arc4rand(canary, sizeof(canary), 0);
314 	destp -= roundup(sizeof(canary), sizeof(void *));
315 	imgp->canary = destp;
316 	error = copyout(canary, (void *)destp, sizeof(canary));
317 	if (error != 0)
318 		return (error);
319 
320 	/* Allocate room for the argument and environment strings. */
321 	destp -= ARG_MAX - imgp->args->stringspace;
322 	destp = rounddown2(destp, sizeof(void *));
323 	ustringp = destp;
324 
325 	if (imgp->auxargs) {
326 		error = imgp->sysent->sv_copyout_auxargs(imgp, &destp);
327 		if (error != 0)
328 			return (error);
329 	}
330 
331 	vectp = (char **)destp;
332 
333 	/*
334 	 * Allocate room for the argv[] and env vectors including the
335 	 * terminating NULL pointers.
336 	 */
337 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
338 
339 	/* vectp also becomes our initial stack base. */
340 	*stack_base = (uintptr_t)vectp;
341 
342 	stringp = imgp->args->begin_argv;
343 	argc = imgp->args->argc;
344 	envc = imgp->args->envc;
345 
346 	/* Copy out strings - arguments and environment. */
347 	error = copyout(stringp, (void *)ustringp,
348 	    ARG_MAX - imgp->args->stringspace);
349 	if (error != 0)
350 		return (error);
351 
352 	/* Fill in "ps_strings" struct for ps, w, etc. */
353 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
354 	    suword(&arginfo->ps_nargvstr, argc) != 0)
355 		return (EFAULT);
356 
357 	/* Fill in argument portion of vector table. */
358 	for (; argc > 0; --argc) {
359 		if (suword(vectp++, ustringp) != 0)
360 			return (EFAULT);
361 		while (*stringp++ != 0)
362 			ustringp++;
363 		ustringp++;
364 	}
365 
366 	/* A null vector table pointer separates the argp's from the envp's. */
367 	if (suword(vectp++, 0) != 0)
368 		return (EFAULT);
369 
370 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
371 	    suword(&arginfo->ps_nenvstr, envc) != 0)
372 		return (EFAULT);
373 
374 	/* Fill in environment portion of vector table. */
375 	for (; envc > 0; --envc) {
376 		if (suword(vectp++, ustringp) != 0)
377 			return (EFAULT);
378 		while (*stringp++ != 0)
379 			ustringp++;
380 		ustringp++;
381 	}
382 
383 	/* The end of the vector table is a null pointer. */
384 	if (suword(vectp, 0) != 0)
385 		return (EFAULT);
386 
387 	return (0);
388 }
389 
390 static void
391 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
392 {
393 	struct thread *td = curthread;
394 	struct proc *p = td->td_proc;
395 	struct sigacts *psp;
396 	struct trapframe *regs;
397 	struct l_rt_sigframe *fp, frame;
398 	int sig, code;
399 	int oonstack;
400 
401 	sig = ksi->ksi_signo;
402 	code = ksi->ksi_code;
403 	PROC_LOCK_ASSERT(p, MA_OWNED);
404 	psp = p->p_sigacts;
405 	mtx_assert(&psp->ps_mtx, MA_OWNED);
406 	regs = td->td_frame;
407 	oonstack = sigonstack(regs->tf_esp);
408 
409 	/* Allocate space for the signal handler context. */
410 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
411 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
412 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
413 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
414 	} else
415 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
416 	mtx_unlock(&psp->ps_mtx);
417 
418 	/* Build the argument list for the signal handler. */
419 	sig = bsd_to_linux_signal(sig);
420 
421 	bzero(&frame, sizeof(frame));
422 
423 	frame.sf_handler = catcher;
424 	frame.sf_sig = sig;
425 	frame.sf_siginfo = &fp->sf_si;
426 	frame.sf_ucontext = &fp->sf_sc;
427 
428 	/* Fill in POSIX parts. */
429 	ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
430 
431 	/* Build the signal context to be used by sigreturn. */
432 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
433 	frame.sf_sc.uc_link = NULL;		/* XXX ??? */
434 
435 	frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
436 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
437 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
438 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
439 	PROC_UNLOCK(p);
440 
441 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
442 
443 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
444 	frame.sf_sc.uc_mcontext.sc_gs     = rgs();
445 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
446 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
447 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
448 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_edi;
449 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_esi;
450 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_ebp;
451 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_ebx;
452 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_esp;
453 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_edx;
454 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_ecx;
455 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_eax;
456 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_eip;
457 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
458 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
459 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
460 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
461 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
462 	frame.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
463 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
464 
465 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
466 		/*
467 		 * Process has trashed its stack; give it an illegal
468 		 * instruction to halt it in its tracks.
469 		 */
470 		PROC_LOCK(p);
471 		sigexit(td, SIGILL);
472 	}
473 
474 	/* Build context to run handler in. */
475 	regs->tf_esp = (int)fp;
476 	regs->tf_eip = linux_rt_sigcode;
477 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
478 	regs->tf_cs = _ucodesel;
479 	regs->tf_ds = _udatasel;
480 	regs->tf_es = _udatasel;
481 	regs->tf_fs = _udatasel;
482 	regs->tf_ss = _udatasel;
483 	PROC_LOCK(p);
484 	mtx_lock(&psp->ps_mtx);
485 }
486 
487 
488 /*
489  * Send an interrupt to process.
490  *
491  * Stack is set up to allow sigcode stored
492  * in u. to call routine, followed by kcall
493  * to sigreturn routine below.  After sigreturn
494  * resets the signal mask, the stack, and the
495  * frame pointer, it returns to the user
496  * specified pc, psl.
497  */
498 static void
499 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
500 {
501 	struct thread *td = curthread;
502 	struct proc *p = td->td_proc;
503 	struct sigacts *psp;
504 	struct trapframe *regs;
505 	struct l_sigframe *fp, frame;
506 	l_sigset_t lmask;
507 	int sig, code;
508 	int oonstack;
509 
510 	PROC_LOCK_ASSERT(p, MA_OWNED);
511 	psp = p->p_sigacts;
512 	sig = ksi->ksi_signo;
513 	code = ksi->ksi_code;
514 	mtx_assert(&psp->ps_mtx, MA_OWNED);
515 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
516 		/* Signal handler installed with SA_SIGINFO. */
517 		linux_rt_sendsig(catcher, ksi, mask);
518 		return;
519 	}
520 	regs = td->td_frame;
521 	oonstack = sigonstack(regs->tf_esp);
522 
523 	/* Allocate space for the signal handler context. */
524 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
525 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
526 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
527 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
528 	} else
529 		fp = (struct l_sigframe *)regs->tf_esp - 1;
530 	mtx_unlock(&psp->ps_mtx);
531 	PROC_UNLOCK(p);
532 
533 	/* Build the argument list for the signal handler. */
534 	sig = bsd_to_linux_signal(sig);
535 
536 	bzero(&frame, sizeof(frame));
537 
538 	frame.sf_handler = catcher;
539 	frame.sf_sig = sig;
540 
541 	bsd_to_linux_sigset(mask, &lmask);
542 
543 	/* Build the signal context to be used by sigreturn. */
544 	frame.sf_sc.sc_mask   = lmask.__mask;
545 	frame.sf_sc.sc_gs     = rgs();
546 	frame.sf_sc.sc_fs     = regs->tf_fs;
547 	frame.sf_sc.sc_es     = regs->tf_es;
548 	frame.sf_sc.sc_ds     = regs->tf_ds;
549 	frame.sf_sc.sc_edi    = regs->tf_edi;
550 	frame.sf_sc.sc_esi    = regs->tf_esi;
551 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
552 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
553 	frame.sf_sc.sc_esp    = regs->tf_esp;
554 	frame.sf_sc.sc_edx    = regs->tf_edx;
555 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
556 	frame.sf_sc.sc_eax    = regs->tf_eax;
557 	frame.sf_sc.sc_eip    = regs->tf_eip;
558 	frame.sf_sc.sc_cs     = regs->tf_cs;
559 	frame.sf_sc.sc_eflags = regs->tf_eflags;
560 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
561 	frame.sf_sc.sc_ss     = regs->tf_ss;
562 	frame.sf_sc.sc_err    = regs->tf_err;
563 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
564 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
565 
566 	frame.sf_extramask[0] = lmask.__mask;
567 
568 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
569 		/*
570 		 * Process has trashed its stack; give it an illegal
571 		 * instruction to halt it in its tracks.
572 		 */
573 		PROC_LOCK(p);
574 		sigexit(td, SIGILL);
575 	}
576 
577 	/* Build context to run handler in. */
578 	regs->tf_esp = (int)fp;
579 	regs->tf_eip = linux_sigcode;
580 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
581 	regs->tf_cs = _ucodesel;
582 	regs->tf_ds = _udatasel;
583 	regs->tf_es = _udatasel;
584 	regs->tf_fs = _udatasel;
585 	regs->tf_ss = _udatasel;
586 	PROC_LOCK(p);
587 	mtx_lock(&psp->ps_mtx);
588 }
589 
590 /*
591  * System call to cleanup state after a signal
592  * has been taken.  Reset signal mask and
593  * stack state from context left by sendsig (above).
594  * Return to previous pc and psl as specified by
595  * context left by sendsig. Check carefully to
596  * make sure that the user has not modified the
597  * psl to gain improper privileges or to cause
598  * a machine fault.
599  */
600 int
601 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
602 {
603 	struct l_sigframe frame;
604 	struct trapframe *regs;
605 	l_sigset_t lmask;
606 	sigset_t bmask;
607 	int eflags;
608 	ksiginfo_t ksi;
609 
610 	regs = td->td_frame;
611 
612 	/*
613 	 * The trampoline code hands us the sigframe.
614 	 * It is unsafe to keep track of it ourselves, in the event that a
615 	 * program jumps out of a signal handler.
616 	 */
617 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
618 		return (EFAULT);
619 
620 	/* Check for security violations. */
621 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
622 	eflags = frame.sf_sc.sc_eflags;
623 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
624 		return (EINVAL);
625 
626 	/*
627 	 * Don't allow users to load a valid privileged %cs.  Let the
628 	 * hardware check for invalid selectors, excess privilege in
629 	 * other selectors, invalid %eip's and invalid %esp's.
630 	 */
631 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
632 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
633 		ksiginfo_init_trap(&ksi);
634 		ksi.ksi_signo = SIGBUS;
635 		ksi.ksi_code = BUS_OBJERR;
636 		ksi.ksi_trapno = T_PROTFLT;
637 		ksi.ksi_addr = (void *)regs->tf_eip;
638 		trapsignal(td, &ksi);
639 		return (EINVAL);
640 	}
641 
642 	lmask.__mask = frame.sf_sc.sc_mask;
643 	linux_to_bsd_sigset(&lmask, &bmask);
644 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
645 
646 	/* Restore signal context. */
647 	/* %gs was restored by the trampoline. */
648 	regs->tf_fs     = frame.sf_sc.sc_fs;
649 	regs->tf_es     = frame.sf_sc.sc_es;
650 	regs->tf_ds     = frame.sf_sc.sc_ds;
651 	regs->tf_edi    = frame.sf_sc.sc_edi;
652 	regs->tf_esi    = frame.sf_sc.sc_esi;
653 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
654 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
655 	regs->tf_edx    = frame.sf_sc.sc_edx;
656 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
657 	regs->tf_eax    = frame.sf_sc.sc_eax;
658 	regs->tf_eip    = frame.sf_sc.sc_eip;
659 	regs->tf_cs     = frame.sf_sc.sc_cs;
660 	regs->tf_eflags = eflags;
661 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
662 	regs->tf_ss     = frame.sf_sc.sc_ss;
663 
664 	return (EJUSTRETURN);
665 }
666 
667 /*
668  * System call to cleanup state after a signal
669  * has been taken.  Reset signal mask and
670  * stack state from context left by rt_sendsig (above).
671  * Return to previous pc and psl as specified by
672  * context left by sendsig. Check carefully to
673  * make sure that the user has not modified the
674  * psl to gain improper privileges or to cause
675  * a machine fault.
676  */
677 int
678 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
679 {
680 	struct l_ucontext uc;
681 	struct l_sigcontext *context;
682 	sigset_t bmask;
683 	l_stack_t *lss;
684 	stack_t ss;
685 	struct trapframe *regs;
686 	int eflags;
687 	ksiginfo_t ksi;
688 
689 	regs = td->td_frame;
690 
691 	/*
692 	 * The trampoline code hands us the ucontext.
693 	 * It is unsafe to keep track of it ourselves, in the event that a
694 	 * program jumps out of a signal handler.
695 	 */
696 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
697 		return (EFAULT);
698 
699 	context = &uc.uc_mcontext;
700 
701 	/* Check for security violations. */
702 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
703 	eflags = context->sc_eflags;
704 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
705 		return (EINVAL);
706 
707 	/*
708 	 * Don't allow users to load a valid privileged %cs.  Let the
709 	 * hardware check for invalid selectors, excess privilege in
710 	 * other selectors, invalid %eip's and invalid %esp's.
711 	 */
712 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
713 	if (!CS_SECURE(context->sc_cs)) {
714 		ksiginfo_init_trap(&ksi);
715 		ksi.ksi_signo = SIGBUS;
716 		ksi.ksi_code = BUS_OBJERR;
717 		ksi.ksi_trapno = T_PROTFLT;
718 		ksi.ksi_addr = (void *)regs->tf_eip;
719 		trapsignal(td, &ksi);
720 		return (EINVAL);
721 	}
722 
723 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
724 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
725 
726 	/* Restore signal context. */
727 	/* %gs was restored by the trampoline. */
728 	regs->tf_fs     = context->sc_fs;
729 	regs->tf_es     = context->sc_es;
730 	regs->tf_ds     = context->sc_ds;
731 	regs->tf_edi    = context->sc_edi;
732 	regs->tf_esi    = context->sc_esi;
733 	regs->tf_ebp    = context->sc_ebp;
734 	regs->tf_ebx    = context->sc_ebx;
735 	regs->tf_edx    = context->sc_edx;
736 	regs->tf_ecx    = context->sc_ecx;
737 	regs->tf_eax    = context->sc_eax;
738 	regs->tf_eip    = context->sc_eip;
739 	regs->tf_cs     = context->sc_cs;
740 	regs->tf_eflags = eflags;
741 	regs->tf_esp    = context->sc_esp_at_signal;
742 	regs->tf_ss     = context->sc_ss;
743 
744 	/* Call sigaltstack & ignore results. */
745 	lss = &uc.uc_stack;
746 	ss.ss_sp = lss->ss_sp;
747 	ss.ss_size = lss->ss_size;
748 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
749 
750 	(void)kern_sigaltstack(td, &ss, NULL);
751 
752 	return (EJUSTRETURN);
753 }
754 
755 static int
756 linux_fetch_syscall_args(struct thread *td)
757 {
758 	struct proc *p;
759 	struct trapframe *frame;
760 	struct syscall_args *sa;
761 
762 	p = td->td_proc;
763 	frame = td->td_frame;
764 	sa = &td->td_sa;
765 
766 	sa->code = frame->tf_eax;
767 	sa->args[0] = frame->tf_ebx;
768 	sa->args[1] = frame->tf_ecx;
769 	sa->args[2] = frame->tf_edx;
770 	sa->args[3] = frame->tf_esi;
771 	sa->args[4] = frame->tf_edi;
772 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
773 
774 	if (sa->code >= p->p_sysent->sv_size)
775 		/* nosys */
776 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
777 	else
778 		sa->callp = &p->p_sysent->sv_table[sa->code];
779 	sa->narg = sa->callp->sy_narg;
780 
781 	td->td_retval[0] = 0;
782 	td->td_retval[1] = frame->tf_edx;
783 
784 	return (0);
785 }
786 
787 /*
788  * exec_setregs may initialize some registers differently than Linux
789  * does, thus potentially confusing Linux binaries. If necessary, we
790  * override the exec_setregs default(s) here.
791  */
792 static void
793 linux_exec_setregs(struct thread *td, struct image_params *imgp,
794     uintptr_t stack)
795 {
796 	struct pcb *pcb = td->td_pcb;
797 
798 	exec_setregs(td, imgp, stack);
799 
800 	/* Linux sets %gs to 0, we default to _udatasel. */
801 	pcb->pcb_gs = 0;
802 	load_gs(0);
803 
804 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
805 }
806 
807 static void
808 linux_get_machine(const char **dst)
809 {
810 
811 	switch (cpu_class) {
812 	case CPUCLASS_686:
813 		*dst = "i686";
814 		break;
815 	case CPUCLASS_586:
816 		*dst = "i586";
817 		break;
818 	case CPUCLASS_486:
819 		*dst = "i486";
820 		break;
821 	default:
822 		*dst = "i386";
823 	}
824 }
825 
826 struct sysentvec linux_sysvec = {
827 	.sv_size	= LINUX_SYS_MAXSYSCALL,
828 	.sv_table	= linux_sysent,
829 	.sv_errsize	= ELAST + 1,
830 	.sv_errtbl	= linux_errtbl,
831 	.sv_transtrap	= linux_translate_traps,
832 	.sv_fixup	= linux_fixup,
833 	.sv_sendsig	= linux_sendsig,
834 	.sv_sigcode	= &_binary_linux_locore_o_start,
835 	.sv_szsigcode	= &linux_szsigcode,
836 	.sv_name	= "Linux a.out",
837 	.sv_coredump	= NULL,
838 	.sv_imgact_try	= linux_exec_imgact_try,
839 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
840 	.sv_minuser	= VM_MIN_ADDRESS,
841 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
842 	.sv_usrstack	= LINUX_USRSTACK,
843 	.sv_psstrings	= PS_STRINGS,
844 	.sv_stackprot	= VM_PROT_ALL,
845 	.sv_copyout_strings = exec_copyout_strings,
846 	.sv_setregs	= linux_exec_setregs,
847 	.sv_fixlimit	= NULL,
848 	.sv_maxssiz	= NULL,
849 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32,
850 	.sv_set_syscall_retval = cpu_set_syscall_retval,
851 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
852 	.sv_syscallnames = NULL,
853 	.sv_shared_page_base = LINUX_SHAREDPAGE,
854 	.sv_shared_page_len = PAGE_SIZE,
855 	.sv_schedtail	= linux_schedtail,
856 	.sv_thread_detach = linux_thread_detach,
857 	.sv_trap	= NULL,
858 };
859 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
860 
861 struct sysentvec elf_linux_sysvec = {
862 	.sv_size	= LINUX_SYS_MAXSYSCALL,
863 	.sv_table	= linux_sysent,
864 	.sv_errsize	= ELAST + 1,
865 	.sv_errtbl	= linux_errtbl,
866 	.sv_transtrap	= linux_translate_traps,
867 	.sv_fixup	= linux_fixup_elf,
868 	.sv_sendsig	= linux_sendsig,
869 	.sv_sigcode	= &_binary_linux_locore_o_start,
870 	.sv_szsigcode	= &linux_szsigcode,
871 	.sv_name	= "Linux ELF",
872 	.sv_coredump	= elf32_coredump,
873 	.sv_imgact_try	= linux_exec_imgact_try,
874 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
875 	.sv_minuser	= VM_MIN_ADDRESS,
876 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
877 	.sv_usrstack	= LINUX_USRSTACK,
878 	.sv_psstrings	= LINUX_PS_STRINGS,
879 	.sv_stackprot	= VM_PROT_ALL,
880 	.sv_copyout_auxargs = linux_copyout_auxargs,
881 	.sv_copyout_strings = linux_copyout_strings,
882 	.sv_setregs	= linux_exec_setregs,
883 	.sv_fixlimit	= NULL,
884 	.sv_maxssiz	= NULL,
885 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP,
886 	.sv_set_syscall_retval = cpu_set_syscall_retval,
887 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
888 	.sv_syscallnames = NULL,
889 	.sv_shared_page_base = LINUX_SHAREDPAGE,
890 	.sv_shared_page_len = PAGE_SIZE,
891 	.sv_schedtail	= linux_schedtail,
892 	.sv_thread_detach = linux_thread_detach,
893 	.sv_trap	= NULL,
894 };
895 
896 static void
897 linux_vdso_install(void *param)
898 {
899 
900 	linux_szsigcode = (&_binary_linux_locore_o_end -
901 	    &_binary_linux_locore_o_start);
902 
903 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
904 		panic("Linux invalid vdso size\n");
905 
906 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
907 
908 	linux_shared_page_obj = __elfN(linux_shared_page_init)
909 	    (&linux_shared_page_mapping);
910 
911 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
912 
913 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
914 	    linux_szsigcode);
915 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
916 }
917 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
918     linux_vdso_install, NULL);
919 
920 static void
921 linux_vdso_deinstall(void *param)
922 {
923 
924 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
925 }
926 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
927     linux_vdso_deinstall, NULL);
928 
929 static char GNU_ABI_VENDOR[] = "GNU";
930 static int GNULINUX_ABI_DESC = 0;
931 
932 static bool
933 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
934 {
935 	const Elf32_Word *desc;
936 	uintptr_t p;
937 
938 	p = (uintptr_t)(note + 1);
939 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
940 
941 	desc = (const Elf32_Word *)p;
942 	if (desc[0] != GNULINUX_ABI_DESC)
943 		return (false);
944 
945 	/*
946 	 * For Linux we encode osrel using the Linux convention of
947 	 * 	(version << 16) | (major << 8) | (minor)
948 	 * See macro in linux_mib.h
949 	 */
950 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
951 
952 	return (true);
953 }
954 
955 static Elf_Brandnote linux_brandnote = {
956 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
957 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
958 	.hdr.n_type	= 1,
959 	.vendor		= GNU_ABI_VENDOR,
960 	.flags		= BN_TRANSLATE_OSREL,
961 	.trans_osrel	= linux_trans_osrel
962 };
963 
964 static Elf32_Brandinfo linux_brand = {
965 	.brand		= ELFOSABI_LINUX,
966 	.machine	= EM_386,
967 	.compat_3_brand	= "Linux",
968 	.emul_path	= "/compat/linux",
969 	.interp_path	= "/lib/ld-linux.so.1",
970 	.sysvec		= &elf_linux_sysvec,
971 	.interp_newpath	= NULL,
972 	.brand_note	= &linux_brandnote,
973 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
974 };
975 
976 static Elf32_Brandinfo linux_glibc2brand = {
977 	.brand		= ELFOSABI_LINUX,
978 	.machine	= EM_386,
979 	.compat_3_brand	= "Linux",
980 	.emul_path	= "/compat/linux",
981 	.interp_path	= "/lib/ld-linux.so.2",
982 	.sysvec		= &elf_linux_sysvec,
983 	.interp_newpath	= NULL,
984 	.brand_note	= &linux_brandnote,
985 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
986 };
987 
988 static Elf32_Brandinfo linux_muslbrand = {
989 	.brand		= ELFOSABI_LINUX,
990 	.machine	= EM_386,
991 	.compat_3_brand	= "Linux",
992 	.emul_path	= "/compat/linux",
993 	.interp_path	= "/lib/ld-musl-i386.so.1",
994 	.sysvec		= &elf_linux_sysvec,
995 	.interp_newpath	= NULL,
996 	.brand_note	= &linux_brandnote,
997 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
998 };
999 
1000 Elf32_Brandinfo *linux_brandlist[] = {
1001 	&linux_brand,
1002 	&linux_glibc2brand,
1003 	&linux_muslbrand,
1004 	NULL
1005 };
1006 
1007 static int
1008 linux_elf_modevent(module_t mod, int type, void *data)
1009 {
1010 	Elf32_Brandinfo **brandinfo;
1011 	int error;
1012 	struct linux_ioctl_handler **lihp;
1013 
1014 	error = 0;
1015 
1016 	switch(type) {
1017 	case MOD_LOAD:
1018 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1019 		     ++brandinfo)
1020 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1021 				error = EINVAL;
1022 		if (error == 0) {
1023 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1024 				linux_ioctl_register_handler(*lihp);
1025 			LIST_INIT(&futex_list);
1026 			mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1027 			linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit,
1028 			      NULL, 1000);
1029 			linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec,
1030 			      NULL, 1000);
1031 			linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor,
1032 			    linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY);
1033 			linux_get_machine(&linux_kplatform);
1034 			linux_szplatform = roundup(strlen(linux_kplatform) + 1,
1035 			    sizeof(char *));
1036 			linux_dev_shm_create();
1037 			linux_osd_jail_register();
1038 			stclohz = (stathz ? stathz : hz);
1039 			if (bootverbose)
1040 				printf("Linux ELF exec handler installed\n");
1041 		} else
1042 			printf("cannot insert Linux ELF brand handler\n");
1043 		break;
1044 	case MOD_UNLOAD:
1045 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1046 		     ++brandinfo)
1047 			if (elf32_brand_inuse(*brandinfo))
1048 				error = EBUSY;
1049 		if (error == 0) {
1050 			for (brandinfo = &linux_brandlist[0];
1051 			     *brandinfo != NULL; ++brandinfo)
1052 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1053 					error = EINVAL;
1054 		}
1055 		if (error == 0) {
1056 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1057 				linux_ioctl_unregister_handler(*lihp);
1058 			mtx_destroy(&futex_mtx);
1059 			EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
1060 			EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
1061 			EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag);
1062 			linux_dev_shm_destroy();
1063 			linux_osd_jail_deregister();
1064 			if (bootverbose)
1065 				printf("Linux ELF exec handler removed\n");
1066 		} else
1067 			printf("Could not deinstall ELF interpreter entry\n");
1068 		break;
1069 	default:
1070 		return (EOPNOTSUPP);
1071 	}
1072 	return (error);
1073 }
1074 
1075 static moduledata_t linux_elf_mod = {
1076 	"linuxelf",
1077 	linux_elf_modevent,
1078 	0
1079 };
1080 
1081 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1082 FEATURE(linux, "Linux 32bit support");
1083