xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision e8d8bef961a50d4dc22501cde4fb9fb0be1b2532)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/imgact_aout.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/signalvar.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysctl.h>
48 #include <sys/sysent.h>
49 #include <sys/sysproto.h>
50 #include <sys/vnode.h>
51 
52 #include <vm/vm.h>
53 #include <vm/pmap.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_param.h>
59 
60 #include <machine/cpu.h>
61 #include <machine/cputypes.h>
62 #include <machine/md_var.h>
63 #include <machine/pcb.h>
64 #include <machine/trap.h>
65 
66 #include <i386/linux/linux.h>
67 #include <i386/linux/linux_proto.h>
68 #include <compat/linux/linux_emul.h>
69 #include <compat/linux/linux_ioctl.h>
70 #include <compat/linux/linux_mib.h>
71 #include <compat/linux/linux_misc.h>
72 #include <compat/linux/linux_signal.h>
73 #include <compat/linux/linux_util.h>
74 #include <compat/linux/linux_vdso.h>
75 
76 MODULE_VERSION(linux, 1);
77 
78 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
79 
80 static int linux_szsigcode;
81 static vm_object_t linux_shared_page_obj;
82 static char *linux_shared_page_mapping;
83 extern char _binary_linux_locore_o_start;
84 extern char _binary_linux_locore_o_end;
85 
86 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
87 
88 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
89 
90 static int	linux_fixup(uintptr_t *stack_base,
91 		    struct image_params *iparams);
92 static int	linux_fixup_elf(uintptr_t *stack_base,
93 		    struct image_params *iparams);
94 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
95 static void	linux_exec_setregs(struct thread *td,
96 		    struct image_params *imgp, uintptr_t stack);
97 static int	linux_copyout_strings(struct image_params *imgp,
98 		    uintptr_t *stack_base);
99 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static void	linux_vdso_install(void *param);
101 static void	linux_vdso_deinstall(void *param);
102 
103 static int linux_szplatform;
104 const char *linux_kplatform;
105 
106 #define LINUX_T_UNKNOWN  255
107 static int _bsd_to_linux_trapcode[] = {
108 	LINUX_T_UNKNOWN,	/* 0 */
109 	6,			/* 1  T_PRIVINFLT */
110 	LINUX_T_UNKNOWN,	/* 2 */
111 	3,			/* 3  T_BPTFLT */
112 	LINUX_T_UNKNOWN,	/* 4 */
113 	LINUX_T_UNKNOWN,	/* 5 */
114 	16,			/* 6  T_ARITHTRAP */
115 	254,			/* 7  T_ASTFLT */
116 	LINUX_T_UNKNOWN,	/* 8 */
117 	13,			/* 9  T_PROTFLT */
118 	1,			/* 10 T_TRCTRAP */
119 	LINUX_T_UNKNOWN,	/* 11 */
120 	14,			/* 12 T_PAGEFLT */
121 	LINUX_T_UNKNOWN,	/* 13 */
122 	17,			/* 14 T_ALIGNFLT */
123 	LINUX_T_UNKNOWN,	/* 15 */
124 	LINUX_T_UNKNOWN,	/* 16 */
125 	LINUX_T_UNKNOWN,	/* 17 */
126 	0,			/* 18 T_DIVIDE */
127 	2,			/* 19 T_NMI */
128 	4,			/* 20 T_OFLOW */
129 	5,			/* 21 T_BOUND */
130 	7,			/* 22 T_DNA */
131 	8,			/* 23 T_DOUBLEFLT */
132 	9,			/* 24 T_FPOPFLT */
133 	10,			/* 25 T_TSSFLT */
134 	11,			/* 26 T_SEGNPFLT */
135 	12,			/* 27 T_STKFLT */
136 	18,			/* 28 T_MCHK */
137 	19,			/* 29 T_XMMFLT */
138 	15			/* 30 T_RESERVED */
139 };
140 #define bsd_to_linux_trapcode(code) \
141     ((code)<nitems(_bsd_to_linux_trapcode)? \
142      _bsd_to_linux_trapcode[(code)]: \
143      LINUX_T_UNKNOWN)
144 
145 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
146 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
147 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
148 
149 /*
150  * If FreeBSD & Linux have a difference of opinion about what a trap
151  * means, deal with it here.
152  *
153  * MPSAFE
154  */
155 static int
156 linux_translate_traps(int signal, int trap_code)
157 {
158 	if (signal != SIGBUS)
159 		return (signal);
160 	switch (trap_code) {
161 	case T_PROTFLT:
162 	case T_TSSFLT:
163 	case T_DOUBLEFLT:
164 	case T_PAGEFLT:
165 		return (SIGSEGV);
166 	default:
167 		return (signal);
168 	}
169 }
170 
171 static int
172 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
173 {
174 	register_t *base, *argv, *envp;
175 
176 	base = (register_t *)*stack_base;
177 	argv = base;
178 	envp = base + (imgp->args->argc + 1);
179 	base--;
180 	suword(base, (intptr_t)envp);
181 	base--;
182 	suword(base, (intptr_t)argv);
183 	base--;
184 	suword(base, imgp->args->argc);
185 	*stack_base = (uintptr_t)base;
186 	return (0);
187 }
188 
189 static int
190 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
191 {
192 	struct proc *p;
193 	Elf32_Auxargs *args;
194 	Elf32_Auxinfo *argarray, *pos;
195 	Elf32_Addr *uplatform;
196 	struct ps_strings *arginfo;
197 	int error, issetugid;
198 
199 	p = imgp->proc;
200 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
201 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
202 	uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
203 	args = (Elf32_Auxargs *)imgp->auxargs;
204 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
205 	    M_WAITOK | M_ZERO);
206 
207 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
208 	    imgp->proc->p_sysent->sv_shared_page_base);
209 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
210 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
211 
212 	/*
213 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
214 	 * as it has appeared in the 2.4.0-rc7 first time.
215 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
216 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
217 	 * is not present.
218 	 * Also see linux_times() implementation.
219 	 */
220 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
221 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
222 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
223 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
224 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
225 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
226 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
227 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
228 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
229 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
230 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
231 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
232 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
233 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
234 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
235 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
236 	if (imgp->execpathp != 0)
237 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
238 	if (args->execfd != -1)
239 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
240 	AUXARGS_ENTRY(pos, AT_NULL, 0);
241 
242 	free(imgp->auxargs, M_TEMP);
243 	imgp->auxargs = NULL;
244 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
245 
246 	error = copyout(argarray, (void *)base,
247 	    sizeof(*argarray) * LINUX_AT_COUNT);
248 	free(argarray, M_TEMP);
249 	return (error);
250 }
251 
252 static int
253 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
254 {
255 	register_t *base;
256 
257 	base = (register_t *)*stack_base;
258 	base--;
259 	if (suword(base, (register_t)imgp->args->argc) == -1)
260 		return (EFAULT);
261 	*stack_base = (uintptr_t)base;
262 	return (0);
263 }
264 
265 /*
266  * Copied from kern/kern_exec.c
267  */
268 static int
269 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
270 {
271 	int argc, envc, error;
272 	char **vectp;
273 	char *stringp;
274 	uintptr_t destp, ustringp;
275 	struct ps_strings *arginfo;
276 	char canary[LINUX_AT_RANDOM_LEN];
277 	size_t execpath_len;
278 	struct proc *p;
279 
280 	/* Calculate string base and vector table pointers. */
281 	p = imgp->proc;
282 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
283 		execpath_len = strlen(imgp->execpath) + 1;
284 	else
285 		execpath_len = 0;
286 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
287 	destp = (uintptr_t)arginfo;
288 
289 	/* Install LINUX_PLATFORM. */
290 	destp -= linux_szplatform;
291 	destp = rounddown2(destp, sizeof(void *));
292 	error = copyout(linux_kplatform, (void *)destp, linux_szplatform);
293 	if (error != 0)
294 		return (error);
295 
296 	if (execpath_len != 0) {
297 		destp -= execpath_len;
298 		destp = rounddown2(destp, sizeof(void *));
299 		imgp->execpathp = (void *)destp;
300 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
301 		if (error != 0)
302 			return (error);
303 	}
304 
305 	/* Prepare the canary for SSP. */
306 	arc4rand(canary, sizeof(canary), 0);
307 	destp -= roundup(sizeof(canary), sizeof(void *));
308 	imgp->canary = (void *)destp;
309 	error = copyout(canary, imgp->canary, sizeof(canary));
310 	if (error != 0)
311 		return (error);
312 
313 	/* Allocate room for the argument and environment strings. */
314 	destp -= ARG_MAX - imgp->args->stringspace;
315 	destp = rounddown2(destp, sizeof(void *));
316 	ustringp = destp;
317 
318 	if (imgp->auxargs) {
319 		/*
320 		 * Allocate room on the stack for the ELF auxargs
321 		 * array.  It has LINUX_AT_COUNT entries.
322 		 */
323 		destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
324 		destp = rounddown2(destp, sizeof(void *));
325 	}
326 
327 	vectp = (char **)destp;
328 
329 	/*
330 	 * Allocate room for the argv[] and env vectors including the
331 	 * terminating NULL pointers.
332 	 */
333 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
334 
335 	/* vectp also becomes our initial stack base. */
336 	*stack_base = (uintptr_t)vectp;
337 
338 	stringp = imgp->args->begin_argv;
339 	argc = imgp->args->argc;
340 	envc = imgp->args->envc;
341 
342 	/* Copy out strings - arguments and environment. */
343 	error = copyout(stringp, (void *)ustringp,
344 	    ARG_MAX - imgp->args->stringspace);
345 	if (error != 0)
346 		return (error);
347 
348 	/* Fill in "ps_strings" struct for ps, w, etc. */
349 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
350 	    suword(&arginfo->ps_nargvstr, argc) != 0)
351 		return (EFAULT);
352 
353 	/* Fill in argument portion of vector table. */
354 	for (; argc > 0; --argc) {
355 		if (suword(vectp++, ustringp) != 0)
356 			return (EFAULT);
357 		while (*stringp++ != 0)
358 			ustringp++;
359 		ustringp++;
360 	}
361 
362 	/* A null vector table pointer separates the argp's from the envp's. */
363 	if (suword(vectp++, 0) != 0)
364 		return (EFAULT);
365 
366 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
367 	    suword(&arginfo->ps_nenvstr, envc) != 0)
368 		return (EFAULT);
369 
370 	/* Fill in environment portion of vector table. */
371 	for (; envc > 0; --envc) {
372 		if (suword(vectp++, ustringp) != 0)
373 			return (EFAULT);
374 		while (*stringp++ != 0)
375 			ustringp++;
376 		ustringp++;
377 	}
378 
379 	/* The end of the vector table is a null pointer. */
380 	if (suword(vectp, 0) != 0)
381 		return (EFAULT);
382 
383 	if (imgp->auxargs) {
384 		vectp++;
385 		error = imgp->sysent->sv_copyout_auxargs(imgp,
386 		    (uintptr_t)vectp);
387 		if (error != 0)
388 			return (error);
389 	}
390 
391 	return (0);
392 }
393 
394 static void
395 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
396 {
397 	struct thread *td = curthread;
398 	struct proc *p = td->td_proc;
399 	struct sigacts *psp;
400 	struct trapframe *regs;
401 	struct l_rt_sigframe *fp, frame;
402 	int sig, code;
403 	int oonstack;
404 
405 	sig = ksi->ksi_signo;
406 	code = ksi->ksi_code;
407 	PROC_LOCK_ASSERT(p, MA_OWNED);
408 	psp = p->p_sigacts;
409 	mtx_assert(&psp->ps_mtx, MA_OWNED);
410 	regs = td->td_frame;
411 	oonstack = sigonstack(regs->tf_esp);
412 
413 	/* Allocate space for the signal handler context. */
414 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
415 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
416 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
417 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
418 	} else
419 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
420 	mtx_unlock(&psp->ps_mtx);
421 
422 	/* Build the argument list for the signal handler. */
423 	sig = bsd_to_linux_signal(sig);
424 
425 	bzero(&frame, sizeof(frame));
426 
427 	frame.sf_handler = catcher;
428 	frame.sf_sig = sig;
429 	frame.sf_siginfo = &fp->sf_si;
430 	frame.sf_ucontext = &fp->sf_sc;
431 
432 	/* Fill in POSIX parts. */
433 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
434 
435 	/* Build the signal context to be used by sigreturn. */
436 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
437 	frame.sf_sc.uc_link = NULL;		/* XXX ??? */
438 
439 	frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
440 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
441 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
442 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
443 	PROC_UNLOCK(p);
444 
445 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
446 
447 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
448 	frame.sf_sc.uc_mcontext.sc_gs     = rgs();
449 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
450 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
451 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
452 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_edi;
453 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_esi;
454 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_ebp;
455 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_ebx;
456 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_esp;
457 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_edx;
458 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_ecx;
459 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_eax;
460 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_eip;
461 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
462 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
463 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
464 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
465 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
466 	frame.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
467 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
468 
469 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
470 		/*
471 		 * Process has trashed its stack; give it an illegal
472 		 * instruction to halt it in its tracks.
473 		 */
474 		PROC_LOCK(p);
475 		sigexit(td, SIGILL);
476 	}
477 
478 	/* Build context to run handler in. */
479 	regs->tf_esp = (int)fp;
480 	regs->tf_eip = linux_rt_sigcode;
481 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
482 	regs->tf_cs = _ucodesel;
483 	regs->tf_ds = _udatasel;
484 	regs->tf_es = _udatasel;
485 	regs->tf_fs = _udatasel;
486 	regs->tf_ss = _udatasel;
487 	PROC_LOCK(p);
488 	mtx_lock(&psp->ps_mtx);
489 }
490 
491 /*
492  * Send an interrupt to process.
493  *
494  * Stack is set up to allow sigcode stored
495  * in u. to call routine, followed by kcall
496  * to sigreturn routine below.  After sigreturn
497  * resets the signal mask, the stack, and the
498  * frame pointer, it returns to the user
499  * specified pc, psl.
500  */
501 static void
502 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
503 {
504 	struct thread *td = curthread;
505 	struct proc *p = td->td_proc;
506 	struct sigacts *psp;
507 	struct trapframe *regs;
508 	struct l_sigframe *fp, frame;
509 	l_sigset_t lmask;
510 	int sig, code;
511 	int oonstack;
512 
513 	PROC_LOCK_ASSERT(p, MA_OWNED);
514 	psp = p->p_sigacts;
515 	sig = ksi->ksi_signo;
516 	code = ksi->ksi_code;
517 	mtx_assert(&psp->ps_mtx, MA_OWNED);
518 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
519 		/* Signal handler installed with SA_SIGINFO. */
520 		linux_rt_sendsig(catcher, ksi, mask);
521 		return;
522 	}
523 	regs = td->td_frame;
524 	oonstack = sigonstack(regs->tf_esp);
525 
526 	/* Allocate space for the signal handler context. */
527 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
528 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
529 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
530 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
531 	} else
532 		fp = (struct l_sigframe *)regs->tf_esp - 1;
533 	mtx_unlock(&psp->ps_mtx);
534 	PROC_UNLOCK(p);
535 
536 	/* Build the argument list for the signal handler. */
537 	sig = bsd_to_linux_signal(sig);
538 
539 	bzero(&frame, sizeof(frame));
540 
541 	frame.sf_handler = catcher;
542 	frame.sf_sig = sig;
543 
544 	bsd_to_linux_sigset(mask, &lmask);
545 
546 	/* Build the signal context to be used by sigreturn. */
547 	frame.sf_sc.sc_mask   = lmask.__mask;
548 	frame.sf_sc.sc_gs     = rgs();
549 	frame.sf_sc.sc_fs     = regs->tf_fs;
550 	frame.sf_sc.sc_es     = regs->tf_es;
551 	frame.sf_sc.sc_ds     = regs->tf_ds;
552 	frame.sf_sc.sc_edi    = regs->tf_edi;
553 	frame.sf_sc.sc_esi    = regs->tf_esi;
554 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
555 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
556 	frame.sf_sc.sc_esp    = regs->tf_esp;
557 	frame.sf_sc.sc_edx    = regs->tf_edx;
558 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
559 	frame.sf_sc.sc_eax    = regs->tf_eax;
560 	frame.sf_sc.sc_eip    = regs->tf_eip;
561 	frame.sf_sc.sc_cs     = regs->tf_cs;
562 	frame.sf_sc.sc_eflags = regs->tf_eflags;
563 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
564 	frame.sf_sc.sc_ss     = regs->tf_ss;
565 	frame.sf_sc.sc_err    = regs->tf_err;
566 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
567 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
568 
569 	frame.sf_extramask[0] = lmask.__mask;
570 
571 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
572 		/*
573 		 * Process has trashed its stack; give it an illegal
574 		 * instruction to halt it in its tracks.
575 		 */
576 		PROC_LOCK(p);
577 		sigexit(td, SIGILL);
578 	}
579 
580 	/* Build context to run handler in. */
581 	regs->tf_esp = (int)fp;
582 	regs->tf_eip = linux_sigcode;
583 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
584 	regs->tf_cs = _ucodesel;
585 	regs->tf_ds = _udatasel;
586 	regs->tf_es = _udatasel;
587 	regs->tf_fs = _udatasel;
588 	regs->tf_ss = _udatasel;
589 	PROC_LOCK(p);
590 	mtx_lock(&psp->ps_mtx);
591 }
592 
593 /*
594  * System call to cleanup state after a signal
595  * has been taken.  Reset signal mask and
596  * stack state from context left by sendsig (above).
597  * Return to previous pc and psl as specified by
598  * context left by sendsig. Check carefully to
599  * make sure that the user has not modified the
600  * psl to gain improper privileges or to cause
601  * a machine fault.
602  */
603 int
604 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
605 {
606 	struct l_sigframe frame;
607 	struct trapframe *regs;
608 	l_sigset_t lmask;
609 	sigset_t bmask;
610 	int eflags;
611 	ksiginfo_t ksi;
612 
613 	regs = td->td_frame;
614 
615 	/*
616 	 * The trampoline code hands us the sigframe.
617 	 * It is unsafe to keep track of it ourselves, in the event that a
618 	 * program jumps out of a signal handler.
619 	 */
620 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
621 		return (EFAULT);
622 
623 	/* Check for security violations. */
624 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
625 	eflags = frame.sf_sc.sc_eflags;
626 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
627 		return (EINVAL);
628 
629 	/*
630 	 * Don't allow users to load a valid privileged %cs.  Let the
631 	 * hardware check for invalid selectors, excess privilege in
632 	 * other selectors, invalid %eip's and invalid %esp's.
633 	 */
634 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
635 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
636 		ksiginfo_init_trap(&ksi);
637 		ksi.ksi_signo = SIGBUS;
638 		ksi.ksi_code = BUS_OBJERR;
639 		ksi.ksi_trapno = T_PROTFLT;
640 		ksi.ksi_addr = (void *)regs->tf_eip;
641 		trapsignal(td, &ksi);
642 		return (EINVAL);
643 	}
644 
645 	lmask.__mask = frame.sf_sc.sc_mask;
646 	linux_to_bsd_sigset(&lmask, &bmask);
647 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
648 
649 	/* Restore signal context. */
650 	/* %gs was restored by the trampoline. */
651 	regs->tf_fs     = frame.sf_sc.sc_fs;
652 	regs->tf_es     = frame.sf_sc.sc_es;
653 	regs->tf_ds     = frame.sf_sc.sc_ds;
654 	regs->tf_edi    = frame.sf_sc.sc_edi;
655 	regs->tf_esi    = frame.sf_sc.sc_esi;
656 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
657 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
658 	regs->tf_edx    = frame.sf_sc.sc_edx;
659 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
660 	regs->tf_eax    = frame.sf_sc.sc_eax;
661 	regs->tf_eip    = frame.sf_sc.sc_eip;
662 	regs->tf_cs     = frame.sf_sc.sc_cs;
663 	regs->tf_eflags = eflags;
664 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
665 	regs->tf_ss     = frame.sf_sc.sc_ss;
666 
667 	return (EJUSTRETURN);
668 }
669 
670 /*
671  * System call to cleanup state after a signal
672  * has been taken.  Reset signal mask and
673  * stack state from context left by rt_sendsig (above).
674  * Return to previous pc and psl as specified by
675  * context left by sendsig. Check carefully to
676  * make sure that the user has not modified the
677  * psl to gain improper privileges or to cause
678  * a machine fault.
679  */
680 int
681 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
682 {
683 	struct l_ucontext uc;
684 	struct l_sigcontext *context;
685 	sigset_t bmask;
686 	l_stack_t *lss;
687 	stack_t ss;
688 	struct trapframe *regs;
689 	int eflags;
690 	ksiginfo_t ksi;
691 
692 	regs = td->td_frame;
693 
694 	/*
695 	 * The trampoline code hands us the ucontext.
696 	 * It is unsafe to keep track of it ourselves, in the event that a
697 	 * program jumps out of a signal handler.
698 	 */
699 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
700 		return (EFAULT);
701 
702 	context = &uc.uc_mcontext;
703 
704 	/* Check for security violations. */
705 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
706 	eflags = context->sc_eflags;
707 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
708 		return (EINVAL);
709 
710 	/*
711 	 * Don't allow users to load a valid privileged %cs.  Let the
712 	 * hardware check for invalid selectors, excess privilege in
713 	 * other selectors, invalid %eip's and invalid %esp's.
714 	 */
715 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
716 	if (!CS_SECURE(context->sc_cs)) {
717 		ksiginfo_init_trap(&ksi);
718 		ksi.ksi_signo = SIGBUS;
719 		ksi.ksi_code = BUS_OBJERR;
720 		ksi.ksi_trapno = T_PROTFLT;
721 		ksi.ksi_addr = (void *)regs->tf_eip;
722 		trapsignal(td, &ksi);
723 		return (EINVAL);
724 	}
725 
726 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
727 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
728 
729 	/* Restore signal context. */
730 	/* %gs was restored by the trampoline. */
731 	regs->tf_fs     = context->sc_fs;
732 	regs->tf_es     = context->sc_es;
733 	regs->tf_ds     = context->sc_ds;
734 	regs->tf_edi    = context->sc_edi;
735 	regs->tf_esi    = context->sc_esi;
736 	regs->tf_ebp    = context->sc_ebp;
737 	regs->tf_ebx    = context->sc_ebx;
738 	regs->tf_edx    = context->sc_edx;
739 	regs->tf_ecx    = context->sc_ecx;
740 	regs->tf_eax    = context->sc_eax;
741 	regs->tf_eip    = context->sc_eip;
742 	regs->tf_cs     = context->sc_cs;
743 	regs->tf_eflags = eflags;
744 	regs->tf_esp    = context->sc_esp_at_signal;
745 	regs->tf_ss     = context->sc_ss;
746 
747 	/* Call sigaltstack & ignore results. */
748 	lss = &uc.uc_stack;
749 	ss.ss_sp = lss->ss_sp;
750 	ss.ss_size = lss->ss_size;
751 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
752 
753 	(void)kern_sigaltstack(td, &ss, NULL);
754 
755 	return (EJUSTRETURN);
756 }
757 
758 static int
759 linux_fetch_syscall_args(struct thread *td)
760 {
761 	struct proc *p;
762 	struct trapframe *frame;
763 	struct syscall_args *sa;
764 
765 	p = td->td_proc;
766 	frame = td->td_frame;
767 	sa = &td->td_sa;
768 
769 	sa->code = frame->tf_eax;
770 	sa->args[0] = frame->tf_ebx;
771 	sa->args[1] = frame->tf_ecx;
772 	sa->args[2] = frame->tf_edx;
773 	sa->args[3] = frame->tf_esi;
774 	sa->args[4] = frame->tf_edi;
775 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
776 
777 	if (sa->code >= p->p_sysent->sv_size)
778 		/* nosys */
779 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
780 	else
781 		sa->callp = &p->p_sysent->sv_table[sa->code];
782 
783 	td->td_retval[0] = 0;
784 	td->td_retval[1] = frame->tf_edx;
785 
786 	return (0);
787 }
788 
789 static void
790 linux_set_syscall_retval(struct thread *td, int error)
791 {
792 	struct trapframe *frame = td->td_frame;
793 
794 	cpu_set_syscall_retval(td, error);
795 
796 	if (__predict_false(error != 0)) {
797 		if (error != ERESTART && error != EJUSTRETURN)
798 			frame->tf_eax = bsd_to_linux_errno(error);
799 	}
800 }
801 
802 /*
803  * exec_setregs may initialize some registers differently than Linux
804  * does, thus potentially confusing Linux binaries. If necessary, we
805  * override the exec_setregs default(s) here.
806  */
807 static void
808 linux_exec_setregs(struct thread *td, struct image_params *imgp,
809     uintptr_t stack)
810 {
811 	struct pcb *pcb = td->td_pcb;
812 
813 	exec_setregs(td, imgp, stack);
814 
815 	/* Linux sets %gs to 0, we default to _udatasel. */
816 	pcb->pcb_gs = 0;
817 	load_gs(0);
818 
819 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
820 }
821 
822 static void
823 linux_get_machine(const char **dst)
824 {
825 
826 	switch (cpu_class) {
827 	case CPUCLASS_686:
828 		*dst = "i686";
829 		break;
830 	case CPUCLASS_586:
831 		*dst = "i586";
832 		break;
833 	case CPUCLASS_486:
834 		*dst = "i486";
835 		break;
836 	default:
837 		*dst = "i386";
838 	}
839 }
840 
841 struct sysentvec linux_sysvec = {
842 	.sv_size	= LINUX_SYS_MAXSYSCALL,
843 	.sv_table	= linux_sysent,
844 	.sv_transtrap	= linux_translate_traps,
845 	.sv_fixup	= linux_fixup,
846 	.sv_sendsig	= linux_sendsig,
847 	.sv_sigcode	= &_binary_linux_locore_o_start,
848 	.sv_szsigcode	= &linux_szsigcode,
849 	.sv_name	= "Linux a.out",
850 	.sv_coredump	= NULL,
851 	.sv_imgact_try	= linux_exec_imgact_try,
852 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
853 	.sv_minuser	= VM_MIN_ADDRESS,
854 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
855 	.sv_usrstack	= LINUX_USRSTACK,
856 	.sv_psstrings	= PS_STRINGS,
857 	.sv_stackprot	= VM_PROT_ALL,
858 	.sv_copyout_strings = exec_copyout_strings,
859 	.sv_setregs	= linux_exec_setregs,
860 	.sv_fixlimit	= NULL,
861 	.sv_maxssiz	= NULL,
862 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32,
863 	.sv_set_syscall_retval = linux_set_syscall_retval,
864 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
865 	.sv_syscallnames = NULL,
866 	.sv_shared_page_base = LINUX_SHAREDPAGE,
867 	.sv_shared_page_len = PAGE_SIZE,
868 	.sv_schedtail	= linux_schedtail,
869 	.sv_thread_detach = linux_thread_detach,
870 	.sv_trap	= NULL,
871 	.sv_onexec	= linux_on_exec,
872 	.sv_onexit	= linux_on_exit,
873 	.sv_ontdexit	= linux_thread_dtor,
874 	.sv_setid_allowed = &linux_setid_allowed_query,
875 };
876 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
877 
878 struct sysentvec elf_linux_sysvec = {
879 	.sv_size	= LINUX_SYS_MAXSYSCALL,
880 	.sv_table	= linux_sysent,
881 	.sv_transtrap	= linux_translate_traps,
882 	.sv_fixup	= linux_fixup_elf,
883 	.sv_sendsig	= linux_sendsig,
884 	.sv_sigcode	= &_binary_linux_locore_o_start,
885 	.sv_szsigcode	= &linux_szsigcode,
886 	.sv_name	= "Linux ELF32",
887 	.sv_coredump	= elf32_coredump,
888 	.sv_imgact_try	= linux_exec_imgact_try,
889 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
890 	.sv_minuser	= VM_MIN_ADDRESS,
891 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
892 	.sv_usrstack	= LINUX_USRSTACK,
893 	.sv_psstrings	= LINUX_PS_STRINGS,
894 	.sv_stackprot	= VM_PROT_ALL,
895 	.sv_copyout_auxargs = linux_copyout_auxargs,
896 	.sv_copyout_strings = linux_copyout_strings,
897 	.sv_setregs	= linux_exec_setregs,
898 	.sv_fixlimit	= NULL,
899 	.sv_maxssiz	= NULL,
900 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP,
901 	.sv_set_syscall_retval = linux_set_syscall_retval,
902 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
903 	.sv_syscallnames = NULL,
904 	.sv_shared_page_base = LINUX_SHAREDPAGE,
905 	.sv_shared_page_len = PAGE_SIZE,
906 	.sv_schedtail	= linux_schedtail,
907 	.sv_thread_detach = linux_thread_detach,
908 	.sv_trap	= NULL,
909 	.sv_onexec	= linux_on_exec,
910 	.sv_onexit	= linux_on_exit,
911 	.sv_ontdexit	= linux_thread_dtor,
912 	.sv_setid_allowed = &linux_setid_allowed_query,
913 };
914 
915 static void
916 linux_vdso_install(void *param)
917 {
918 
919 	linux_szsigcode = (&_binary_linux_locore_o_end -
920 	    &_binary_linux_locore_o_start);
921 
922 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
923 		panic("Linux invalid vdso size\n");
924 
925 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
926 
927 	linux_shared_page_obj = __elfN(linux_shared_page_init)
928 	    (&linux_shared_page_mapping);
929 
930 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
931 
932 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
933 	    linux_szsigcode);
934 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
935 }
936 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
937     linux_vdso_install, NULL);
938 
939 static void
940 linux_vdso_deinstall(void *param)
941 {
942 
943 	__elfN(linux_shared_page_fini)(linux_shared_page_obj,
944 	    linux_shared_page_mapping);
945 }
946 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
947     linux_vdso_deinstall, NULL);
948 
949 static char GNU_ABI_VENDOR[] = "GNU";
950 static int GNULINUX_ABI_DESC = 0;
951 
952 static bool
953 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
954 {
955 	const Elf32_Word *desc;
956 	uintptr_t p;
957 
958 	p = (uintptr_t)(note + 1);
959 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
960 
961 	desc = (const Elf32_Word *)p;
962 	if (desc[0] != GNULINUX_ABI_DESC)
963 		return (false);
964 
965 	/*
966 	 * For Linux we encode osrel using the Linux convention of
967 	 * 	(version << 16) | (major << 8) | (minor)
968 	 * See macro in linux_mib.h
969 	 */
970 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
971 
972 	return (true);
973 }
974 
975 static Elf_Brandnote linux_brandnote = {
976 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
977 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
978 	.hdr.n_type	= 1,
979 	.vendor		= GNU_ABI_VENDOR,
980 	.flags		= BN_TRANSLATE_OSREL,
981 	.trans_osrel	= linux_trans_osrel
982 };
983 
984 static Elf32_Brandinfo linux_brand = {
985 	.brand		= ELFOSABI_LINUX,
986 	.machine	= EM_386,
987 	.compat_3_brand	= "Linux",
988 	.emul_path	= linux_emul_path,
989 	.interp_path	= "/lib/ld-linux.so.1",
990 	.sysvec		= &elf_linux_sysvec,
991 	.interp_newpath	= NULL,
992 	.brand_note	= &linux_brandnote,
993 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
994 };
995 
996 static Elf32_Brandinfo linux_glibc2brand = {
997 	.brand		= ELFOSABI_LINUX,
998 	.machine	= EM_386,
999 	.compat_3_brand	= "Linux",
1000 	.emul_path	= linux_emul_path,
1001 	.interp_path	= "/lib/ld-linux.so.2",
1002 	.sysvec		= &elf_linux_sysvec,
1003 	.interp_newpath	= NULL,
1004 	.brand_note	= &linux_brandnote,
1005 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1006 };
1007 
1008 static Elf32_Brandinfo linux_muslbrand = {
1009 	.brand		= ELFOSABI_LINUX,
1010 	.machine	= EM_386,
1011 	.compat_3_brand	= "Linux",
1012 	.emul_path	= linux_emul_path,
1013 	.interp_path	= "/lib/ld-musl-i386.so.1",
1014 	.sysvec		= &elf_linux_sysvec,
1015 	.interp_newpath	= NULL,
1016 	.brand_note	= &linux_brandnote,
1017 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1018 };
1019 
1020 Elf32_Brandinfo *linux_brandlist[] = {
1021 	&linux_brand,
1022 	&linux_glibc2brand,
1023 	&linux_muslbrand,
1024 	NULL
1025 };
1026 
1027 static int
1028 linux_elf_modevent(module_t mod, int type, void *data)
1029 {
1030 	Elf32_Brandinfo **brandinfo;
1031 	int error;
1032 	struct linux_ioctl_handler **lihp;
1033 
1034 	error = 0;
1035 
1036 	switch(type) {
1037 	case MOD_LOAD:
1038 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1039 		     ++brandinfo)
1040 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1041 				error = EINVAL;
1042 		if (error == 0) {
1043 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1044 				linux_ioctl_register_handler(*lihp);
1045 			LIST_INIT(&futex_list);
1046 			mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1047 			linux_get_machine(&linux_kplatform);
1048 			linux_szplatform = roundup(strlen(linux_kplatform) + 1,
1049 			    sizeof(char *));
1050 			linux_dev_shm_create();
1051 			linux_osd_jail_register();
1052 			stclohz = (stathz ? stathz : hz);
1053 			if (bootverbose)
1054 				printf("Linux ELF exec handler installed\n");
1055 		} else
1056 			printf("cannot insert Linux ELF brand handler\n");
1057 		break;
1058 	case MOD_UNLOAD:
1059 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1060 		     ++brandinfo)
1061 			if (elf32_brand_inuse(*brandinfo))
1062 				error = EBUSY;
1063 		if (error == 0) {
1064 			for (brandinfo = &linux_brandlist[0];
1065 			     *brandinfo != NULL; ++brandinfo)
1066 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1067 					error = EINVAL;
1068 		}
1069 		if (error == 0) {
1070 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1071 				linux_ioctl_unregister_handler(*lihp);
1072 			mtx_destroy(&futex_mtx);
1073 			linux_dev_shm_destroy();
1074 			linux_osd_jail_deregister();
1075 			if (bootverbose)
1076 				printf("Linux ELF exec handler removed\n");
1077 		} else
1078 			printf("Could not deinstall ELF interpreter entry\n");
1079 		break;
1080 	default:
1081 		return (EOPNOTSUPP);
1082 	}
1083 	return (error);
1084 }
1085 
1086 static moduledata_t linux_elf_mod = {
1087 	"linuxelf",
1088 	linux_elf_modevent,
1089 	0
1090 };
1091 
1092 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1093 FEATURE(linux, "Linux 32bit support");
1094