xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 9034852c84a13f0e3b5527e1c886ca94b2863b2b)
1 /*-
2  * Copyright (c) 2013 Dmitry Chagin
3  * Copyright (c) 2004 Tim J. Robbins
4  * Copyright (c) 2003 Peter Wemm
5  * Copyright (c) 2002 Doug Rabson
6  * Copyright (c) 1998-1999 Andrew Gallatin
7  * Copyright (c) 1994-1996 Søren Schmidt
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_compat.h"
38 
39 #define	__ELF_WORD_SIZE	64
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/exec.h>
44 #include <sys/fcntl.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/ktr.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/module.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/resourcevar.h>
55 #include <sys/signalvar.h>
56 #include <sys/sysctl.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
59 #include <sys/sysproto.h>
60 #include <sys/vnode.h>
61 #include <sys/eventhandler.h>
62 
63 #include <vm/vm.h>
64 #include <vm/pmap.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_param.h>
70 
71 #include <machine/cpu.h>
72 #include <machine/md_var.h>
73 #include <machine/pcb.h>
74 #include <machine/specialreg.h>
75 
76 #include <amd64/linux/linux.h>
77 #include <amd64/linux/linux_proto.h>
78 #include <compat/linux/linux_emul.h>
79 #include <compat/linux/linux_futex.h>
80 #include <compat/linux/linux_ioctl.h>
81 #include <compat/linux/linux_mib.h>
82 #include <compat/linux/linux_misc.h>
83 #include <compat/linux/linux_signal.h>
84 #include <compat/linux/linux_sysproto.h>
85 #include <compat/linux/linux_util.h>
86 #include <compat/linux/linux_vdso.h>
87 
88 MODULE_VERSION(linux64, 1);
89 
90 #if BYTE_ORDER == LITTLE_ENDIAN
91 #define SHELLMAGIC      0x2123 /* #! */
92 #else
93 #define SHELLMAGIC      0x2321
94 #endif
95 
96 #if defined(DEBUG)
97 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
98 	    CTLTYPE_STRING | CTLFLAG_RW,
99 	    0, 0, linux_sysctl_debug, "A",
100 	    "Linux 64 debugging control");
101 #endif
102 
103 /*
104  * Allow the this functions to use the ldebug() facility
105  * even though they are not syscalls themselves. Map them
106  * to syscall 0. This is slightly less bogus than using
107  * ldebug(sigreturn).
108  */
109 #define	LINUX_SYS_linux_rt_sendsig	0
110 
111 const char *linux_kplatform;
112 static int linux_szsigcode;
113 static vm_object_t linux_shared_page_obj;
114 static char *linux_shared_page_mapping;
115 extern char _binary_linux_locore_o_start;
116 extern char _binary_linux_locore_o_end;
117 
118 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
119 
120 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
121 
122 static register_t * linux_copyout_strings(struct image_params *imgp);
123 static int	elf_linux_fixup(register_t **stack_base,
124 		    struct image_params *iparams);
125 static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
126 static void	linux_vdso_install(void *param);
127 static void	linux_vdso_deinstall(void *param);
128 static void	linux_set_syscall_retval(struct thread *td, int error);
129 static int	linux_fetch_syscall_args(struct thread *td, struct syscall_args *sa);
130 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
131 		    u_long stack);
132 
133 /*
134  * Linux syscalls return negative errno's, we do positive and map them
135  * Reference:
136  *   FreeBSD: src/sys/sys/errno.h
137  *   Linux:   linux-2.6.17.8/include/asm-generic/errno-base.h
138  *            linux-2.6.17.8/include/asm-generic/errno.h
139  */
140 static int bsd_to_linux_errno[ELAST + 1] = {
141 	-0,  -1,  -2,  -3,  -4,  -5,  -6,  -7,  -8,  -9,
142 	-10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
143 	-20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
144 	-30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
145 	-90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
146 	-100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
147 	-110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
148 	-116, -66,  -6,  -6,  -6,  -6,  -6, -37, -38,  -9,
149 	  -6,  -6, -43, -42, -75,-125, -84, -95, -16, -74,
150 	 -72, -67, -71
151 };
152 
153 #define LINUX_T_UNKNOWN  255
154 static int _bsd_to_linux_trapcode[] = {
155 	LINUX_T_UNKNOWN,	/* 0 */
156 	6,			/* 1  T_PRIVINFLT */
157 	LINUX_T_UNKNOWN,	/* 2 */
158 	3,			/* 3  T_BPTFLT */
159 	LINUX_T_UNKNOWN,	/* 4 */
160 	LINUX_T_UNKNOWN,	/* 5 */
161 	16,			/* 6  T_ARITHTRAP */
162 	254,			/* 7  T_ASTFLT */
163 	LINUX_T_UNKNOWN,	/* 8 */
164 	13,			/* 9  T_PROTFLT */
165 	1,			/* 10 T_TRCTRAP */
166 	LINUX_T_UNKNOWN,	/* 11 */
167 	14,			/* 12 T_PAGEFLT */
168 	LINUX_T_UNKNOWN,	/* 13 */
169 	17,			/* 14 T_ALIGNFLT */
170 	LINUX_T_UNKNOWN,	/* 15 */
171 	LINUX_T_UNKNOWN,	/* 16 */
172 	LINUX_T_UNKNOWN,	/* 17 */
173 	0,			/* 18 T_DIVIDE */
174 	2,			/* 19 T_NMI */
175 	4,			/* 20 T_OFLOW */
176 	5,			/* 21 T_BOUND */
177 	7,			/* 22 T_DNA */
178 	8,			/* 23 T_DOUBLEFLT */
179 	9,			/* 24 T_FPOPFLT */
180 	10,			/* 25 T_TSSFLT */
181 	11,			/* 26 T_SEGNPFLT */
182 	12,			/* 27 T_STKFLT */
183 	18,			/* 28 T_MCHK */
184 	19,			/* 29 T_XMMFLT */
185 	15			/* 30 T_RESERVED */
186 };
187 #define bsd_to_linux_trapcode(code) \
188     ((code)<sizeof(_bsd_to_linux_trapcode)/sizeof(*_bsd_to_linux_trapcode)? \
189      _bsd_to_linux_trapcode[(code)]: \
190      LINUX_T_UNKNOWN)
191 
192 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
193 LINUX_VDSO_SYM_CHAR(linux_platform);
194 
195 /*
196  * If FreeBSD & Linux have a difference of opinion about what a trap
197  * means, deal with it here.
198  *
199  * MPSAFE
200  */
201 static int
202 translate_traps(int signal, int trap_code)
203 {
204 
205 	if (signal != SIGBUS)
206 		return signal;
207 	switch (trap_code) {
208 	case T_PROTFLT:
209 	case T_TSSFLT:
210 	case T_DOUBLEFLT:
211 	case T_PAGEFLT:
212 		return SIGSEGV;
213 	default:
214 		return signal;
215 	}
216 }
217 
218 static int
219 linux_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
220 {
221 	struct proc *p;
222 	struct trapframe *frame;
223 
224 	p = td->td_proc;
225 	frame = td->td_frame;
226 
227 	sa->args[0] = frame->tf_rdi;
228 	sa->args[1] = frame->tf_rsi;
229 	sa->args[2] = frame->tf_rdx;
230 	sa->args[3] = frame->tf_rcx;
231 	sa->args[4] = frame->tf_r8;
232 	sa->args[5] = frame->tf_r9;
233 	sa->code = frame->tf_rax;
234 
235 	if (sa->code >= p->p_sysent->sv_size)
236 		/* nosys */
237 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
238 	else
239 		sa->callp = &p->p_sysent->sv_table[sa->code];
240 	sa->narg = sa->callp->sy_narg;
241 
242 	td->td_retval[0] = 0;
243 	return (0);
244 }
245 
246 static void
247 linux_set_syscall_retval(struct thread *td, int error)
248 {
249 	struct trapframe *frame = td->td_frame;
250 
251 	/*
252 	 * On Linux only %rcx and %r11 values are not preserved across
253 	 * the syscall.
254 	 * So, do not clobber %rdx and %r10
255 	 */
256 	td->td_retval[1] = frame->tf_rdx;
257 	frame->tf_r10 = frame->tf_rcx;
258 
259 	cpu_set_syscall_retval(td, error);
260 
261 	 /* Restore all registers. */
262 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
263 }
264 
265 static int
266 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
267 {
268 	Elf_Auxargs *args;
269 	Elf_Addr *base;
270 	Elf_Addr *pos;
271 	struct ps_strings *arginfo;
272 	struct proc *p;
273 
274 	p = imgp->proc;
275 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
276 
277 	KASSERT(curthread->td_proc == imgp->proc,
278 	    ("unsafe elf_linux_fixup(), should be curproc"));
279 	base = (Elf64_Addr *)*stack_base;
280 	args = (Elf64_Auxargs *)imgp->auxargs;
281 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
282 
283 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
284 	    imgp->proc->p_sysent->sv_shared_page_base);
285 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
286 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
287 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
288 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
289 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
290 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
291 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
292 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
293 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
294 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
295 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
296 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
297 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
298 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, 0);
299 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
300 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
301 	if (imgp->execpathp != 0)
302 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
303 	if (args->execfd != -1)
304 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
305 	AUXARGS_ENTRY(pos, AT_NULL, 0);
306 	free(imgp->auxargs, M_TEMP);
307 	imgp->auxargs = NULL;
308 
309 	base--;
310 	suword(base, (uint64_t)imgp->args->argc);
311 
312 	*stack_base = (register_t *)base;
313 	return (0);
314 }
315 
316 /*
317  * Copy strings out to the new process address space, constructing new arg
318  * and env vector tables. Return a pointer to the base so that it can be used
319  * as the initial stack pointer.
320  */
321 static register_t *
322 linux_copyout_strings(struct image_params *imgp)
323 {
324 	int argc, envc;
325 	char **vectp;
326 	char *stringp, *destp;
327 	register_t *stack_base;
328 	struct ps_strings *arginfo;
329 	char canary[LINUX_AT_RANDOM_LEN];
330 	size_t execpath_len;
331 	struct proc *p;
332 
333 	/*
334 	 * Calculate string base and vector table pointers.
335 	 */
336 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
337 		execpath_len = strlen(imgp->execpath) + 1;
338 	else
339 		execpath_len = 0;
340 
341 	p = imgp->proc;
342 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
343 	destp =	(caddr_t)arginfo - SPARE_USRSPACE -
344 	    roundup(sizeof(canary), sizeof(char *)) -
345 	    roundup(execpath_len, sizeof(char *)) -
346 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
347 
348 	if (execpath_len != 0) {
349 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
350 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
351 	}
352 
353 	/*
354 	 * Prepare the canary for SSP.
355 	 */
356 	arc4rand(canary, sizeof(canary), 0);
357 	imgp->canary = (uintptr_t)arginfo -
358 	    roundup(execpath_len, sizeof(char *)) -
359 	    roundup(sizeof(canary), sizeof(char *));
360 	copyout(canary, (void *)imgp->canary, sizeof(canary));
361 
362 	/*
363 	 * If we have a valid auxargs ptr, prepare some room
364 	 * on the stack.
365 	 */
366 	if (imgp->auxargs) {
367 		/*
368 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
369 		 * lower compatibility.
370 		 */
371 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
372 		    (LINUX_AT_COUNT * 2);
373 
374 		/*
375 		 * The '+ 2' is for the null pointers at the end of each of
376 		 * the arg and env vector sets,and imgp->auxarg_size is room
377 		 * for argument of Runtime loader.
378 		 */
379 		vectp = (char **)(destp - (imgp->args->argc +
380 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
381 
382 	} else {
383 		/*
384 		 * The '+ 2' is for the null pointers at the end of each of
385 		 * the arg and env vector sets
386 		 */
387 		vectp = (char **)(destp - (imgp->args->argc +
388 		    imgp->args->envc + 2) * sizeof(char *));
389 	}
390 
391 	/*
392 	 * vectp also becomes our initial stack base
393 	 */
394 	stack_base = (register_t *)vectp;
395 
396 	stringp = imgp->args->begin_argv;
397 	argc = imgp->args->argc;
398 	envc = imgp->args->envc;
399 
400 	/*
401 	 * Copy out strings - arguments and environment.
402 	 */
403 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
404 
405 	/*
406 	 * Fill in "ps_strings" struct for ps, w, etc.
407 	 */
408 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
409 	suword(&arginfo->ps_nargvstr, argc);
410 
411 	/*
412 	 * Fill in argument portion of vector table.
413 	 */
414 	for (; argc > 0; --argc) {
415 		suword(vectp++, (long)(intptr_t)destp);
416 		while (*stringp++ != 0)
417 			destp++;
418 		destp++;
419 	}
420 
421 	/* a null vector table pointer separates the argp's from the envp's */
422 	suword(vectp++, 0);
423 
424 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
425 	suword(&arginfo->ps_nenvstr, envc);
426 
427 	/*
428 	 * Fill in environment portion of vector table.
429 	 */
430 	for (; envc > 0; --envc) {
431 		suword(vectp++, (long)(intptr_t)destp);
432 		while (*stringp++ != 0)
433 			destp++;
434 		destp++;
435 	}
436 
437 	/* end of vector table is a null pointer */
438 	suword(vectp, 0);
439 	return (stack_base);
440 }
441 
442 /*
443  * Reset registers to default values on exec.
444  */
445 static void
446 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
447 {
448 	struct trapframe *regs = td->td_frame;
449 	struct pcb *pcb = td->td_pcb;
450 
451 	mtx_lock(&dt_lock);
452 	if (td->td_proc->p_md.md_ldt != NULL)
453 		user_ldt_free(td);
454 	else
455 		mtx_unlock(&dt_lock);
456 
457 	pcb->pcb_fsbase = 0;
458 	pcb->pcb_gsbase = 0;
459 	clear_pcb_flags(pcb, PCB_32BIT);
460 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
461 	set_pcb_flags(pcb, PCB_FULL_IRET);
462 
463 	bzero((char *)regs, sizeof(struct trapframe));
464 	regs->tf_rip = imgp->entry_addr;
465 	regs->tf_rsp = stack;
466 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
467 	regs->tf_ss = _udatasel;
468 	regs->tf_cs = _ucodesel;
469 	regs->tf_ds = _udatasel;
470 	regs->tf_es = _udatasel;
471 	regs->tf_fs = _ufssel;
472 	regs->tf_gs = _ugssel;
473 	regs->tf_flags = TF_HASSEGS;
474 
475 	/*
476 	 * Reset the hardware debug registers if they were in use.
477 	 * They won't have any meaning for the newly exec'd process.
478 	 */
479 	if (pcb->pcb_flags & PCB_DBREGS) {
480 		pcb->pcb_dr0 = 0;
481 		pcb->pcb_dr1 = 0;
482 		pcb->pcb_dr2 = 0;
483 		pcb->pcb_dr3 = 0;
484 		pcb->pcb_dr6 = 0;
485 		pcb->pcb_dr7 = 0;
486 		if (pcb == curpcb) {
487 			/*
488 			 * Clear the debug registers on the running
489 			 * CPU, otherwise they will end up affecting
490 			 * the next process we switch to.
491 			 */
492 			reset_dbregs();
493 		}
494 		clear_pcb_flags(pcb, PCB_DBREGS);
495 	}
496 
497 	/*
498 	 * Drop the FP state if we hold it, so that the process gets a
499 	 * clean FP state if it uses the FPU again.
500 	 */
501 	fpstate_drop(td);
502 }
503 
504 /*
505  * Copied from amd64/amd64/machdep.c
506  *
507  * XXX fpu state need? don't think so
508  */
509 int
510 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
511 {
512 	struct proc *p;
513 	struct l_ucontext uc;
514 	struct l_sigcontext *context;
515 	struct trapframe *regs;
516 	unsigned long rflags;
517 	int error;
518 	ksiginfo_t ksi;
519 
520 	regs = td->td_frame;
521 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
522 	if (error != 0)
523 		return (error);
524 
525 	p = td->td_proc;
526 	context = &uc.uc_mcontext;
527 	rflags = context->sc_rflags;
528 
529 	/*
530 	 * Don't allow users to change privileged or reserved flags.
531 	 */
532 	/*
533 	 * XXX do allow users to change the privileged flag PSL_RF.
534 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
535 	 * should sometimes set it there too.  tf_rflags is kept in
536 	 * the signal context during signal handling and there is no
537 	 * other place to remember it, so the PSL_RF bit may be
538 	 * corrupted by the signal handler without us knowing.
539 	 * Corruption of the PSL_RF bit at worst causes one more or
540 	 * one less debugger trap, so allowing it is fairly harmless.
541 	 */
542 
543 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
544 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
545 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
546 		return (EINVAL);
547 	}
548 
549 	/*
550 	 * Don't allow users to load a valid privileged %cs.  Let the
551 	 * hardware check for invalid selectors, excess privilege in
552 	 * other selectors, invalid %eip's and invalid %esp's.
553 	 */
554 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
555 	if (!CS_SECURE(context->sc_cs)) {
556 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
557 		ksiginfo_init_trap(&ksi);
558 		ksi.ksi_signo = SIGBUS;
559 		ksi.ksi_code = BUS_OBJERR;
560 		ksi.ksi_trapno = T_PROTFLT;
561 		ksi.ksi_addr = (void *)regs->tf_rip;
562 		trapsignal(td, &ksi);
563 		return (EINVAL);
564 	}
565 
566 	PROC_LOCK(p);
567 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
568 	SIG_CANTMASK(td->td_sigmask);
569 	signotify(td);
570 	PROC_UNLOCK(p);
571 
572 	regs->tf_rdi    = context->sc_rdi;
573 	regs->tf_rsi    = context->sc_rsi;
574 	regs->tf_rdx    = context->sc_rdx;
575 	regs->tf_rbp    = context->sc_rbp;
576 	regs->tf_rbx    = context->sc_rbx;
577 	regs->tf_rcx    = context->sc_rcx;
578 	regs->tf_rax    = context->sc_rax;
579 	regs->tf_rip    = context->sc_rip;
580 	regs->tf_rsp    = context->sc_rsp;
581 	regs->tf_r8     = context->sc_r8;
582 	regs->tf_r9     = context->sc_r9;
583 	regs->tf_r10    = context->sc_r10;
584 	regs->tf_r11    = context->sc_r11;
585 	regs->tf_r12    = context->sc_r12;
586 	regs->tf_r13    = context->sc_r13;
587 	regs->tf_r14    = context->sc_r14;
588 	regs->tf_r15    = context->sc_r15;
589 	regs->tf_cs     = context->sc_cs;
590 	regs->tf_err    = context->sc_err;
591 	regs->tf_rflags = rflags;
592 
593 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
594 	return (EJUSTRETURN);
595 }
596 
597 /*
598  * copied from amd64/amd64/machdep.c
599  *
600  * Send an interrupt to process.
601  */
602 static void
603 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
604 {
605 	struct l_rt_sigframe sf, *sfp;
606 	struct proc *p;
607 	struct thread *td;
608 	struct sigacts *psp;
609 	caddr_t sp;
610 	struct trapframe *regs;
611 	int sig, code;
612 	int oonstack;
613 
614 	td = curthread;
615 	p = td->td_proc;
616 	PROC_LOCK_ASSERT(p, MA_OWNED);
617 	sig = ksi->ksi_signo;
618 	psp = p->p_sigacts;
619 	code = ksi->ksi_code;
620 	mtx_assert(&psp->ps_mtx, MA_OWNED);
621 	regs = td->td_frame;
622 	oonstack = sigonstack(regs->tf_rsp);
623 
624 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
625 	    catcher, sig, mask, code);
626 
627 	/* Allocate space for the signal handler context. */
628 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
629 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
630 		sp = td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
631 		    sizeof(struct l_rt_sigframe);
632 	} else
633 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
634 	/* Align to 16 bytes. */
635 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
636 	mtx_unlock(&psp->ps_mtx);
637 
638 	/* Translate the signal. */
639 	sig = bsd_to_linux_signal(sig);
640 
641 	/* Save user context. */
642 	bzero(&sf, sizeof(sf));
643 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
644 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
645 
646 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
647 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
648 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
649 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
650 	PROC_UNLOCK(p);
651 
652 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
653 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
654 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
655 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
656 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
657 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
658 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
659 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
660 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
661 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
662 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
663 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
664 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
665 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
666 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
667 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
668 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
669 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
670 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
671 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
672 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
673 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
674 
675 	/* Build the argument list for the signal handler. */
676 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
677 	regs->tf_rax = 0;
678 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
679 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
680 
681 	sf.sf_handler = catcher;
682 	/* Fill in POSIX parts */
683 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
684 
685 	/*
686 	 * Copy the sigframe out to the user's stack.
687 	 */
688 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
689 #ifdef DEBUG
690 		printf("process %ld has trashed its stack\n", (long)p->p_pid);
691 #endif
692 		PROC_LOCK(p);
693 		sigexit(td, SIGILL);
694 	}
695 
696 	regs->tf_rsp = (long)sfp;
697 	regs->tf_rip = linux_rt_sigcode;
698 	regs->tf_rflags &= ~(PSL_T | PSL_D);
699 	regs->tf_cs = _ucodesel;
700 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
701 	PROC_LOCK(p);
702 	mtx_lock(&psp->ps_mtx);
703 }
704 
705 /*
706  * If a linux binary is exec'ing something, try this image activator
707  * first.  We override standard shell script execution in order to
708  * be able to modify the interpreter path.  We only do this if a linux
709  * binary is doing the exec, so we do not create an EXEC module for it.
710  */
711 static int exec_linux_imgact_try(struct image_params *iparams);
712 
713 static int
714 exec_linux_imgact_try(struct image_params *imgp)
715 {
716 	const char *head = (const char *)imgp->image_header;
717 	char *rpath;
718 	int error = -1, len;
719 
720 	/*
721 	 * The interpreter for shell scripts run from a linux binary needs
722 	 * to be located in /compat/linux if possible in order to recursively
723 	 * maintain linux path emulation.
724 	 */
725 	if (((const short *)head)[0] == SHELLMAGIC) {
726 		/*
727 		 * Run our normal shell image activator.  If it succeeds
728 		 * attempt to use the alternate path for the interpreter.
729 		 * If an alternate path is found, use our stringspace
730 		 * to store it.
731 		 */
732 		if ((error = exec_shell_imgact(imgp)) == 0) {
733 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
734 			    imgp->interpreter_name, UIO_SYSSPACE,
735 			    &rpath, 0, AT_FDCWD);
736 			if (rpath != NULL) {
737 				len = strlen(rpath) + 1;
738 
739 				if (len <= MAXSHELLCMDLEN)
740 					memcpy(imgp->interpreter_name,
741 					    rpath, len);
742 				free(rpath, M_TEMP);
743 			}
744 		}
745 	}
746 	return(error);
747 }
748 
749 struct sysentvec elf_linux_sysvec = {
750 	.sv_size	= LINUX_SYS_MAXSYSCALL,
751 	.sv_table	= linux_sysent,
752 	.sv_mask	= 0,
753 	.sv_sigsize	= 0,
754 	.sv_sigtbl	= NULL,
755 	.sv_errsize	= ELAST + 1,
756 	.sv_errtbl	= bsd_to_linux_errno,
757 	.sv_transtrap	= translate_traps,
758 	.sv_fixup	= elf_linux_fixup,
759 	.sv_sendsig	= linux_rt_sendsig,
760 	.sv_sigcode	= &_binary_linux_locore_o_start,
761 	.sv_szsigcode	= &linux_szsigcode,
762 	.sv_prepsyscall	= NULL,
763 	.sv_name	= "Linux ELF64",
764 	.sv_coredump	= elf64_coredump,
765 	.sv_imgact_try	= exec_linux_imgact_try,
766 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
767 	.sv_pagesize	= PAGE_SIZE,
768 	.sv_minuser	= VM_MIN_ADDRESS,
769 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
770 	.sv_usrstack	= USRSTACK,
771 	.sv_psstrings	= PS_STRINGS,
772 	.sv_stackprot	= VM_PROT_ALL,
773 	.sv_copyout_strings = linux_copyout_strings,
774 	.sv_setregs	= linux_exec_setregs,
775 	.sv_fixlimit	= NULL,
776 	.sv_maxssiz	= NULL,
777 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
778 	.sv_set_syscall_retval = linux_set_syscall_retval,
779 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
780 	.sv_syscallnames = NULL,
781 	.sv_shared_page_base = SHAREDPAGE,
782 	.sv_shared_page_len = PAGE_SIZE,
783 	.sv_schedtail	= linux_schedtail,
784 	.sv_thread_detach = linux_thread_detach
785 };
786 
787 static void
788 linux_vdso_install(void *param)
789 {
790 
791 	linux_szsigcode = (&_binary_linux_locore_o_end -
792 	    &_binary_linux_locore_o_start);
793 
794 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
795 		panic("Linux invalid vdso size\n");
796 
797 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
798 
799 	linux_shared_page_obj = __elfN(linux_shared_page_init)
800 	    (&linux_shared_page_mapping);
801 
802 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec, SHAREDPAGE);
803 
804 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
805 	    linux_szsigcode);
806 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
807 
808 	linux_kplatform = linux_shared_page_mapping +
809 	    (linux_platform - (caddr_t)SHAREDPAGE);
810 }
811 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
812     (sysinit_cfunc_t)linux_vdso_install, NULL);
813 
814 static void
815 linux_vdso_deinstall(void *param)
816 {
817 
818 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
819 };
820 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
821     (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
822 
823 static char GNULINUX_ABI_VENDOR[] = "GNU";
824 static int GNULINUX_ABI_DESC = 0;
825 
826 static boolean_t
827 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
828 {
829 	const Elf32_Word *desc;
830 	uintptr_t p;
831 
832 	p = (uintptr_t)(note + 1);
833 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
834 
835 	desc = (const Elf32_Word *)p;
836 	if (desc[0] != GNULINUX_ABI_DESC)
837 		return (FALSE);
838 
839 	/*
840 	 * For linux we encode osrel as follows (see linux_mib.c):
841 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
842 	 */
843 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
844 
845 	return (TRUE);
846 }
847 
848 static Elf_Brandnote linux64_brandnote = {
849 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
850 	.hdr.n_descsz	= 16,
851 	.hdr.n_type	= 1,
852 	.vendor		= GNULINUX_ABI_VENDOR,
853 	.flags		= BN_TRANSLATE_OSREL,
854 	.trans_osrel	= linux_trans_osrel
855 };
856 
857 static Elf64_Brandinfo linux_glibc2brand = {
858 	.brand		= ELFOSABI_LINUX,
859 	.machine	= EM_X86_64,
860 	.compat_3_brand	= "Linux",
861 	.emul_path	= "/compat/linux",
862 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
863 	.sysvec		= &elf_linux_sysvec,
864 	.interp_newpath	= NULL,
865 	.brand_note	= &linux64_brandnote,
866 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
867 };
868 
869 static Elf64_Brandinfo linux_glibc2brandshort = {
870 	.brand		= ELFOSABI_LINUX,
871 	.machine	= EM_X86_64,
872 	.compat_3_brand	= "Linux",
873 	.emul_path	= "/compat/linux",
874 	.interp_path	= "/lib64/ld-linux.so.2",
875 	.sysvec		= &elf_linux_sysvec,
876 	.interp_newpath	= NULL,
877 	.brand_note	= &linux64_brandnote,
878 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
879 };
880 
881 Elf64_Brandinfo *linux_brandlist[] = {
882 	&linux_glibc2brand,
883 	&linux_glibc2brandshort,
884 	NULL
885 };
886 
887 static int
888 linux64_elf_modevent(module_t mod, int type, void *data)
889 {
890 	Elf64_Brandinfo **brandinfo;
891 	int error;
892 	struct linux_ioctl_handler **lihp;
893 
894 	error = 0;
895 
896 	switch(type) {
897 	case MOD_LOAD:
898 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
899 		     ++brandinfo)
900 			if (elf64_insert_brand_entry(*brandinfo) < 0)
901 				error = EINVAL;
902 		if (error == 0) {
903 			SET_FOREACH(lihp, linux_ioctl_handler_set)
904 				linux_ioctl_register_handler(*lihp);
905 			LIST_INIT(&futex_list);
906 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
907 			stclohz = (stathz ? stathz : hz);
908 			if (bootverbose)
909 				printf("Linux x86-64 ELF exec handler installed\n");
910 		} else
911 			printf("cannot insert Linux x86-64 ELF brand handler\n");
912 		break;
913 	case MOD_UNLOAD:
914 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
915 		     ++brandinfo)
916 			if (elf64_brand_inuse(*brandinfo))
917 				error = EBUSY;
918 		if (error == 0) {
919 			for (brandinfo = &linux_brandlist[0];
920 			     *brandinfo != NULL; ++brandinfo)
921 				if (elf64_remove_brand_entry(*brandinfo) < 0)
922 					error = EINVAL;
923 		}
924 		if (error == 0) {
925 			SET_FOREACH(lihp, linux_ioctl_handler_set)
926 				linux_ioctl_unregister_handler(*lihp);
927 			mtx_destroy(&futex_mtx);
928 			if (bootverbose)
929 				printf("Linux ELF exec handler removed\n");
930 		} else
931 			printf("Could not deinstall ELF interpreter entry\n");
932 		break;
933 	default:
934 		return (EOPNOTSUPP);
935 	}
936 	return (error);
937 }
938 
939 static moduledata_t linux64_elf_mod = {
940 	"linux64elf",
941 	linux64_elf_modevent,
942 	0
943 };
944 
945 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
946 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
947