1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2004 Tim J. Robbins
5  * Copyright (c) 2003 Peter Wemm
6  * Copyright (c) 2002 Doug Rabson
7  * Copyright (c) 1998-1999 Andrew Gallatin
8  * Copyright (c) 1994-1996 Søren Schmidt
9  * All rights reserved.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer
16  *    in this position and unchanged.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. The name of the author may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 #include "opt_compat.h"
38 
39 #ifndef COMPAT_FREEBSD32
40 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!"
41 #endif
42 
43 #define	__ELF_WORD_SIZE	32
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/exec.h>
48 #include <sys/fcntl.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/module.h>
55 #include <sys/mutex.h>
56 #include <sys/proc.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysent.h>
62 #include <sys/sysproto.h>
63 #include <sys/vnode.h>
64 #include <sys/eventhandler.h>
65 
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_param.h>
73 
74 #include <machine/cpu.h>
75 #include <machine/md_var.h>
76 #include <machine/pcb.h>
77 #include <machine/specialreg.h>
78 
79 #include <amd64/linux32/linux.h>
80 #include <amd64/linux32/linux32_proto.h>
81 #include <compat/linux/linux_emul.h>
82 #include <compat/linux/linux_futex.h>
83 #include <compat/linux/linux_ioctl.h>
84 #include <compat/linux/linux_mib.h>
85 #include <compat/linux/linux_misc.h>
86 #include <compat/linux/linux_signal.h>
87 #include <compat/linux/linux_util.h>
88 #include <compat/linux/linux_vdso.h>
89 
90 MODULE_VERSION(linux, 1);
91 
92 #define	AUXARGS_ENTRY_32(pos, id, val)	\
93 	do {				\
94 		suword32(pos++, id);	\
95 		suword32(pos++, val);	\
96 	} while (0)
97 
98 #if BYTE_ORDER == LITTLE_ENDIAN
99 #define SHELLMAGIC      0x2123 /* #! */
100 #else
101 #define SHELLMAGIC      0x2321
102 #endif
103 
104 /*
105  * Allow the sendsig functions to use the ldebug() facility even though they
106  * are not syscalls themselves.  Map them to syscall 0.  This is slightly less
107  * bogus than using ldebug(sigreturn).
108  */
109 #define	LINUX32_SYS_linux_rt_sendsig	0
110 #define	LINUX32_SYS_linux_sendsig	0
111 
112 const char *linux_kplatform;
113 static int linux_szsigcode;
114 static vm_object_t linux_shared_page_obj;
115 static char *linux_shared_page_mapping;
116 extern char _binary_linux32_locore_o_start;
117 extern char _binary_linux32_locore_o_end;
118 
119 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
120 
121 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
122 
123 static int	elf_linux_fixup(register_t **stack_base,
124 		    struct image_params *iparams);
125 static register_t *linux_copyout_strings(struct image_params *imgp);
126 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
127 static void	exec_linux_setregs(struct thread *td,
128 				   struct image_params *imgp, u_long stack);
129 static void	linux32_fixlimit(struct rlimit *rl, int which);
130 static bool	linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
131 static void	linux_vdso_install(void *param);
132 static void	linux_vdso_deinstall(void *param);
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 struct linux32_ps_strings {
194 	u_int32_t ps_argvstr;	/* first of 0 or more argument strings */
195 	u_int ps_nargvstr;	/* the number of argument strings */
196 	u_int32_t ps_envstr;	/* first of 0 or more environment strings */
197 	u_int ps_nenvstr;	/* the number of environment strings */
198 };
199 
200 LINUX_VDSO_SYM_INTPTR(linux32_sigcode);
201 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode);
202 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall);
203 LINUX_VDSO_SYM_CHAR(linux_platform);
204 
205 /*
206  * If FreeBSD & Linux have a difference of opinion about what a trap
207  * means, deal with it here.
208  *
209  * MPSAFE
210  */
211 static int
212 translate_traps(int signal, int trap_code)
213 {
214 	if (signal != SIGBUS)
215 		return signal;
216 	switch (trap_code) {
217 	case T_PROTFLT:
218 	case T_TSSFLT:
219 	case T_DOUBLEFLT:
220 	case T_PAGEFLT:
221 		return SIGSEGV;
222 	default:
223 		return signal;
224 	}
225 }
226 
227 static int
228 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
229 {
230 	Elf32_Auxargs *args;
231 	Elf32_Addr *base;
232 	Elf32_Addr *pos;
233 	struct linux32_ps_strings *arginfo;
234 	int issetugid;
235 
236 	arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
237 
238 	KASSERT(curthread->td_proc == imgp->proc,
239 	    ("unsafe elf_linux_fixup(), should be curproc"));
240 	base = (Elf32_Addr *)*stack_base;
241 	args = (Elf32_Auxargs *)imgp->auxargs;
242 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
243 
244 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
245 	AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO_EHDR,
246 	    imgp->proc->p_sysent->sv_shared_page_base);
247 	AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO, linux32_vsyscall);
248 	AUXARGS_ENTRY_32(pos, LINUX_AT_HWCAP, cpu_feature);
249 
250 	/*
251 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
252 	 * as it has appeared in the 2.4.0-rc7 first time.
253 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
254 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
255 	 * is not present.
256 	 * Also see linux_times() implementation.
257 	 */
258 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
259 		AUXARGS_ENTRY_32(pos, LINUX_AT_CLKTCK, stclohz);
260 	AUXARGS_ENTRY_32(pos, AT_PHDR, args->phdr);
261 	AUXARGS_ENTRY_32(pos, AT_PHENT, args->phent);
262 	AUXARGS_ENTRY_32(pos, AT_PHNUM, args->phnum);
263 	AUXARGS_ENTRY_32(pos, AT_PAGESZ, args->pagesz);
264 	AUXARGS_ENTRY_32(pos, AT_FLAGS, args->flags);
265 	AUXARGS_ENTRY_32(pos, AT_ENTRY, args->entry);
266 	AUXARGS_ENTRY_32(pos, AT_BASE, args->base);
267 	AUXARGS_ENTRY_32(pos, LINUX_AT_SECURE, issetugid);
268 	AUXARGS_ENTRY_32(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
269 	AUXARGS_ENTRY_32(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
270 	AUXARGS_ENTRY_32(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
271 	AUXARGS_ENTRY_32(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
272 	AUXARGS_ENTRY_32(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
273 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
274 	if (imgp->execpathp != 0)
275 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
276 	if (args->execfd != -1)
277 		AUXARGS_ENTRY_32(pos, AT_EXECFD, args->execfd);
278 	AUXARGS_ENTRY_32(pos, AT_NULL, 0);
279 
280 	free(imgp->auxargs, M_TEMP);
281 	imgp->auxargs = NULL;
282 
283 	base--;
284 	suword32(base, (uint32_t)imgp->args->argc);
285 	*stack_base = (register_t *)base;
286 	return (0);
287 }
288 
289 static void
290 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
291 {
292 	struct thread *td = curthread;
293 	struct proc *p = td->td_proc;
294 	struct sigacts *psp;
295 	struct trapframe *regs;
296 	struct l_rt_sigframe *fp, frame;
297 	int oonstack;
298 	int sig;
299 	int code;
300 
301 	sig = ksi->ksi_signo;
302 	code = ksi->ksi_code;
303 	PROC_LOCK_ASSERT(p, MA_OWNED);
304 	psp = p->p_sigacts;
305 	mtx_assert(&psp->ps_mtx, MA_OWNED);
306 	regs = td->td_frame;
307 	oonstack = sigonstack(regs->tf_rsp);
308 
309 #ifdef DEBUG
310 	if (ldebug(rt_sendsig))
311 		printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
312 		    catcher, sig, (void*)mask, code);
313 #endif
314 	/* Allocate space for the signal handler context. */
315 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
316 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
317 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
318 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
319 	} else
320 		fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
321 	mtx_unlock(&psp->ps_mtx);
322 
323 	/* Build the argument list for the signal handler. */
324 	sig = bsd_to_linux_signal(sig);
325 
326 	bzero(&frame, sizeof(frame));
327 
328 	frame.sf_handler = PTROUT(catcher);
329 	frame.sf_sig = sig;
330 	frame.sf_siginfo = PTROUT(&fp->sf_si);
331 	frame.sf_ucontext = PTROUT(&fp->sf_sc);
332 
333 	/* Fill in POSIX parts. */
334 	ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
335 
336 	/*
337 	 * Build the signal context to be used by sigreturn and libgcc unwind.
338 	 */
339 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
340 	frame.sf_sc.uc_link = 0;		/* XXX ??? */
341 
342 	frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
343 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
344 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
345 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
346 	PROC_UNLOCK(p);
347 
348 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
349 
350 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
351 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_rdi;
352 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_rsi;
353 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_rbp;
354 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_rbx;
355 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_rsp;
356 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_rdx;
357 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_rcx;
358 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_rax;
359 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_rip;
360 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
361 	frame.sf_sc.uc_mcontext.sc_gs     = regs->tf_gs;
362 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
363 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
364 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
365 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
366 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
367 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
368 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
369 	frame.sf_sc.uc_mcontext.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
370 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
371 
372 #ifdef DEBUG
373 	if (ldebug(rt_sendsig))
374 		printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"),
375 		    frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
376 		    td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
377 #endif
378 
379 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
380 		/*
381 		 * Process has trashed its stack; give it an illegal
382 		 * instruction to halt it in its tracks.
383 		 */
384 #ifdef DEBUG
385 		if (ldebug(rt_sendsig))
386 			printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
387 			    fp, oonstack);
388 #endif
389 		PROC_LOCK(p);
390 		sigexit(td, SIGILL);
391 	}
392 
393 	/* Build context to run handler in. */
394 	regs->tf_rsp = PTROUT(fp);
395 	regs->tf_rip = linux32_rt_sigcode;
396 	regs->tf_rflags &= ~(PSL_T | PSL_D);
397 	regs->tf_cs = _ucode32sel;
398 	regs->tf_ss = _udatasel;
399 	regs->tf_ds = _udatasel;
400 	regs->tf_es = _udatasel;
401 	regs->tf_fs = _ufssel;
402 	regs->tf_gs = _ugssel;
403 	regs->tf_flags = TF_HASSEGS;
404 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
405 	PROC_LOCK(p);
406 	mtx_lock(&psp->ps_mtx);
407 }
408 
409 
410 /*
411  * Send an interrupt to process.
412  *
413  * Stack is set up to allow sigcode stored
414  * in u. to call routine, followed by kcall
415  * to sigreturn routine below.  After sigreturn
416  * resets the signal mask, the stack, and the
417  * frame pointer, it returns to the user
418  * specified pc, psl.
419  */
420 static void
421 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
422 {
423 	struct thread *td = curthread;
424 	struct proc *p = td->td_proc;
425 	struct sigacts *psp;
426 	struct trapframe *regs;
427 	struct l_sigframe *fp, frame;
428 	l_sigset_t lmask;
429 	int oonstack;
430 	int sig, code;
431 
432 	sig = ksi->ksi_signo;
433 	code = ksi->ksi_code;
434 	PROC_LOCK_ASSERT(p, MA_OWNED);
435 	psp = p->p_sigacts;
436 	mtx_assert(&psp->ps_mtx, MA_OWNED);
437 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
438 		/* Signal handler installed with SA_SIGINFO. */
439 		linux_rt_sendsig(catcher, ksi, mask);
440 		return;
441 	}
442 
443 	regs = td->td_frame;
444 	oonstack = sigonstack(regs->tf_rsp);
445 
446 #ifdef DEBUG
447 	if (ldebug(sendsig))
448 		printf(ARGS(sendsig, "%p, %d, %p, %u"),
449 		    catcher, sig, (void*)mask, code);
450 #endif
451 
452 	/* Allocate space for the signal handler context. */
453 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
454 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
455 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
456 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
457 	} else
458 		fp = (struct l_sigframe *)regs->tf_rsp - 1;
459 	mtx_unlock(&psp->ps_mtx);
460 	PROC_UNLOCK(p);
461 
462 	/* Build the argument list for the signal handler. */
463 	sig = bsd_to_linux_signal(sig);
464 
465 	bzero(&frame, sizeof(frame));
466 
467 	frame.sf_handler = PTROUT(catcher);
468 	frame.sf_sig = sig;
469 
470 	bsd_to_linux_sigset(mask, &lmask);
471 
472 	/* Build the signal context to be used by sigreturn. */
473 	frame.sf_sc.sc_mask   = lmask.__mask;
474 	frame.sf_sc.sc_gs     = regs->tf_gs;
475 	frame.sf_sc.sc_fs     = regs->tf_fs;
476 	frame.sf_sc.sc_es     = regs->tf_es;
477 	frame.sf_sc.sc_ds     = regs->tf_ds;
478 	frame.sf_sc.sc_edi    = regs->tf_rdi;
479 	frame.sf_sc.sc_esi    = regs->tf_rsi;
480 	frame.sf_sc.sc_ebp    = regs->tf_rbp;
481 	frame.sf_sc.sc_ebx    = regs->tf_rbx;
482 	frame.sf_sc.sc_esp    = regs->tf_rsp;
483 	frame.sf_sc.sc_edx    = regs->tf_rdx;
484 	frame.sf_sc.sc_ecx    = regs->tf_rcx;
485 	frame.sf_sc.sc_eax    = regs->tf_rax;
486 	frame.sf_sc.sc_eip    = regs->tf_rip;
487 	frame.sf_sc.sc_cs     = regs->tf_cs;
488 	frame.sf_sc.sc_eflags = regs->tf_rflags;
489 	frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
490 	frame.sf_sc.sc_ss     = regs->tf_ss;
491 	frame.sf_sc.sc_err    = regs->tf_err;
492 	frame.sf_sc.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
493 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
494 
495 	frame.sf_extramask[0] = lmask.__mask;
496 
497 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
498 		/*
499 		 * Process has trashed its stack; give it an illegal
500 		 * instruction to halt it in its tracks.
501 		 */
502 		PROC_LOCK(p);
503 		sigexit(td, SIGILL);
504 	}
505 
506 	/* Build context to run handler in. */
507 	regs->tf_rsp = PTROUT(fp);
508 	regs->tf_rip = linux32_sigcode;
509 	regs->tf_rflags &= ~(PSL_T | PSL_D);
510 	regs->tf_cs = _ucode32sel;
511 	regs->tf_ss = _udatasel;
512 	regs->tf_ds = _udatasel;
513 	regs->tf_es = _udatasel;
514 	regs->tf_fs = _ufssel;
515 	regs->tf_gs = _ugssel;
516 	regs->tf_flags = TF_HASSEGS;
517 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
518 	PROC_LOCK(p);
519 	mtx_lock(&psp->ps_mtx);
520 }
521 
522 /*
523  * System call to cleanup state after a signal
524  * has been taken.  Reset signal mask and
525  * stack state from context left by sendsig (above).
526  * Return to previous pc and psl as specified by
527  * context left by sendsig. Check carefully to
528  * make sure that the user has not modified the
529  * psl to gain improper privileges or to cause
530  * a machine fault.
531  */
532 int
533 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
534 {
535 	struct l_sigframe frame;
536 	struct trapframe *regs;
537 	sigset_t bmask;
538 	l_sigset_t lmask;
539 	int eflags;
540 	ksiginfo_t ksi;
541 
542 	regs = td->td_frame;
543 
544 #ifdef DEBUG
545 	if (ldebug(sigreturn))
546 		printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
547 #endif
548 	/*
549 	 * The trampoline code hands us the sigframe.
550 	 * It is unsafe to keep track of it ourselves, in the event that a
551 	 * program jumps out of a signal handler.
552 	 */
553 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
554 		return (EFAULT);
555 
556 	/* Check for security violations. */
557 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
558 	eflags = frame.sf_sc.sc_eflags;
559 	if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
560 		return(EINVAL);
561 
562 	/*
563 	 * Don't allow users to load a valid privileged %cs.  Let the
564 	 * hardware check for invalid selectors, excess privilege in
565 	 * other selectors, invalid %eip's and invalid %esp's.
566 	 */
567 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
568 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
569 		ksiginfo_init_trap(&ksi);
570 		ksi.ksi_signo = SIGBUS;
571 		ksi.ksi_code = BUS_OBJERR;
572 		ksi.ksi_trapno = T_PROTFLT;
573 		ksi.ksi_addr = (void *)regs->tf_rip;
574 		trapsignal(td, &ksi);
575 		return(EINVAL);
576 	}
577 
578 	lmask.__mask = frame.sf_sc.sc_mask;
579 	lmask.__mask = frame.sf_extramask[0];
580 	linux_to_bsd_sigset(&lmask, &bmask);
581 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
582 
583 	/* Restore signal context. */
584 	regs->tf_rdi    = frame.sf_sc.sc_edi;
585 	regs->tf_rsi    = frame.sf_sc.sc_esi;
586 	regs->tf_rbp    = frame.sf_sc.sc_ebp;
587 	regs->tf_rbx    = frame.sf_sc.sc_ebx;
588 	regs->tf_rdx    = frame.sf_sc.sc_edx;
589 	regs->tf_rcx    = frame.sf_sc.sc_ecx;
590 	regs->tf_rax    = frame.sf_sc.sc_eax;
591 	regs->tf_rip    = frame.sf_sc.sc_eip;
592 	regs->tf_cs     = frame.sf_sc.sc_cs;
593 	regs->tf_ds     = frame.sf_sc.sc_ds;
594 	regs->tf_es     = frame.sf_sc.sc_es;
595 	regs->tf_fs     = frame.sf_sc.sc_fs;
596 	regs->tf_gs     = frame.sf_sc.sc_gs;
597 	regs->tf_rflags = eflags;
598 	regs->tf_rsp    = frame.sf_sc.sc_esp_at_signal;
599 	regs->tf_ss     = frame.sf_sc.sc_ss;
600 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
601 
602 	return (EJUSTRETURN);
603 }
604 
605 /*
606  * System call to cleanup state after a signal
607  * has been taken.  Reset signal mask and
608  * stack state from context left by rt_sendsig (above).
609  * Return to previous pc and psl as specified by
610  * context left by sendsig. Check carefully to
611  * make sure that the user has not modified the
612  * psl to gain improper privileges or to cause
613  * a machine fault.
614  */
615 int
616 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
617 {
618 	struct l_ucontext uc;
619 	struct l_sigcontext *context;
620 	sigset_t bmask;
621 	l_stack_t *lss;
622 	stack_t ss;
623 	struct trapframe *regs;
624 	int eflags;
625 	ksiginfo_t ksi;
626 
627 	regs = td->td_frame;
628 
629 #ifdef DEBUG
630 	if (ldebug(rt_sigreturn))
631 		printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
632 #endif
633 	/*
634 	 * The trampoline code hands us the ucontext.
635 	 * It is unsafe to keep track of it ourselves, in the event that a
636 	 * program jumps out of a signal handler.
637 	 */
638 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
639 		return (EFAULT);
640 
641 	context = &uc.uc_mcontext;
642 
643 	/* Check for security violations. */
644 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
645 	eflags = context->sc_eflags;
646 	if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
647 		return(EINVAL);
648 
649 	/*
650 	 * Don't allow users to load a valid privileged %cs.  Let the
651 	 * hardware check for invalid selectors, excess privilege in
652 	 * other selectors, invalid %eip's and invalid %esp's.
653 	 */
654 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
655 	if (!CS_SECURE(context->sc_cs)) {
656 		ksiginfo_init_trap(&ksi);
657 		ksi.ksi_signo = SIGBUS;
658 		ksi.ksi_code = BUS_OBJERR;
659 		ksi.ksi_trapno = T_PROTFLT;
660 		ksi.ksi_addr = (void *)regs->tf_rip;
661 		trapsignal(td, &ksi);
662 		return(EINVAL);
663 	}
664 
665 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
666 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
667 
668 	/*
669 	 * Restore signal context
670 	 */
671 	regs->tf_gs	= context->sc_gs;
672 	regs->tf_fs	= context->sc_fs;
673 	regs->tf_es	= context->sc_es;
674 	regs->tf_ds	= context->sc_ds;
675 	regs->tf_rdi    = context->sc_edi;
676 	regs->tf_rsi    = context->sc_esi;
677 	regs->tf_rbp    = context->sc_ebp;
678 	regs->tf_rbx    = context->sc_ebx;
679 	regs->tf_rdx    = context->sc_edx;
680 	regs->tf_rcx    = context->sc_ecx;
681 	regs->tf_rax    = context->sc_eax;
682 	regs->tf_rip    = context->sc_eip;
683 	regs->tf_cs     = context->sc_cs;
684 	regs->tf_rflags = eflags;
685 	regs->tf_rsp    = context->sc_esp_at_signal;
686 	regs->tf_ss     = context->sc_ss;
687 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
688 
689 	/*
690 	 * call sigaltstack & ignore results..
691 	 */
692 	lss = &uc.uc_stack;
693 	ss.ss_sp = PTRIN(lss->ss_sp);
694 	ss.ss_size = lss->ss_size;
695 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
696 
697 #ifdef DEBUG
698 	if (ldebug(rt_sigreturn))
699 		printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"),
700 		    ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
701 #endif
702 	(void)kern_sigaltstack(td, &ss, NULL);
703 
704 	return (EJUSTRETURN);
705 }
706 
707 static int
708 linux32_fetch_syscall_args(struct thread *td)
709 {
710 	struct proc *p;
711 	struct trapframe *frame;
712 	struct syscall_args *sa;
713 
714 	p = td->td_proc;
715 	frame = td->td_frame;
716 	sa = &td->td_sa;
717 
718 	sa->args[0] = frame->tf_rbx;
719 	sa->args[1] = frame->tf_rcx;
720 	sa->args[2] = frame->tf_rdx;
721 	sa->args[3] = frame->tf_rsi;
722 	sa->args[4] = frame->tf_rdi;
723 	sa->args[5] = frame->tf_rbp;	/* Unconfirmed */
724 	sa->code = frame->tf_rax;
725 
726 	if (sa->code >= p->p_sysent->sv_size)
727 		/* nosys */
728 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
729 	else
730 		sa->callp = &p->p_sysent->sv_table[sa->code];
731 	sa->narg = sa->callp->sy_narg;
732 
733 	td->td_retval[0] = 0;
734 	td->td_retval[1] = frame->tf_rdx;
735 
736 	return (0);
737 }
738 
739 /*
740  * If a Linux binary is exec'ing something, try this image activator
741  * first.  We override standard shell script execution in order to
742  * be able to modify the interpreter path.  We only do this if a Linux
743  * binary is doing the exec, so we do not create an EXEC module for it.
744  */
745 static int	exec_linux_imgact_try(struct image_params *iparams);
746 
747 static int
748 exec_linux_imgact_try(struct image_params *imgp)
749 {
750 	const char *head = (const char *)imgp->image_header;
751 	char *rpath;
752 	int error = -1;
753 
754 	/*
755 	* The interpreter for shell scripts run from a Linux binary needs
756 	* to be located in /compat/linux if possible in order to recursively
757 	* maintain Linux path emulation.
758 	*/
759 	if (((const short *)head)[0] == SHELLMAGIC) {
760 		/*
761 		* Run our normal shell image activator.  If it succeeds attempt
762 		* to use the alternate path for the interpreter.  If an
763 		* alternate * path is found, use our stringspace to store it.
764 		*/
765 		if ((error = exec_shell_imgact(imgp)) == 0) {
766 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
767 			    imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
768 			    AT_FDCWD);
769 			if (rpath != NULL)
770 				imgp->args->fname_buf =
771 				    imgp->interpreter_name = rpath;
772 		}
773 	}
774 	return (error);
775 }
776 
777 /*
778  * Clear registers on exec
779  * XXX copied from ia32_signal.c.
780  */
781 static void
782 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack)
783 {
784 	struct trapframe *regs = td->td_frame;
785 	struct pcb *pcb = td->td_pcb;
786 
787 	if (td->td_proc->p_md.md_ldt != NULL)
788 		user_ldt_free(td);
789 
790 	critical_enter();
791 	wrmsr(MSR_FSBASE, 0);
792 	wrmsr(MSR_KGSBASE, 0);	/* User value while we're in the kernel */
793 	pcb->pcb_fsbase = 0;
794 	pcb->pcb_gsbase = 0;
795 	critical_exit();
796 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
797 
798 	bzero((char *)regs, sizeof(struct trapframe));
799 	regs->tf_rip = imgp->entry_addr;
800 	regs->tf_rsp = stack;
801 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
802 	regs->tf_gs = _ugssel;
803 	regs->tf_fs = _ufssel;
804 	regs->tf_es = _udatasel;
805 	regs->tf_ds = _udatasel;
806 	regs->tf_ss = _udatasel;
807 	regs->tf_flags = TF_HASSEGS;
808 	regs->tf_cs = _ucode32sel;
809 	regs->tf_rbx = imgp->ps_strings;
810 
811 	fpstate_drop(td);
812 
813 	/* Do full restore on return so that we can change to a different %cs */
814 	set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
815 }
816 
817 /*
818  * XXX copied from ia32_sysvec.c.
819  */
820 static register_t *
821 linux_copyout_strings(struct image_params *imgp)
822 {
823 	int argc, envc;
824 	u_int32_t *vectp;
825 	char *stringp, *destp;
826 	u_int32_t *stack_base;
827 	struct linux32_ps_strings *arginfo;
828 	char canary[LINUX_AT_RANDOM_LEN];
829 	size_t execpath_len;
830 
831 	/* Calculate string base and vector table pointers. */
832 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
833 		execpath_len = strlen(imgp->execpath) + 1;
834 	else
835 		execpath_len = 0;
836 
837 	arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
838 	destp =	(caddr_t)arginfo - SPARE_USRSPACE -
839 	    roundup(sizeof(canary), sizeof(char *)) -
840 	    roundup(execpath_len, sizeof(char *)) -
841 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
842 
843 	if (execpath_len != 0) {
844 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
845 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
846 	}
847 
848 	/* Prepare the canary for SSP. */
849 	arc4rand(canary, sizeof(canary), 0);
850 	imgp->canary = (uintptr_t)arginfo -
851 	    roundup(execpath_len, sizeof(char *)) -
852 	    roundup(sizeof(canary), sizeof(char *));
853 	copyout(canary, (void *)imgp->canary, sizeof(canary));
854 
855 	/* If we have a valid auxargs ptr, prepare some room on the stack. */
856 	if (imgp->auxargs) {
857 		/*
858 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
859 		 * lower compatibility.
860 		 */
861 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
862 		    (LINUX_AT_COUNT * 2);
863 		/*
864 		 * The '+ 2' is for the null pointers at the end of each of
865 		 * the arg and env vector sets,and imgp->auxarg_size is room
866 		 * for argument of Runtime loader.
867 		 */
868 		vectp = (u_int32_t *) (destp - (imgp->args->argc +
869 		    imgp->args->envc + 2 + imgp->auxarg_size) *
870 		    sizeof(u_int32_t));
871 
872 	} else
873 		/*
874 		 * The '+ 2' is for the null pointers at the end of each of
875 		 * the arg and env vector sets
876 		 */
877 		vectp = (u_int32_t *)(destp - (imgp->args->argc +
878 		    imgp->args->envc + 2) * sizeof(u_int32_t));
879 
880 	/* vectp also becomes our initial stack base. */
881 	stack_base = vectp;
882 
883 	stringp = imgp->args->begin_argv;
884 	argc = imgp->args->argc;
885 	envc = imgp->args->envc;
886 	/* Copy out strings - arguments and environment. */
887 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
888 
889 	/* Fill in "ps_strings" struct for ps, w, etc. */
890 	suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp);
891 	suword32(&arginfo->ps_nargvstr, argc);
892 
893 	/* Fill in argument portion of vector table. */
894 	for (; argc > 0; --argc) {
895 		suword32(vectp++, (uint32_t)(intptr_t)destp);
896 		while (*stringp++ != 0)
897 			destp++;
898 		destp++;
899 	}
900 
901 	/* A null vector table pointer separates the argp's from the envp's. */
902 	suword32(vectp++, 0);
903 
904 	suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp);
905 	suword32(&arginfo->ps_nenvstr, envc);
906 
907 	/* Fill in environment portion of vector table. */
908 	for (; envc > 0; --envc) {
909 		suword32(vectp++, (uint32_t)(intptr_t)destp);
910 		while (*stringp++ != 0)
911 			destp++;
912 		destp++;
913 	}
914 
915 	/* The end of the vector table is a null pointer. */
916 	suword32(vectp, 0);
917 
918 	return ((register_t *)stack_base);
919 }
920 
921 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0,
922     "32-bit Linux emulation");
923 
924 static u_long	linux32_maxdsiz = LINUX32_MAXDSIZ;
925 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
926     &linux32_maxdsiz, 0, "");
927 static u_long	linux32_maxssiz = LINUX32_MAXSSIZ;
928 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
929     &linux32_maxssiz, 0, "");
930 static u_long	linux32_maxvmem = LINUX32_MAXVMEM;
931 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
932     &linux32_maxvmem, 0, "");
933 
934 #if defined(DEBUG)
935 SYSCTL_PROC(_compat_linux32, OID_AUTO, debug, CTLTYPE_STRING | CTLFLAG_RW, 0, 0,
936     linux_sysctl_debug, "A", "Linux debugging control");
937 #endif
938 
939 static void
940 linux32_fixlimit(struct rlimit *rl, int which)
941 {
942 
943 	switch (which) {
944 	case RLIMIT_DATA:
945 		if (linux32_maxdsiz != 0) {
946 			if (rl->rlim_cur > linux32_maxdsiz)
947 				rl->rlim_cur = linux32_maxdsiz;
948 			if (rl->rlim_max > linux32_maxdsiz)
949 				rl->rlim_max = linux32_maxdsiz;
950 		}
951 		break;
952 	case RLIMIT_STACK:
953 		if (linux32_maxssiz != 0) {
954 			if (rl->rlim_cur > linux32_maxssiz)
955 				rl->rlim_cur = linux32_maxssiz;
956 			if (rl->rlim_max > linux32_maxssiz)
957 				rl->rlim_max = linux32_maxssiz;
958 		}
959 		break;
960 	case RLIMIT_VMEM:
961 		if (linux32_maxvmem != 0) {
962 			if (rl->rlim_cur > linux32_maxvmem)
963 				rl->rlim_cur = linux32_maxvmem;
964 			if (rl->rlim_max > linux32_maxvmem)
965 				rl->rlim_max = linux32_maxvmem;
966 		}
967 		break;
968 	}
969 }
970 
971 struct sysentvec elf_linux_sysvec = {
972 	.sv_size	= LINUX32_SYS_MAXSYSCALL,
973 	.sv_table	= linux32_sysent,
974 	.sv_mask	= 0,
975 	.sv_errsize	= ELAST + 1,
976 	.sv_errtbl	= bsd_to_linux_errno,
977 	.sv_transtrap	= translate_traps,
978 	.sv_fixup	= elf_linux_fixup,
979 	.sv_sendsig	= linux_sendsig,
980 	.sv_sigcode	= &_binary_linux32_locore_o_start,
981 	.sv_szsigcode	= &linux_szsigcode,
982 	.sv_name	= "Linux ELF32",
983 	.sv_coredump	= elf32_coredump,
984 	.sv_imgact_try	= exec_linux_imgact_try,
985 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
986 	.sv_pagesize	= PAGE_SIZE,
987 	.sv_minuser	= VM_MIN_ADDRESS,
988 	.sv_maxuser	= LINUX32_MAXUSER,
989 	.sv_usrstack	= LINUX32_USRSTACK,
990 	.sv_psstrings	= LINUX32_PS_STRINGS,
991 	.sv_stackprot	= VM_PROT_ALL,
992 	.sv_copyout_strings = linux_copyout_strings,
993 	.sv_setregs	= exec_linux_setregs,
994 	.sv_fixlimit	= linux32_fixlimit,
995 	.sv_maxssiz	= &linux32_maxssiz,
996 	.sv_flags	= SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP,
997 	.sv_set_syscall_retval = cpu_set_syscall_retval,
998 	.sv_fetch_syscall_args = linux32_fetch_syscall_args,
999 	.sv_syscallnames = NULL,
1000 	.sv_shared_page_base = LINUX32_SHAREDPAGE,
1001 	.sv_shared_page_len = PAGE_SIZE,
1002 	.sv_schedtail	= linux_schedtail,
1003 	.sv_thread_detach = linux_thread_detach,
1004 	.sv_trap	= NULL,
1005 };
1006 
1007 static void
1008 linux_vdso_install(void *param)
1009 {
1010 
1011 	linux_szsigcode = (&_binary_linux32_locore_o_end -
1012 	    &_binary_linux32_locore_o_start);
1013 
1014 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
1015 		panic("Linux invalid vdso size\n");
1016 
1017 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
1018 
1019 	linux_shared_page_obj = __elfN(linux_shared_page_init)
1020 	    (&linux_shared_page_mapping);
1021 
1022 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
1023 
1024 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
1025 	    linux_szsigcode);
1026 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
1027 
1028 	linux_kplatform = linux_shared_page_mapping +
1029 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
1030 }
1031 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
1032     (sysinit_cfunc_t)linux_vdso_install, NULL);
1033 
1034 static void
1035 linux_vdso_deinstall(void *param)
1036 {
1037 
1038 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
1039 };
1040 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1041     (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
1042 
1043 static char GNU_ABI_VENDOR[] = "GNU";
1044 static int GNULINUX_ABI_DESC = 0;
1045 
1046 static bool
1047 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
1048 {
1049 	const Elf32_Word *desc;
1050 	uintptr_t p;
1051 
1052 	p = (uintptr_t)(note + 1);
1053 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1054 
1055 	desc = (const Elf32_Word *)p;
1056 	if (desc[0] != GNULINUX_ABI_DESC)
1057 		return (false);
1058 
1059 	/*
1060 	 * For Linux we encode osrel as follows (see linux_mib.c):
1061 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
1062 	 */
1063 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1064 
1065 	return (true);
1066 }
1067 
1068 static Elf_Brandnote linux32_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	= linux32_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	= "/compat/linux",
1082 	.interp_path	= "/lib/ld-linux.so.1",
1083 	.sysvec		= &elf_linux_sysvec,
1084 	.interp_newpath	= NULL,
1085 	.brand_note	= &linux32_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	= "/compat/linux",
1094 	.interp_path	= "/lib/ld-linux.so.2",
1095 	.sysvec		= &elf_linux_sysvec,
1096 	.interp_newpath	= NULL,
1097 	.brand_note	= &linux32_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	= "/compat/linux",
1106 	.interp_path	= "/lib/ld-musl-i386.so.1",
1107 	.sysvec		= &elf_linux_sysvec,
1108 	.interp_newpath	= NULL,
1109 	.brand_note	= &linux32_brandnote,
1110 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1111 };
1112 
1113 Elf32_Brandinfo *linux_brandlist[] = {
1114 	&linux_brand,
1115 	&linux_glibc2brand,
1116 	&linux_muslbrand,
1117 	NULL
1118 };
1119 
1120 static int
1121 linux_elf_modevent(module_t mod, int type, void *data)
1122 {
1123 	Elf32_Brandinfo **brandinfo;
1124 	int error;
1125 	struct linux_ioctl_handler **lihp;
1126 
1127 	error = 0;
1128 
1129 	switch(type) {
1130 	case MOD_LOAD:
1131 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1132 		     ++brandinfo)
1133 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1134 				error = EINVAL;
1135 		if (error == 0) {
1136 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1137 				linux_ioctl_register_handler(*lihp);
1138 			LIST_INIT(&futex_list);
1139 			mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1140 			stclohz = (stathz ? stathz : hz);
1141 			if (bootverbose)
1142 				printf("Linux ELF exec handler installed\n");
1143 		} else
1144 			printf("cannot insert Linux ELF brand handler\n");
1145 		break;
1146 	case MOD_UNLOAD:
1147 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1148 		     ++brandinfo)
1149 			if (elf32_brand_inuse(*brandinfo))
1150 				error = EBUSY;
1151 		if (error == 0) {
1152 			for (brandinfo = &linux_brandlist[0];
1153 			     *brandinfo != NULL; ++brandinfo)
1154 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1155 					error = EINVAL;
1156 		}
1157 		if (error == 0) {
1158 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1159 				linux_ioctl_unregister_handler(*lihp);
1160 			mtx_destroy(&futex_mtx);
1161 			if (bootverbose)
1162 				printf("Linux ELF exec handler removed\n");
1163 		} else
1164 			printf("Could not deinstall ELF interpreter entry\n");
1165 		break;
1166 	default:
1167 		return (EOPNOTSUPP);
1168 	}
1169 	return (error);
1170 }
1171 
1172 static moduledata_t linux_elf_mod = {
1173 	"linuxelf",
1174 	linux_elf_modevent,
1175 	0
1176 };
1177 
1178 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1179 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1180 FEATURE(linux, "Linux 32bit support");
1181