xref: /freebsd/sys/amd64/linux32/linux32_sysvec.c (revision 058ac3e8063366dafa634d9107642e12b038bf09)
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 "opt_compat.h"
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #ifndef COMPAT_FREEBSD32
41 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!"
42 #endif
43 
44 #define	__ELF_WORD_SIZE	32
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/exec.h>
49 #include <sys/fcntl.h>
50 #include <sys/imgact.h>
51 #include <sys/imgact_elf.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/module.h>
56 #include <sys/mutex.h>
57 #include <sys/proc.h>
58 #include <sys/resourcevar.h>
59 #include <sys/stddef.h>
60 #include <sys/signalvar.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysent.h>
64 #include <sys/sysproto.h>
65 #include <sys/vnode.h>
66 #include <sys/eventhandler.h>
67 
68 #include <vm/vm.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_param.h>
75 
76 #include <machine/cpu.h>
77 #include <machine/md_var.h>
78 #include <machine/pcb.h>
79 #include <machine/specialreg.h>
80 #include <machine/trap.h>
81 
82 #include <x86/linux/linux_x86.h>
83 #include <amd64/linux32/linux.h>
84 #include <amd64/linux32/linux32_proto.h>
85 #include <compat/linux/linux_emul.h>
86 #include <compat/linux/linux_fork.h>
87 #include <compat/linux/linux_ioctl.h>
88 #include <compat/linux/linux_mib.h>
89 #include <compat/linux/linux_misc.h>
90 #include <compat/linux/linux_signal.h>
91 #include <compat/linux/linux_util.h>
92 #include <compat/linux/linux_vdso.h>
93 
94 #include <x86/linux/linux_x86_sigframe.h>
95 
96 MODULE_VERSION(linux, 1);
97 
98 #define	LINUX32_MAXUSER		((1ul << 32) - PAGE_SIZE)
99 #define	LINUX32_VDSOPAGE_SIZE	PAGE_SIZE * 2
100 #define	LINUX32_VDSOPAGE	(LINUX32_MAXUSER - LINUX32_VDSOPAGE_SIZE)
101 #define	LINUX32_SHAREDPAGE	(LINUX32_VDSOPAGE - PAGE_SIZE)
102 				/*
103 				 * PAGE_SIZE - the size
104 				 * of the native SHAREDPAGE
105 				 */
106 #define	LINUX32_USRSTACK	LINUX32_SHAREDPAGE
107 
108 static int linux_szsigcode;
109 static vm_object_t linux_vdso_obj;
110 static char *linux_vdso_mapping;
111 extern char _binary_linux32_vdso_so_o_start;
112 extern char _binary_linux32_vdso_so_o_end;
113 static vm_offset_t linux_vdso_base;
114 
115 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
116 extern const char *linux32_syscallnames[];
117 
118 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
119 
120 static int	linux_fixup_elf(uintptr_t *stack_base,
121 		    struct image_params *iparams);
122 static int	linux_copyout_strings(struct image_params *imgp,
123 		    uintptr_t *stack_base);
124 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
125 static void	linux_exec_setregs(struct thread *td,
126 				   struct image_params *imgp, uintptr_t stack);
127 static void	linux_exec_sysvec_init(void *param);
128 static int	linux_on_exec_vmspace(struct proc *p,
129 		    struct image_params *imgp);
130 static void	linux32_fixlimit(struct rlimit *rl, int which);
131 static bool	linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
132 static void	linux_vdso_install(const void *param);
133 static void	linux_vdso_deinstall(const void *param);
134 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
135 static void	linux32_set_fork_retval(struct thread *td);
136 static void	linux32_set_syscall_retval(struct thread *td, int error);
137 
138 struct linux32_ps_strings {
139 	u_int32_t ps_argvstr;	/* first of 0 or more argument strings */
140 	u_int ps_nargvstr;	/* the number of argument strings */
141 	u_int32_t ps_envstr;	/* first of 0 or more environment strings */
142 	u_int ps_nenvstr;	/* the number of environment strings */
143 };
144 #define	LINUX32_PS_STRINGS	(LINUX32_USRSTACK - \
145 				    sizeof(struct linux32_ps_strings))
146 
147 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
148 LINUX_VDSO_SYM_INTPTR(linux32_vdso_sigcode);
149 LINUX_VDSO_SYM_INTPTR(linux32_vdso_rt_sigcode);
150 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
151 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
152 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
153 LINUX_VDSO_SYM_CHAR(linux_platform);
154 
155 static int
156 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
157 {
158 	Elf32_Auxargs *args;
159 	Elf32_Auxinfo *argarray, *pos;
160 	int error, issetugid;
161 
162 	args = (Elf32_Auxargs *)imgp->auxargs;
163 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
164 	    M_WAITOK | M_ZERO);
165 
166 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
167 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, __kernel_vsyscall);
168 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
169 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
170 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
171 
172 	/*
173 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
174 	 * as it has appeared in the 2.4.0-rc7 first time.
175 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
176 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
177 	 * is not present.
178 	 * Also see linux_times() implementation.
179 	 */
180 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
181 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
182 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
183 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
184 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
185 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
186 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
187 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
188 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
189 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
190 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
191 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
192 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
193 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
194 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
195 	if (imgp->execpathp != 0)
196 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
197 	if (args->execfd != -1)
198 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
199 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
200 	AUXARGS_ENTRY(pos, AT_NULL, 0);
201 
202 	free(imgp->auxargs, M_TEMP);
203 	imgp->auxargs = NULL;
204 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
205 
206 	error = copyout(argarray, (void *)base,
207 	    sizeof(*argarray) * LINUX_AT_COUNT);
208 	free(argarray, M_TEMP);
209 	return (error);
210 }
211 
212 static int
213 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
214 {
215 	Elf32_Addr *base;
216 
217 	base = (Elf32_Addr *)*stack_base;
218 	base--;
219 	if (suword32(base, (uint32_t)imgp->args->argc) == -1)
220 		return (EFAULT);
221 	*stack_base = (uintptr_t)base;
222 	return (0);
223 }
224 
225 static void
226 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
227 {
228 	struct thread *td = curthread;
229 	struct proc *p = td->td_proc;
230 	struct sigacts *psp;
231 	struct trapframe *regs;
232 	struct l_rt_sigframe *fp, frame;
233 	int oonstack;
234 	int sig;
235 	int code;
236 
237 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
238 	code = ksi->ksi_code;
239 	PROC_LOCK_ASSERT(p, MA_OWNED);
240 	psp = p->p_sigacts;
241 	mtx_assert(&psp->ps_mtx, MA_OWNED);
242 	regs = td->td_frame;
243 	oonstack = sigonstack(regs->tf_rsp);
244 
245 	/* Allocate space for the signal handler context. */
246 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
247 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
248 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
249 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
250 	} else
251 		fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
252 	mtx_unlock(&psp->ps_mtx);
253 
254 	/* Build the argument list for the signal handler. */
255 	sig = bsd_to_linux_signal(sig);
256 
257 	bzero(&frame, sizeof(frame));
258 
259 	frame.sf_sig = sig;
260 	frame.sf_siginfo = PTROUT(&fp->sf_si);
261 	frame.sf_ucontext = PTROUT(&fp->sf_uc);
262 
263 	/* Fill in POSIX parts. */
264 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
265 
266 	/*
267 	 * Build the signal context to be used by sigreturn and libgcc unwind.
268 	 */
269 	frame.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
270 	frame.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
271 	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
272 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
273 	PROC_UNLOCK(p);
274 
275 	bsd_to_linux_sigset(mask, &frame.sf_uc.uc_sigmask);
276 
277 	frame.sf_uc.uc_mcontext.sc_mask   = frame.sf_uc.uc_sigmask.__mask;
278 	frame.sf_uc.uc_mcontext.sc_edi    = regs->tf_rdi;
279 	frame.sf_uc.uc_mcontext.sc_esi    = regs->tf_rsi;
280 	frame.sf_uc.uc_mcontext.sc_ebp    = regs->tf_rbp;
281 	frame.sf_uc.uc_mcontext.sc_ebx    = regs->tf_rbx;
282 	frame.sf_uc.uc_mcontext.sc_esp    = regs->tf_rsp;
283 	frame.sf_uc.uc_mcontext.sc_edx    = regs->tf_rdx;
284 	frame.sf_uc.uc_mcontext.sc_ecx    = regs->tf_rcx;
285 	frame.sf_uc.uc_mcontext.sc_eax    = regs->tf_rax;
286 	frame.sf_uc.uc_mcontext.sc_eip    = regs->tf_rip;
287 	frame.sf_uc.uc_mcontext.sc_cs     = regs->tf_cs;
288 	frame.sf_uc.uc_mcontext.sc_gs     = regs->tf_gs;
289 	frame.sf_uc.uc_mcontext.sc_fs     = regs->tf_fs;
290 	frame.sf_uc.uc_mcontext.sc_es     = regs->tf_es;
291 	frame.sf_uc.uc_mcontext.sc_ds     = regs->tf_ds;
292 	frame.sf_uc.uc_mcontext.sc_eflags = regs->tf_rflags;
293 	frame.sf_uc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
294 	frame.sf_uc.uc_mcontext.sc_ss     = regs->tf_ss;
295 	frame.sf_uc.uc_mcontext.sc_err    = regs->tf_err;
296 	frame.sf_uc.uc_mcontext.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
297 	frame.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
298 
299 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
300 		/*
301 		 * Process has trashed its stack; give it an illegal
302 		 * instruction to halt it in its tracks.
303 		 */
304 		PROC_LOCK(p);
305 		sigexit(td, SIGILL);
306 	}
307 
308 	/* Build context to run handler in. */
309 	regs->tf_rsp = PTROUT(fp);
310 	regs->tf_rip = linux32_vdso_rt_sigcode;
311 	regs->tf_rdi = PTROUT(catcher);
312 	regs->tf_rflags &= ~(PSL_T | PSL_D);
313 	regs->tf_cs = _ucode32sel;
314 	regs->tf_ss = _udatasel;
315 	regs->tf_ds = _udatasel;
316 	regs->tf_es = _udatasel;
317 	regs->tf_fs = _ufssel;
318 	regs->tf_gs = _ugssel;
319 	regs->tf_flags = TF_HASSEGS;
320 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
321 	PROC_LOCK(p);
322 	mtx_lock(&psp->ps_mtx);
323 }
324 
325 /*
326  * Send an interrupt to process.
327  *
328  * Stack is set up to allow sigcode stored
329  * in u. to call routine, followed by kcall
330  * to sigreturn routine below.  After sigreturn
331  * resets the signal mask, the stack, and the
332  * frame pointer, it returns to the user
333  * specified pc, psl.
334  */
335 static void
336 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
337 {
338 	struct thread *td = curthread;
339 	struct proc *p = td->td_proc;
340 	struct sigacts *psp;
341 	struct trapframe *regs;
342 	struct l_sigframe *fp, frame;
343 	l_sigset_t lmask;
344 	int oonstack;
345 	int sig, code;
346 
347 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
348 	code = ksi->ksi_code;
349 	PROC_LOCK_ASSERT(p, MA_OWNED);
350 	psp = p->p_sigacts;
351 	mtx_assert(&psp->ps_mtx, MA_OWNED);
352 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
353 		/* Signal handler installed with SA_SIGINFO. */
354 		linux_rt_sendsig(catcher, ksi, mask);
355 		return;
356 	}
357 
358 	regs = td->td_frame;
359 	oonstack = sigonstack(regs->tf_rsp);
360 
361 	/* Allocate space for the signal handler context. */
362 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
363 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
364 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
365 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
366 	} else
367 		fp = (struct l_sigframe *)regs->tf_rsp - 1;
368 	mtx_unlock(&psp->ps_mtx);
369 	PROC_UNLOCK(p);
370 
371 	/* Build the argument list for the signal handler. */
372 	sig = bsd_to_linux_signal(sig);
373 
374 	bzero(&frame, sizeof(frame));
375 
376 	frame.sf_sig = sig;
377 	frame.sf_sigmask = *mask;
378 	bsd_to_linux_sigset(mask, &lmask);
379 
380 	/* Build the signal context to be used by sigreturn. */
381 	frame.sf_sc.sc_mask   = lmask.__mask;
382 	frame.sf_sc.sc_gs     = regs->tf_gs;
383 	frame.sf_sc.sc_fs     = regs->tf_fs;
384 	frame.sf_sc.sc_es     = regs->tf_es;
385 	frame.sf_sc.sc_ds     = regs->tf_ds;
386 	frame.sf_sc.sc_edi    = regs->tf_rdi;
387 	frame.sf_sc.sc_esi    = regs->tf_rsi;
388 	frame.sf_sc.sc_ebp    = regs->tf_rbp;
389 	frame.sf_sc.sc_ebx    = regs->tf_rbx;
390 	frame.sf_sc.sc_esp    = regs->tf_rsp;
391 	frame.sf_sc.sc_edx    = regs->tf_rdx;
392 	frame.sf_sc.sc_ecx    = regs->tf_rcx;
393 	frame.sf_sc.sc_eax    = regs->tf_rax;
394 	frame.sf_sc.sc_eip    = regs->tf_rip;
395 	frame.sf_sc.sc_cs     = regs->tf_cs;
396 	frame.sf_sc.sc_eflags = regs->tf_rflags;
397 	frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
398 	frame.sf_sc.sc_ss     = regs->tf_ss;
399 	frame.sf_sc.sc_err    = regs->tf_err;
400 	frame.sf_sc.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
401 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
402 
403 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
404 		/*
405 		 * Process has trashed its stack; give it an illegal
406 		 * instruction to halt it in its tracks.
407 		 */
408 		PROC_LOCK(p);
409 		sigexit(td, SIGILL);
410 	}
411 
412 	/* Build context to run handler in. */
413 	regs->tf_rsp = PTROUT(fp);
414 	regs->tf_rip = linux32_vdso_sigcode;
415 	regs->tf_rdi = PTROUT(catcher);
416 	regs->tf_rflags &= ~(PSL_T | PSL_D);
417 	regs->tf_cs = _ucode32sel;
418 	regs->tf_ss = _udatasel;
419 	regs->tf_ds = _udatasel;
420 	regs->tf_es = _udatasel;
421 	regs->tf_fs = _ufssel;
422 	regs->tf_gs = _ugssel;
423 	regs->tf_flags = TF_HASSEGS;
424 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
425 	PROC_LOCK(p);
426 	mtx_lock(&psp->ps_mtx);
427 }
428 
429 /*
430  * System call to cleanup state after a signal
431  * has been taken.  Reset signal mask and
432  * stack state from context left by sendsig (above).
433  * Return to previous pc and psl as specified by
434  * context left by sendsig. Check carefully to
435  * make sure that the user has not modified the
436  * psl to gain improper privileges or to cause
437  * a machine fault.
438  */
439 int
440 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
441 {
442 	struct l_sigframe frame;
443 	struct trapframe *regs;
444 	int eflags;
445 	ksiginfo_t ksi;
446 
447 	regs = td->td_frame;
448 
449 	/*
450 	 * The trampoline code hands us the sigframe.
451 	 * It is unsafe to keep track of it ourselves, in the event that a
452 	 * program jumps out of a signal handler.
453 	 */
454 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
455 		return (EFAULT);
456 
457 	/* Check for security violations. */
458 	eflags = frame.sf_sc.sc_eflags;
459 	if (!EFL_SECURE(eflags, regs->tf_rflags))
460 		return(EINVAL);
461 
462 	/*
463 	 * Don't allow users to load a valid privileged %cs.  Let the
464 	 * hardware check for invalid selectors, excess privilege in
465 	 * other selectors, invalid %eip's and invalid %esp's.
466 	 */
467 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
468 		ksiginfo_init_trap(&ksi);
469 		ksi.ksi_signo = SIGBUS;
470 		ksi.ksi_code = BUS_OBJERR;
471 		ksi.ksi_trapno = T_PROTFLT;
472 		ksi.ksi_addr = (void *)regs->tf_rip;
473 		trapsignal(td, &ksi);
474 		return(EINVAL);
475 	}
476 
477 	kern_sigprocmask(td, SIG_SETMASK, &frame.sf_sigmask, NULL, 0);
478 
479 	/* Restore signal context. */
480 	regs->tf_rdi    = frame.sf_sc.sc_edi;
481 	regs->tf_rsi    = frame.sf_sc.sc_esi;
482 	regs->tf_rbp    = frame.sf_sc.sc_ebp;
483 	regs->tf_rbx    = frame.sf_sc.sc_ebx;
484 	regs->tf_rdx    = frame.sf_sc.sc_edx;
485 	regs->tf_rcx    = frame.sf_sc.sc_ecx;
486 	regs->tf_rax    = frame.sf_sc.sc_eax;
487 	regs->tf_rip    = frame.sf_sc.sc_eip;
488 	regs->tf_cs     = frame.sf_sc.sc_cs;
489 	regs->tf_ds     = frame.sf_sc.sc_ds;
490 	regs->tf_es     = frame.sf_sc.sc_es;
491 	regs->tf_fs     = frame.sf_sc.sc_fs;
492 	regs->tf_gs     = frame.sf_sc.sc_gs;
493 	regs->tf_rflags = eflags;
494 	regs->tf_rsp    = frame.sf_sc.sc_esp_at_signal;
495 	regs->tf_ss     = frame.sf_sc.sc_ss;
496 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
497 
498 	return (EJUSTRETURN);
499 }
500 
501 /*
502  * System call to cleanup state after a signal
503  * has been taken.  Reset signal mask and
504  * stack state from context left by rt_sendsig (above).
505  * Return to previous pc and psl as specified by
506  * context left by sendsig. Check carefully to
507  * make sure that the user has not modified the
508  * psl to gain improper privileges or to cause
509  * a machine fault.
510  */
511 int
512 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
513 {
514 	struct l_ucontext uc;
515 	struct l_sigcontext *context;
516 	sigset_t bmask;
517 	l_stack_t *lss;
518 	stack_t ss;
519 	struct trapframe *regs;
520 	int eflags;
521 	ksiginfo_t ksi;
522 
523 	regs = td->td_frame;
524 
525 	/*
526 	 * The trampoline code hands us the ucontext.
527 	 * It is unsafe to keep track of it ourselves, in the event that a
528 	 * program jumps out of a signal handler.
529 	 */
530 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
531 		return (EFAULT);
532 
533 	context = &uc.uc_mcontext;
534 
535 	/* Check for security violations. */
536 	eflags = context->sc_eflags;
537 	if (!EFL_SECURE(eflags, regs->tf_rflags))
538 		return(EINVAL);
539 
540 	/*
541 	 * Don't allow users to load a valid privileged %cs.  Let the
542 	 * hardware check for invalid selectors, excess privilege in
543 	 * other selectors, invalid %eip's and invalid %esp's.
544 	 */
545 	if (!CS_SECURE(context->sc_cs)) {
546 		ksiginfo_init_trap(&ksi);
547 		ksi.ksi_signo = SIGBUS;
548 		ksi.ksi_code = BUS_OBJERR;
549 		ksi.ksi_trapno = T_PROTFLT;
550 		ksi.ksi_addr = (void *)regs->tf_rip;
551 		trapsignal(td, &ksi);
552 		return(EINVAL);
553 	}
554 
555 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
556 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
557 
558 	/*
559 	 * Restore signal context
560 	 */
561 	regs->tf_gs	= context->sc_gs;
562 	regs->tf_fs	= context->sc_fs;
563 	regs->tf_es	= context->sc_es;
564 	regs->tf_ds	= context->sc_ds;
565 	regs->tf_rdi    = context->sc_edi;
566 	regs->tf_rsi    = context->sc_esi;
567 	regs->tf_rbp    = context->sc_ebp;
568 	regs->tf_rbx    = context->sc_ebx;
569 	regs->tf_rdx    = context->sc_edx;
570 	regs->tf_rcx    = context->sc_ecx;
571 	regs->tf_rax    = context->sc_eax;
572 	regs->tf_rip    = context->sc_eip;
573 	regs->tf_cs     = context->sc_cs;
574 	regs->tf_rflags = eflags;
575 	regs->tf_rsp    = context->sc_esp_at_signal;
576 	regs->tf_ss     = context->sc_ss;
577 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
578 
579 	/*
580 	 * call sigaltstack & ignore results..
581 	 */
582 	lss = &uc.uc_stack;
583 	ss.ss_sp = PTRIN(lss->ss_sp);
584 	ss.ss_size = lss->ss_size;
585 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
586 
587 	(void)kern_sigaltstack(td, &ss, NULL);
588 
589 	return (EJUSTRETURN);
590 }
591 
592 static int
593 linux32_fetch_syscall_args(struct thread *td)
594 {
595 	struct proc *p;
596 	struct trapframe *frame;
597 	struct syscall_args *sa;
598 
599 	p = td->td_proc;
600 	frame = td->td_frame;
601 	sa = &td->td_sa;
602 
603 	sa->args[0] = frame->tf_rbx;
604 	sa->args[1] = frame->tf_rcx;
605 	sa->args[2] = frame->tf_rdx;
606 	sa->args[3] = frame->tf_rsi;
607 	sa->args[4] = frame->tf_rdi;
608 	sa->args[5] = frame->tf_rbp;	/* Unconfirmed */
609 	sa->code = frame->tf_rax;
610 	sa->original_code = sa->code;
611 
612 	if (sa->code >= p->p_sysent->sv_size)
613 		/* nosys */
614 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
615 	else
616 		sa->callp = &p->p_sysent->sv_table[sa->code];
617 
618 	td->td_retval[0] = 0;
619 	td->td_retval[1] = frame->tf_rdx;
620 
621 	return (0);
622 }
623 
624 static void
625 linux32_set_syscall_retval(struct thread *td, int error)
626 {
627 	struct trapframe *frame = td->td_frame;
628 
629 	cpu_set_syscall_retval(td, error);
630 
631 	if (__predict_false(error != 0)) {
632 		if (error != ERESTART && error != EJUSTRETURN)
633 			frame->tf_rax = bsd_to_linux_errno(error);
634 	}
635 }
636 
637 static void
638 linux32_set_fork_retval(struct thread *td)
639 {
640 	struct trapframe *frame = td->td_frame;
641 
642 	frame->tf_rax = 0;
643 }
644 
645 /*
646  * Clear registers on exec
647  * XXX copied from ia32_signal.c.
648  */
649 static void
650 linux_exec_setregs(struct thread *td, struct image_params *imgp,
651     uintptr_t stack)
652 {
653 	struct trapframe *regs = td->td_frame;
654 	struct pcb *pcb = td->td_pcb;
655 	register_t saved_rflags;
656 
657 	regs = td->td_frame;
658 	pcb = td->td_pcb;
659 
660 	if (td->td_proc->p_md.md_ldt != NULL)
661 		user_ldt_free(td);
662 
663 	critical_enter();
664 	wrmsr(MSR_FSBASE, 0);
665 	wrmsr(MSR_KGSBASE, 0);	/* User value while we're in the kernel */
666 	pcb->pcb_fsbase = 0;
667 	pcb->pcb_gsbase = 0;
668 	critical_exit();
669 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
670 
671 	saved_rflags = regs->tf_rflags & PSL_T;
672 	bzero((char *)regs, sizeof(struct trapframe));
673 	regs->tf_rip = imgp->entry_addr;
674 	regs->tf_rsp = stack;
675 	regs->tf_rflags = PSL_USER | saved_rflags;
676 	regs->tf_gs = _ugssel;
677 	regs->tf_fs = _ufssel;
678 	regs->tf_es = _udatasel;
679 	regs->tf_ds = _udatasel;
680 	regs->tf_ss = _udatasel;
681 	regs->tf_flags = TF_HASSEGS;
682 	regs->tf_cs = _ucode32sel;
683 	regs->tf_rbx = (register_t)imgp->ps_strings;
684 
685 	x86_clear_dbregs(pcb);
686 
687 	fpstate_drop(td);
688 
689 	/* Do full restore on return so that we can change to a different %cs */
690 	set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
691 }
692 
693 /*
694  * XXX copied from ia32_sysvec.c.
695  */
696 static int
697 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
698 {
699 	int argc, envc, error;
700 	u_int32_t *vectp;
701 	char *stringp;
702 	uintptr_t destp, ustringp;
703 	struct linux32_ps_strings *arginfo;
704 	char canary[LINUX_AT_RANDOM_LEN];
705 	size_t execpath_len;
706 
707 	arginfo = (struct linux32_ps_strings *)PROC_PS_STRINGS(imgp->proc);
708 	destp = (uintptr_t)arginfo;
709 
710 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
711 		execpath_len = strlen(imgp->execpath) + 1;
712 		destp -= execpath_len;
713 		destp = rounddown2(destp, sizeof(uint32_t));
714 		imgp->execpathp = (void *)destp;
715 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
716 		if (error != 0)
717 			return (error);
718 	}
719 
720 	/* Prepare the canary for SSP. */
721 	arc4rand(canary, sizeof(canary), 0);
722 	destp -= roundup(sizeof(canary), sizeof(uint32_t));
723 	imgp->canary = (void *)destp;
724 	error = copyout(canary, imgp->canary, sizeof(canary));
725 	if (error != 0)
726 		return (error);
727 
728 	/* Allocate room for the argument and environment strings. */
729 	destp -= ARG_MAX - imgp->args->stringspace;
730 	destp = rounddown2(destp, sizeof(uint32_t));
731 	ustringp = destp;
732 
733 	if (imgp->auxargs) {
734 		/*
735 		 * Allocate room on the stack for the ELF auxargs
736 		 * array.  It has LINUX_AT_COUNT entries.
737 		 */
738 		destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
739 		destp = rounddown2(destp, sizeof(uint32_t));
740 	}
741 
742 	vectp = (uint32_t *)destp;
743 
744 	/*
745 	 * Allocate room for the argv[] and env vectors including the
746 	 * terminating NULL pointers.
747 	 */
748 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
749 
750 	/* vectp also becomes our initial stack base. */
751 	*stack_base = (uintptr_t)vectp;
752 
753 	stringp = imgp->args->begin_argv;
754 	argc = imgp->args->argc;
755 	envc = imgp->args->envc;
756 
757 	/* Copy out strings - arguments and environment. */
758 	error = copyout(stringp, (void *)ustringp,
759 	    ARG_MAX - imgp->args->stringspace);
760 	if (error != 0)
761 		return (error);
762 
763 	/* Fill in "ps_strings" struct for ps, w, etc. */
764 	if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 ||
765 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
766 		return (EFAULT);
767 
768 	/* Fill in argument portion of vector table. */
769 	for (; argc > 0; --argc) {
770 		if (suword32(vectp++, ustringp) != 0)
771 			return (EFAULT);
772 		while (*stringp++ != 0)
773 			ustringp++;
774 		ustringp++;
775 	}
776 
777 	/* A null vector table pointer separates the argp's from the envp's. */
778 	if (suword32(vectp++, 0) != 0)
779 		return (EFAULT);
780 
781 	if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 ||
782 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
783 		return (EFAULT);
784 
785 	/* Fill in environment portion of vector table. */
786 	for (; envc > 0; --envc) {
787 		if (suword32(vectp++, ustringp) != 0)
788 			return (EFAULT);
789 		while (*stringp++ != 0)
790 			ustringp++;
791 		ustringp++;
792 	}
793 
794 	/* The end of the vector table is a null pointer. */
795 	if (suword32(vectp, 0) != 0)
796 		return (EFAULT);
797 
798 	if (imgp->auxargs) {
799 		vectp++;
800 		error = imgp->sysent->sv_copyout_auxargs(imgp,
801 		    (uintptr_t)vectp);
802 		if (error != 0)
803 			return (error);
804 	}
805 
806 	return (0);
807 }
808 
809 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
810     "32-bit Linux emulation");
811 
812 static u_long	linux32_maxdsiz = LINUX32_MAXDSIZ;
813 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
814     &linux32_maxdsiz, 0, "");
815 static u_long	linux32_maxssiz = LINUX32_MAXSSIZ;
816 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
817     &linux32_maxssiz, 0, "");
818 static u_long	linux32_maxvmem = LINUX32_MAXVMEM;
819 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
820     &linux32_maxvmem, 0, "");
821 bool linux32_emulate_i386 = false;
822 SYSCTL_BOOL(_compat_linux32, OID_AUTO, emulate_i386, CTLFLAG_RWTUN,
823     &linux32_emulate_i386, 0, "Emulate the real i386");
824 
825 static void
826 linux32_fixlimit(struct rlimit *rl, int which)
827 {
828 
829 	switch (which) {
830 	case RLIMIT_DATA:
831 		if (linux32_maxdsiz != 0) {
832 			if (rl->rlim_cur > linux32_maxdsiz)
833 				rl->rlim_cur = linux32_maxdsiz;
834 			if (rl->rlim_max > linux32_maxdsiz)
835 				rl->rlim_max = linux32_maxdsiz;
836 		}
837 		break;
838 	case RLIMIT_STACK:
839 		if (linux32_maxssiz != 0) {
840 			if (rl->rlim_cur > linux32_maxssiz)
841 				rl->rlim_cur = linux32_maxssiz;
842 			if (rl->rlim_max > linux32_maxssiz)
843 				rl->rlim_max = linux32_maxssiz;
844 		}
845 		break;
846 	case RLIMIT_VMEM:
847 		if (linux32_maxvmem != 0) {
848 			if (rl->rlim_cur > linux32_maxvmem)
849 				rl->rlim_cur = linux32_maxvmem;
850 			if (rl->rlim_max > linux32_maxvmem)
851 				rl->rlim_max = linux32_maxvmem;
852 		}
853 		break;
854 	}
855 }
856 
857 struct sysentvec elf_linux_sysvec = {
858 	.sv_size	= LINUX32_SYS_MAXSYSCALL,
859 	.sv_table	= linux32_sysent,
860 	.sv_fixup	= linux_fixup_elf,
861 	.sv_sendsig	= linux_sendsig,
862 	.sv_sigcode	= &_binary_linux32_vdso_so_o_start,
863 	.sv_szsigcode	= &linux_szsigcode,
864 	.sv_name	= "Linux ELF32",
865 	.sv_coredump	= elf32_coredump,
866 	.sv_elf_core_osabi = ELFOSABI_NONE,
867 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
868 	.sv_elf_core_prepare_notes = linux32_prepare_notes,
869 	.sv_imgact_try	= linux_exec_imgact_try,
870 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
871 	.sv_minuser	= VM_MIN_ADDRESS,
872 	.sv_maxuser	= LINUX32_MAXUSER,
873 	.sv_usrstack	= LINUX32_USRSTACK,
874 	.sv_psstrings	= LINUX32_PS_STRINGS,
875 	.sv_psstringssz	= sizeof(struct linux32_ps_strings),
876 	.sv_stackprot	= VM_PROT_ALL,
877 	.sv_copyout_auxargs = linux_copyout_auxargs,
878 	.sv_copyout_strings = linux_copyout_strings,
879 	.sv_setregs	= linux_exec_setregs,
880 	.sv_fixlimit	= linux32_fixlimit,
881 	.sv_maxssiz	= &linux32_maxssiz,
882 	.sv_flags	= SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP |
883 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
884 	.sv_set_syscall_retval = linux32_set_syscall_retval,
885 	.sv_fetch_syscall_args = linux32_fetch_syscall_args,
886 	.sv_syscallnames = linux32_syscallnames,
887 	.sv_shared_page_base = LINUX32_SHAREDPAGE,
888 	.sv_shared_page_len = PAGE_SIZE,
889 	.sv_schedtail	= linux_schedtail,
890 	.sv_thread_detach = linux_thread_detach,
891 	.sv_trap	= NULL,
892 	.sv_onexec	= linux_on_exec_vmspace,
893 	.sv_onexit	= linux_on_exit,
894 	.sv_ontdexit	= linux_thread_dtor,
895 	.sv_setid_allowed = &linux_setid_allowed_query,
896 	.sv_set_fork_retval = linux32_set_fork_retval,
897 };
898 
899 static int
900 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
901 {
902 	int error;
903 
904 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
905 	    LINUX32_VDSOPAGE_SIZE, imgp);
906 	if (error == 0)
907 		linux_on_exec(p, imgp);
908 	return (error);
909 }
910 
911 /*
912  * linux_vdso_install() and linux_exec_sysvec_init() must be called
913  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
914  */
915 static void
916 linux_exec_sysvec_init(void *param)
917 {
918 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
919 	struct sysentvec *sv;
920 	ptrdiff_t tkoff;
921 
922 	sv = param;
923 	/* Fill timekeep_base */
924 	exec_sysvec_init(sv);
925 
926 	tkoff = kern_timekeep_base - linux_vdso_base;
927 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
928 	*ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
929 
930 	tkoff = kern_tsc_selector - linux_vdso_base;
931 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
932 	*ktsc_selector = linux_vdso_tsc_selector_idx();
933 	if (bootverbose)
934 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
935 
936 	tkoff = kern_cpu_selector - linux_vdso_base;
937 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
938 	*ktsc_selector = linux_vdso_cpu_selector_idx();
939 	if (bootverbose)
940 		printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector);
941 }
942 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
943     linux_exec_sysvec_init, &elf_linux_sysvec);
944 
945 static void
946 linux_vdso_install(const void *param)
947 {
948 	char *vdso_start = &_binary_linux32_vdso_so_o_start;
949 	char *vdso_end = &_binary_linux32_vdso_so_o_end;
950 
951 	linux_szsigcode = vdso_end - vdso_start;
952 	MPASS(linux_szsigcode <= LINUX32_VDSOPAGE_SIZE);
953 
954 	linux_vdso_base = LINUX32_VDSOPAGE;
955 
956 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
957 
958 	linux_vdso_obj = __elfN(linux_shared_page_init)
959 	    (&linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
960 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
961 
962 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
963 }
964 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
965     linux_vdso_install, NULL);
966 
967 static void
968 linux_vdso_deinstall(const void *param)
969 {
970 
971 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
972 	    linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
973 }
974 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
975     linux_vdso_deinstall, NULL);
976 
977 static void
978 linux_vdso_reloc(char *mapping, Elf_Addr offset)
979 {
980 	const Elf_Shdr *shdr;
981 	const Elf_Rel *rel;
982 	const Elf_Ehdr *ehdr;
983 	Elf32_Addr *where;
984 	Elf_Size rtype, symidx;
985 	Elf32_Addr addr, addend;
986 	int i, relcnt;
987 
988 	MPASS(offset != 0);
989 
990 	relcnt = 0;
991 	ehdr = (const Elf_Ehdr *)mapping;
992 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
993 	for (i = 0; i < ehdr->e_shnum; i++)
994 	{
995 		switch (shdr[i].sh_type) {
996 		case SHT_REL:
997 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
998 			relcnt = shdr[i].sh_size / sizeof(*rel);
999 			break;
1000 		case SHT_RELA:
1001 			printf("Linux i386 vDSO: unexpected Rela section\n");
1002 			break;
1003 		}
1004 	}
1005 
1006 	for (i = 0; i < relcnt; i++, rel++) {
1007 		where = (Elf32_Addr *)(mapping + rel->r_offset);
1008 		addend = *where;
1009 		rtype = ELF_R_TYPE(rel->r_info);
1010 		symidx = ELF_R_SYM(rel->r_info);
1011 
1012 		switch (rtype) {
1013 		case R_386_NONE:	/* none */
1014 			break;
1015 
1016 		case R_386_RELATIVE:	/* B + A */
1017 			addr = (Elf32_Addr)PTROUT(offset + addend);
1018 			if (*where != addr)
1019 				*where = addr;
1020 			break;
1021 
1022 		case R_386_IRELATIVE:
1023 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
1024 			    "symbol index %ld\n", (intmax_t)symidx);
1025 			break;
1026 		default:
1027 			printf("Linux i386 vDSO: unexpected relocation type %ld, "
1028 			    "symbol index %ld\n", (intmax_t)rtype, (intmax_t)symidx);
1029 		}
1030 	}
1031 }
1032 
1033 static char GNU_ABI_VENDOR[] = "GNU";
1034 static int GNULINUX_ABI_DESC = 0;
1035 
1036 static bool
1037 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
1038 {
1039 	const Elf32_Word *desc;
1040 	uintptr_t p;
1041 
1042 	p = (uintptr_t)(note + 1);
1043 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1044 
1045 	desc = (const Elf32_Word *)p;
1046 	if (desc[0] != GNULINUX_ABI_DESC)
1047 		return (false);
1048 
1049 	/*
1050 	 * For Linux we encode osrel using the Linux convention of
1051 	 * 	(version << 16) | (major << 8) | (minor)
1052 	 * See macro in linux_mib.h
1053 	 */
1054 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
1055 
1056 	return (true);
1057 }
1058 
1059 static Elf_Brandnote linux32_brandnote = {
1060 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
1061 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
1062 	.hdr.n_type	= 1,
1063 	.vendor		= GNU_ABI_VENDOR,
1064 	.flags		= BN_TRANSLATE_OSREL,
1065 	.trans_osrel	= linux32_trans_osrel
1066 };
1067 
1068 static Elf32_Brandinfo linux_brand = {
1069 	.brand		= ELFOSABI_LINUX,
1070 	.machine	= EM_386,
1071 	.compat_3_brand	= "Linux",
1072 	.emul_path	= linux_emul_path,
1073 	.interp_path	= "/lib/ld-linux.so.1",
1074 	.sysvec		= &elf_linux_sysvec,
1075 	.interp_newpath	= NULL,
1076 	.brand_note	= &linux32_brandnote,
1077 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1078 };
1079 
1080 static Elf32_Brandinfo linux_glibc2brand = {
1081 	.brand		= ELFOSABI_LINUX,
1082 	.machine	= EM_386,
1083 	.compat_3_brand	= "Linux",
1084 	.emul_path	= linux_emul_path,
1085 	.interp_path	= "/lib/ld-linux.so.2",
1086 	.sysvec		= &elf_linux_sysvec,
1087 	.interp_newpath	= NULL,
1088 	.brand_note	= &linux32_brandnote,
1089 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1090 };
1091 
1092 static Elf32_Brandinfo linux_muslbrand = {
1093 	.brand		= ELFOSABI_LINUX,
1094 	.machine	= EM_386,
1095 	.compat_3_brand	= "Linux",
1096 	.emul_path	= linux_emul_path,
1097 	.interp_path	= "/lib/ld-musl-i386.so.1",
1098 	.sysvec		= &elf_linux_sysvec,
1099 	.interp_newpath	= NULL,
1100 	.brand_note	= &linux32_brandnote,
1101 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
1102 			    LINUX_BI_FUTEX_REQUEUE
1103 };
1104 
1105 Elf32_Brandinfo *linux_brandlist[] = {
1106 	&linux_brand,
1107 	&linux_glibc2brand,
1108 	&linux_muslbrand,
1109 	NULL
1110 };
1111 
1112 static int
1113 linux_elf_modevent(module_t mod, int type, void *data)
1114 {
1115 	Elf32_Brandinfo **brandinfo;
1116 	int error;
1117 	struct linux_ioctl_handler **lihp;
1118 
1119 	error = 0;
1120 
1121 	switch(type) {
1122 	case MOD_LOAD:
1123 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1124 		     ++brandinfo)
1125 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1126 				error = EINVAL;
1127 		if (error == 0) {
1128 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1129 				linux32_ioctl_register_handler(*lihp);
1130 			stclohz = (stathz ? stathz : hz);
1131 			if (bootverbose)
1132 				printf("Linux i386 ELF exec handler installed\n");
1133 		} else
1134 			printf("cannot insert Linux i386 ELF brand handler\n");
1135 		break;
1136 	case MOD_UNLOAD:
1137 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1138 		     ++brandinfo)
1139 			if (elf32_brand_inuse(*brandinfo))
1140 				error = EBUSY;
1141 		if (error == 0) {
1142 			for (brandinfo = &linux_brandlist[0];
1143 			     *brandinfo != NULL; ++brandinfo)
1144 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1145 					error = EINVAL;
1146 		}
1147 		if (error == 0) {
1148 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1149 				linux32_ioctl_unregister_handler(*lihp);
1150 			if (bootverbose)
1151 				printf("Linux i386 ELF exec handler removed\n");
1152 		} else
1153 			printf("Could not deinstall Linux i386 ELF interpreter entry\n");
1154 		break;
1155 	default:
1156 		return (EOPNOTSUPP);
1157 	}
1158 	return (error);
1159 }
1160 
1161 static moduledata_t linux_elf_mod = {
1162 	"linuxelf",
1163 	linux_elf_modevent,
1164 	0
1165 };
1166 
1167 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1168 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1169 FEATURE(linux, "Linux 32bit support");
1170