xref: /freebsd/sys/amd64/linux32/linux32_sysvec.c (revision 5e801ac66d24704442eba426ed13c3effb8a34e7)
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 MODULE_VERSION(linux, 1);
95 
96 #define	LINUX32_MAXUSER		((1ul << 32) - PAGE_SIZE)
97 #define	LINUX32_VDSOPAGE_SIZE	PAGE_SIZE * 2
98 #define	LINUX32_VDSOPAGE	(LINUX32_MAXUSER - LINUX32_VDSOPAGE_SIZE)
99 #define	LINUX32_SHAREDPAGE	(LINUX32_VDSOPAGE - PAGE_SIZE)
100 				/*
101 				 * PAGE_SIZE - the size
102 				 * of the native SHAREDPAGE
103 				 */
104 #define	LINUX32_USRSTACK	LINUX32_SHAREDPAGE
105 
106 static int linux_szsigcode;
107 static vm_object_t linux_vdso_obj;
108 static char *linux_vdso_mapping;
109 extern char _binary_linux32_vdso_so_o_start;
110 extern char _binary_linux32_vdso_so_o_end;
111 static vm_offset_t linux_vdso_base;
112 
113 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
114 
115 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
116 
117 static int	linux_fixup_elf(uintptr_t *stack_base,
118 		    struct image_params *iparams);
119 static int	linux_copyout_strings(struct image_params *imgp,
120 		    uintptr_t *stack_base);
121 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
122 static void	linux_exec_setregs(struct thread *td,
123 				   struct image_params *imgp, uintptr_t stack);
124 static void	linux_exec_sysvec_init(void *param);
125 static int	linux_on_exec_vmspace(struct proc *p,
126 		    struct image_params *imgp);
127 static void	linux32_fixlimit(struct rlimit *rl, int which);
128 static bool	linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
129 static void	linux_vdso_install(const void *param);
130 static void	linux_vdso_deinstall(const void *param);
131 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
132 static void	linux32_set_fork_retval(struct thread *td);
133 static void	linux32_set_syscall_retval(struct thread *td, int error);
134 
135 #define LINUX_T_UNKNOWN  255
136 static int _bsd_to_linux_trapcode[] = {
137 	LINUX_T_UNKNOWN,	/* 0 */
138 	6,			/* 1  T_PRIVINFLT */
139 	LINUX_T_UNKNOWN,	/* 2 */
140 	3,			/* 3  T_BPTFLT */
141 	LINUX_T_UNKNOWN,	/* 4 */
142 	LINUX_T_UNKNOWN,	/* 5 */
143 	16,			/* 6  T_ARITHTRAP */
144 	254,			/* 7  T_ASTFLT */
145 	LINUX_T_UNKNOWN,	/* 8 */
146 	13,			/* 9  T_PROTFLT */
147 	1,			/* 10 T_TRCTRAP */
148 	LINUX_T_UNKNOWN,	/* 11 */
149 	14,			/* 12 T_PAGEFLT */
150 	LINUX_T_UNKNOWN,	/* 13 */
151 	17,			/* 14 T_ALIGNFLT */
152 	LINUX_T_UNKNOWN,	/* 15 */
153 	LINUX_T_UNKNOWN,	/* 16 */
154 	LINUX_T_UNKNOWN,	/* 17 */
155 	0,			/* 18 T_DIVIDE */
156 	2,			/* 19 T_NMI */
157 	4,			/* 20 T_OFLOW */
158 	5,			/* 21 T_BOUND */
159 	7,			/* 22 T_DNA */
160 	8,			/* 23 T_DOUBLEFLT */
161 	9,			/* 24 T_FPOPFLT */
162 	10,			/* 25 T_TSSFLT */
163 	11,			/* 26 T_SEGNPFLT */
164 	12,			/* 27 T_STKFLT */
165 	18,			/* 28 T_MCHK */
166 	19,			/* 29 T_XMMFLT */
167 	15			/* 30 T_RESERVED */
168 };
169 #define bsd_to_linux_trapcode(code) \
170     ((code)<nitems(_bsd_to_linux_trapcode)? \
171      _bsd_to_linux_trapcode[(code)]: \
172      LINUX_T_UNKNOWN)
173 
174 struct linux32_ps_strings {
175 	u_int32_t ps_argvstr;	/* first of 0 or more argument strings */
176 	u_int ps_nargvstr;	/* the number of argument strings */
177 	u_int32_t ps_envstr;	/* first of 0 or more environment strings */
178 	u_int ps_nenvstr;	/* the number of environment strings */
179 };
180 #define	LINUX32_PS_STRINGS	(LINUX32_USRSTACK - \
181 				    sizeof(struct linux32_ps_strings))
182 
183 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
184 LINUX_VDSO_SYM_INTPTR(__kernel_sigreturn);
185 LINUX_VDSO_SYM_INTPTR(__kernel_rt_sigreturn);
186 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
187 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
188 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
189 LINUX_VDSO_SYM_CHAR(linux_platform);
190 
191 /*
192  * If FreeBSD & Linux have a difference of opinion about what a trap
193  * means, deal with it here.
194  *
195  * MPSAFE
196  */
197 static int
198 linux_translate_traps(int signal, int trap_code)
199 {
200 	if (signal != SIGBUS)
201 		return (signal);
202 	switch (trap_code) {
203 	case T_PROTFLT:
204 	case T_TSSFLT:
205 	case T_DOUBLEFLT:
206 	case T_PAGEFLT:
207 		return (SIGSEGV);
208 	default:
209 		return (signal);
210 	}
211 }
212 
213 static int
214 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
215 {
216 	Elf32_Auxargs *args;
217 	Elf32_Auxinfo *argarray, *pos;
218 	int error, issetugid;
219 
220 	args = (Elf32_Auxargs *)imgp->auxargs;
221 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
222 	    M_WAITOK | M_ZERO);
223 
224 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
225 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, __kernel_vsyscall);
226 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
227 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
228 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
229 
230 	/*
231 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
232 	 * as it has appeared in the 2.4.0-rc7 first time.
233 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
234 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
235 	 * is not present.
236 	 * Also see linux_times() implementation.
237 	 */
238 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
239 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
240 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
241 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
242 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
243 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
244 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
245 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
246 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
247 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
248 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
249 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
250 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
251 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
252 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
253 	if (imgp->execpathp != 0)
254 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
255 	if (args->execfd != -1)
256 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
257 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
258 	AUXARGS_ENTRY(pos, AT_NULL, 0);
259 
260 	free(imgp->auxargs, M_TEMP);
261 	imgp->auxargs = NULL;
262 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
263 
264 	error = copyout(argarray, (void *)base,
265 	    sizeof(*argarray) * LINUX_AT_COUNT);
266 	free(argarray, M_TEMP);
267 	return (error);
268 }
269 
270 static int
271 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
272 {
273 	Elf32_Addr *base;
274 
275 	base = (Elf32_Addr *)*stack_base;
276 	base--;
277 	if (suword32(base, (uint32_t)imgp->args->argc) == -1)
278 		return (EFAULT);
279 	*stack_base = (uintptr_t)base;
280 	return (0);
281 }
282 
283 static void
284 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
285 {
286 	struct thread *td = curthread;
287 	struct proc *p = td->td_proc;
288 	struct sigacts *psp;
289 	struct trapframe *regs;
290 	struct l_rt_sigframe *fp, frame;
291 	int oonstack;
292 	int sig;
293 	int code;
294 
295 	sig = ksi->ksi_signo;
296 	code = ksi->ksi_code;
297 	PROC_LOCK_ASSERT(p, MA_OWNED);
298 	psp = p->p_sigacts;
299 	mtx_assert(&psp->ps_mtx, MA_OWNED);
300 	regs = td->td_frame;
301 	oonstack = sigonstack(regs->tf_rsp);
302 
303 	/* Allocate space for the signal handler context. */
304 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
305 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
306 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
307 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
308 	} else
309 		fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
310 	mtx_unlock(&psp->ps_mtx);
311 
312 	/* Build the argument list for the signal handler. */
313 	sig = bsd_to_linux_signal(sig);
314 
315 	bzero(&frame, sizeof(frame));
316 
317 	frame.sf_handler = PTROUT(catcher);
318 	frame.sf_sig = sig;
319 	frame.sf_siginfo = PTROUT(&fp->sf_si);
320 	frame.sf_ucontext = PTROUT(&fp->sf_sc);
321 
322 	/* Fill in POSIX parts. */
323 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
324 
325 	/*
326 	 * Build the signal context to be used by sigreturn and libgcc unwind.
327 	 */
328 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
329 	frame.sf_sc.uc_link = 0;		/* XXX ??? */
330 
331 	frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
332 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
333 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
334 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
335 	PROC_UNLOCK(p);
336 
337 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
338 
339 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
340 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_rdi;
341 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_rsi;
342 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_rbp;
343 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_rbx;
344 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_rsp;
345 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_rdx;
346 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_rcx;
347 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_rax;
348 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_rip;
349 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
350 	frame.sf_sc.uc_mcontext.sc_gs     = regs->tf_gs;
351 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
352 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
353 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
354 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
355 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
356 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
357 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
358 	frame.sf_sc.uc_mcontext.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
359 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
360 
361 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
362 		/*
363 		 * Process has trashed its stack; give it an illegal
364 		 * instruction to halt it in its tracks.
365 		 */
366 		PROC_LOCK(p);
367 		sigexit(td, SIGILL);
368 	}
369 
370 	/* Build context to run handler in. */
371 	regs->tf_rsp = PTROUT(fp);
372 	regs->tf_rip = __kernel_rt_sigreturn;
373 	regs->tf_rflags &= ~(PSL_T | PSL_D);
374 	regs->tf_cs = _ucode32sel;
375 	regs->tf_ss = _udatasel;
376 	regs->tf_ds = _udatasel;
377 	regs->tf_es = _udatasel;
378 	regs->tf_fs = _ufssel;
379 	regs->tf_gs = _ugssel;
380 	regs->tf_flags = TF_HASSEGS;
381 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
382 	PROC_LOCK(p);
383 	mtx_lock(&psp->ps_mtx);
384 }
385 
386 /*
387  * Send an interrupt to process.
388  *
389  * Stack is set up to allow sigcode stored
390  * in u. to call routine, followed by kcall
391  * to sigreturn routine below.  After sigreturn
392  * resets the signal mask, the stack, and the
393  * frame pointer, it returns to the user
394  * specified pc, psl.
395  */
396 static void
397 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
398 {
399 	struct thread *td = curthread;
400 	struct proc *p = td->td_proc;
401 	struct sigacts *psp;
402 	struct trapframe *regs;
403 	struct l_sigframe *fp, frame;
404 	l_sigset_t lmask;
405 	int oonstack;
406 	int sig, code;
407 
408 	sig = ksi->ksi_signo;
409 	code = ksi->ksi_code;
410 	PROC_LOCK_ASSERT(p, MA_OWNED);
411 	psp = p->p_sigacts;
412 	mtx_assert(&psp->ps_mtx, MA_OWNED);
413 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
414 		/* Signal handler installed with SA_SIGINFO. */
415 		linux_rt_sendsig(catcher, ksi, mask);
416 		return;
417 	}
418 
419 	regs = td->td_frame;
420 	oonstack = sigonstack(regs->tf_rsp);
421 
422 	/* Allocate space for the signal handler context. */
423 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
424 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
425 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
426 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
427 	} else
428 		fp = (struct l_sigframe *)regs->tf_rsp - 1;
429 	mtx_unlock(&psp->ps_mtx);
430 	PROC_UNLOCK(p);
431 
432 	/* Build the argument list for the signal handler. */
433 	sig = bsd_to_linux_signal(sig);
434 
435 	bzero(&frame, sizeof(frame));
436 
437 	frame.sf_handler = PTROUT(catcher);
438 	frame.sf_sig = sig;
439 
440 	bsd_to_linux_sigset(mask, &lmask);
441 
442 	/* Build the signal context to be used by sigreturn. */
443 	frame.sf_sc.sc_mask   = lmask.__mask;
444 	frame.sf_sc.sc_gs     = regs->tf_gs;
445 	frame.sf_sc.sc_fs     = regs->tf_fs;
446 	frame.sf_sc.sc_es     = regs->tf_es;
447 	frame.sf_sc.sc_ds     = regs->tf_ds;
448 	frame.sf_sc.sc_edi    = regs->tf_rdi;
449 	frame.sf_sc.sc_esi    = regs->tf_rsi;
450 	frame.sf_sc.sc_ebp    = regs->tf_rbp;
451 	frame.sf_sc.sc_ebx    = regs->tf_rbx;
452 	frame.sf_sc.sc_esp    = regs->tf_rsp;
453 	frame.sf_sc.sc_edx    = regs->tf_rdx;
454 	frame.sf_sc.sc_ecx    = regs->tf_rcx;
455 	frame.sf_sc.sc_eax    = regs->tf_rax;
456 	frame.sf_sc.sc_eip    = regs->tf_rip;
457 	frame.sf_sc.sc_cs     = regs->tf_cs;
458 	frame.sf_sc.sc_eflags = regs->tf_rflags;
459 	frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
460 	frame.sf_sc.sc_ss     = regs->tf_ss;
461 	frame.sf_sc.sc_err    = regs->tf_err;
462 	frame.sf_sc.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
463 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
464 
465 	frame.sf_extramask[0] = lmask.__mask;
466 
467 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
468 		/*
469 		 * Process has trashed its stack; give it an illegal
470 		 * instruction to halt it in its tracks.
471 		 */
472 		PROC_LOCK(p);
473 		sigexit(td, SIGILL);
474 	}
475 
476 	/* Build context to run handler in. */
477 	regs->tf_rsp = PTROUT(fp);
478 	regs->tf_rip = __kernel_sigreturn;
479 	regs->tf_rflags &= ~(PSL_T | PSL_D);
480 	regs->tf_cs = _ucode32sel;
481 	regs->tf_ss = _udatasel;
482 	regs->tf_ds = _udatasel;
483 	regs->tf_es = _udatasel;
484 	regs->tf_fs = _ufssel;
485 	regs->tf_gs = _ugssel;
486 	regs->tf_flags = TF_HASSEGS;
487 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
488 	PROC_LOCK(p);
489 	mtx_lock(&psp->ps_mtx);
490 }
491 
492 /*
493  * System call to cleanup state after a signal
494  * has been taken.  Reset signal mask and
495  * stack state from context left by sendsig (above).
496  * Return to previous pc and psl as specified by
497  * context left by sendsig. Check carefully to
498  * make sure that the user has not modified the
499  * psl to gain improper privileges or to cause
500  * a machine fault.
501  */
502 int
503 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
504 {
505 	struct l_sigframe frame;
506 	struct trapframe *regs;
507 	sigset_t bmask;
508 	l_sigset_t lmask;
509 	int eflags;
510 	ksiginfo_t ksi;
511 
512 	regs = td->td_frame;
513 
514 	/*
515 	 * The trampoline code hands us the sigframe.
516 	 * It is unsafe to keep track of it ourselves, in the event that a
517 	 * program jumps out of a signal handler.
518 	 */
519 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
520 		return (EFAULT);
521 
522 	/* Check for security violations. */
523 	eflags = frame.sf_sc.sc_eflags;
524 	if (!EFL_SECURE(eflags, regs->tf_rflags))
525 		return(EINVAL);
526 
527 	/*
528 	 * Don't allow users to load a valid privileged %cs.  Let the
529 	 * hardware check for invalid selectors, excess privilege in
530 	 * other selectors, invalid %eip's and invalid %esp's.
531 	 */
532 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
533 		ksiginfo_init_trap(&ksi);
534 		ksi.ksi_signo = SIGBUS;
535 		ksi.ksi_code = BUS_OBJERR;
536 		ksi.ksi_trapno = T_PROTFLT;
537 		ksi.ksi_addr = (void *)regs->tf_rip;
538 		trapsignal(td, &ksi);
539 		return(EINVAL);
540 	}
541 
542 	lmask.__mask = frame.sf_sc.sc_mask;
543 	lmask.__mask = frame.sf_extramask[0];
544 	linux_to_bsd_sigset(&lmask, &bmask);
545 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
546 
547 	/* Restore signal context. */
548 	regs->tf_rdi    = frame.sf_sc.sc_edi;
549 	regs->tf_rsi    = frame.sf_sc.sc_esi;
550 	regs->tf_rbp    = frame.sf_sc.sc_ebp;
551 	regs->tf_rbx    = frame.sf_sc.sc_ebx;
552 	regs->tf_rdx    = frame.sf_sc.sc_edx;
553 	regs->tf_rcx    = frame.sf_sc.sc_ecx;
554 	regs->tf_rax    = frame.sf_sc.sc_eax;
555 	regs->tf_rip    = frame.sf_sc.sc_eip;
556 	regs->tf_cs     = frame.sf_sc.sc_cs;
557 	regs->tf_ds     = frame.sf_sc.sc_ds;
558 	regs->tf_es     = frame.sf_sc.sc_es;
559 	regs->tf_fs     = frame.sf_sc.sc_fs;
560 	regs->tf_gs     = frame.sf_sc.sc_gs;
561 	regs->tf_rflags = eflags;
562 	regs->tf_rsp    = frame.sf_sc.sc_esp_at_signal;
563 	regs->tf_ss     = frame.sf_sc.sc_ss;
564 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
565 
566 	return (EJUSTRETURN);
567 }
568 
569 /*
570  * System call to cleanup state after a signal
571  * has been taken.  Reset signal mask and
572  * stack state from context left by rt_sendsig (above).
573  * Return to previous pc and psl as specified by
574  * context left by sendsig. Check carefully to
575  * make sure that the user has not modified the
576  * psl to gain improper privileges or to cause
577  * a machine fault.
578  */
579 int
580 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
581 {
582 	struct l_ucontext uc;
583 	struct l_sigcontext *context;
584 	sigset_t bmask;
585 	l_stack_t *lss;
586 	stack_t ss;
587 	struct trapframe *regs;
588 	int eflags;
589 	ksiginfo_t ksi;
590 
591 	regs = td->td_frame;
592 
593 	/*
594 	 * The trampoline code hands us the ucontext.
595 	 * It is unsafe to keep track of it ourselves, in the event that a
596 	 * program jumps out of a signal handler.
597 	 */
598 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
599 		return (EFAULT);
600 
601 	context = &uc.uc_mcontext;
602 
603 	/* Check for security violations. */
604 	eflags = context->sc_eflags;
605 	if (!EFL_SECURE(eflags, regs->tf_rflags))
606 		return(EINVAL);
607 
608 	/*
609 	 * Don't allow users to load a valid privileged %cs.  Let the
610 	 * hardware check for invalid selectors, excess privilege in
611 	 * other selectors, invalid %eip's and invalid %esp's.
612 	 */
613 	if (!CS_SECURE(context->sc_cs)) {
614 		ksiginfo_init_trap(&ksi);
615 		ksi.ksi_signo = SIGBUS;
616 		ksi.ksi_code = BUS_OBJERR;
617 		ksi.ksi_trapno = T_PROTFLT;
618 		ksi.ksi_addr = (void *)regs->tf_rip;
619 		trapsignal(td, &ksi);
620 		return(EINVAL);
621 	}
622 
623 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
624 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
625 
626 	/*
627 	 * Restore signal context
628 	 */
629 	regs->tf_gs	= context->sc_gs;
630 	regs->tf_fs	= context->sc_fs;
631 	regs->tf_es	= context->sc_es;
632 	regs->tf_ds	= context->sc_ds;
633 	regs->tf_rdi    = context->sc_edi;
634 	regs->tf_rsi    = context->sc_esi;
635 	regs->tf_rbp    = context->sc_ebp;
636 	regs->tf_rbx    = context->sc_ebx;
637 	regs->tf_rdx    = context->sc_edx;
638 	regs->tf_rcx    = context->sc_ecx;
639 	regs->tf_rax    = context->sc_eax;
640 	regs->tf_rip    = context->sc_eip;
641 	regs->tf_cs     = context->sc_cs;
642 	regs->tf_rflags = eflags;
643 	regs->tf_rsp    = context->sc_esp_at_signal;
644 	regs->tf_ss     = context->sc_ss;
645 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
646 
647 	/*
648 	 * call sigaltstack & ignore results..
649 	 */
650 	lss = &uc.uc_stack;
651 	ss.ss_sp = PTRIN(lss->ss_sp);
652 	ss.ss_size = lss->ss_size;
653 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
654 
655 	(void)kern_sigaltstack(td, &ss, NULL);
656 
657 	return (EJUSTRETURN);
658 }
659 
660 static int
661 linux32_fetch_syscall_args(struct thread *td)
662 {
663 	struct proc *p;
664 	struct trapframe *frame;
665 	struct syscall_args *sa;
666 
667 	p = td->td_proc;
668 	frame = td->td_frame;
669 	sa = &td->td_sa;
670 
671 	sa->args[0] = frame->tf_rbx;
672 	sa->args[1] = frame->tf_rcx;
673 	sa->args[2] = frame->tf_rdx;
674 	sa->args[3] = frame->tf_rsi;
675 	sa->args[4] = frame->tf_rdi;
676 	sa->args[5] = frame->tf_rbp;	/* Unconfirmed */
677 	sa->code = frame->tf_rax;
678 	sa->original_code = sa->code;
679 
680 	if (sa->code >= p->p_sysent->sv_size)
681 		/* nosys */
682 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
683 	else
684 		sa->callp = &p->p_sysent->sv_table[sa->code];
685 
686 	td->td_retval[0] = 0;
687 	td->td_retval[1] = frame->tf_rdx;
688 
689 	return (0);
690 }
691 
692 static void
693 linux32_set_syscall_retval(struct thread *td, int error)
694 {
695 	struct trapframe *frame = td->td_frame;
696 
697 	cpu_set_syscall_retval(td, error);
698 
699 	if (__predict_false(error != 0)) {
700 		if (error != ERESTART && error != EJUSTRETURN)
701 			frame->tf_rax = bsd_to_linux_errno(error);
702 	}
703 }
704 
705 static void
706 linux32_set_fork_retval(struct thread *td)
707 {
708 	struct trapframe *frame = td->td_frame;
709 
710 	frame->tf_rax = 0;
711 }
712 
713 /*
714  * Clear registers on exec
715  * XXX copied from ia32_signal.c.
716  */
717 static void
718 linux_exec_setregs(struct thread *td, struct image_params *imgp,
719     uintptr_t stack)
720 {
721 	struct trapframe *regs = td->td_frame;
722 	struct pcb *pcb = td->td_pcb;
723 	register_t saved_rflags;
724 
725 	regs = td->td_frame;
726 	pcb = td->td_pcb;
727 
728 	if (td->td_proc->p_md.md_ldt != NULL)
729 		user_ldt_free(td);
730 
731 	critical_enter();
732 	wrmsr(MSR_FSBASE, 0);
733 	wrmsr(MSR_KGSBASE, 0);	/* User value while we're in the kernel */
734 	pcb->pcb_fsbase = 0;
735 	pcb->pcb_gsbase = 0;
736 	critical_exit();
737 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
738 
739 	saved_rflags = regs->tf_rflags & PSL_T;
740 	bzero((char *)regs, sizeof(struct trapframe));
741 	regs->tf_rip = imgp->entry_addr;
742 	regs->tf_rsp = stack;
743 	regs->tf_rflags = PSL_USER | saved_rflags;
744 	regs->tf_gs = _ugssel;
745 	regs->tf_fs = _ufssel;
746 	regs->tf_es = _udatasel;
747 	regs->tf_ds = _udatasel;
748 	regs->tf_ss = _udatasel;
749 	regs->tf_flags = TF_HASSEGS;
750 	regs->tf_cs = _ucode32sel;
751 	regs->tf_rbx = (register_t)imgp->ps_strings;
752 
753 	x86_clear_dbregs(pcb);
754 
755 	fpstate_drop(td);
756 
757 	/* Do full restore on return so that we can change to a different %cs */
758 	set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
759 }
760 
761 /*
762  * XXX copied from ia32_sysvec.c.
763  */
764 static int
765 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
766 {
767 	int argc, envc, error;
768 	u_int32_t *vectp;
769 	char *stringp;
770 	uintptr_t destp, ustringp;
771 	struct linux32_ps_strings *arginfo;
772 	char canary[LINUX_AT_RANDOM_LEN];
773 	size_t execpath_len;
774 
775 	arginfo = (struct linux32_ps_strings *)PROC_PS_STRINGS(imgp->proc);
776 	destp = (uintptr_t)arginfo;
777 
778 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
779 		execpath_len = strlen(imgp->execpath) + 1;
780 		destp -= execpath_len;
781 		destp = rounddown2(destp, sizeof(uint32_t));
782 		imgp->execpathp = (void *)destp;
783 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
784 		if (error != 0)
785 			return (error);
786 	}
787 
788 	/* Prepare the canary for SSP. */
789 	arc4rand(canary, sizeof(canary), 0);
790 	destp -= roundup(sizeof(canary), sizeof(uint32_t));
791 	imgp->canary = (void *)destp;
792 	error = copyout(canary, imgp->canary, sizeof(canary));
793 	if (error != 0)
794 		return (error);
795 
796 	/* Allocate room for the argument and environment strings. */
797 	destp -= ARG_MAX - imgp->args->stringspace;
798 	destp = rounddown2(destp, sizeof(uint32_t));
799 	ustringp = destp;
800 
801 	if (imgp->auxargs) {
802 		/*
803 		 * Allocate room on the stack for the ELF auxargs
804 		 * array.  It has LINUX_AT_COUNT entries.
805 		 */
806 		destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
807 		destp = rounddown2(destp, sizeof(uint32_t));
808 	}
809 
810 	vectp = (uint32_t *)destp;
811 
812 	/*
813 	 * Allocate room for the argv[] and env vectors including the
814 	 * terminating NULL pointers.
815 	 */
816 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
817 
818 	/* vectp also becomes our initial stack base. */
819 	*stack_base = (uintptr_t)vectp;
820 
821 	stringp = imgp->args->begin_argv;
822 	argc = imgp->args->argc;
823 	envc = imgp->args->envc;
824 
825 	/* Copy out strings - arguments and environment. */
826 	error = copyout(stringp, (void *)ustringp,
827 	    ARG_MAX - imgp->args->stringspace);
828 	if (error != 0)
829 		return (error);
830 
831 	/* Fill in "ps_strings" struct for ps, w, etc. */
832 	if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 ||
833 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
834 		return (EFAULT);
835 
836 	/* Fill in argument portion of vector table. */
837 	for (; argc > 0; --argc) {
838 		if (suword32(vectp++, ustringp) != 0)
839 			return (EFAULT);
840 		while (*stringp++ != 0)
841 			ustringp++;
842 		ustringp++;
843 	}
844 
845 	/* A null vector table pointer separates the argp's from the envp's. */
846 	if (suword32(vectp++, 0) != 0)
847 		return (EFAULT);
848 
849 	if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 ||
850 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
851 		return (EFAULT);
852 
853 	/* Fill in environment portion of vector table. */
854 	for (; envc > 0; --envc) {
855 		if (suword32(vectp++, ustringp) != 0)
856 			return (EFAULT);
857 		while (*stringp++ != 0)
858 			ustringp++;
859 		ustringp++;
860 	}
861 
862 	/* The end of the vector table is a null pointer. */
863 	if (suword32(vectp, 0) != 0)
864 		return (EFAULT);
865 
866 	if (imgp->auxargs) {
867 		vectp++;
868 		error = imgp->sysent->sv_copyout_auxargs(imgp,
869 		    (uintptr_t)vectp);
870 		if (error != 0)
871 			return (error);
872 	}
873 
874 	return (0);
875 }
876 
877 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
878     "32-bit Linux emulation");
879 
880 static u_long	linux32_maxdsiz = LINUX32_MAXDSIZ;
881 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
882     &linux32_maxdsiz, 0, "");
883 static u_long	linux32_maxssiz = LINUX32_MAXSSIZ;
884 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
885     &linux32_maxssiz, 0, "");
886 static u_long	linux32_maxvmem = LINUX32_MAXVMEM;
887 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
888     &linux32_maxvmem, 0, "");
889 bool linux32_emulate_i386 = false;
890 SYSCTL_BOOL(_compat_linux32, OID_AUTO, emulate_i386, CTLFLAG_RWTUN,
891     &linux32_emulate_i386, 0, "Emulate the real i386");
892 
893 static void
894 linux32_fixlimit(struct rlimit *rl, int which)
895 {
896 
897 	switch (which) {
898 	case RLIMIT_DATA:
899 		if (linux32_maxdsiz != 0) {
900 			if (rl->rlim_cur > linux32_maxdsiz)
901 				rl->rlim_cur = linux32_maxdsiz;
902 			if (rl->rlim_max > linux32_maxdsiz)
903 				rl->rlim_max = linux32_maxdsiz;
904 		}
905 		break;
906 	case RLIMIT_STACK:
907 		if (linux32_maxssiz != 0) {
908 			if (rl->rlim_cur > linux32_maxssiz)
909 				rl->rlim_cur = linux32_maxssiz;
910 			if (rl->rlim_max > linux32_maxssiz)
911 				rl->rlim_max = linux32_maxssiz;
912 		}
913 		break;
914 	case RLIMIT_VMEM:
915 		if (linux32_maxvmem != 0) {
916 			if (rl->rlim_cur > linux32_maxvmem)
917 				rl->rlim_cur = linux32_maxvmem;
918 			if (rl->rlim_max > linux32_maxvmem)
919 				rl->rlim_max = linux32_maxvmem;
920 		}
921 		break;
922 	}
923 }
924 
925 struct sysentvec elf_linux_sysvec = {
926 	.sv_size	= LINUX32_SYS_MAXSYSCALL,
927 	.sv_table	= linux32_sysent,
928 	.sv_transtrap	= linux_translate_traps,
929 	.sv_fixup	= linux_fixup_elf,
930 	.sv_sendsig	= linux_sendsig,
931 	.sv_sigcode	= &_binary_linux32_vdso_so_o_start,
932 	.sv_szsigcode	= &linux_szsigcode,
933 	.sv_name	= "Linux ELF32",
934 	.sv_coredump	= elf32_coredump,
935 	.sv_elf_core_osabi = ELFOSABI_NONE,
936 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
937 	.sv_elf_core_prepare_notes = linux32_prepare_notes,
938 	.sv_imgact_try	= linux_exec_imgact_try,
939 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
940 	.sv_minuser	= VM_MIN_ADDRESS,
941 	.sv_maxuser	= LINUX32_MAXUSER,
942 	.sv_usrstack	= LINUX32_USRSTACK,
943 	.sv_psstrings	= LINUX32_PS_STRINGS,
944 	.sv_psstringssz	= sizeof(struct linux32_ps_strings),
945 	.sv_stackprot	= VM_PROT_ALL,
946 	.sv_copyout_auxargs = linux_copyout_auxargs,
947 	.sv_copyout_strings = linux_copyout_strings,
948 	.sv_setregs	= linux_exec_setregs,
949 	.sv_fixlimit	= linux32_fixlimit,
950 	.sv_maxssiz	= &linux32_maxssiz,
951 	.sv_flags	= SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP |
952 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
953 	.sv_set_syscall_retval = linux32_set_syscall_retval,
954 	.sv_fetch_syscall_args = linux32_fetch_syscall_args,
955 	.sv_syscallnames = NULL,
956 	.sv_shared_page_base = LINUX32_SHAREDPAGE,
957 	.sv_shared_page_len = PAGE_SIZE,
958 	.sv_schedtail	= linux_schedtail,
959 	.sv_thread_detach = linux_thread_detach,
960 	.sv_trap	= NULL,
961 	.sv_onexec	= linux_on_exec_vmspace,
962 	.sv_onexit	= linux_on_exit,
963 	.sv_ontdexit	= linux_thread_dtor,
964 	.sv_setid_allowed = &linux_setid_allowed_query,
965 	.sv_set_fork_retval = linux32_set_fork_retval,
966 };
967 
968 static int
969 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
970 {
971 	int error;
972 
973 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
974 	    LINUX32_VDSOPAGE_SIZE, imgp);
975 	if (error == 0)
976 		linux_on_exec(p, imgp);
977 	return (error);
978 }
979 
980 /*
981  * linux_vdso_install() and linux_exec_sysvec_init() must be called
982  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
983  */
984 static void
985 linux_exec_sysvec_init(void *param)
986 {
987 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
988 	struct sysentvec *sv;
989 	ptrdiff_t tkoff;
990 
991 	sv = param;
992 	/* Fill timekeep_base */
993 	exec_sysvec_init(sv);
994 
995 	tkoff = kern_timekeep_base - linux_vdso_base;
996 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
997 	*ktimekeep_base = sv->sv_timekeep_base;
998 
999 	tkoff = kern_tsc_selector - linux_vdso_base;
1000 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
1001 	*ktsc_selector = linux_vdso_tsc_selector_idx();
1002 	if (bootverbose)
1003 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
1004 
1005 	tkoff = kern_cpu_selector - linux_vdso_base;
1006 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
1007 	*ktsc_selector = linux_vdso_cpu_selector_idx();
1008 	if (bootverbose)
1009 		printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector);
1010 }
1011 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
1012     linux_exec_sysvec_init, &elf_linux_sysvec);
1013 
1014 static void
1015 linux_vdso_install(const void *param)
1016 {
1017 	char *vdso_start = &_binary_linux32_vdso_so_o_start;
1018 	char *vdso_end = &_binary_linux32_vdso_so_o_end;
1019 
1020 	linux_szsigcode = vdso_end - vdso_start;
1021 	MPASS(linux_szsigcode <= LINUX32_VDSOPAGE_SIZE);
1022 
1023 	linux_vdso_base = LINUX32_VDSOPAGE;
1024 
1025 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
1026 
1027 	linux_vdso_obj = __elfN(linux_shared_page_init)
1028 	    (&linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
1029 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
1030 
1031 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
1032 }
1033 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
1034     linux_vdso_install, NULL);
1035 
1036 static void
1037 linux_vdso_deinstall(const void *param)
1038 {
1039 
1040 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
1041 	    linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
1042 }
1043 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1044     linux_vdso_deinstall, NULL);
1045 
1046 static void
1047 linux_vdso_reloc(char *mapping, Elf_Addr offset)
1048 {
1049 	const Elf_Shdr *shdr;
1050 	const Elf_Rel *rel;
1051 	const Elf_Ehdr *ehdr;
1052 	Elf32_Addr *where;
1053 	Elf_Size rtype, symidx;
1054 	Elf32_Addr addr, addend;
1055 	int i, relcnt;
1056 
1057 	MPASS(offset != 0);
1058 
1059 	relcnt = 0;
1060 	ehdr = (const Elf_Ehdr *)mapping;
1061 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
1062 	for (i = 0; i < ehdr->e_shnum; i++)
1063 	{
1064 		switch (shdr[i].sh_type) {
1065 		case SHT_REL:
1066 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
1067 			relcnt = shdr[i].sh_size / sizeof(*rel);
1068 			break;
1069 		case SHT_RELA:
1070 			printf("Linux i386 vDSO: unexpected Rela section\n");
1071 			break;
1072 		}
1073 	}
1074 
1075 	for (i = 0; i < relcnt; i++, rel++) {
1076 		where = (Elf32_Addr *)(mapping + rel->r_offset);
1077 		addend = *where;
1078 		rtype = ELF_R_TYPE(rel->r_info);
1079 		symidx = ELF_R_SYM(rel->r_info);
1080 
1081 		switch (rtype) {
1082 		case R_386_NONE:	/* none */
1083 			break;
1084 
1085 		case R_386_RELATIVE:	/* B + A */
1086 			addr = (Elf32_Addr)PTROUT(offset + addend);
1087 			if (*where != addr)
1088 				*where = addr;
1089 			break;
1090 
1091 		case R_386_IRELATIVE:
1092 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
1093 			    "symbol index %ld\n", (intmax_t)symidx);
1094 			break;
1095 		default:
1096 			printf("Linux i386 vDSO: unexpected relocation type %ld, "
1097 			    "symbol index %ld\n", (intmax_t)rtype, (intmax_t)symidx);
1098 		}
1099 	}
1100 }
1101 
1102 static char GNU_ABI_VENDOR[] = "GNU";
1103 static int GNULINUX_ABI_DESC = 0;
1104 
1105 static bool
1106 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
1107 {
1108 	const Elf32_Word *desc;
1109 	uintptr_t p;
1110 
1111 	p = (uintptr_t)(note + 1);
1112 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1113 
1114 	desc = (const Elf32_Word *)p;
1115 	if (desc[0] != GNULINUX_ABI_DESC)
1116 		return (false);
1117 
1118 	/*
1119 	 * For Linux we encode osrel using the Linux convention of
1120 	 * 	(version << 16) | (major << 8) | (minor)
1121 	 * See macro in linux_mib.h
1122 	 */
1123 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
1124 
1125 	return (true);
1126 }
1127 
1128 static Elf_Brandnote linux32_brandnote = {
1129 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
1130 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
1131 	.hdr.n_type	= 1,
1132 	.vendor		= GNU_ABI_VENDOR,
1133 	.flags		= BN_TRANSLATE_OSREL,
1134 	.trans_osrel	= linux32_trans_osrel
1135 };
1136 
1137 static Elf32_Brandinfo linux_brand = {
1138 	.brand		= ELFOSABI_LINUX,
1139 	.machine	= EM_386,
1140 	.compat_3_brand	= "Linux",
1141 	.emul_path	= linux_emul_path,
1142 	.interp_path	= "/lib/ld-linux.so.1",
1143 	.sysvec		= &elf_linux_sysvec,
1144 	.interp_newpath	= NULL,
1145 	.brand_note	= &linux32_brandnote,
1146 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1147 };
1148 
1149 static Elf32_Brandinfo linux_glibc2brand = {
1150 	.brand		= ELFOSABI_LINUX,
1151 	.machine	= EM_386,
1152 	.compat_3_brand	= "Linux",
1153 	.emul_path	= linux_emul_path,
1154 	.interp_path	= "/lib/ld-linux.so.2",
1155 	.sysvec		= &elf_linux_sysvec,
1156 	.interp_newpath	= NULL,
1157 	.brand_note	= &linux32_brandnote,
1158 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1159 };
1160 
1161 static Elf32_Brandinfo linux_muslbrand = {
1162 	.brand		= ELFOSABI_LINUX,
1163 	.machine	= EM_386,
1164 	.compat_3_brand	= "Linux",
1165 	.emul_path	= linux_emul_path,
1166 	.interp_path	= "/lib/ld-musl-i386.so.1",
1167 	.sysvec		= &elf_linux_sysvec,
1168 	.interp_newpath	= NULL,
1169 	.brand_note	= &linux32_brandnote,
1170 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
1171 			    LINUX_BI_FUTEX_REQUEUE
1172 };
1173 
1174 Elf32_Brandinfo *linux_brandlist[] = {
1175 	&linux_brand,
1176 	&linux_glibc2brand,
1177 	&linux_muslbrand,
1178 	NULL
1179 };
1180 
1181 static int
1182 linux_elf_modevent(module_t mod, int type, void *data)
1183 {
1184 	Elf32_Brandinfo **brandinfo;
1185 	int error;
1186 	struct linux_ioctl_handler **lihp;
1187 
1188 	error = 0;
1189 
1190 	switch(type) {
1191 	case MOD_LOAD:
1192 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1193 		     ++brandinfo)
1194 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1195 				error = EINVAL;
1196 		if (error == 0) {
1197 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1198 				linux32_ioctl_register_handler(*lihp);
1199 			stclohz = (stathz ? stathz : hz);
1200 			if (bootverbose)
1201 				printf("Linux i386 ELF exec handler installed\n");
1202 		} else
1203 			printf("cannot insert Linux i386 ELF brand handler\n");
1204 		break;
1205 	case MOD_UNLOAD:
1206 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1207 		     ++brandinfo)
1208 			if (elf32_brand_inuse(*brandinfo))
1209 				error = EBUSY;
1210 		if (error == 0) {
1211 			for (brandinfo = &linux_brandlist[0];
1212 			     *brandinfo != NULL; ++brandinfo)
1213 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1214 					error = EINVAL;
1215 		}
1216 		if (error == 0) {
1217 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1218 				linux32_ioctl_unregister_handler(*lihp);
1219 			if (bootverbose)
1220 				printf("Linux i386 ELF exec handler removed\n");
1221 		} else
1222 			printf("Could not deinstall Linux i386 ELF interpreter entry\n");
1223 		break;
1224 	default:
1225 		return (EOPNOTSUPP);
1226 	}
1227 	return (error);
1228 }
1229 
1230 static moduledata_t linux_elf_mod = {
1231 	"linuxelf",
1232 	linux_elf_modevent,
1233 	0
1234 };
1235 
1236 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1237 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1238 FEATURE(linux, "Linux 32bit support");
1239