xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision 13ec1e3155c7e9bf037b12af186351b7fa9b9450)
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/malloc.h>
41 #include <sys/module.h>
42 #include <sys/proc.h>
43 #include <sys/stddef.h>
44 #include <sys/signalvar.h>
45 #include <sys/syscallsubr.h>
46 #include <sys/sysctl.h>
47 #include <sys/sysent.h>
48 #include <sys/sysproto.h>
49 #include <sys/vnode.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_extern.h>
54 #include <vm/vm_map.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_param.h>
58 
59 #include <machine/cpu.h>
60 #include <machine/cputypes.h>
61 #include <machine/md_var.h>
62 #include <machine/pcb.h>
63 #include <machine/trap.h>
64 
65 #include <x86/linux/linux_x86.h>
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_fork.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_VDSOPAGE_SIZE	PAGE_SIZE * 2
80 #define	LINUX_VDSOPAGE		(VM_MAXUSER_ADDRESS - LINUX_VDSOPAGE_SIZE)
81 #define	LINUX_SHAREDPAGE	(LINUX_VDSOPAGE - PAGE_SIZE)
82 				/*
83 				 * PAGE_SIZE - the size
84 				 * of the native SHAREDPAGE
85 				 */
86 #define	LINUX_USRSTACK		LINUX_SHAREDPAGE
87 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
88 
89 static int linux_szsigcode;
90 static vm_object_t linux_vdso_obj;
91 static char *linux_vdso_mapping;
92 extern char _binary_linux_vdso_so_o_start;
93 extern char _binary_linux_vdso_so_o_end;
94 static vm_offset_t linux_vdso_base;
95 
96 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
97 
98 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
99 
100 static int	linux_fixup(uintptr_t *stack_base,
101 		    struct image_params *iparams);
102 static int	linux_fixup_elf(uintptr_t *stack_base,
103 		    struct image_params *iparams);
104 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
105 static void	linux_exec_setregs(struct thread *td,
106 		    struct image_params *imgp, uintptr_t stack);
107 static void	linux_exec_sysvec_init(void *param);
108 static int	linux_on_exec_vmspace(struct proc *p,
109 		    struct image_params *imgp);
110 static int	linux_copyout_strings(struct image_params *imgp,
111 		    uintptr_t *stack_base);
112 static void	linux_set_fork_retval(struct thread *td);
113 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
114 static void	linux_vdso_install(const void *param);
115 static void	linux_vdso_deinstall(const void *param);
116 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
117 
118 #define LINUX_T_UNKNOWN  255
119 static int _bsd_to_linux_trapcode[] = {
120 	LINUX_T_UNKNOWN,	/* 0 */
121 	6,			/* 1  T_PRIVINFLT */
122 	LINUX_T_UNKNOWN,	/* 2 */
123 	3,			/* 3  T_BPTFLT */
124 	LINUX_T_UNKNOWN,	/* 4 */
125 	LINUX_T_UNKNOWN,	/* 5 */
126 	16,			/* 6  T_ARITHTRAP */
127 	254,			/* 7  T_ASTFLT */
128 	LINUX_T_UNKNOWN,	/* 8 */
129 	13,			/* 9  T_PROTFLT */
130 	1,			/* 10 T_TRCTRAP */
131 	LINUX_T_UNKNOWN,	/* 11 */
132 	14,			/* 12 T_PAGEFLT */
133 	LINUX_T_UNKNOWN,	/* 13 */
134 	17,			/* 14 T_ALIGNFLT */
135 	LINUX_T_UNKNOWN,	/* 15 */
136 	LINUX_T_UNKNOWN,	/* 16 */
137 	LINUX_T_UNKNOWN,	/* 17 */
138 	0,			/* 18 T_DIVIDE */
139 	2,			/* 19 T_NMI */
140 	4,			/* 20 T_OFLOW */
141 	5,			/* 21 T_BOUND */
142 	7,			/* 22 T_DNA */
143 	8,			/* 23 T_DOUBLEFLT */
144 	9,			/* 24 T_FPOPFLT */
145 	10,			/* 25 T_TSSFLT */
146 	11,			/* 26 T_SEGNPFLT */
147 	12,			/* 27 T_STKFLT */
148 	18,			/* 28 T_MCHK */
149 	19,			/* 29 T_XMMFLT */
150 	15			/* 30 T_RESERVED */
151 };
152 #define bsd_to_linux_trapcode(code) \
153     ((code)<nitems(_bsd_to_linux_trapcode)? \
154      _bsd_to_linux_trapcode[(code)]: \
155      LINUX_T_UNKNOWN)
156 
157 LINUX_VDSO_SYM_CHAR(linux_platform);
158 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
159 LINUX_VDSO_SYM_INTPTR(__kernel_sigreturn);
160 LINUX_VDSO_SYM_INTPTR(__kernel_rt_sigreturn);
161 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
162 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
163 
164 /*
165  * If FreeBSD & Linux have a difference of opinion about what a trap
166  * means, deal with it here.
167  *
168  * MPSAFE
169  */
170 static int
171 linux_translate_traps(int signal, int trap_code)
172 {
173 	if (signal != SIGBUS)
174 		return (signal);
175 	switch (trap_code) {
176 	case T_PROTFLT:
177 	case T_TSSFLT:
178 	case T_DOUBLEFLT:
179 	case T_PAGEFLT:
180 		return (SIGSEGV);
181 	default:
182 		return (signal);
183 	}
184 }
185 
186 static int
187 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
188 {
189 	register_t *base, *argv, *envp;
190 
191 	base = (register_t *)*stack_base;
192 	argv = base;
193 	envp = base + (imgp->args->argc + 1);
194 	base--;
195 	suword(base, (intptr_t)envp);
196 	base--;
197 	suword(base, (intptr_t)argv);
198 	base--;
199 	suword(base, imgp->args->argc);
200 	*stack_base = (uintptr_t)base;
201 	return (0);
202 }
203 
204 static int
205 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
206 {
207 	struct proc *p;
208 	Elf32_Auxargs *args;
209 	Elf32_Auxinfo *argarray, *pos;
210 	struct ps_strings *arginfo;
211 	int error, issetugid;
212 
213 	p = imgp->proc;
214 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
215 	arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
216 	args = (Elf32_Auxargs *)imgp->auxargs;
217 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
218 	    M_WAITOK | M_ZERO);
219 
220 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
221 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, __kernel_vsyscall);
222 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
223 
224 	/*
225 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
226 	 * as it has appeared in the 2.4.0-rc7 first time.
227 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
228 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
229 	 * is not present.
230 	 * Also see linux_times() implementation.
231 	 */
232 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
233 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
234 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
235 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
236 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
237 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
238 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
239 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
240 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
241 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
242 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
243 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
244 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
245 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
246 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
247 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
248 	if (imgp->execpathp != 0)
249 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
250 	if (args->execfd != -1)
251 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
252 	AUXARGS_ENTRY(pos, AT_NULL, 0);
253 
254 	free(imgp->auxargs, M_TEMP);
255 	imgp->auxargs = NULL;
256 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
257 
258 	error = copyout(argarray, (void *)base,
259 	    sizeof(*argarray) * LINUX_AT_COUNT);
260 	free(argarray, M_TEMP);
261 	return (error);
262 }
263 
264 static int
265 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
266 {
267 	register_t *base;
268 
269 	base = (register_t *)*stack_base;
270 	base--;
271 	if (suword(base, (register_t)imgp->args->argc) == -1)
272 		return (EFAULT);
273 	*stack_base = (uintptr_t)base;
274 	return (0);
275 }
276 
277 /*
278  * Copied from kern/kern_exec.c
279  */
280 static int
281 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
282 {
283 	int argc, envc, error;
284 	char **vectp;
285 	char *stringp;
286 	uintptr_t destp, ustringp;
287 	struct ps_strings *arginfo;
288 	char canary[LINUX_AT_RANDOM_LEN];
289 	size_t execpath_len;
290 	struct proc *p;
291 
292 	p = imgp->proc;
293 	arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
294 	destp = (uintptr_t)arginfo;
295 
296 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
297 		execpath_len = strlen(imgp->execpath) + 1;
298 		destp -= execpath_len;
299 		destp = rounddown2(destp, sizeof(void *));
300 		imgp->execpathp = (void *)destp;
301 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
302 		if (error != 0)
303 			return (error);
304 	}
305 
306 	/* Prepare the canary for SSP. */
307 	arc4rand(canary, sizeof(canary), 0);
308 	destp -= roundup(sizeof(canary), sizeof(void *));
309 	imgp->canary = (void *)destp;
310 	error = copyout(canary, imgp->canary, sizeof(canary));
311 	if (error != 0)
312 		return (error);
313 
314 	/* Allocate room for the argument and environment strings. */
315 	destp -= ARG_MAX - imgp->args->stringspace;
316 	destp = rounddown2(destp, sizeof(void *));
317 	ustringp = destp;
318 
319 	if (imgp->auxargs) {
320 		/*
321 		 * Allocate room on the stack for the ELF auxargs
322 		 * array.  It has LINUX_AT_COUNT entries.
323 		 */
324 		destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
325 		destp = rounddown2(destp, sizeof(void *));
326 	}
327 
328 	vectp = (char **)destp;
329 
330 	/*
331 	 * Allocate room for the argv[] and env vectors including the
332 	 * terminating NULL pointers.
333 	 */
334 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
335 
336 	/* vectp also becomes our initial stack base. */
337 	*stack_base = (uintptr_t)vectp;
338 
339 	stringp = imgp->args->begin_argv;
340 	argc = imgp->args->argc;
341 	envc = imgp->args->envc;
342 
343 	/* Copy out strings - arguments and environment. */
344 	error = copyout(stringp, (void *)ustringp,
345 	    ARG_MAX - imgp->args->stringspace);
346 	if (error != 0)
347 		return (error);
348 
349 	/* Fill in "ps_strings" struct for ps, w, etc. */
350 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
351 	    suword(&arginfo->ps_nargvstr, argc) != 0)
352 		return (EFAULT);
353 
354 	/* Fill in argument portion of vector table. */
355 	for (; argc > 0; --argc) {
356 		if (suword(vectp++, ustringp) != 0)
357 			return (EFAULT);
358 		while (*stringp++ != 0)
359 			ustringp++;
360 		ustringp++;
361 	}
362 
363 	/* A null vector table pointer separates the argp's from the envp's. */
364 	if (suword(vectp++, 0) != 0)
365 		return (EFAULT);
366 
367 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
368 	    suword(&arginfo->ps_nenvstr, envc) != 0)
369 		return (EFAULT);
370 
371 	/* Fill in environment portion of vector table. */
372 	for (; envc > 0; --envc) {
373 		if (suword(vectp++, ustringp) != 0)
374 			return (EFAULT);
375 		while (*stringp++ != 0)
376 			ustringp++;
377 		ustringp++;
378 	}
379 
380 	/* The end of the vector table is a null pointer. */
381 	if (suword(vectp, 0) != 0)
382 		return (EFAULT);
383 
384 	if (imgp->auxargs) {
385 		vectp++;
386 		error = imgp->sysent->sv_copyout_auxargs(imgp,
387 		    (uintptr_t)vectp);
388 		if (error != 0)
389 			return (error);
390 	}
391 
392 	return (0);
393 }
394 
395 static void
396 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
397 {
398 	struct thread *td = curthread;
399 	struct proc *p = td->td_proc;
400 	struct sigacts *psp;
401 	struct trapframe *regs;
402 	struct l_rt_sigframe *fp, frame;
403 	int sig, code;
404 	int oonstack;
405 
406 	sig = ksi->ksi_signo;
407 	code = ksi->ksi_code;
408 	PROC_LOCK_ASSERT(p, MA_OWNED);
409 	psp = p->p_sigacts;
410 	mtx_assert(&psp->ps_mtx, MA_OWNED);
411 	regs = td->td_frame;
412 	oonstack = sigonstack(regs->tf_esp);
413 
414 	/* Allocate space for the signal handler context. */
415 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
416 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
417 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
418 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
419 	} else
420 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
421 	mtx_unlock(&psp->ps_mtx);
422 
423 	/* Build the argument list for the signal handler. */
424 	sig = bsd_to_linux_signal(sig);
425 
426 	bzero(&frame, sizeof(frame));
427 
428 	frame.sf_handler = catcher;
429 	frame.sf_sig = sig;
430 	frame.sf_siginfo = &fp->sf_si;
431 	frame.sf_ucontext = &fp->sf_sc;
432 
433 	/* Fill in POSIX parts. */
434 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
435 
436 	/* Build the signal context to be used by sigreturn. */
437 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
438 	frame.sf_sc.uc_link = NULL;		/* XXX ??? */
439 
440 	frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
441 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
442 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
443 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
444 	PROC_UNLOCK(p);
445 
446 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
447 
448 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
449 	frame.sf_sc.uc_mcontext.sc_gs     = rgs();
450 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
451 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
452 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
453 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_edi;
454 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_esi;
455 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_ebp;
456 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_ebx;
457 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_esp;
458 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_edx;
459 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_ecx;
460 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_eax;
461 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_eip;
462 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
463 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
464 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
465 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
466 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
467 	frame.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
468 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
469 
470 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
471 		/*
472 		 * Process has trashed its stack; give it an illegal
473 		 * instruction to halt it in its tracks.
474 		 */
475 		PROC_LOCK(p);
476 		sigexit(td, SIGILL);
477 	}
478 
479 	/* Build context to run handler in. */
480 	regs->tf_esp = (int)fp;
481 	regs->tf_eip = __kernel_rt_sigreturn;
482 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
483 	regs->tf_cs = _ucodesel;
484 	regs->tf_ds = _udatasel;
485 	regs->tf_es = _udatasel;
486 	regs->tf_fs = _udatasel;
487 	regs->tf_ss = _udatasel;
488 	PROC_LOCK(p);
489 	mtx_lock(&psp->ps_mtx);
490 }
491 
492 /*
493  * Send an interrupt to process.
494  *
495  * Stack is set up to allow sigcode stored
496  * in u. to call routine, followed by kcall
497  * to sigreturn routine below.  After sigreturn
498  * resets the signal mask, the stack, and the
499  * frame pointer, it returns to the user
500  * specified pc, psl.
501  */
502 static void
503 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
504 {
505 	struct thread *td = curthread;
506 	struct proc *p = td->td_proc;
507 	struct sigacts *psp;
508 	struct trapframe *regs;
509 	struct l_sigframe *fp, frame;
510 	l_sigset_t lmask;
511 	int sig, code;
512 	int oonstack;
513 
514 	PROC_LOCK_ASSERT(p, MA_OWNED);
515 	psp = p->p_sigacts;
516 	sig = ksi->ksi_signo;
517 	code = ksi->ksi_code;
518 	mtx_assert(&psp->ps_mtx, MA_OWNED);
519 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
520 		/* Signal handler installed with SA_SIGINFO. */
521 		linux_rt_sendsig(catcher, ksi, mask);
522 		return;
523 	}
524 	regs = td->td_frame;
525 	oonstack = sigonstack(regs->tf_esp);
526 
527 	/* Allocate space for the signal handler context. */
528 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
529 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
530 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
531 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
532 	} else
533 		fp = (struct l_sigframe *)regs->tf_esp - 1;
534 	mtx_unlock(&psp->ps_mtx);
535 	PROC_UNLOCK(p);
536 
537 	/* Build the argument list for the signal handler. */
538 	sig = bsd_to_linux_signal(sig);
539 
540 	bzero(&frame, sizeof(frame));
541 
542 	frame.sf_handler = catcher;
543 	frame.sf_sig = sig;
544 
545 	bsd_to_linux_sigset(mask, &lmask);
546 
547 	/* Build the signal context to be used by sigreturn. */
548 	frame.sf_sc.sc_mask   = lmask.__mask;
549 	frame.sf_sc.sc_gs     = rgs();
550 	frame.sf_sc.sc_fs     = regs->tf_fs;
551 	frame.sf_sc.sc_es     = regs->tf_es;
552 	frame.sf_sc.sc_ds     = regs->tf_ds;
553 	frame.sf_sc.sc_edi    = regs->tf_edi;
554 	frame.sf_sc.sc_esi    = regs->tf_esi;
555 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
556 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
557 	frame.sf_sc.sc_esp    = regs->tf_esp;
558 	frame.sf_sc.sc_edx    = regs->tf_edx;
559 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
560 	frame.sf_sc.sc_eax    = regs->tf_eax;
561 	frame.sf_sc.sc_eip    = regs->tf_eip;
562 	frame.sf_sc.sc_cs     = regs->tf_cs;
563 	frame.sf_sc.sc_eflags = regs->tf_eflags;
564 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
565 	frame.sf_sc.sc_ss     = regs->tf_ss;
566 	frame.sf_sc.sc_err    = regs->tf_err;
567 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
568 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
569 
570 	frame.sf_extramask[0] = lmask.__mask;
571 
572 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
573 		/*
574 		 * Process has trashed its stack; give it an illegal
575 		 * instruction to halt it in its tracks.
576 		 */
577 		PROC_LOCK(p);
578 		sigexit(td, SIGILL);
579 	}
580 
581 	/* Build context to run handler in. */
582 	regs->tf_esp = (int)fp;
583 	regs->tf_eip = __kernel_sigreturn;
584 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
585 	regs->tf_cs = _ucodesel;
586 	regs->tf_ds = _udatasel;
587 	regs->tf_es = _udatasel;
588 	regs->tf_fs = _udatasel;
589 	regs->tf_ss = _udatasel;
590 	PROC_LOCK(p);
591 	mtx_lock(&psp->ps_mtx);
592 }
593 
594 /*
595  * System call to cleanup state after a signal
596  * has been taken.  Reset signal mask and
597  * stack state from context left by sendsig (above).
598  * Return to previous pc and psl as specified by
599  * context left by sendsig. Check carefully to
600  * make sure that the user has not modified the
601  * psl to gain improper privileges or to cause
602  * a machine fault.
603  */
604 int
605 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
606 {
607 	struct l_sigframe frame;
608 	struct trapframe *regs;
609 	l_sigset_t lmask;
610 	sigset_t bmask;
611 	int eflags;
612 	ksiginfo_t ksi;
613 
614 	regs = td->td_frame;
615 
616 	/*
617 	 * The trampoline code hands us the sigframe.
618 	 * It is unsafe to keep track of it ourselves, in the event that a
619 	 * program jumps out of a signal handler.
620 	 */
621 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
622 		return (EFAULT);
623 
624 	/* Check for security violations. */
625 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
626 	eflags = frame.sf_sc.sc_eflags;
627 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
628 		return (EINVAL);
629 
630 	/*
631 	 * Don't allow users to load a valid privileged %cs.  Let the
632 	 * hardware check for invalid selectors, excess privilege in
633 	 * other selectors, invalid %eip's and invalid %esp's.
634 	 */
635 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
636 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
637 		ksiginfo_init_trap(&ksi);
638 		ksi.ksi_signo = SIGBUS;
639 		ksi.ksi_code = BUS_OBJERR;
640 		ksi.ksi_trapno = T_PROTFLT;
641 		ksi.ksi_addr = (void *)regs->tf_eip;
642 		trapsignal(td, &ksi);
643 		return (EINVAL);
644 	}
645 
646 	lmask.__mask = frame.sf_sc.sc_mask;
647 	linux_to_bsd_sigset(&lmask, &bmask);
648 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
649 
650 	/* Restore signal context. */
651 	/* %gs was restored by the trampoline. */
652 	regs->tf_fs     = frame.sf_sc.sc_fs;
653 	regs->tf_es     = frame.sf_sc.sc_es;
654 	regs->tf_ds     = frame.sf_sc.sc_ds;
655 	regs->tf_edi    = frame.sf_sc.sc_edi;
656 	regs->tf_esi    = frame.sf_sc.sc_esi;
657 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
658 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
659 	regs->tf_edx    = frame.sf_sc.sc_edx;
660 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
661 	regs->tf_eax    = frame.sf_sc.sc_eax;
662 	regs->tf_eip    = frame.sf_sc.sc_eip;
663 	regs->tf_cs     = frame.sf_sc.sc_cs;
664 	regs->tf_eflags = eflags;
665 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
666 	regs->tf_ss     = frame.sf_sc.sc_ss;
667 
668 	return (EJUSTRETURN);
669 }
670 
671 /*
672  * System call to cleanup state after a signal
673  * has been taken.  Reset signal mask and
674  * stack state from context left by rt_sendsig (above).
675  * Return to previous pc and psl as specified by
676  * context left by sendsig. Check carefully to
677  * make sure that the user has not modified the
678  * psl to gain improper privileges or to cause
679  * a machine fault.
680  */
681 int
682 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
683 {
684 	struct l_ucontext uc;
685 	struct l_sigcontext *context;
686 	sigset_t bmask;
687 	l_stack_t *lss;
688 	stack_t ss;
689 	struct trapframe *regs;
690 	int eflags;
691 	ksiginfo_t ksi;
692 
693 	regs = td->td_frame;
694 
695 	/*
696 	 * The trampoline code hands us the ucontext.
697 	 * It is unsafe to keep track of it ourselves, in the event that a
698 	 * program jumps out of a signal handler.
699 	 */
700 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
701 		return (EFAULT);
702 
703 	context = &uc.uc_mcontext;
704 
705 	/* Check for security violations. */
706 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
707 	eflags = context->sc_eflags;
708 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
709 		return (EINVAL);
710 
711 	/*
712 	 * Don't allow users to load a valid privileged %cs.  Let the
713 	 * hardware check for invalid selectors, excess privilege in
714 	 * other selectors, invalid %eip's and invalid %esp's.
715 	 */
716 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
717 	if (!CS_SECURE(context->sc_cs)) {
718 		ksiginfo_init_trap(&ksi);
719 		ksi.ksi_signo = SIGBUS;
720 		ksi.ksi_code = BUS_OBJERR;
721 		ksi.ksi_trapno = T_PROTFLT;
722 		ksi.ksi_addr = (void *)regs->tf_eip;
723 		trapsignal(td, &ksi);
724 		return (EINVAL);
725 	}
726 
727 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
728 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
729 
730 	/* Restore signal context. */
731 	/* %gs was restored by the trampoline. */
732 	regs->tf_fs     = context->sc_fs;
733 	regs->tf_es     = context->sc_es;
734 	regs->tf_ds     = context->sc_ds;
735 	regs->tf_edi    = context->sc_edi;
736 	regs->tf_esi    = context->sc_esi;
737 	regs->tf_ebp    = context->sc_ebp;
738 	regs->tf_ebx    = context->sc_ebx;
739 	regs->tf_edx    = context->sc_edx;
740 	regs->tf_ecx    = context->sc_ecx;
741 	regs->tf_eax    = context->sc_eax;
742 	regs->tf_eip    = context->sc_eip;
743 	regs->tf_cs     = context->sc_cs;
744 	regs->tf_eflags = eflags;
745 	regs->tf_esp    = context->sc_esp_at_signal;
746 	regs->tf_ss     = context->sc_ss;
747 
748 	/* Call sigaltstack & ignore results. */
749 	lss = &uc.uc_stack;
750 	ss.ss_sp = PTRIN(lss->ss_sp);
751 	ss.ss_size = lss->ss_size;
752 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
753 
754 	(void)kern_sigaltstack(td, &ss, NULL);
755 
756 	return (EJUSTRETURN);
757 }
758 
759 static int
760 linux_fetch_syscall_args(struct thread *td)
761 {
762 	struct proc *p;
763 	struct trapframe *frame;
764 	struct syscall_args *sa;
765 
766 	p = td->td_proc;
767 	frame = td->td_frame;
768 	sa = &td->td_sa;
769 
770 	sa->code = frame->tf_eax;
771 	sa->original_code = sa->code;
772 	sa->args[0] = frame->tf_ebx;
773 	sa->args[1] = frame->tf_ecx;
774 	sa->args[2] = frame->tf_edx;
775 	sa->args[3] = frame->tf_esi;
776 	sa->args[4] = frame->tf_edi;
777 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
778 
779 	if (sa->code >= p->p_sysent->sv_size)
780 		/* nosys */
781 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
782 	else
783 		sa->callp = &p->p_sysent->sv_table[sa->code];
784 
785 	td->td_retval[0] = 0;
786 	td->td_retval[1] = frame->tf_edx;
787 
788 	return (0);
789 }
790 
791 static void
792 linux_set_syscall_retval(struct thread *td, int error)
793 {
794 	struct trapframe *frame = td->td_frame;
795 
796 	cpu_set_syscall_retval(td, error);
797 
798 	if (__predict_false(error != 0)) {
799 		if (error != ERESTART && error != EJUSTRETURN)
800 			frame->tf_eax = bsd_to_linux_errno(error);
801 	}
802 }
803 
804 static void
805 linux_set_fork_retval(struct thread *td)
806 {
807 	struct trapframe *frame = td->td_frame;
808 
809 	frame->tf_eax = 0;
810 }
811 
812 /*
813  * exec_setregs may initialize some registers differently than Linux
814  * does, thus potentially confusing Linux binaries. If necessary, we
815  * override the exec_setregs default(s) here.
816  */
817 static void
818 linux_exec_setregs(struct thread *td, struct image_params *imgp,
819     uintptr_t stack)
820 {
821 	struct pcb *pcb = td->td_pcb;
822 
823 	exec_setregs(td, imgp, stack);
824 
825 	/* Linux sets %gs to 0, we default to _udatasel. */
826 	pcb->pcb_gs = 0;
827 	load_gs(0);
828 
829 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
830 }
831 
832 struct sysentvec linux_sysvec = {
833 	.sv_size	= LINUX_SYS_MAXSYSCALL,
834 	.sv_table	= linux_sysent,
835 	.sv_transtrap	= linux_translate_traps,
836 	.sv_fixup	= linux_fixup,
837 	.sv_sendsig	= linux_sendsig,
838 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
839 	.sv_szsigcode	= &linux_szsigcode,
840 	.sv_name	= "Linux a.out",
841 	.sv_coredump	= NULL,
842 	.sv_imgact_try	= linux_exec_imgact_try,
843 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
844 	.sv_minuser	= VM_MIN_ADDRESS,
845 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
846 	.sv_usrstack	= LINUX_USRSTACK,
847 	.sv_psstrings	= PS_STRINGS,
848 	.sv_psstringssz	= sizeof(struct ps_strings),
849 	.sv_stackprot	= VM_PROT_ALL,
850 	.sv_copyout_strings = exec_copyout_strings,
851 	.sv_setregs	= linux_exec_setregs,
852 	.sv_fixlimit	= NULL,
853 	.sv_maxssiz	= NULL,
854 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32 |
855 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
856 	.sv_set_syscall_retval = linux_set_syscall_retval,
857 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
858 	.sv_syscallnames = NULL,
859 	.sv_schedtail	= linux_schedtail,
860 	.sv_thread_detach = linux_thread_detach,
861 	.sv_trap	= NULL,
862 	.sv_onexec	= linux_on_exec_vmspace,
863 	.sv_onexit	= linux_on_exit,
864 	.sv_ontdexit	= linux_thread_dtor,
865 	.sv_setid_allowed = &linux_setid_allowed_query,
866 	.sv_set_fork_retval = linux_set_fork_retval,
867 };
868 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
869 
870 struct sysentvec elf_linux_sysvec = {
871 	.sv_size	= LINUX_SYS_MAXSYSCALL,
872 	.sv_table	= linux_sysent,
873 	.sv_transtrap	= linux_translate_traps,
874 	.sv_fixup	= linux_fixup_elf,
875 	.sv_sendsig	= linux_sendsig,
876 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
877 	.sv_szsigcode	= &linux_szsigcode,
878 	.sv_name	= "Linux ELF32",
879 	.sv_coredump	= elf32_coredump,
880 	.sv_elf_core_osabi = ELFOSABI_FREEBSD,
881 	.sv_elf_core_abi_vendor = FREEBSD_ABI_VENDOR,
882 	.sv_elf_core_prepare_notes = elf32_prepare_notes,
883 	.sv_imgact_try	= linux_exec_imgact_try,
884 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
885 	.sv_minuser	= VM_MIN_ADDRESS,
886 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
887 	.sv_usrstack	= LINUX_USRSTACK,
888 	.sv_psstrings	= LINUX_PS_STRINGS,
889 	.sv_psstringssz	= sizeof(struct ps_strings),
890 	.sv_stackprot	= VM_PROT_ALL,
891 	.sv_copyout_auxargs = linux_copyout_auxargs,
892 	.sv_copyout_strings = linux_copyout_strings,
893 	.sv_setregs	= linux_exec_setregs,
894 	.sv_fixlimit	= NULL,
895 	.sv_maxssiz	= NULL,
896 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP |
897 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
898 	.sv_set_syscall_retval = linux_set_syscall_retval,
899 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
900 	.sv_syscallnames = NULL,
901 	.sv_shared_page_base = LINUX_SHAREDPAGE,
902 	.sv_shared_page_len = PAGE_SIZE,
903 	.sv_schedtail	= linux_schedtail,
904 	.sv_thread_detach = linux_thread_detach,
905 	.sv_trap	= NULL,
906 	.sv_onexec	= linux_on_exec_vmspace,
907 	.sv_onexit	= linux_on_exit,
908 	.sv_ontdexit	= linux_thread_dtor,
909 	.sv_setid_allowed = &linux_setid_allowed_query,
910 	.sv_set_fork_retval = linux_set_fork_retval,
911 };
912 
913 static int
914 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
915 {
916 	int error = 0;
917 
918 	if (SV_PROC_FLAG(p, SV_SHP) != 0)
919 		error = linux_map_vdso(p, linux_vdso_obj,
920 		    linux_vdso_base, LINUX_VDSOPAGE_SIZE, imgp);
921 	if (error == 0)
922 		linux_on_exec(p, imgp);
923 	return (error);
924 }
925 
926 /*
927  * linux_vdso_install() and linux_exec_sysvec_init() must be called
928  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
929  */
930 static void
931 linux_exec_sysvec_init(void *param)
932 {
933 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
934 	struct sysentvec *sv;
935 	ptrdiff_t tkoff;
936 
937 	sv = param;
938 	/* Fill timekeep_base */
939 	exec_sysvec_init(sv);
940 
941 	tkoff = kern_timekeep_base - linux_vdso_base;
942 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
943 	*ktimekeep_base = sv->sv_timekeep_base;
944 
945 	tkoff = kern_tsc_selector - linux_vdso_base;
946 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
947 	*ktsc_selector = linux_vdso_tsc_selector_idx();
948 	if (bootverbose)
949 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
950 }
951 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
952     linux_exec_sysvec_init, &elf_linux_sysvec);
953 
954 static void
955 linux_vdso_install(const void *param)
956 {
957 	char *vdso_start = &_binary_linux_vdso_so_o_start;
958 	char *vdso_end = &_binary_linux_vdso_so_o_end;
959 
960 	linux_szsigcode = vdso_end - vdso_start;
961 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
962 
963 	linux_vdso_base = LINUX_VDSOPAGE;
964 
965 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
966 
967 	linux_vdso_obj = __elfN(linux_shared_page_init)
968 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
969 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
970 
971 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
972 }
973 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
974     linux_vdso_install, NULL);
975 
976 static void
977 linux_vdso_deinstall(const void *param)
978 {
979 
980 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
981 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
982 }
983 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
984     linux_vdso_deinstall, NULL);
985 
986 static void
987 linux_vdso_reloc(char *mapping, Elf_Addr offset)
988 {
989 	const Elf_Shdr *shdr;
990 	const Elf_Rel *rel;
991 	const Elf_Ehdr *ehdr;
992 	Elf_Addr *where;
993 	Elf_Size rtype, symidx;
994 	Elf_Addr addr, addend;
995 	int i, relcnt;
996 
997 	MPASS(offset != 0);
998 
999 	relcnt = 0;
1000 	ehdr = (const Elf_Ehdr *)mapping;
1001 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
1002 	for (i = 0; i < ehdr->e_shnum; i++)
1003 	{
1004 		switch (shdr[i].sh_type) {
1005 		case SHT_REL:
1006 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
1007 			relcnt = shdr[i].sh_size / sizeof(*rel);
1008 			break;
1009 		case SHT_RELA:
1010 			printf("Linux i386 vDSO: unexpected Rela section\n");
1011 			break;
1012 		}
1013 	}
1014 
1015 	for (i = 0; i < relcnt; i++, rel++) {
1016 		where = (Elf_Addr *)(mapping + rel->r_offset);
1017 		addend = *where;
1018 		rtype = ELF_R_TYPE(rel->r_info);
1019 		symidx = ELF_R_SYM(rel->r_info);
1020 
1021 		switch (rtype) {
1022 		case R_386_NONE:	/* none */
1023 			break;
1024 
1025 		case R_386_RELATIVE:	/* B + A */
1026 			addr = (Elf_Addr)PTROUT(offset + addend);
1027 			if (*where != addr)
1028 				*where = addr;
1029 			break;
1030 
1031 		case R_386_IRELATIVE:
1032 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
1033 			    "symbol index %d\n", symidx);
1034 			break;
1035 		default:
1036 			printf("Linux i386 vDSO: unexpected relocation type %d, "
1037 			    "symbol index %d\n", rtype, symidx);
1038 		}
1039 	}
1040 }
1041 
1042 static char GNU_ABI_VENDOR[] = "GNU";
1043 static int GNULINUX_ABI_DESC = 0;
1044 
1045 static bool
1046 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
1047 {
1048 	const Elf32_Word *desc;
1049 	uintptr_t p;
1050 
1051 	p = (uintptr_t)(note + 1);
1052 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1053 
1054 	desc = (const Elf32_Word *)p;
1055 	if (desc[0] != GNULINUX_ABI_DESC)
1056 		return (false);
1057 
1058 	/*
1059 	 * For Linux we encode osrel using the Linux convention of
1060 	 * 	(version << 16) | (major << 8) | (minor)
1061 	 * See macro in linux_mib.h
1062 	 */
1063 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
1064 
1065 	return (true);
1066 }
1067 
1068 static Elf_Brandnote linux_brandnote = {
1069 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
1070 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
1071 	.hdr.n_type	= 1,
1072 	.vendor		= GNU_ABI_VENDOR,
1073 	.flags		= BN_TRANSLATE_OSREL,
1074 	.trans_osrel	= linux_trans_osrel
1075 };
1076 
1077 static Elf32_Brandinfo linux_brand = {
1078 	.brand		= ELFOSABI_LINUX,
1079 	.machine	= EM_386,
1080 	.compat_3_brand	= "Linux",
1081 	.emul_path	= linux_emul_path,
1082 	.interp_path	= "/lib/ld-linux.so.1",
1083 	.sysvec		= &elf_linux_sysvec,
1084 	.interp_newpath	= NULL,
1085 	.brand_note	= &linux_brandnote,
1086 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1087 };
1088 
1089 static Elf32_Brandinfo linux_glibc2brand = {
1090 	.brand		= ELFOSABI_LINUX,
1091 	.machine	= EM_386,
1092 	.compat_3_brand	= "Linux",
1093 	.emul_path	= linux_emul_path,
1094 	.interp_path	= "/lib/ld-linux.so.2",
1095 	.sysvec		= &elf_linux_sysvec,
1096 	.interp_newpath	= NULL,
1097 	.brand_note	= &linux_brandnote,
1098 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1099 };
1100 
1101 static Elf32_Brandinfo linux_muslbrand = {
1102 	.brand		= ELFOSABI_LINUX,
1103 	.machine	= EM_386,
1104 	.compat_3_brand	= "Linux",
1105 	.emul_path	= linux_emul_path,
1106 	.interp_path	= "/lib/ld-musl-i386.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 			    LINUX_BI_FUTEX_REQUEUE
1112 };
1113 
1114 Elf32_Brandinfo *linux_brandlist[] = {
1115 	&linux_brand,
1116 	&linux_glibc2brand,
1117 	&linux_muslbrand,
1118 	NULL
1119 };
1120 
1121 static int
1122 linux_elf_modevent(module_t mod, int type, void *data)
1123 {
1124 	Elf32_Brandinfo **brandinfo;
1125 	int error;
1126 	struct linux_ioctl_handler **lihp;
1127 
1128 	error = 0;
1129 
1130 	switch(type) {
1131 	case MOD_LOAD:
1132 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1133 		     ++brandinfo)
1134 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1135 				error = EINVAL;
1136 		if (error == 0) {
1137 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1138 				linux_ioctl_register_handler(*lihp);
1139 			linux_dev_shm_create();
1140 			linux_osd_jail_register();
1141 			stclohz = (stathz ? stathz : hz);
1142 			if (bootverbose)
1143 				printf("Linux ELF exec handler installed\n");
1144 		} else
1145 			printf("cannot insert Linux ELF brand handler\n");
1146 		break;
1147 	case MOD_UNLOAD:
1148 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1149 		     ++brandinfo)
1150 			if (elf32_brand_inuse(*brandinfo))
1151 				error = EBUSY;
1152 		if (error == 0) {
1153 			for (brandinfo = &linux_brandlist[0];
1154 			     *brandinfo != NULL; ++brandinfo)
1155 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1156 					error = EINVAL;
1157 		}
1158 		if (error == 0) {
1159 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1160 				linux_ioctl_unregister_handler(*lihp);
1161 			linux_dev_shm_destroy();
1162 			linux_osd_jail_deregister();
1163 			if (bootverbose)
1164 				printf("Linux ELF exec handler removed\n");
1165 		} else
1166 			printf("Could not deinstall ELF interpreter entry\n");
1167 		break;
1168 	default:
1169 		return (EOPNOTSUPP);
1170 	}
1171 	return (error);
1172 }
1173 
1174 static moduledata_t linux_elf_mod = {
1175 	"linuxelf",
1176 	linux_elf_modevent,
1177 	0
1178 };
1179 
1180 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1181 FEATURE(linux, "Linux 32bit support");
1182