xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision 48c779cdecb5f803e5fe5d761987e976ca9609db)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/imgact_aout.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/signalvar.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysctl.h>
48 #include <sys/sysent.h>
49 #include <sys/sysproto.h>
50 #include <sys/vnode.h>
51 #include <sys/eventhandler.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_extern.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_param.h>
60 
61 #include <machine/cpu.h>
62 #include <machine/cputypes.h>
63 #include <machine/md_var.h>
64 #include <machine/pcb.h>
65 #include <machine/trap.h>
66 
67 #include <i386/linux/linux.h>
68 #include <i386/linux/linux_proto.h>
69 #include <compat/linux/linux_emul.h>
70 #include <compat/linux/linux_ioctl.h>
71 #include <compat/linux/linux_mib.h>
72 #include <compat/linux/linux_misc.h>
73 #include <compat/linux/linux_signal.h>
74 #include <compat/linux/linux_util.h>
75 #include <compat/linux/linux_vdso.h>
76 
77 MODULE_VERSION(linux, 1);
78 
79 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
80 
81 static int linux_szsigcode;
82 static vm_object_t linux_shared_page_obj;
83 static char *linux_shared_page_mapping;
84 extern char _binary_linux_locore_o_start;
85 extern char _binary_linux_locore_o_end;
86 
87 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
88 
89 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
90 
91 static int	linux_fixup(register_t **stack_base,
92 		    struct image_params *iparams);
93 static int	linux_fixup_elf(register_t **stack_base,
94 		    struct image_params *iparams);
95 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
96 static void	linux_exec_setregs(struct thread *td,
97 		    struct image_params *imgp, u_long stack);
98 static register_t *linux_copyout_strings(struct image_params *imgp);
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 static eventhandler_tag linux_exit_tag;
107 static eventhandler_tag linux_exec_tag;
108 static eventhandler_tag linux_thread_dtor_tag;
109 
110 #define LINUX_T_UNKNOWN  255
111 static int _bsd_to_linux_trapcode[] = {
112 	LINUX_T_UNKNOWN,	/* 0 */
113 	6,			/* 1  T_PRIVINFLT */
114 	LINUX_T_UNKNOWN,	/* 2 */
115 	3,			/* 3  T_BPTFLT */
116 	LINUX_T_UNKNOWN,	/* 4 */
117 	LINUX_T_UNKNOWN,	/* 5 */
118 	16,			/* 6  T_ARITHTRAP */
119 	254,			/* 7  T_ASTFLT */
120 	LINUX_T_UNKNOWN,	/* 8 */
121 	13,			/* 9  T_PROTFLT */
122 	1,			/* 10 T_TRCTRAP */
123 	LINUX_T_UNKNOWN,	/* 11 */
124 	14,			/* 12 T_PAGEFLT */
125 	LINUX_T_UNKNOWN,	/* 13 */
126 	17,			/* 14 T_ALIGNFLT */
127 	LINUX_T_UNKNOWN,	/* 15 */
128 	LINUX_T_UNKNOWN,	/* 16 */
129 	LINUX_T_UNKNOWN,	/* 17 */
130 	0,			/* 18 T_DIVIDE */
131 	2,			/* 19 T_NMI */
132 	4,			/* 20 T_OFLOW */
133 	5,			/* 21 T_BOUND */
134 	7,			/* 22 T_DNA */
135 	8,			/* 23 T_DOUBLEFLT */
136 	9,			/* 24 T_FPOPFLT */
137 	10,			/* 25 T_TSSFLT */
138 	11,			/* 26 T_SEGNPFLT */
139 	12,			/* 27 T_STKFLT */
140 	18,			/* 28 T_MCHK */
141 	19,			/* 29 T_XMMFLT */
142 	15			/* 30 T_RESERVED */
143 };
144 #define bsd_to_linux_trapcode(code) \
145     ((code)<nitems(_bsd_to_linux_trapcode)? \
146      _bsd_to_linux_trapcode[(code)]: \
147      LINUX_T_UNKNOWN)
148 
149 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
150 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
151 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
152 
153 /*
154  * If FreeBSD & Linux have a difference of opinion about what a trap
155  * means, deal with it here.
156  *
157  * MPSAFE
158  */
159 static int
160 linux_translate_traps(int signal, int trap_code)
161 {
162 	if (signal != SIGBUS)
163 		return (signal);
164 	switch (trap_code) {
165 	case T_PROTFLT:
166 	case T_TSSFLT:
167 	case T_DOUBLEFLT:
168 	case T_PAGEFLT:
169 		return (SIGSEGV);
170 	default:
171 		return (signal);
172 	}
173 }
174 
175 static int
176 linux_fixup(register_t **stack_base, struct image_params *imgp)
177 {
178 	register_t *argv, *envp;
179 
180 	argv = *stack_base;
181 	envp = *stack_base + (imgp->args->argc + 1);
182 	(*stack_base)--;
183 	suword(*stack_base, (intptr_t)(void *)envp);
184 	(*stack_base)--;
185 	suword(*stack_base, (intptr_t)(void *)argv);
186 	(*stack_base)--;
187 	suword(*stack_base, imgp->args->argc);
188 	return (0);
189 }
190 
191 static int
192 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
193 {
194 	struct proc *p;
195 	Elf32_Auxargs *args;
196 	Elf32_Auxinfo *argarray, *pos;
197 	Elf32_Addr *auxbase, *uplatform;
198 	struct ps_strings *arginfo;
199 	int error, issetugid;
200 
201 	KASSERT(curthread->td_proc == imgp->proc,
202 	    ("unsafe linux_fixup_elf(), should be curproc"));
203 
204 	p = imgp->proc;
205 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
206 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
207 	uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
208 	args = (Elf32_Auxargs *)imgp->auxargs;
209 	auxbase = *stack_base + imgp->args->argc + 1 + imgp->args->envc + 1;
210 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
211 	    M_WAITOK | M_ZERO);
212 
213 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
214 	    imgp->proc->p_sysent->sv_shared_page_base);
215 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
216 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
217 
218 	/*
219 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
220 	 * as it has appeared in the 2.4.0-rc7 first time.
221 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
222 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
223 	 * is not present.
224 	 * Also see linux_times() implementation.
225 	 */
226 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
227 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
228 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
229 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
230 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
231 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
232 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
233 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
234 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
235 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
236 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
237 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
238 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
239 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
240 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
241 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
242 	if (imgp->execpathp != 0)
243 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
244 	if (args->execfd != -1)
245 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
246 	AUXARGS_ENTRY(pos, AT_NULL, 0);
247 
248 	free(imgp->auxargs, M_TEMP);
249 	imgp->auxargs = NULL;
250 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
251 
252 	error = copyout(argarray, auxbase, sizeof(*argarray) * LINUX_AT_COUNT);
253 	free(argarray, M_TEMP);
254 	if (error != 0)
255 		return (error);
256 
257 	(*stack_base)--;
258 	if (suword(*stack_base, (register_t)imgp->args->argc) == -1)
259 		return (EFAULT);
260 	return (0);
261 }
262 
263 /*
264  * Copied from kern/kern_exec.c
265  */
266 static register_t *
267 linux_copyout_strings(struct image_params *imgp)
268 {
269 	int argc, envc;
270 	char **vectp;
271 	char *stringp, *destp;
272 	register_t *stack_base;
273 	struct ps_strings *arginfo;
274 	char canary[LINUX_AT_RANDOM_LEN];
275 	size_t execpath_len;
276 	struct proc *p;
277 
278 	/* Calculate string base and vector table pointers. */
279 	p = imgp->proc;
280 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
281 		execpath_len = strlen(imgp->execpath) + 1;
282 	else
283 		execpath_len = 0;
284 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
285 	destp = (caddr_t)arginfo - SPARE_USRSPACE - linux_szplatform -
286 	    roundup(sizeof(canary), sizeof(char *)) -
287 	    roundup(execpath_len, sizeof(char *)) -
288 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
289 
290 	/* Install LINUX_PLATFORM. */
291 	copyout(linux_kplatform, ((caddr_t)arginfo - linux_szplatform),
292 	    linux_szplatform);
293 
294 	if (execpath_len != 0) {
295 		imgp->execpathp = (uintptr_t)arginfo -
296 		linux_szplatform - execpath_len;
297 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
298 	}
299 
300 	/* Prepare the canary for SSP. */
301 	arc4rand(canary, sizeof(canary), 0);
302 	imgp->canary = (uintptr_t)arginfo - linux_szplatform -
303 	    roundup(execpath_len, sizeof(char *)) -
304 	    roundup(sizeof(canary), sizeof(char *));
305 	copyout(canary, (void *)imgp->canary, sizeof(canary));
306 
307 	vectp = (char **)destp;
308 	if (imgp->auxargs) {
309 		/*
310 		 * Allocate room on the stack for the ELF auxargs
311 		 * array.  It has LINUX_AT_COUNT entries.
312 		 */
313 		vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf32_Auxinfo),
314 		    sizeof(*vectp));
315 	}
316 
317 	/*
318 	 * Allocate room for the argv[] and env vectors including the
319 	 * terminating NULL pointers.
320 	 */
321 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
322 
323 	/* vectp also becomes our initial stack base. */
324 	stack_base = (register_t *)vectp;
325 
326 	stringp = imgp->args->begin_argv;
327 	argc = imgp->args->argc;
328 	envc = imgp->args->envc;
329 
330 	/* Copy out strings - arguments and environment. */
331 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
332 
333 	/* Fill in "ps_strings" struct for ps, w, etc. */
334 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
335 	suword(&arginfo->ps_nargvstr, argc);
336 
337 	/* Fill in argument portion of vector table. */
338 	for (; argc > 0; --argc) {
339 		suword(vectp++, (long)(intptr_t)destp);
340 		while (*stringp++ != 0)
341 			destp++;
342 		destp++;
343 	}
344 
345 	/* A null vector table pointer separates the argp's from the envp's. */
346 	suword(vectp++, 0);
347 
348 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
349 	suword(&arginfo->ps_nenvstr, envc);
350 
351 	/* Fill in environment portion of vector table. */
352 	for (; envc > 0; --envc) {
353 		suword(vectp++, (long)(intptr_t)destp);
354 		while (*stringp++ != 0)
355 			destp++;
356 		destp++;
357 	}
358 
359 	/* The end of the vector table is a null pointer. */
360 	suword(vectp, 0);
361 
362 	return (stack_base);
363 }
364 
365 static void
366 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
367 {
368 	struct thread *td = curthread;
369 	struct proc *p = td->td_proc;
370 	struct sigacts *psp;
371 	struct trapframe *regs;
372 	struct l_rt_sigframe *fp, frame;
373 	int sig, code;
374 	int oonstack;
375 
376 	sig = ksi->ksi_signo;
377 	code = ksi->ksi_code;
378 	PROC_LOCK_ASSERT(p, MA_OWNED);
379 	psp = p->p_sigacts;
380 	mtx_assert(&psp->ps_mtx, MA_OWNED);
381 	regs = td->td_frame;
382 	oonstack = sigonstack(regs->tf_esp);
383 
384 	/* Allocate space for the signal handler context. */
385 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
386 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
387 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
388 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
389 	} else
390 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
391 	mtx_unlock(&psp->ps_mtx);
392 
393 	/* Build the argument list for the signal handler. */
394 	sig = bsd_to_linux_signal(sig);
395 
396 	bzero(&frame, sizeof(frame));
397 
398 	frame.sf_handler = catcher;
399 	frame.sf_sig = sig;
400 	frame.sf_siginfo = &fp->sf_si;
401 	frame.sf_ucontext = &fp->sf_sc;
402 
403 	/* Fill in POSIX parts. */
404 	ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
405 
406 	/* Build the signal context to be used by sigreturn. */
407 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
408 	frame.sf_sc.uc_link = NULL;		/* XXX ??? */
409 
410 	frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
411 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
412 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
413 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
414 	PROC_UNLOCK(p);
415 
416 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
417 
418 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
419 	frame.sf_sc.uc_mcontext.sc_gs     = rgs();
420 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
421 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
422 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
423 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_edi;
424 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_esi;
425 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_ebp;
426 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_ebx;
427 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_esp;
428 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_edx;
429 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_ecx;
430 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_eax;
431 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_eip;
432 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
433 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
434 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
435 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
436 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
437 	frame.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
438 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
439 
440 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
441 		/*
442 		 * Process has trashed its stack; give it an illegal
443 		 * instruction to halt it in its tracks.
444 		 */
445 		PROC_LOCK(p);
446 		sigexit(td, SIGILL);
447 	}
448 
449 	/* Build context to run handler in. */
450 	regs->tf_esp = (int)fp;
451 	regs->tf_eip = linux_rt_sigcode;
452 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
453 	regs->tf_cs = _ucodesel;
454 	regs->tf_ds = _udatasel;
455 	regs->tf_es = _udatasel;
456 	regs->tf_fs = _udatasel;
457 	regs->tf_ss = _udatasel;
458 	PROC_LOCK(p);
459 	mtx_lock(&psp->ps_mtx);
460 }
461 
462 
463 /*
464  * Send an interrupt to process.
465  *
466  * Stack is set up to allow sigcode stored
467  * in u. to call routine, followed by kcall
468  * to sigreturn routine below.  After sigreturn
469  * resets the signal mask, the stack, and the
470  * frame pointer, it returns to the user
471  * specified pc, psl.
472  */
473 static void
474 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
475 {
476 	struct thread *td = curthread;
477 	struct proc *p = td->td_proc;
478 	struct sigacts *psp;
479 	struct trapframe *regs;
480 	struct l_sigframe *fp, frame;
481 	l_sigset_t lmask;
482 	int sig, code;
483 	int oonstack;
484 
485 	PROC_LOCK_ASSERT(p, MA_OWNED);
486 	psp = p->p_sigacts;
487 	sig = ksi->ksi_signo;
488 	code = ksi->ksi_code;
489 	mtx_assert(&psp->ps_mtx, MA_OWNED);
490 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
491 		/* Signal handler installed with SA_SIGINFO. */
492 		linux_rt_sendsig(catcher, ksi, mask);
493 		return;
494 	}
495 	regs = td->td_frame;
496 	oonstack = sigonstack(regs->tf_esp);
497 
498 	/* Allocate space for the signal handler context. */
499 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
500 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
501 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
502 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
503 	} else
504 		fp = (struct l_sigframe *)regs->tf_esp - 1;
505 	mtx_unlock(&psp->ps_mtx);
506 	PROC_UNLOCK(p);
507 
508 	/* Build the argument list for the signal handler. */
509 	sig = bsd_to_linux_signal(sig);
510 
511 	bzero(&frame, sizeof(frame));
512 
513 	frame.sf_handler = catcher;
514 	frame.sf_sig = sig;
515 
516 	bsd_to_linux_sigset(mask, &lmask);
517 
518 	/* Build the signal context to be used by sigreturn. */
519 	frame.sf_sc.sc_mask   = lmask.__mask;
520 	frame.sf_sc.sc_gs     = rgs();
521 	frame.sf_sc.sc_fs     = regs->tf_fs;
522 	frame.sf_sc.sc_es     = regs->tf_es;
523 	frame.sf_sc.sc_ds     = regs->tf_ds;
524 	frame.sf_sc.sc_edi    = regs->tf_edi;
525 	frame.sf_sc.sc_esi    = regs->tf_esi;
526 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
527 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
528 	frame.sf_sc.sc_esp    = regs->tf_esp;
529 	frame.sf_sc.sc_edx    = regs->tf_edx;
530 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
531 	frame.sf_sc.sc_eax    = regs->tf_eax;
532 	frame.sf_sc.sc_eip    = regs->tf_eip;
533 	frame.sf_sc.sc_cs     = regs->tf_cs;
534 	frame.sf_sc.sc_eflags = regs->tf_eflags;
535 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
536 	frame.sf_sc.sc_ss     = regs->tf_ss;
537 	frame.sf_sc.sc_err    = regs->tf_err;
538 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
539 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
540 
541 	frame.sf_extramask[0] = lmask.__mask;
542 
543 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
544 		/*
545 		 * Process has trashed its stack; give it an illegal
546 		 * instruction to halt it in its tracks.
547 		 */
548 		PROC_LOCK(p);
549 		sigexit(td, SIGILL);
550 	}
551 
552 	/* Build context to run handler in. */
553 	regs->tf_esp = (int)fp;
554 	regs->tf_eip = linux_sigcode;
555 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
556 	regs->tf_cs = _ucodesel;
557 	regs->tf_ds = _udatasel;
558 	regs->tf_es = _udatasel;
559 	regs->tf_fs = _udatasel;
560 	regs->tf_ss = _udatasel;
561 	PROC_LOCK(p);
562 	mtx_lock(&psp->ps_mtx);
563 }
564 
565 /*
566  * System call to cleanup state after a signal
567  * has been taken.  Reset signal mask and
568  * stack state from context left by sendsig (above).
569  * Return to previous pc and psl as specified by
570  * context left by sendsig. Check carefully to
571  * make sure that the user has not modified the
572  * psl to gain improper privileges or to cause
573  * a machine fault.
574  */
575 int
576 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
577 {
578 	struct l_sigframe frame;
579 	struct trapframe *regs;
580 	l_sigset_t lmask;
581 	sigset_t bmask;
582 	int eflags;
583 	ksiginfo_t ksi;
584 
585 	regs = td->td_frame;
586 
587 	/*
588 	 * The trampoline code hands us the sigframe.
589 	 * It is unsafe to keep track of it ourselves, in the event that a
590 	 * program jumps out of a signal handler.
591 	 */
592 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
593 		return (EFAULT);
594 
595 	/* Check for security violations. */
596 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
597 	eflags = frame.sf_sc.sc_eflags;
598 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
599 		return (EINVAL);
600 
601 	/*
602 	 * Don't allow users to load a valid privileged %cs.  Let the
603 	 * hardware check for invalid selectors, excess privilege in
604 	 * other selectors, invalid %eip's and invalid %esp's.
605 	 */
606 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
607 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
608 		ksiginfo_init_trap(&ksi);
609 		ksi.ksi_signo = SIGBUS;
610 		ksi.ksi_code = BUS_OBJERR;
611 		ksi.ksi_trapno = T_PROTFLT;
612 		ksi.ksi_addr = (void *)regs->tf_eip;
613 		trapsignal(td, &ksi);
614 		return (EINVAL);
615 	}
616 
617 	lmask.__mask = frame.sf_sc.sc_mask;
618 	linux_to_bsd_sigset(&lmask, &bmask);
619 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
620 
621 	/* Restore signal context. */
622 	/* %gs was restored by the trampoline. */
623 	regs->tf_fs     = frame.sf_sc.sc_fs;
624 	regs->tf_es     = frame.sf_sc.sc_es;
625 	regs->tf_ds     = frame.sf_sc.sc_ds;
626 	regs->tf_edi    = frame.sf_sc.sc_edi;
627 	regs->tf_esi    = frame.sf_sc.sc_esi;
628 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
629 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
630 	regs->tf_edx    = frame.sf_sc.sc_edx;
631 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
632 	regs->tf_eax    = frame.sf_sc.sc_eax;
633 	regs->tf_eip    = frame.sf_sc.sc_eip;
634 	regs->tf_cs     = frame.sf_sc.sc_cs;
635 	regs->tf_eflags = eflags;
636 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
637 	regs->tf_ss     = frame.sf_sc.sc_ss;
638 
639 	return (EJUSTRETURN);
640 }
641 
642 /*
643  * System call to cleanup state after a signal
644  * has been taken.  Reset signal mask and
645  * stack state from context left by rt_sendsig (above).
646  * Return to previous pc and psl as specified by
647  * context left by sendsig. Check carefully to
648  * make sure that the user has not modified the
649  * psl to gain improper privileges or to cause
650  * a machine fault.
651  */
652 int
653 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
654 {
655 	struct l_ucontext uc;
656 	struct l_sigcontext *context;
657 	sigset_t bmask;
658 	l_stack_t *lss;
659 	stack_t ss;
660 	struct trapframe *regs;
661 	int eflags;
662 	ksiginfo_t ksi;
663 
664 	regs = td->td_frame;
665 
666 	/*
667 	 * The trampoline code hands us the ucontext.
668 	 * It is unsafe to keep track of it ourselves, in the event that a
669 	 * program jumps out of a signal handler.
670 	 */
671 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
672 		return (EFAULT);
673 
674 	context = &uc.uc_mcontext;
675 
676 	/* Check for security violations. */
677 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
678 	eflags = context->sc_eflags;
679 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
680 		return (EINVAL);
681 
682 	/*
683 	 * Don't allow users to load a valid privileged %cs.  Let the
684 	 * hardware check for invalid selectors, excess privilege in
685 	 * other selectors, invalid %eip's and invalid %esp's.
686 	 */
687 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
688 	if (!CS_SECURE(context->sc_cs)) {
689 		ksiginfo_init_trap(&ksi);
690 		ksi.ksi_signo = SIGBUS;
691 		ksi.ksi_code = BUS_OBJERR;
692 		ksi.ksi_trapno = T_PROTFLT;
693 		ksi.ksi_addr = (void *)regs->tf_eip;
694 		trapsignal(td, &ksi);
695 		return (EINVAL);
696 	}
697 
698 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
699 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
700 
701 	/* Restore signal context. */
702 	/* %gs was restored by the trampoline. */
703 	regs->tf_fs     = context->sc_fs;
704 	regs->tf_es     = context->sc_es;
705 	regs->tf_ds     = context->sc_ds;
706 	regs->tf_edi    = context->sc_edi;
707 	regs->tf_esi    = context->sc_esi;
708 	regs->tf_ebp    = context->sc_ebp;
709 	regs->tf_ebx    = context->sc_ebx;
710 	regs->tf_edx    = context->sc_edx;
711 	regs->tf_ecx    = context->sc_ecx;
712 	regs->tf_eax    = context->sc_eax;
713 	regs->tf_eip    = context->sc_eip;
714 	regs->tf_cs     = context->sc_cs;
715 	regs->tf_eflags = eflags;
716 	regs->tf_esp    = context->sc_esp_at_signal;
717 	regs->tf_ss     = context->sc_ss;
718 
719 	/* Call sigaltstack & ignore results. */
720 	lss = &uc.uc_stack;
721 	ss.ss_sp = lss->ss_sp;
722 	ss.ss_size = lss->ss_size;
723 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
724 
725 	(void)kern_sigaltstack(td, &ss, NULL);
726 
727 	return (EJUSTRETURN);
728 }
729 
730 static int
731 linux_fetch_syscall_args(struct thread *td)
732 {
733 	struct proc *p;
734 	struct trapframe *frame;
735 	struct syscall_args *sa;
736 
737 	p = td->td_proc;
738 	frame = td->td_frame;
739 	sa = &td->td_sa;
740 
741 	sa->code = frame->tf_eax;
742 	sa->args[0] = frame->tf_ebx;
743 	sa->args[1] = frame->tf_ecx;
744 	sa->args[2] = frame->tf_edx;
745 	sa->args[3] = frame->tf_esi;
746 	sa->args[4] = frame->tf_edi;
747 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
748 
749 	if (sa->code >= p->p_sysent->sv_size)
750 		/* nosys */
751 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
752 	else
753 		sa->callp = &p->p_sysent->sv_table[sa->code];
754 	sa->narg = sa->callp->sy_narg;
755 
756 	td->td_retval[0] = 0;
757 	td->td_retval[1] = frame->tf_edx;
758 
759 	return (0);
760 }
761 
762 /*
763  * exec_setregs may initialize some registers differently than Linux
764  * does, thus potentially confusing Linux binaries. If necessary, we
765  * override the exec_setregs default(s) here.
766  */
767 static void
768 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
769 {
770 	struct pcb *pcb = td->td_pcb;
771 
772 	exec_setregs(td, imgp, stack);
773 
774 	/* Linux sets %gs to 0, we default to _udatasel. */
775 	pcb->pcb_gs = 0;
776 	load_gs(0);
777 
778 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
779 }
780 
781 static void
782 linux_get_machine(const char **dst)
783 {
784 
785 	switch (cpu_class) {
786 	case CPUCLASS_686:
787 		*dst = "i686";
788 		break;
789 	case CPUCLASS_586:
790 		*dst = "i586";
791 		break;
792 	case CPUCLASS_486:
793 		*dst = "i486";
794 		break;
795 	default:
796 		*dst = "i386";
797 	}
798 }
799 
800 struct sysentvec linux_sysvec = {
801 	.sv_size	= LINUX_SYS_MAXSYSCALL,
802 	.sv_table	= linux_sysent,
803 	.sv_errsize	= ELAST + 1,
804 	.sv_errtbl	= linux_errtbl,
805 	.sv_transtrap	= linux_translate_traps,
806 	.sv_fixup	= linux_fixup,
807 	.sv_sendsig	= linux_sendsig,
808 	.sv_sigcode	= &_binary_linux_locore_o_start,
809 	.sv_szsigcode	= &linux_szsigcode,
810 	.sv_name	= "Linux a.out",
811 	.sv_coredump	= NULL,
812 	.sv_imgact_try	= linux_exec_imgact_try,
813 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
814 	.sv_minuser	= VM_MIN_ADDRESS,
815 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
816 	.sv_usrstack	= LINUX_USRSTACK,
817 	.sv_psstrings	= PS_STRINGS,
818 	.sv_stackprot	= VM_PROT_ALL,
819 	.sv_copyout_strings = exec_copyout_strings,
820 	.sv_setregs	= linux_exec_setregs,
821 	.sv_fixlimit	= NULL,
822 	.sv_maxssiz	= NULL,
823 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32,
824 	.sv_set_syscall_retval = cpu_set_syscall_retval,
825 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
826 	.sv_syscallnames = NULL,
827 	.sv_shared_page_base = LINUX_SHAREDPAGE,
828 	.sv_shared_page_len = PAGE_SIZE,
829 	.sv_schedtail	= linux_schedtail,
830 	.sv_thread_detach = linux_thread_detach,
831 	.sv_trap	= NULL,
832 };
833 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
834 
835 struct sysentvec elf_linux_sysvec = {
836 	.sv_size	= LINUX_SYS_MAXSYSCALL,
837 	.sv_table	= linux_sysent,
838 	.sv_errsize	= ELAST + 1,
839 	.sv_errtbl	= linux_errtbl,
840 	.sv_transtrap	= linux_translate_traps,
841 	.sv_fixup	= linux_fixup_elf,
842 	.sv_sendsig	= linux_sendsig,
843 	.sv_sigcode	= &_binary_linux_locore_o_start,
844 	.sv_szsigcode	= &linux_szsigcode,
845 	.sv_name	= "Linux ELF",
846 	.sv_coredump	= elf32_coredump,
847 	.sv_imgact_try	= linux_exec_imgact_try,
848 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
849 	.sv_minuser	= VM_MIN_ADDRESS,
850 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
851 	.sv_usrstack	= LINUX_USRSTACK,
852 	.sv_psstrings	= LINUX_PS_STRINGS,
853 	.sv_stackprot	= VM_PROT_ALL,
854 	.sv_copyout_strings = linux_copyout_strings,
855 	.sv_setregs	= linux_exec_setregs,
856 	.sv_fixlimit	= NULL,
857 	.sv_maxssiz	= NULL,
858 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP,
859 	.sv_set_syscall_retval = cpu_set_syscall_retval,
860 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
861 	.sv_syscallnames = NULL,
862 	.sv_shared_page_base = LINUX_SHAREDPAGE,
863 	.sv_shared_page_len = PAGE_SIZE,
864 	.sv_schedtail	= linux_schedtail,
865 	.sv_thread_detach = linux_thread_detach,
866 	.sv_trap	= NULL,
867 };
868 
869 static void
870 linux_vdso_install(void *param)
871 {
872 
873 	linux_szsigcode = (&_binary_linux_locore_o_end -
874 	    &_binary_linux_locore_o_start);
875 
876 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
877 		panic("Linux invalid vdso size\n");
878 
879 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
880 
881 	linux_shared_page_obj = __elfN(linux_shared_page_init)
882 	    (&linux_shared_page_mapping);
883 
884 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
885 
886 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
887 	    linux_szsigcode);
888 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
889 }
890 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
891     linux_vdso_install, NULL);
892 
893 static void
894 linux_vdso_deinstall(void *param)
895 {
896 
897 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
898 }
899 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
900     linux_vdso_deinstall, NULL);
901 
902 static char GNU_ABI_VENDOR[] = "GNU";
903 static int GNULINUX_ABI_DESC = 0;
904 
905 static bool
906 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
907 {
908 	const Elf32_Word *desc;
909 	uintptr_t p;
910 
911 	p = (uintptr_t)(note + 1);
912 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
913 
914 	desc = (const Elf32_Word *)p;
915 	if (desc[0] != GNULINUX_ABI_DESC)
916 		return (false);
917 
918 	/*
919 	 * For Linux we encode osrel using the Linux convention of
920 	 * 	(version << 16) | (major << 8) | (minor)
921 	 * See macro in linux_mib.h
922 	 */
923 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
924 
925 	return (true);
926 }
927 
928 static Elf_Brandnote linux_brandnote = {
929 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
930 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
931 	.hdr.n_type	= 1,
932 	.vendor		= GNU_ABI_VENDOR,
933 	.flags		= BN_TRANSLATE_OSREL,
934 	.trans_osrel	= linux_trans_osrel
935 };
936 
937 static Elf32_Brandinfo linux_brand = {
938 	.brand		= ELFOSABI_LINUX,
939 	.machine	= EM_386,
940 	.compat_3_brand	= "Linux",
941 	.emul_path	= "/compat/linux",
942 	.interp_path	= "/lib/ld-linux.so.1",
943 	.sysvec		= &elf_linux_sysvec,
944 	.interp_newpath	= NULL,
945 	.brand_note	= &linux_brandnote,
946 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
947 };
948 
949 static Elf32_Brandinfo linux_glibc2brand = {
950 	.brand		= ELFOSABI_LINUX,
951 	.machine	= EM_386,
952 	.compat_3_brand	= "Linux",
953 	.emul_path	= "/compat/linux",
954 	.interp_path	= "/lib/ld-linux.so.2",
955 	.sysvec		= &elf_linux_sysvec,
956 	.interp_newpath	= NULL,
957 	.brand_note	= &linux_brandnote,
958 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
959 };
960 
961 static Elf32_Brandinfo linux_muslbrand = {
962 	.brand		= ELFOSABI_LINUX,
963 	.machine	= EM_386,
964 	.compat_3_brand	= "Linux",
965 	.emul_path	= "/compat/linux",
966 	.interp_path	= "/lib/ld-musl-i386.so.1",
967 	.sysvec		= &elf_linux_sysvec,
968 	.interp_newpath	= NULL,
969 	.brand_note	= &linux_brandnote,
970 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
971 };
972 
973 Elf32_Brandinfo *linux_brandlist[] = {
974 	&linux_brand,
975 	&linux_glibc2brand,
976 	&linux_muslbrand,
977 	NULL
978 };
979 
980 static int
981 linux_elf_modevent(module_t mod, int type, void *data)
982 {
983 	Elf32_Brandinfo **brandinfo;
984 	int error;
985 	struct linux_ioctl_handler **lihp;
986 
987 	error = 0;
988 
989 	switch(type) {
990 	case MOD_LOAD:
991 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
992 		     ++brandinfo)
993 			if (elf32_insert_brand_entry(*brandinfo) < 0)
994 				error = EINVAL;
995 		if (error == 0) {
996 			SET_FOREACH(lihp, linux_ioctl_handler_set)
997 				linux_ioctl_register_handler(*lihp);
998 			LIST_INIT(&futex_list);
999 			mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1000 			linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit,
1001 			      NULL, 1000);
1002 			linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec,
1003 			      NULL, 1000);
1004 			linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor,
1005 			    linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY);
1006 			linux_get_machine(&linux_kplatform);
1007 			linux_szplatform = roundup(strlen(linux_kplatform) + 1,
1008 			    sizeof(char *));
1009 			linux_dev_shm_create();
1010 			linux_osd_jail_register();
1011 			stclohz = (stathz ? stathz : hz);
1012 			if (bootverbose)
1013 				printf("Linux ELF exec handler installed\n");
1014 		} else
1015 			printf("cannot insert Linux ELF brand handler\n");
1016 		break;
1017 	case MOD_UNLOAD:
1018 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1019 		     ++brandinfo)
1020 			if (elf32_brand_inuse(*brandinfo))
1021 				error = EBUSY;
1022 		if (error == 0) {
1023 			for (brandinfo = &linux_brandlist[0];
1024 			     *brandinfo != NULL; ++brandinfo)
1025 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1026 					error = EINVAL;
1027 		}
1028 		if (error == 0) {
1029 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1030 				linux_ioctl_unregister_handler(*lihp);
1031 			mtx_destroy(&futex_mtx);
1032 			EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
1033 			EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
1034 			EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag);
1035 			linux_dev_shm_destroy();
1036 			linux_osd_jail_deregister();
1037 			if (bootverbose)
1038 				printf("Linux ELF exec handler removed\n");
1039 		} else
1040 			printf("Could not deinstall ELF interpreter entry\n");
1041 		break;
1042 	default:
1043 		return (EOPNOTSUPP);
1044 	}
1045 	return (error);
1046 }
1047 
1048 static moduledata_t linux_elf_mod = {
1049 	"linuxelf",
1050 	linux_elf_modevent,
1051 	0
1052 };
1053 
1054 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1055 FEATURE(linux, "Linux 32bit support");
1056