xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision 2f68cdb94454035989fa370bcb68db2f4e96b1ed)
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_futex.h>
71 #include <compat/linux/linux_ioctl.h>
72 #include <compat/linux/linux_mib.h>
73 #include <compat/linux/linux_misc.h>
74 #include <compat/linux/linux_signal.h>
75 #include <compat/linux/linux_util.h>
76 #include <compat/linux/linux_vdso.h>
77 
78 MODULE_VERSION(linux, 1);
79 
80 #if BYTE_ORDER == LITTLE_ENDIAN
81 #define SHELLMAGIC      0x2123 /* #! */
82 #else
83 #define SHELLMAGIC      0x2321
84 #endif
85 
86 #if defined(DEBUG)
87 SYSCTL_PROC(_compat_linux, OID_AUTO, debug, CTLTYPE_STRING | CTLFLAG_RW, 0, 0,
88     linux_sysctl_debug, "A", "Linux debugging control");
89 #endif
90 
91 /*
92  * Allow the sendsig functions to use the ldebug() facility even though they
93  * are not syscalls themselves.  Map them to syscall 0.  This is slightly less
94  * bogus than using ldebug(sigreturn).
95  */
96 #define	LINUX_SYS_linux_rt_sendsig	0
97 #define	LINUX_SYS_linux_sendsig		0
98 
99 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
100 
101 static int linux_szsigcode;
102 static vm_object_t linux_shared_page_obj;
103 static char *linux_shared_page_mapping;
104 extern char _binary_linux_locore_o_start;
105 extern char _binary_linux_locore_o_end;
106 
107 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
108 
109 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
110 
111 static int	linux_fixup(register_t **stack_base,
112 		    struct image_params *iparams);
113 static int	linux_fixup_elf(register_t **stack_base,
114 		    struct image_params *iparams);
115 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
116 static int	linux_exec_imgact_try(struct image_params *iparams);
117 static void	linux_exec_setregs(struct thread *td,
118 		    struct image_params *imgp, u_long stack);
119 static register_t *linux_copyout_strings(struct image_params *imgp);
120 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
121 static void	linux_vdso_install(void *param);
122 static void	linux_vdso_deinstall(void *param);
123 
124 static int linux_szplatform;
125 const char *linux_kplatform;
126 
127 static eventhandler_tag linux_exit_tag;
128 static eventhandler_tag linux_exec_tag;
129 static eventhandler_tag linux_thread_dtor_tag;
130 
131 #define LINUX_T_UNKNOWN  255
132 static int _bsd_to_linux_trapcode[] = {
133 	LINUX_T_UNKNOWN,	/* 0 */
134 	6,			/* 1  T_PRIVINFLT */
135 	LINUX_T_UNKNOWN,	/* 2 */
136 	3,			/* 3  T_BPTFLT */
137 	LINUX_T_UNKNOWN,	/* 4 */
138 	LINUX_T_UNKNOWN,	/* 5 */
139 	16,			/* 6  T_ARITHTRAP */
140 	254,			/* 7  T_ASTFLT */
141 	LINUX_T_UNKNOWN,	/* 8 */
142 	13,			/* 9  T_PROTFLT */
143 	1,			/* 10 T_TRCTRAP */
144 	LINUX_T_UNKNOWN,	/* 11 */
145 	14,			/* 12 T_PAGEFLT */
146 	LINUX_T_UNKNOWN,	/* 13 */
147 	17,			/* 14 T_ALIGNFLT */
148 	LINUX_T_UNKNOWN,	/* 15 */
149 	LINUX_T_UNKNOWN,	/* 16 */
150 	LINUX_T_UNKNOWN,	/* 17 */
151 	0,			/* 18 T_DIVIDE */
152 	2,			/* 19 T_NMI */
153 	4,			/* 20 T_OFLOW */
154 	5,			/* 21 T_BOUND */
155 	7,			/* 22 T_DNA */
156 	8,			/* 23 T_DOUBLEFLT */
157 	9,			/* 24 T_FPOPFLT */
158 	10,			/* 25 T_TSSFLT */
159 	11,			/* 26 T_SEGNPFLT */
160 	12,			/* 27 T_STKFLT */
161 	18,			/* 28 T_MCHK */
162 	19,			/* 29 T_XMMFLT */
163 	15			/* 30 T_RESERVED */
164 };
165 #define bsd_to_linux_trapcode(code) \
166     ((code)<nitems(_bsd_to_linux_trapcode)? \
167      _bsd_to_linux_trapcode[(code)]: \
168      LINUX_T_UNKNOWN)
169 
170 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
171 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
172 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
173 
174 /*
175  * If FreeBSD & Linux have a difference of opinion about what a trap
176  * means, deal with it here.
177  *
178  * MPSAFE
179  */
180 static int
181 linux_translate_traps(int signal, int trap_code)
182 {
183 	if (signal != SIGBUS)
184 		return (signal);
185 	switch (trap_code) {
186 	case T_PROTFLT:
187 	case T_TSSFLT:
188 	case T_DOUBLEFLT:
189 	case T_PAGEFLT:
190 		return (SIGSEGV);
191 	default:
192 		return (signal);
193 	}
194 }
195 
196 static int
197 linux_fixup(register_t **stack_base, struct image_params *imgp)
198 {
199 	register_t *argv, *envp;
200 
201 	argv = *stack_base;
202 	envp = *stack_base + (imgp->args->argc + 1);
203 	(*stack_base)--;
204 	suword(*stack_base, (intptr_t)(void *)envp);
205 	(*stack_base)--;
206 	suword(*stack_base, (intptr_t)(void *)argv);
207 	(*stack_base)--;
208 	suword(*stack_base, imgp->args->argc);
209 	return (0);
210 }
211 
212 static int
213 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
214 {
215 	struct proc *p;
216 	Elf32_Auxargs *args;
217 	Elf32_Addr *uplatform;
218 	struct ps_strings *arginfo;
219 	register_t *pos;
220 	int issetugid;
221 
222 	KASSERT(curthread->td_proc == imgp->proc,
223 	    ("unsafe linux_fixup_elf(), should be curproc"));
224 
225 	p = imgp->proc;
226 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
227 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
228 	uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
229 	args = (Elf32_Auxargs *)imgp->auxargs;
230 	pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
231 
232 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
233 	    imgp->proc->p_sysent->sv_shared_page_base);
234 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
235 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
236 
237 	/*
238 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
239 	 * as it has appeared in the 2.4.0-rc7 first time.
240 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
241 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
242 	 * is not present.
243 	 * Also see linux_times() implementation.
244 	 */
245 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
246 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
247 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
248 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
249 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
250 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
251 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
252 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
253 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
254 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
255 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
256 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
257 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
258 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
259 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
260 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
261 	if (imgp->execpathp != 0)
262 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
263 	if (args->execfd != -1)
264 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
265 	AUXARGS_ENTRY(pos, AT_NULL, 0);
266 
267 	free(imgp->auxargs, M_TEMP);
268 	imgp->auxargs = NULL;
269 
270 	(*stack_base)--;
271 	suword(*stack_base, (register_t)imgp->args->argc);
272 	return (0);
273 }
274 
275 /*
276  * Copied from kern/kern_exec.c
277  */
278 static register_t *
279 linux_copyout_strings(struct image_params *imgp)
280 {
281 	int argc, envc;
282 	char **vectp;
283 	char *stringp, *destp;
284 	register_t *stack_base;
285 	struct ps_strings *arginfo;
286 	char canary[LINUX_AT_RANDOM_LEN];
287 	size_t execpath_len;
288 	struct proc *p;
289 
290 	/* Calculate string base and vector table pointers. */
291 	p = imgp->proc;
292 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
293 		execpath_len = strlen(imgp->execpath) + 1;
294 	else
295 		execpath_len = 0;
296 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
297 	destp = (caddr_t)arginfo - SPARE_USRSPACE - linux_szplatform -
298 	    roundup(sizeof(canary), sizeof(char *)) -
299 	    roundup(execpath_len, sizeof(char *)) -
300 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
301 
302 	/* Install LINUX_PLATFORM. */
303 	copyout(linux_kplatform, ((caddr_t)arginfo - linux_szplatform),
304 	    linux_szplatform);
305 
306 	if (execpath_len != 0) {
307 		imgp->execpathp = (uintptr_t)arginfo -
308 		linux_szplatform - execpath_len;
309 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
310 	}
311 
312 	/* Prepare the canary for SSP. */
313 	arc4rand(canary, sizeof(canary), 0);
314 	imgp->canary = (uintptr_t)arginfo - linux_szplatform -
315 	    roundup(execpath_len, sizeof(char *)) -
316 	    roundup(sizeof(canary), sizeof(char *));
317 	copyout(canary, (void *)imgp->canary, sizeof(canary));
318 
319 	/* If we have a valid auxargs ptr, prepare some room on the stack. */
320 	if (imgp->auxargs) {
321 		/*
322 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
323 		 * lower compatibility.
324 		 */
325 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
326 		    (LINUX_AT_COUNT * 2);
327 		/*
328 		 * The '+ 2' is for the null pointers at the end of each of
329 		 * the arg and env vector sets,and imgp->auxarg_size is room
330 		 * for argument of Runtime loader.
331 		 */
332 		vectp = (char **)(destp - (imgp->args->argc +
333 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
334 	} else {
335 		/*
336 		 * The '+ 2' is for the null pointers at the end of each of
337 		 * the arg and env vector sets
338 		 */
339 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
340 		    sizeof(char *));
341 	}
342 
343 	/* vectp also becomes our initial stack base. */
344 	stack_base = (register_t *)vectp;
345 
346 	stringp = imgp->args->begin_argv;
347 	argc = imgp->args->argc;
348 	envc = imgp->args->envc;
349 
350 	/* Copy out strings - arguments and environment. */
351 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
352 
353 	/* Fill in "ps_strings" struct for ps, w, etc. */
354 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
355 	suword(&arginfo->ps_nargvstr, argc);
356 
357 	/* Fill in argument portion of vector table. */
358 	for (; argc > 0; --argc) {
359 		suword(vectp++, (long)(intptr_t)destp);
360 		while (*stringp++ != 0)
361 			destp++;
362 		destp++;
363 	}
364 
365 	/* A null vector table pointer separates the argp's from the envp's. */
366 	suword(vectp++, 0);
367 
368 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
369 	suword(&arginfo->ps_nenvstr, envc);
370 
371 	/* Fill in environment portion of vector table. */
372 	for (; envc > 0; --envc) {
373 		suword(vectp++, (long)(intptr_t)destp);
374 		while (*stringp++ != 0)
375 			destp++;
376 		destp++;
377 	}
378 
379 	/* The end of the vector table is a null pointer. */
380 	suword(vectp, 0);
381 
382 	return (stack_base);
383 }
384 
385 static void
386 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
387 {
388 	struct thread *td = curthread;
389 	struct proc *p = td->td_proc;
390 	struct sigacts *psp;
391 	struct trapframe *regs;
392 	struct l_rt_sigframe *fp, frame;
393 	int sig, code;
394 	int oonstack;
395 
396 	sig = ksi->ksi_signo;
397 	code = ksi->ksi_code;
398 	PROC_LOCK_ASSERT(p, MA_OWNED);
399 	psp = p->p_sigacts;
400 	mtx_assert(&psp->ps_mtx, MA_OWNED);
401 	regs = td->td_frame;
402 	oonstack = sigonstack(regs->tf_esp);
403 
404 #ifdef DEBUG
405 	if (ldebug(rt_sendsig))
406 		printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
407 		    catcher, sig, (void*)mask, code);
408 #endif
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 #ifdef DEBUG
466 	if (ldebug(rt_sendsig))
467 		printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
468 		    frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
469 		    td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
470 #endif
471 
472 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
473 		/*
474 		 * Process has trashed its stack; give it an illegal
475 		 * instruction to halt it in its tracks.
476 		 */
477 #ifdef DEBUG
478 		if (ldebug(rt_sendsig))
479 			printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
480 			    fp, oonstack);
481 #endif
482 		PROC_LOCK(p);
483 		sigexit(td, SIGILL);
484 	}
485 
486 	/* Build context to run handler in. */
487 	regs->tf_esp = (int)fp;
488 	regs->tf_eip = linux_rt_sigcode;
489 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
490 	regs->tf_cs = _ucodesel;
491 	regs->tf_ds = _udatasel;
492 	regs->tf_es = _udatasel;
493 	regs->tf_fs = _udatasel;
494 	regs->tf_ss = _udatasel;
495 	PROC_LOCK(p);
496 	mtx_lock(&psp->ps_mtx);
497 }
498 
499 
500 /*
501  * Send an interrupt to process.
502  *
503  * Stack is set up to allow sigcode stored
504  * in u. to call routine, followed by kcall
505  * to sigreturn routine below.  After sigreturn
506  * resets the signal mask, the stack, and the
507  * frame pointer, it returns to the user
508  * specified pc, psl.
509  */
510 static void
511 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
512 {
513 	struct thread *td = curthread;
514 	struct proc *p = td->td_proc;
515 	struct sigacts *psp;
516 	struct trapframe *regs;
517 	struct l_sigframe *fp, frame;
518 	l_sigset_t lmask;
519 	int sig, code;
520 	int oonstack;
521 
522 	PROC_LOCK_ASSERT(p, MA_OWNED);
523 	psp = p->p_sigacts;
524 	sig = ksi->ksi_signo;
525 	code = ksi->ksi_code;
526 	mtx_assert(&psp->ps_mtx, MA_OWNED);
527 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
528 		/* Signal handler installed with SA_SIGINFO. */
529 		linux_rt_sendsig(catcher, ksi, mask);
530 		return;
531 	}
532 	regs = td->td_frame;
533 	oonstack = sigonstack(regs->tf_esp);
534 
535 #ifdef DEBUG
536 	if (ldebug(sendsig))
537 		printf(ARGS(sendsig, "%p, %d, %p, %u"),
538 		    catcher, sig, (void*)mask, code);
539 #endif
540 
541 	/* Allocate space for the signal handler context. */
542 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
543 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
544 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
545 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
546 	} else
547 		fp = (struct l_sigframe *)regs->tf_esp - 1;
548 	mtx_unlock(&psp->ps_mtx);
549 	PROC_UNLOCK(p);
550 
551 	/* Build the argument list for the signal handler. */
552 	sig = bsd_to_linux_signal(sig);
553 
554 	bzero(&frame, sizeof(frame));
555 
556 	frame.sf_handler = catcher;
557 	frame.sf_sig = sig;
558 
559 	bsd_to_linux_sigset(mask, &lmask);
560 
561 	/* Build the signal context to be used by sigreturn. */
562 	frame.sf_sc.sc_mask   = lmask.__mask;
563 	frame.sf_sc.sc_gs     = rgs();
564 	frame.sf_sc.sc_fs     = regs->tf_fs;
565 	frame.sf_sc.sc_es     = regs->tf_es;
566 	frame.sf_sc.sc_ds     = regs->tf_ds;
567 	frame.sf_sc.sc_edi    = regs->tf_edi;
568 	frame.sf_sc.sc_esi    = regs->tf_esi;
569 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
570 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
571 	frame.sf_sc.sc_esp    = regs->tf_esp;
572 	frame.sf_sc.sc_edx    = regs->tf_edx;
573 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
574 	frame.sf_sc.sc_eax    = regs->tf_eax;
575 	frame.sf_sc.sc_eip    = regs->tf_eip;
576 	frame.sf_sc.sc_cs     = regs->tf_cs;
577 	frame.sf_sc.sc_eflags = regs->tf_eflags;
578 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
579 	frame.sf_sc.sc_ss     = regs->tf_ss;
580 	frame.sf_sc.sc_err    = regs->tf_err;
581 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
582 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
583 
584 	frame.sf_extramask[0] = lmask.__mask;
585 
586 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
587 		/*
588 		 * Process has trashed its stack; give it an illegal
589 		 * instruction to halt it in its tracks.
590 		 */
591 		PROC_LOCK(p);
592 		sigexit(td, SIGILL);
593 	}
594 
595 	/* Build context to run handler in. */
596 	regs->tf_esp = (int)fp;
597 	regs->tf_eip = linux_sigcode;
598 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
599 	regs->tf_cs = _ucodesel;
600 	regs->tf_ds = _udatasel;
601 	regs->tf_es = _udatasel;
602 	regs->tf_fs = _udatasel;
603 	regs->tf_ss = _udatasel;
604 	PROC_LOCK(p);
605 	mtx_lock(&psp->ps_mtx);
606 }
607 
608 /*
609  * System call to cleanup state after a signal
610  * has been taken.  Reset signal mask and
611  * stack state from context left by sendsig (above).
612  * Return to previous pc and psl as specified by
613  * context left by sendsig. Check carefully to
614  * make sure that the user has not modified the
615  * psl to gain improper privileges or to cause
616  * a machine fault.
617  */
618 int
619 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
620 {
621 	struct l_sigframe frame;
622 	struct trapframe *regs;
623 	l_sigset_t lmask;
624 	sigset_t bmask;
625 	int eflags;
626 	ksiginfo_t ksi;
627 
628 	regs = td->td_frame;
629 
630 #ifdef DEBUG
631 	if (ldebug(sigreturn))
632 		printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
633 #endif
634 	/*
635 	 * The trampoline code hands us the sigframe.
636 	 * It is unsafe to keep track of it ourselves, in the event that a
637 	 * program jumps out of a signal handler.
638 	 */
639 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
640 		return (EFAULT);
641 
642 	/* Check for security violations. */
643 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
644 	eflags = frame.sf_sc.sc_eflags;
645 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
646 		return (EINVAL);
647 
648 	/*
649 	 * Don't allow users to load a valid privileged %cs.  Let the
650 	 * hardware check for invalid selectors, excess privilege in
651 	 * other selectors, invalid %eip's and invalid %esp's.
652 	 */
653 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
654 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
655 		ksiginfo_init_trap(&ksi);
656 		ksi.ksi_signo = SIGBUS;
657 		ksi.ksi_code = BUS_OBJERR;
658 		ksi.ksi_trapno = T_PROTFLT;
659 		ksi.ksi_addr = (void *)regs->tf_eip;
660 		trapsignal(td, &ksi);
661 		return (EINVAL);
662 	}
663 
664 	lmask.__mask = frame.sf_sc.sc_mask;
665 	linux_to_bsd_sigset(&lmask, &bmask);
666 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
667 
668 	/* Restore signal context. */
669 	/* %gs was restored by the trampoline. */
670 	regs->tf_fs     = frame.sf_sc.sc_fs;
671 	regs->tf_es     = frame.sf_sc.sc_es;
672 	regs->tf_ds     = frame.sf_sc.sc_ds;
673 	regs->tf_edi    = frame.sf_sc.sc_edi;
674 	regs->tf_esi    = frame.sf_sc.sc_esi;
675 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
676 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
677 	regs->tf_edx    = frame.sf_sc.sc_edx;
678 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
679 	regs->tf_eax    = frame.sf_sc.sc_eax;
680 	regs->tf_eip    = frame.sf_sc.sc_eip;
681 	regs->tf_cs     = frame.sf_sc.sc_cs;
682 	regs->tf_eflags = eflags;
683 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
684 	regs->tf_ss     = frame.sf_sc.sc_ss;
685 
686 	return (EJUSTRETURN);
687 }
688 
689 /*
690  * System call to cleanup state after a signal
691  * has been taken.  Reset signal mask and
692  * stack state from context left by rt_sendsig (above).
693  * Return to previous pc and psl as specified by
694  * context left by sendsig. Check carefully to
695  * make sure that the user has not modified the
696  * psl to gain improper privileges or to cause
697  * a machine fault.
698  */
699 int
700 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
701 {
702 	struct l_ucontext uc;
703 	struct l_sigcontext *context;
704 	sigset_t bmask;
705 	l_stack_t *lss;
706 	stack_t ss;
707 	struct trapframe *regs;
708 	int eflags;
709 	ksiginfo_t ksi;
710 
711 	regs = td->td_frame;
712 
713 #ifdef DEBUG
714 	if (ldebug(rt_sigreturn))
715 		printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
716 #endif
717 	/*
718 	 * The trampoline code hands us the ucontext.
719 	 * It is unsafe to keep track of it ourselves, in the event that a
720 	 * program jumps out of a signal handler.
721 	 */
722 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
723 		return (EFAULT);
724 
725 	context = &uc.uc_mcontext;
726 
727 	/* Check for security violations. */
728 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
729 	eflags = context->sc_eflags;
730 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
731 		return (EINVAL);
732 
733 	/*
734 	 * Don't allow users to load a valid privileged %cs.  Let the
735 	 * hardware check for invalid selectors, excess privilege in
736 	 * other selectors, invalid %eip's and invalid %esp's.
737 	 */
738 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
739 	if (!CS_SECURE(context->sc_cs)) {
740 		ksiginfo_init_trap(&ksi);
741 		ksi.ksi_signo = SIGBUS;
742 		ksi.ksi_code = BUS_OBJERR;
743 		ksi.ksi_trapno = T_PROTFLT;
744 		ksi.ksi_addr = (void *)regs->tf_eip;
745 		trapsignal(td, &ksi);
746 		return (EINVAL);
747 	}
748 
749 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
750 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
751 
752 	/* Restore signal context. */
753 	/* %gs was restored by the trampoline. */
754 	regs->tf_fs     = context->sc_fs;
755 	regs->tf_es     = context->sc_es;
756 	regs->tf_ds     = context->sc_ds;
757 	regs->tf_edi    = context->sc_edi;
758 	regs->tf_esi    = context->sc_esi;
759 	regs->tf_ebp    = context->sc_ebp;
760 	regs->tf_ebx    = context->sc_ebx;
761 	regs->tf_edx    = context->sc_edx;
762 	regs->tf_ecx    = context->sc_ecx;
763 	regs->tf_eax    = context->sc_eax;
764 	regs->tf_eip    = context->sc_eip;
765 	regs->tf_cs     = context->sc_cs;
766 	regs->tf_eflags = eflags;
767 	regs->tf_esp    = context->sc_esp_at_signal;
768 	regs->tf_ss     = context->sc_ss;
769 
770 	/* Call sigaltstack & ignore results. */
771 	lss = &uc.uc_stack;
772 	ss.ss_sp = lss->ss_sp;
773 	ss.ss_size = lss->ss_size;
774 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
775 
776 #ifdef DEBUG
777 	if (ldebug(rt_sigreturn))
778 		printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
779 		    ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
780 #endif
781 	(void)kern_sigaltstack(td, &ss, NULL);
782 
783 	return (EJUSTRETURN);
784 }
785 
786 static int
787 linux_fetch_syscall_args(struct thread *td)
788 {
789 	struct proc *p;
790 	struct trapframe *frame;
791 	struct syscall_args *sa;
792 
793 	p = td->td_proc;
794 	frame = td->td_frame;
795 	sa = &td->td_sa;
796 
797 	sa->code = frame->tf_eax;
798 	sa->args[0] = frame->tf_ebx;
799 	sa->args[1] = frame->tf_ecx;
800 	sa->args[2] = frame->tf_edx;
801 	sa->args[3] = frame->tf_esi;
802 	sa->args[4] = frame->tf_edi;
803 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
804 
805 	if (sa->code >= p->p_sysent->sv_size)
806 		/* nosys */
807 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
808 	else
809 		sa->callp = &p->p_sysent->sv_table[sa->code];
810 	sa->narg = sa->callp->sy_narg;
811 
812 	td->td_retval[0] = 0;
813 	td->td_retval[1] = frame->tf_edx;
814 
815 	return (0);
816 }
817 
818 /*
819  * If a Linux binary is exec'ing something, try this image activator
820  * first.  We override standard shell script execution in order to
821  * be able to modify the interpreter path.  We only do this if a Linux
822  * binary is doing the exec, so we do not create an EXEC module for it.
823  */
824 static int
825 linux_exec_imgact_try(struct image_params *imgp)
826 {
827 	const char *head = (const char *)imgp->image_header;
828 	char *rpath;
829 	int error = -1;
830 
831 	/*
832 	 * The interpreter for shell scripts run from a Linux binary needs
833 	 * to be located in /compat/linux if possible in order to recursively
834 	 * maintain Linux path emulation.
835 	 */
836 	if (((const short *)head)[0] == SHELLMAGIC) {
837 		/*
838 		 * Run our normal shell image activator.  If it succeeds then
839 		 * attempt to use the alternate path for the interpreter.  If
840 		 * an alternate path is found, use our stringspace to store it.
841 		 */
842 		if ((error = exec_shell_imgact(imgp)) == 0) {
843 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
844 			    imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
845 			    AT_FDCWD);
846 			if (rpath != NULL)
847 				imgp->args->fname_buf =
848 				    imgp->interpreter_name = rpath;
849 		}
850 	}
851 	return (error);
852 }
853 
854 /*
855  * exec_setregs may initialize some registers differently than Linux
856  * does, thus potentially confusing Linux binaries. If necessary, we
857  * override the exec_setregs default(s) here.
858  */
859 static void
860 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
861 {
862 	struct pcb *pcb = td->td_pcb;
863 
864 	exec_setregs(td, imgp, stack);
865 
866 	/* Linux sets %gs to 0, we default to _udatasel. */
867 	pcb->pcb_gs = 0;
868 	load_gs(0);
869 
870 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
871 }
872 
873 static void
874 linux_get_machine(const char **dst)
875 {
876 
877 	switch (cpu_class) {
878 	case CPUCLASS_686:
879 		*dst = "i686";
880 		break;
881 	case CPUCLASS_586:
882 		*dst = "i586";
883 		break;
884 	case CPUCLASS_486:
885 		*dst = "i486";
886 		break;
887 	default:
888 		*dst = "i386";
889 	}
890 }
891 
892 struct sysentvec linux_sysvec = {
893 	.sv_size	= LINUX_SYS_MAXSYSCALL,
894 	.sv_table	= linux_sysent,
895 	.sv_mask	= 0,
896 	.sv_errsize	= ELAST + 1,
897 	.sv_errtbl	= linux_errtbl,
898 	.sv_transtrap	= linux_translate_traps,
899 	.sv_fixup	= linux_fixup,
900 	.sv_sendsig	= linux_sendsig,
901 	.sv_sigcode	= &_binary_linux_locore_o_start,
902 	.sv_szsigcode	= &linux_szsigcode,
903 	.sv_name	= "Linux a.out",
904 	.sv_coredump	= NULL,
905 	.sv_imgact_try	= linux_exec_imgact_try,
906 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
907 	.sv_pagesize	= PAGE_SIZE,
908 	.sv_minuser	= VM_MIN_ADDRESS,
909 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
910 	.sv_usrstack	= LINUX_USRSTACK,
911 	.sv_psstrings	= PS_STRINGS,
912 	.sv_stackprot	= VM_PROT_ALL,
913 	.sv_copyout_strings = exec_copyout_strings,
914 	.sv_setregs	= linux_exec_setregs,
915 	.sv_fixlimit	= NULL,
916 	.sv_maxssiz	= NULL,
917 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32,
918 	.sv_set_syscall_retval = cpu_set_syscall_retval,
919 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
920 	.sv_syscallnames = NULL,
921 	.sv_shared_page_base = LINUX_SHAREDPAGE,
922 	.sv_shared_page_len = PAGE_SIZE,
923 	.sv_schedtail	= linux_schedtail,
924 	.sv_thread_detach = linux_thread_detach,
925 	.sv_trap	= NULL,
926 };
927 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
928 
929 struct sysentvec elf_linux_sysvec = {
930 	.sv_size	= LINUX_SYS_MAXSYSCALL,
931 	.sv_table	= linux_sysent,
932 	.sv_mask	= 0,
933 	.sv_errsize	= ELAST + 1,
934 	.sv_errtbl	= linux_errtbl,
935 	.sv_transtrap	= linux_translate_traps,
936 	.sv_fixup	= linux_fixup_elf,
937 	.sv_sendsig	= linux_sendsig,
938 	.sv_sigcode	= &_binary_linux_locore_o_start,
939 	.sv_szsigcode	= &linux_szsigcode,
940 	.sv_name	= "Linux ELF",
941 	.sv_coredump	= elf32_coredump,
942 	.sv_imgact_try	= linux_exec_imgact_try,
943 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
944 	.sv_pagesize	= PAGE_SIZE,
945 	.sv_minuser	= VM_MIN_ADDRESS,
946 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
947 	.sv_usrstack	= LINUX_USRSTACK,
948 	.sv_psstrings	= LINUX_PS_STRINGS,
949 	.sv_stackprot	= VM_PROT_ALL,
950 	.sv_copyout_strings = linux_copyout_strings,
951 	.sv_setregs	= linux_exec_setregs,
952 	.sv_fixlimit	= NULL,
953 	.sv_maxssiz	= NULL,
954 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP,
955 	.sv_set_syscall_retval = cpu_set_syscall_retval,
956 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
957 	.sv_syscallnames = NULL,
958 	.sv_shared_page_base = LINUX_SHAREDPAGE,
959 	.sv_shared_page_len = PAGE_SIZE,
960 	.sv_schedtail	= linux_schedtail,
961 	.sv_thread_detach = linux_thread_detach,
962 	.sv_trap	= NULL,
963 };
964 
965 static void
966 linux_vdso_install(void *param)
967 {
968 
969 	linux_szsigcode = (&_binary_linux_locore_o_end -
970 	    &_binary_linux_locore_o_start);
971 
972 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
973 		panic("Linux invalid vdso size\n");
974 
975 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
976 
977 	linux_shared_page_obj = __elfN(linux_shared_page_init)
978 	    (&linux_shared_page_mapping);
979 
980 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
981 
982 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
983 	    linux_szsigcode);
984 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
985 }
986 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
987     linux_vdso_install, NULL);
988 
989 static void
990 linux_vdso_deinstall(void *param)
991 {
992 
993 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
994 }
995 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
996     linux_vdso_deinstall, NULL);
997 
998 static char GNU_ABI_VENDOR[] = "GNU";
999 static int GNULINUX_ABI_DESC = 0;
1000 
1001 static bool
1002 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
1003 {
1004 	const Elf32_Word *desc;
1005 	uintptr_t p;
1006 
1007 	p = (uintptr_t)(note + 1);
1008 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1009 
1010 	desc = (const Elf32_Word *)p;
1011 	if (desc[0] != GNULINUX_ABI_DESC)
1012 		return (false);
1013 
1014 	/*
1015 	 * For Linux we encode osrel as follows (see linux_mib.c):
1016 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
1017 	 */
1018 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1019 
1020 	return (true);
1021 }
1022 
1023 static Elf_Brandnote linux_brandnote = {
1024 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
1025 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
1026 	.hdr.n_type	= 1,
1027 	.vendor		= GNU_ABI_VENDOR,
1028 	.flags		= BN_TRANSLATE_OSREL,
1029 	.trans_osrel	= linux_trans_osrel
1030 };
1031 
1032 static Elf32_Brandinfo linux_brand = {
1033 	.brand		= ELFOSABI_LINUX,
1034 	.machine	= EM_386,
1035 	.compat_3_brand	= "Linux",
1036 	.emul_path	= "/compat/linux",
1037 	.interp_path	= "/lib/ld-linux.so.1",
1038 	.sysvec		= &elf_linux_sysvec,
1039 	.interp_newpath	= NULL,
1040 	.brand_note	= &linux_brandnote,
1041 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1042 };
1043 
1044 static Elf32_Brandinfo linux_glibc2brand = {
1045 	.brand		= ELFOSABI_LINUX,
1046 	.machine	= EM_386,
1047 	.compat_3_brand	= "Linux",
1048 	.emul_path	= "/compat/linux",
1049 	.interp_path	= "/lib/ld-linux.so.2",
1050 	.sysvec		= &elf_linux_sysvec,
1051 	.interp_newpath	= NULL,
1052 	.brand_note	= &linux_brandnote,
1053 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1054 };
1055 
1056 static Elf32_Brandinfo linux_muslbrand = {
1057 	.brand		= ELFOSABI_LINUX,
1058 	.machine	= EM_386,
1059 	.compat_3_brand	= "Linux",
1060 	.emul_path	= "/compat/linux",
1061 	.interp_path	= "/lib/ld-musl-i386.so.1",
1062 	.sysvec		= &elf_linux_sysvec,
1063 	.interp_newpath	= NULL,
1064 	.brand_note	= &linux_brandnote,
1065 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1066 };
1067 
1068 Elf32_Brandinfo *linux_brandlist[] = {
1069 	&linux_brand,
1070 	&linux_glibc2brand,
1071 	&linux_muslbrand,
1072 	NULL
1073 };
1074 
1075 static int
1076 linux_elf_modevent(module_t mod, int type, void *data)
1077 {
1078 	Elf32_Brandinfo **brandinfo;
1079 	int error;
1080 	struct linux_ioctl_handler **lihp;
1081 
1082 	error = 0;
1083 
1084 	switch(type) {
1085 	case MOD_LOAD:
1086 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1087 		     ++brandinfo)
1088 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1089 				error = EINVAL;
1090 		if (error == 0) {
1091 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1092 				linux_ioctl_register_handler(*lihp);
1093 			LIST_INIT(&futex_list);
1094 			mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1095 			linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit,
1096 			      NULL, 1000);
1097 			linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec,
1098 			      NULL, 1000);
1099 			linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor,
1100 			    linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY);
1101 			linux_get_machine(&linux_kplatform);
1102 			linux_szplatform = roundup(strlen(linux_kplatform) + 1,
1103 			    sizeof(char *));
1104 			linux_osd_jail_register();
1105 			stclohz = (stathz ? stathz : hz);
1106 			if (bootverbose)
1107 				printf("Linux ELF exec handler installed\n");
1108 		} else
1109 			printf("cannot insert Linux ELF brand handler\n");
1110 		break;
1111 	case MOD_UNLOAD:
1112 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1113 		     ++brandinfo)
1114 			if (elf32_brand_inuse(*brandinfo))
1115 				error = EBUSY;
1116 		if (error == 0) {
1117 			for (brandinfo = &linux_brandlist[0];
1118 			     *brandinfo != NULL; ++brandinfo)
1119 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1120 					error = EINVAL;
1121 		}
1122 		if (error == 0) {
1123 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1124 				linux_ioctl_unregister_handler(*lihp);
1125 			mtx_destroy(&futex_mtx);
1126 			EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
1127 			EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
1128 			EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag);
1129 			linux_osd_jail_deregister();
1130 			if (bootverbose)
1131 				printf("Linux ELF exec handler removed\n");
1132 		} else
1133 			printf("Could not deinstall ELF interpreter entry\n");
1134 		break;
1135 	default:
1136 		return (EOPNOTSUPP);
1137 	}
1138 	return (error);
1139 }
1140 
1141 static moduledata_t linux_elf_mod = {
1142 	"linuxelf",
1143 	linux_elf_modevent,
1144 	0
1145 };
1146 
1147 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1148 FEATURE(linux, "Linux 32bit support");
1149