xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision 97cb52fa9aefd90fad38790fded50905aeeb9b9e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/exec.h>
37 #include <sys/fcntl.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_aout.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/module.h>
45 #include <sys/mutex.h>
46 #include <sys/proc.h>
47 #include <sys/signalvar.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysctl.h>
50 #include <sys/sysent.h>
51 #include <sys/sysproto.h>
52 #include <sys/vnode.h>
53 #include <sys/eventhandler.h>
54 
55 #include <vm/vm.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_extern.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_object.h>
60 #include <vm/vm_page.h>
61 #include <vm/vm_param.h>
62 
63 #include <machine/cpu.h>
64 #include <machine/cputypes.h>
65 #include <machine/md_var.h>
66 #include <machine/pcb.h>
67 
68 #include <i386/linux/linux.h>
69 #include <i386/linux/linux_proto.h>
70 #include <compat/linux/linux_emul.h>
71 #include <compat/linux/linux_futex.h>
72 #include <compat/linux/linux_ioctl.h>
73 #include <compat/linux/linux_mib.h>
74 #include <compat/linux/linux_misc.h>
75 #include <compat/linux/linux_signal.h>
76 #include <compat/linux/linux_util.h>
77 #include <compat/linux/linux_vdso.h>
78 
79 MODULE_VERSION(linux, 1);
80 
81 #if BYTE_ORDER == LITTLE_ENDIAN
82 #define SHELLMAGIC      0x2123 /* #! */
83 #else
84 #define SHELLMAGIC      0x2321
85 #endif
86 
87 #if defined(DEBUG)
88 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
89             CTLTYPE_STRING | CTLFLAG_RW,
90             0, 0, linux_sysctl_debug, "A",
91             "Linux debugging control");
92 #endif
93 
94 /*
95  * Allow the sendsig functions to use the ldebug() facility
96  * even though they are not syscalls themselves. Map them
97  * to syscall 0. This is slightly less bogus than using
98  * ldebug(sigreturn).
99  */
100 #define	LINUX_SYS_linux_rt_sendsig	0
101 #define	LINUX_SYS_linux_sendsig		0
102 
103 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
104 
105 static int linux_szsigcode;
106 static vm_object_t linux_shared_page_obj;
107 static char *linux_shared_page_mapping;
108 extern char _binary_linux_locore_o_start;
109 extern char _binary_linux_locore_o_end;
110 
111 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
112 
113 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
114 
115 static int	linux_fixup(register_t **stack_base,
116 		    struct image_params *iparams);
117 static int	elf_linux_fixup(register_t **stack_base,
118 		    struct image_params *iparams);
119 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
120 static void	exec_linux_setregs(struct thread *td,
121 		    struct image_params *imgp, u_long stack);
122 static register_t *linux_copyout_strings(struct image_params *imgp);
123 static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
124 static void	linux_vdso_install(void *param);
125 static void	linux_vdso_deinstall(void *param);
126 
127 static int linux_szplatform;
128 const char *linux_kplatform;
129 
130 static eventhandler_tag linux_exit_tag;
131 static eventhandler_tag linux_exec_tag;
132 static eventhandler_tag linux_thread_dtor_tag;
133 
134 /*
135  * Linux syscalls return negative errno's, we do positive and map them
136  * Reference:
137  *   FreeBSD: src/sys/sys/errno.h
138  *   Linux:   linux-2.6.17.8/include/asm-generic/errno-base.h
139  *            linux-2.6.17.8/include/asm-generic/errno.h
140  */
141 static int bsd_to_linux_errno[ELAST + 1] = {
142 	-0,  -1,  -2,  -3,  -4,  -5,  -6,  -7,  -8,  -9,
143 	-10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
144 	-20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
145 	-30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
146 	-90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
147 	-100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
148 	-110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
149 	-116, -66,  -6,  -6,  -6,  -6,  -6, -37, -38,  -9,
150 	  -6,  -6, -43, -42, -75,-125, -84, -61, -16, -74,
151 	 -72, -67, -71
152 };
153 
154 #define LINUX_T_UNKNOWN  255
155 static int _bsd_to_linux_trapcode[] = {
156 	LINUX_T_UNKNOWN,	/* 0 */
157 	6,			/* 1  T_PRIVINFLT */
158 	LINUX_T_UNKNOWN,	/* 2 */
159 	3,			/* 3  T_BPTFLT */
160 	LINUX_T_UNKNOWN,	/* 4 */
161 	LINUX_T_UNKNOWN,	/* 5 */
162 	16,			/* 6  T_ARITHTRAP */
163 	254,			/* 7  T_ASTFLT */
164 	LINUX_T_UNKNOWN,	/* 8 */
165 	13,			/* 9  T_PROTFLT */
166 	1,			/* 10 T_TRCTRAP */
167 	LINUX_T_UNKNOWN,	/* 11 */
168 	14,			/* 12 T_PAGEFLT */
169 	LINUX_T_UNKNOWN,	/* 13 */
170 	17,			/* 14 T_ALIGNFLT */
171 	LINUX_T_UNKNOWN,	/* 15 */
172 	LINUX_T_UNKNOWN,	/* 16 */
173 	LINUX_T_UNKNOWN,	/* 17 */
174 	0,			/* 18 T_DIVIDE */
175 	2,			/* 19 T_NMI */
176 	4,			/* 20 T_OFLOW */
177 	5,			/* 21 T_BOUND */
178 	7,			/* 22 T_DNA */
179 	8,			/* 23 T_DOUBLEFLT */
180 	9,			/* 24 T_FPOPFLT */
181 	10,			/* 25 T_TSSFLT */
182 	11,			/* 26 T_SEGNPFLT */
183 	12,			/* 27 T_STKFLT */
184 	18,			/* 28 T_MCHK */
185 	19,			/* 29 T_XMMFLT */
186 	15			/* 30 T_RESERVED */
187 };
188 #define bsd_to_linux_trapcode(code) \
189     ((code)<nitems(_bsd_to_linux_trapcode)? \
190      _bsd_to_linux_trapcode[(code)]: \
191      LINUX_T_UNKNOWN)
192 
193 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
194 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
195 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
196 
197 /*
198  * If FreeBSD & Linux have a difference of opinion about what a trap
199  * means, deal with it here.
200  *
201  * MPSAFE
202  */
203 static int
204 translate_traps(int signal, int trap_code)
205 {
206 	if (signal != SIGBUS)
207 		return (signal);
208 	switch (trap_code) {
209 	case T_PROTFLT:
210 	case T_TSSFLT:
211 	case T_DOUBLEFLT:
212 	case T_PAGEFLT:
213 		return (SIGSEGV);
214 	default:
215 		return (signal);
216 	}
217 }
218 
219 static int
220 linux_fixup(register_t **stack_base, struct image_params *imgp)
221 {
222 	register_t *argv, *envp;
223 
224 	argv = *stack_base;
225 	envp = *stack_base + (imgp->args->argc + 1);
226 	(*stack_base)--;
227 	suword(*stack_base, (intptr_t)(void *)envp);
228 	(*stack_base)--;
229 	suword(*stack_base, (intptr_t)(void *)argv);
230 	(*stack_base)--;
231 	suword(*stack_base, imgp->args->argc);
232 	return (0);
233 }
234 
235 static int
236 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
237 {
238 	struct proc *p;
239 	Elf32_Auxargs *args;
240 	Elf32_Addr *uplatform;
241 	struct ps_strings *arginfo;
242 	register_t *pos;
243 	int issetugid;
244 
245 	KASSERT(curthread->td_proc == imgp->proc,
246 	    ("unsafe elf_linux_fixup(), should be curproc"));
247 
248 	p = imgp->proc;
249 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
250 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
251 	uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
252 	args = (Elf32_Auxargs *)imgp->auxargs;
253 	pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
254 
255 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
256 	    imgp->proc->p_sysent->sv_shared_page_base);
257 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
258 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
259 
260 	/*
261 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
262 	 * as it has appeared in the 2.4.0-rc7 first time.
263 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
264 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
265 	 * is not present.
266 	 * Also see linux_times() implementation.
267 	 */
268 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
269 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
270 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
271 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
272 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
273 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
274 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
275 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
276 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
277 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
278 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
279 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
280 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
281 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
282 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
283 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
284 	if (imgp->execpathp != 0)
285 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
286 	if (args->execfd != -1)
287 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
288 	AUXARGS_ENTRY(pos, AT_NULL, 0);
289 
290 	free(imgp->auxargs, M_TEMP);
291 	imgp->auxargs = NULL;
292 
293 	(*stack_base)--;
294 	suword(*stack_base, (register_t)imgp->args->argc);
295 	return (0);
296 }
297 
298 /*
299  * Copied from kern/kern_exec.c
300  */
301 static register_t *
302 linux_copyout_strings(struct image_params *imgp)
303 {
304 	int argc, envc;
305 	char **vectp;
306 	char *stringp, *destp;
307 	register_t *stack_base;
308 	struct ps_strings *arginfo;
309 	char canary[LINUX_AT_RANDOM_LEN];
310 	size_t execpath_len;
311 	struct proc *p;
312 
313 	/*
314 	 * Calculate string base and vector table pointers.
315 	 */
316 	p = imgp->proc;
317 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
318 		execpath_len = strlen(imgp->execpath) + 1;
319 	else
320 		execpath_len = 0;
321 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
322 	destp = (caddr_t)arginfo - SPARE_USRSPACE - linux_szplatform -
323 	    roundup(sizeof(canary), sizeof(char *)) -
324 	    roundup(execpath_len, sizeof(char *)) -
325 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
326 
327 	/*
328 	 * install LINUX_PLATFORM
329 	 */
330 	copyout(linux_kplatform, ((caddr_t)arginfo - linux_szplatform),
331 	    linux_szplatform);
332 
333 	if (execpath_len != 0) {
334 		imgp->execpathp = (uintptr_t)arginfo -
335 		linux_szplatform - execpath_len;
336 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
337 	}
338 
339 	/*
340 	 * Prepare the canary for SSP.
341 	 */
342 	arc4rand(canary, sizeof(canary), 0);
343 	imgp->canary = (uintptr_t)arginfo - linux_szplatform -
344 	    roundup(execpath_len, sizeof(char *)) -
345 	    roundup(sizeof(canary), sizeof(char *));
346 	copyout(canary, (void *)imgp->canary, sizeof(canary));
347 
348 	/*
349 	 * If we have a valid auxargs ptr, prepare some room
350 	 * on the stack.
351 	 */
352 	if (imgp->auxargs) {
353 		/*
354 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
355 		 * lower compatibility.
356 		 */
357 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
358 		    (LINUX_AT_COUNT * 2);
359 		/*
360 		 * The '+ 2' is for the null pointers at the end of each of
361 		 * the arg and env vector sets,and imgp->auxarg_size is room
362 		 * for argument of Runtime loader.
363 		 */
364 		vectp = (char **)(destp - (imgp->args->argc +
365 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
366 	} else {
367 		/*
368 		 * The '+ 2' is for the null pointers at the end of each of
369 		 * the arg and env vector sets
370 		 */
371 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
372 		    sizeof(char *));
373 	}
374 
375 	/*
376 	 * vectp also becomes our initial stack base
377 	 */
378 	stack_base = (register_t *)vectp;
379 
380 	stringp = imgp->args->begin_argv;
381 	argc = imgp->args->argc;
382 	envc = imgp->args->envc;
383 
384 	/*
385 	 * Copy out strings - arguments and environment.
386 	 */
387 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
388 
389 	/*
390 	 * Fill in "ps_strings" struct for ps, w, etc.
391 	 */
392 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
393 	suword(&arginfo->ps_nargvstr, argc);
394 
395 	/*
396 	 * Fill in argument portion of vector table.
397 	 */
398 	for (; argc > 0; --argc) {
399 		suword(vectp++, (long)(intptr_t)destp);
400 		while (*stringp++ != 0)
401 			destp++;
402 		destp++;
403 	}
404 
405 	/* a null vector table pointer separates the argp's from the envp's */
406 	suword(vectp++, 0);
407 
408 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
409 	suword(&arginfo->ps_nenvstr, envc);
410 
411 	/*
412 	 * Fill in environment portion of vector table.
413 	 */
414 	for (; envc > 0; --envc) {
415 		suword(vectp++, (long)(intptr_t)destp);
416 		while (*stringp++ != 0)
417 			destp++;
418 		destp++;
419 	}
420 
421 	/* end of vector table is a null pointer */
422 	suword(vectp, 0);
423 
424 	return (stack_base);
425 }
426 
427 static void
428 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
429 {
430 	struct thread *td = curthread;
431 	struct proc *p = td->td_proc;
432 	struct sigacts *psp;
433 	struct trapframe *regs;
434 	struct l_rt_sigframe *fp, frame;
435 	int sig, code;
436 	int oonstack;
437 
438 	sig = ksi->ksi_signo;
439 	code = ksi->ksi_code;
440 	PROC_LOCK_ASSERT(p, MA_OWNED);
441 	psp = p->p_sigacts;
442 	mtx_assert(&psp->ps_mtx, MA_OWNED);
443 	regs = td->td_frame;
444 	oonstack = sigonstack(regs->tf_esp);
445 
446 #ifdef DEBUG
447 	if (ldebug(rt_sendsig))
448 		printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
449 		    catcher, sig, (void*)mask, code);
450 #endif
451 	/*
452 	 * Allocate space for the signal handler context.
453 	 */
454 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
455 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
456 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
457 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
458 	} else
459 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
460 	mtx_unlock(&psp->ps_mtx);
461 
462 	/*
463 	 * Build the argument list for the signal handler.
464 	 */
465 	sig = bsd_to_linux_signal(sig);
466 
467 	bzero(&frame, sizeof(frame));
468 
469 	frame.sf_handler = catcher;
470 	frame.sf_sig = sig;
471 	frame.sf_siginfo = &fp->sf_si;
472 	frame.sf_ucontext = &fp->sf_sc;
473 
474 	/* Fill in POSIX parts */
475 	ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
476 
477 	/*
478 	 * Build the signal context to be used by sigreturn.
479 	 */
480 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
481 	frame.sf_sc.uc_link = NULL;		/* XXX ??? */
482 
483 	frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
484 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
485 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
486 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
487 	PROC_UNLOCK(p);
488 
489 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
490 
491 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
492 	frame.sf_sc.uc_mcontext.sc_gs     = rgs();
493 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
494 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
495 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
496 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_edi;
497 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_esi;
498 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_ebp;
499 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_ebx;
500 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_esp;
501 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_edx;
502 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_ecx;
503 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_eax;
504 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_eip;
505 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
506 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
507 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
508 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
509 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
510 	frame.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
511 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
512 
513 #ifdef DEBUG
514 	if (ldebug(rt_sendsig))
515 		printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
516 		    frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
517 		    td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
518 #endif
519 
520 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
521 		/*
522 		 * Process has trashed its stack; give it an illegal
523 		 * instruction to halt it in its tracks.
524 		 */
525 #ifdef DEBUG
526 		if (ldebug(rt_sendsig))
527 			printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
528 			    fp, oonstack);
529 #endif
530 		PROC_LOCK(p);
531 		sigexit(td, SIGILL);
532 	}
533 
534 	/*
535 	 * Build context to run handler in.
536 	 */
537 	regs->tf_esp = (int)fp;
538 	regs->tf_eip = linux_rt_sigcode;
539 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
540 	regs->tf_cs = _ucodesel;
541 	regs->tf_ds = _udatasel;
542 	regs->tf_es = _udatasel;
543 	regs->tf_fs = _udatasel;
544 	regs->tf_ss = _udatasel;
545 	PROC_LOCK(p);
546 	mtx_lock(&psp->ps_mtx);
547 }
548 
549 
550 /*
551  * Send an interrupt to process.
552  *
553  * Stack is set up to allow sigcode stored
554  * in u. to call routine, followed by kcall
555  * to sigreturn routine below.  After sigreturn
556  * resets the signal mask, the stack, and the
557  * frame pointer, it returns to the user
558  * specified pc, psl.
559  */
560 static void
561 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
562 {
563 	struct thread *td = curthread;
564 	struct proc *p = td->td_proc;
565 	struct sigacts *psp;
566 	struct trapframe *regs;
567 	struct l_sigframe *fp, frame;
568 	l_sigset_t lmask;
569 	int sig, code;
570 	int oonstack;
571 
572 	PROC_LOCK_ASSERT(p, MA_OWNED);
573 	psp = p->p_sigacts;
574 	sig = ksi->ksi_signo;
575 	code = ksi->ksi_code;
576 	mtx_assert(&psp->ps_mtx, MA_OWNED);
577 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
578 		/* Signal handler installed with SA_SIGINFO. */
579 		linux_rt_sendsig(catcher, ksi, mask);
580 		return;
581 	}
582 	regs = td->td_frame;
583 	oonstack = sigonstack(regs->tf_esp);
584 
585 #ifdef DEBUG
586 	if (ldebug(sendsig))
587 		printf(ARGS(sendsig, "%p, %d, %p, %u"),
588 		    catcher, sig, (void*)mask, code);
589 #endif
590 
591 	/*
592 	 * Allocate space for the signal handler context.
593 	 */
594 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
595 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
596 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
597 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
598 	} else
599 		fp = (struct l_sigframe *)regs->tf_esp - 1;
600 	mtx_unlock(&psp->ps_mtx);
601 	PROC_UNLOCK(p);
602 
603 	/*
604 	 * Build the argument list for the signal handler.
605 	 */
606 	sig = bsd_to_linux_signal(sig);
607 
608 	bzero(&frame, sizeof(frame));
609 
610 	frame.sf_handler = catcher;
611 	frame.sf_sig = sig;
612 
613 	bsd_to_linux_sigset(mask, &lmask);
614 
615 	/*
616 	 * Build the signal context to be used by sigreturn.
617 	 */
618 	frame.sf_sc.sc_mask   = lmask.__mask;
619 	frame.sf_sc.sc_gs     = rgs();
620 	frame.sf_sc.sc_fs     = regs->tf_fs;
621 	frame.sf_sc.sc_es     = regs->tf_es;
622 	frame.sf_sc.sc_ds     = regs->tf_ds;
623 	frame.sf_sc.sc_edi    = regs->tf_edi;
624 	frame.sf_sc.sc_esi    = regs->tf_esi;
625 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
626 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
627 	frame.sf_sc.sc_esp    = regs->tf_esp;
628 	frame.sf_sc.sc_edx    = regs->tf_edx;
629 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
630 	frame.sf_sc.sc_eax    = regs->tf_eax;
631 	frame.sf_sc.sc_eip    = regs->tf_eip;
632 	frame.sf_sc.sc_cs     = regs->tf_cs;
633 	frame.sf_sc.sc_eflags = regs->tf_eflags;
634 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
635 	frame.sf_sc.sc_ss     = regs->tf_ss;
636 	frame.sf_sc.sc_err    = regs->tf_err;
637 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
638 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
639 
640 	frame.sf_extramask[0] = lmask.__mask;
641 
642 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
643 		/*
644 		 * Process has trashed its stack; give it an illegal
645 		 * instruction to halt it in its tracks.
646 		 */
647 		PROC_LOCK(p);
648 		sigexit(td, SIGILL);
649 	}
650 
651 	/*
652 	 * Build context to run handler in.
653 	 */
654 	regs->tf_esp = (int)fp;
655 	regs->tf_eip = linux_sigcode;
656 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
657 	regs->tf_cs = _ucodesel;
658 	regs->tf_ds = _udatasel;
659 	regs->tf_es = _udatasel;
660 	regs->tf_fs = _udatasel;
661 	regs->tf_ss = _udatasel;
662 	PROC_LOCK(p);
663 	mtx_lock(&psp->ps_mtx);
664 }
665 
666 /*
667  * System call to cleanup state after a signal
668  * has been taken.  Reset signal mask and
669  * stack state from context left by sendsig (above).
670  * Return to previous pc and psl as specified by
671  * context left by sendsig. Check carefully to
672  * make sure that the user has not modified the
673  * psl to gain improper privileges or to cause
674  * a machine fault.
675  */
676 int
677 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
678 {
679 	struct l_sigframe frame;
680 	struct trapframe *regs;
681 	l_sigset_t lmask;
682 	sigset_t bmask;
683 	int eflags;
684 	ksiginfo_t ksi;
685 
686 	regs = td->td_frame;
687 
688 #ifdef DEBUG
689 	if (ldebug(sigreturn))
690 		printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
691 #endif
692 	/*
693 	 * The trampoline code hands us the sigframe.
694 	 * It is unsafe to keep track of it ourselves, in the event that a
695 	 * program jumps out of a signal handler.
696 	 */
697 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
698 		return (EFAULT);
699 
700 	/*
701 	 * Check for security violations.
702 	 */
703 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
704 	eflags = frame.sf_sc.sc_eflags;
705 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
706 		return (EINVAL);
707 
708 	/*
709 	 * Don't allow users to load a valid privileged %cs.  Let the
710 	 * hardware check for invalid selectors, excess privilege in
711 	 * other selectors, invalid %eip's and invalid %esp's.
712 	 */
713 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
714 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
715 		ksiginfo_init_trap(&ksi);
716 		ksi.ksi_signo = SIGBUS;
717 		ksi.ksi_code = BUS_OBJERR;
718 		ksi.ksi_trapno = T_PROTFLT;
719 		ksi.ksi_addr = (void *)regs->tf_eip;
720 		trapsignal(td, &ksi);
721 		return (EINVAL);
722 	}
723 
724 	lmask.__mask = frame.sf_sc.sc_mask;
725 	linux_to_bsd_sigset(&lmask, &bmask);
726 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
727 
728 	/*
729 	 * Restore signal context.
730 	 */
731 	/* %gs was restored by the trampoline. */
732 	regs->tf_fs     = frame.sf_sc.sc_fs;
733 	regs->tf_es     = frame.sf_sc.sc_es;
734 	regs->tf_ds     = frame.sf_sc.sc_ds;
735 	regs->tf_edi    = frame.sf_sc.sc_edi;
736 	regs->tf_esi    = frame.sf_sc.sc_esi;
737 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
738 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
739 	regs->tf_edx    = frame.sf_sc.sc_edx;
740 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
741 	regs->tf_eax    = frame.sf_sc.sc_eax;
742 	regs->tf_eip    = frame.sf_sc.sc_eip;
743 	regs->tf_cs     = frame.sf_sc.sc_cs;
744 	regs->tf_eflags = eflags;
745 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
746 	regs->tf_ss     = frame.sf_sc.sc_ss;
747 
748 	return (EJUSTRETURN);
749 }
750 
751 /*
752  * System call to cleanup state after a signal
753  * has been taken.  Reset signal mask and
754  * stack state from context left by rt_sendsig (above).
755  * Return to previous pc and psl as specified by
756  * context left by sendsig. Check carefully to
757  * make sure that the user has not modified the
758  * psl to gain improper privileges or to cause
759  * a machine fault.
760  */
761 int
762 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
763 {
764 	struct l_ucontext uc;
765 	struct l_sigcontext *context;
766 	sigset_t bmask;
767 	l_stack_t *lss;
768 	stack_t ss;
769 	struct trapframe *regs;
770 	int eflags;
771 	ksiginfo_t ksi;
772 
773 	regs = td->td_frame;
774 
775 #ifdef DEBUG
776 	if (ldebug(rt_sigreturn))
777 		printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
778 #endif
779 	/*
780 	 * The trampoline code hands us the ucontext.
781 	 * It is unsafe to keep track of it ourselves, in the event that a
782 	 * program jumps out of a signal handler.
783 	 */
784 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
785 		return (EFAULT);
786 
787 	context = &uc.uc_mcontext;
788 
789 	/*
790 	 * Check for security violations.
791 	 */
792 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
793 	eflags = context->sc_eflags;
794 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
795 		return (EINVAL);
796 
797 	/*
798 	 * Don't allow users to load a valid privileged %cs.  Let the
799 	 * hardware check for invalid selectors, excess privilege in
800 	 * other selectors, invalid %eip's and invalid %esp's.
801 	 */
802 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
803 	if (!CS_SECURE(context->sc_cs)) {
804 		ksiginfo_init_trap(&ksi);
805 		ksi.ksi_signo = SIGBUS;
806 		ksi.ksi_code = BUS_OBJERR;
807 		ksi.ksi_trapno = T_PROTFLT;
808 		ksi.ksi_addr = (void *)regs->tf_eip;
809 		trapsignal(td, &ksi);
810 		return (EINVAL);
811 	}
812 
813 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
814 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
815 
816 	/*
817 	 * Restore signal context
818 	 */
819 	/* %gs was restored by the trampoline. */
820 	regs->tf_fs     = context->sc_fs;
821 	regs->tf_es     = context->sc_es;
822 	regs->tf_ds     = context->sc_ds;
823 	regs->tf_edi    = context->sc_edi;
824 	regs->tf_esi    = context->sc_esi;
825 	regs->tf_ebp    = context->sc_ebp;
826 	regs->tf_ebx    = context->sc_ebx;
827 	regs->tf_edx    = context->sc_edx;
828 	regs->tf_ecx    = context->sc_ecx;
829 	regs->tf_eax    = context->sc_eax;
830 	regs->tf_eip    = context->sc_eip;
831 	regs->tf_cs     = context->sc_cs;
832 	regs->tf_eflags = eflags;
833 	regs->tf_esp    = context->sc_esp_at_signal;
834 	regs->tf_ss     = context->sc_ss;
835 
836 	/*
837 	 * call sigaltstack & ignore results..
838 	 */
839 	lss = &uc.uc_stack;
840 	ss.ss_sp = lss->ss_sp;
841 	ss.ss_size = lss->ss_size;
842 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
843 
844 #ifdef DEBUG
845 	if (ldebug(rt_sigreturn))
846 		printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
847 		    ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
848 #endif
849 	(void)kern_sigaltstack(td, &ss, NULL);
850 
851 	return (EJUSTRETURN);
852 }
853 
854 static int
855 linux_fetch_syscall_args(struct thread *td)
856 {
857 	struct proc *p;
858 	struct trapframe *frame;
859 	struct syscall_args *sa;
860 
861 	p = td->td_proc;
862 	frame = td->td_frame;
863 	sa = &td->td_sa;
864 
865 	sa->code = frame->tf_eax;
866 	sa->args[0] = frame->tf_ebx;
867 	sa->args[1] = frame->tf_ecx;
868 	sa->args[2] = frame->tf_edx;
869 	sa->args[3] = frame->tf_esi;
870 	sa->args[4] = frame->tf_edi;
871 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
872 
873 	if (sa->code >= p->p_sysent->sv_size)
874 		/* nosys */
875 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
876  	else
877  		sa->callp = &p->p_sysent->sv_table[sa->code];
878 	sa->narg = sa->callp->sy_narg;
879 
880 	td->td_retval[0] = 0;
881 	td->td_retval[1] = frame->tf_edx;
882 
883 	return (0);
884 }
885 
886 /*
887  * If a linux binary is exec'ing something, try this image activator
888  * first.  We override standard shell script execution in order to
889  * be able to modify the interpreter path.  We only do this if a linux
890  * binary is doing the exec, so we do not create an EXEC module for it.
891  */
892 static int	exec_linux_imgact_try(struct image_params *iparams);
893 
894 static int
895 exec_linux_imgact_try(struct image_params *imgp)
896 {
897     const char *head = (const char *)imgp->image_header;
898     char *rpath;
899     int error = -1;
900 
901     /*
902      * The interpreter for shell scripts run from a linux binary needs
903      * to be located in /compat/linux if possible in order to recursively
904      * maintain linux path emulation.
905      */
906     if (((const short *)head)[0] == SHELLMAGIC) {
907 	    /*
908 	     * Run our normal shell image activator.  If it succeeds attempt
909 	     * to use the alternate path for the interpreter.  If an alternate
910 	     * path is found, use our stringspace to store it.
911 	     */
912 	    if ((error = exec_shell_imgact(imgp)) == 0) {
913 		    linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
914 			imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, AT_FDCWD);
915 		    if (rpath != NULL)
916 			    imgp->args->fname_buf =
917 				imgp->interpreter_name = rpath;
918 	    }
919     }
920     return (error);
921 }
922 
923 /*
924  * exec_setregs may initialize some registers differently than Linux
925  * does, thus potentially confusing Linux binaries. If necessary, we
926  * override the exec_setregs default(s) here.
927  */
928 static void
929 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack)
930 {
931 	struct pcb *pcb = td->td_pcb;
932 
933 	exec_setregs(td, imgp, stack);
934 
935 	/* Linux sets %gs to 0, we default to _udatasel */
936 	pcb->pcb_gs = 0;
937 	load_gs(0);
938 
939 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
940 }
941 
942 static void
943 linux_get_machine(const char **dst)
944 {
945 
946 	switch (cpu_class) {
947 	case CPUCLASS_686:
948 		*dst = "i686";
949 		break;
950 	case CPUCLASS_586:
951 		*dst = "i586";
952 		break;
953 	case CPUCLASS_486:
954 		*dst = "i486";
955 		break;
956 	default:
957 		*dst = "i386";
958 	}
959 }
960 
961 struct sysentvec linux_sysvec = {
962 	.sv_size	= LINUX_SYS_MAXSYSCALL,
963 	.sv_table	= linux_sysent,
964 	.sv_mask	= 0,
965 	.sv_errsize	= ELAST + 1,
966 	.sv_errtbl	= bsd_to_linux_errno,
967 	.sv_transtrap	= translate_traps,
968 	.sv_fixup	= linux_fixup,
969 	.sv_sendsig	= linux_sendsig,
970 	.sv_sigcode	= &_binary_linux_locore_o_start,
971 	.sv_szsigcode	= &linux_szsigcode,
972 	.sv_name	= "Linux a.out",
973 	.sv_coredump	= NULL,
974 	.sv_imgact_try	= exec_linux_imgact_try,
975 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
976 	.sv_pagesize	= PAGE_SIZE,
977 	.sv_minuser	= VM_MIN_ADDRESS,
978 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
979 	.sv_usrstack	= LINUX_USRSTACK,
980 	.sv_psstrings	= PS_STRINGS,
981 	.sv_stackprot	= VM_PROT_ALL,
982 	.sv_copyout_strings = exec_copyout_strings,
983 	.sv_setregs	= exec_linux_setregs,
984 	.sv_fixlimit	= NULL,
985 	.sv_maxssiz	= NULL,
986 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32,
987 	.sv_set_syscall_retval = cpu_set_syscall_retval,
988 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
989 	.sv_syscallnames = NULL,
990 	.sv_shared_page_base = LINUX_SHAREDPAGE,
991 	.sv_shared_page_len = PAGE_SIZE,
992 	.sv_schedtail	= linux_schedtail,
993 	.sv_thread_detach = linux_thread_detach,
994 	.sv_trap	= NULL,
995 };
996 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
997 
998 struct sysentvec elf_linux_sysvec = {
999 	.sv_size	= LINUX_SYS_MAXSYSCALL,
1000 	.sv_table	= linux_sysent,
1001 	.sv_mask	= 0,
1002 	.sv_errsize	= ELAST + 1,
1003 	.sv_errtbl	= bsd_to_linux_errno,
1004 	.sv_transtrap	= translate_traps,
1005 	.sv_fixup	= elf_linux_fixup,
1006 	.sv_sendsig	= linux_sendsig,
1007 	.sv_sigcode	= &_binary_linux_locore_o_start,
1008 	.sv_szsigcode	= &linux_szsigcode,
1009 	.sv_name	= "Linux ELF",
1010 	.sv_coredump	= elf32_coredump,
1011 	.sv_imgact_try	= exec_linux_imgact_try,
1012 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
1013 	.sv_pagesize	= PAGE_SIZE,
1014 	.sv_minuser	= VM_MIN_ADDRESS,
1015 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
1016 	.sv_usrstack	= LINUX_USRSTACK,
1017 	.sv_psstrings	= LINUX_PS_STRINGS,
1018 	.sv_stackprot	= VM_PROT_ALL,
1019 	.sv_copyout_strings = linux_copyout_strings,
1020 	.sv_setregs	= exec_linux_setregs,
1021 	.sv_fixlimit	= NULL,
1022 	.sv_maxssiz	= NULL,
1023 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP,
1024 	.sv_set_syscall_retval = cpu_set_syscall_retval,
1025 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
1026 	.sv_syscallnames = NULL,
1027 	.sv_shared_page_base = LINUX_SHAREDPAGE,
1028 	.sv_shared_page_len = PAGE_SIZE,
1029 	.sv_schedtail	= linux_schedtail,
1030 	.sv_thread_detach = linux_thread_detach,
1031 	.sv_trap	= NULL,
1032 };
1033 
1034 static void
1035 linux_vdso_install(void *param)
1036 {
1037 
1038 	linux_szsigcode = (&_binary_linux_locore_o_end -
1039 	    &_binary_linux_locore_o_start);
1040 
1041 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
1042 		panic("Linux invalid vdso size\n");
1043 
1044 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
1045 
1046 	linux_shared_page_obj = __elfN(linux_shared_page_init)
1047 	    (&linux_shared_page_mapping);
1048 
1049 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
1050 
1051 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
1052 	    linux_szsigcode);
1053 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
1054 }
1055 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
1056     (sysinit_cfunc_t)linux_vdso_install, NULL);
1057 
1058 static void
1059 linux_vdso_deinstall(void *param)
1060 {
1061 
1062 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
1063 };
1064 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1065     (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
1066 
1067 static char GNU_ABI_VENDOR[] = "GNU";
1068 static int GNULINUX_ABI_DESC = 0;
1069 
1070 static boolean_t
1071 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
1072 {
1073 	const Elf32_Word *desc;
1074 	uintptr_t p;
1075 
1076 	p = (uintptr_t)(note + 1);
1077 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1078 
1079 	desc = (const Elf32_Word *)p;
1080 	if (desc[0] != GNULINUX_ABI_DESC)
1081 		return (FALSE);
1082 
1083 	/*
1084 	 * For linux we encode osrel as follows (see linux_mib.c):
1085 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
1086 	 */
1087 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1088 
1089 	return (TRUE);
1090 }
1091 
1092 static Elf_Brandnote linux_brandnote = {
1093 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
1094 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
1095 	.hdr.n_type	= 1,
1096 	.vendor		= GNU_ABI_VENDOR,
1097 	.flags		= BN_TRANSLATE_OSREL,
1098 	.trans_osrel	= linux_trans_osrel
1099 };
1100 
1101 static Elf32_Brandinfo linux_brand = {
1102 	.brand		= ELFOSABI_LINUX,
1103 	.machine	= EM_386,
1104 	.compat_3_brand	= "Linux",
1105 	.emul_path	= "/compat/linux",
1106 	.interp_path	= "/lib/ld-linux.so.1",
1107 	.sysvec		= &elf_linux_sysvec,
1108 	.interp_newpath	= NULL,
1109 	.brand_note	= &linux_brandnote,
1110 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1111 };
1112 
1113 static Elf32_Brandinfo linux_glibc2brand = {
1114 	.brand		= ELFOSABI_LINUX,
1115 	.machine	= EM_386,
1116 	.compat_3_brand	= "Linux",
1117 	.emul_path	= "/compat/linux",
1118 	.interp_path	= "/lib/ld-linux.so.2",
1119 	.sysvec		= &elf_linux_sysvec,
1120 	.interp_newpath	= NULL,
1121 	.brand_note	= &linux_brandnote,
1122 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1123 };
1124 
1125 static Elf32_Brandinfo linux_muslbrand = {
1126 	.brand		= ELFOSABI_LINUX,
1127 	.machine	= EM_386,
1128 	.compat_3_brand	= "Linux",
1129 	.emul_path	= "/compat/linux",
1130 	.interp_path	= "/lib/ld-musl-i386.so.1",
1131 	.sysvec		= &elf_linux_sysvec,
1132 	.interp_newpath	= NULL,
1133 	.brand_note	= &linux_brandnote,
1134 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1135 };
1136 
1137 Elf32_Brandinfo *linux_brandlist[] = {
1138 	&linux_brand,
1139 	&linux_glibc2brand,
1140 	&linux_muslbrand,
1141 	NULL
1142 };
1143 
1144 static int
1145 linux_elf_modevent(module_t mod, int type, void *data)
1146 {
1147 	Elf32_Brandinfo **brandinfo;
1148 	int error;
1149 	struct linux_ioctl_handler **lihp;
1150 
1151 	error = 0;
1152 
1153 	switch(type) {
1154 	case MOD_LOAD:
1155 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1156 		     ++brandinfo)
1157 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1158 				error = EINVAL;
1159 		if (error == 0) {
1160 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1161 				linux_ioctl_register_handler(*lihp);
1162 			LIST_INIT(&futex_list);
1163 			mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1164 			linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit,
1165 			      NULL, 1000);
1166 			linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec,
1167 			      NULL, 1000);
1168 			linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor,
1169 			    linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY);
1170 			linux_get_machine(&linux_kplatform);
1171 			linux_szplatform = roundup(strlen(linux_kplatform) + 1,
1172 			    sizeof(char *));
1173 			linux_osd_jail_register();
1174 			stclohz = (stathz ? stathz : hz);
1175 			if (bootverbose)
1176 				printf("Linux ELF exec handler installed\n");
1177 		} else
1178 			printf("cannot insert Linux ELF brand handler\n");
1179 		break;
1180 	case MOD_UNLOAD:
1181 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1182 		     ++brandinfo)
1183 			if (elf32_brand_inuse(*brandinfo))
1184 				error = EBUSY;
1185 		if (error == 0) {
1186 			for (brandinfo = &linux_brandlist[0];
1187 			     *brandinfo != NULL; ++brandinfo)
1188 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1189 					error = EINVAL;
1190 		}
1191 		if (error == 0) {
1192 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1193 				linux_ioctl_unregister_handler(*lihp);
1194 			mtx_destroy(&futex_mtx);
1195 			EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
1196 			EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
1197 			EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag);
1198 			linux_osd_jail_deregister();
1199 			if (bootverbose)
1200 				printf("Linux ELF exec handler removed\n");
1201 		} else
1202 			printf("Could not deinstall ELF interpreter entry\n");
1203 		break;
1204 	default:
1205 		return (EOPNOTSUPP);
1206 	}
1207 	return (error);
1208 }
1209 
1210 static moduledata_t linux_elf_mod = {
1211 	"linuxelf",
1212 	linux_elf_modevent,
1213 	0
1214 };
1215 
1216 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1217 FEATURE(linux, "Linux 32bit support");
1218