xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 49b49cda41feabe3439f7318e8bf40e3896c7bf4)
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 static int	linux_vsyscall(struct thread *td);
133 
134 /*
135  * Linux syscalls return negative errno's, we do positive and map them
136  * Reference:
137  *   FreeBSD: src/sys/sys/errno.h
138  *   Linux:   linux-2.6.17.8/include/asm-generic/errno-base.h
139  *            linux-2.6.17.8/include/asm-generic/errno.h
140  */
141 static int bsd_to_linux_errno[ELAST + 1] = {
142 	-0,  -1,  -2,  -3,  -4,  -5,  -6,  -7,  -8,  -9,
143 	-10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
144 	-20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
145 	-30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
146 	-90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
147 	-100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
148 	-110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
149 	-116, -66,  -6,  -6,  -6,  -6,  -6, -37, -38,  -9,
150 	  -6,  -6, -43, -42, -75,-125, -84, -95, -16, -74,
151 	 -72, -67, -71
152 };
153 
154 #define LINUX_T_UNKNOWN  255
155 static int _bsd_to_linux_trapcode[] = {
156 	LINUX_T_UNKNOWN,	/* 0 */
157 	6,			/* 1  T_PRIVINFLT */
158 	LINUX_T_UNKNOWN,	/* 2 */
159 	3,			/* 3  T_BPTFLT */
160 	LINUX_T_UNKNOWN,	/* 4 */
161 	LINUX_T_UNKNOWN,	/* 5 */
162 	16,			/* 6  T_ARITHTRAP */
163 	254,			/* 7  T_ASTFLT */
164 	LINUX_T_UNKNOWN,	/* 8 */
165 	13,			/* 9  T_PROTFLT */
166 	1,			/* 10 T_TRCTRAP */
167 	LINUX_T_UNKNOWN,	/* 11 */
168 	14,			/* 12 T_PAGEFLT */
169 	LINUX_T_UNKNOWN,	/* 13 */
170 	17,			/* 14 T_ALIGNFLT */
171 	LINUX_T_UNKNOWN,	/* 15 */
172 	LINUX_T_UNKNOWN,	/* 16 */
173 	LINUX_T_UNKNOWN,	/* 17 */
174 	0,			/* 18 T_DIVIDE */
175 	2,			/* 19 T_NMI */
176 	4,			/* 20 T_OFLOW */
177 	5,			/* 21 T_BOUND */
178 	7,			/* 22 T_DNA */
179 	8,			/* 23 T_DOUBLEFLT */
180 	9,			/* 24 T_FPOPFLT */
181 	10,			/* 25 T_TSSFLT */
182 	11,			/* 26 T_SEGNPFLT */
183 	12,			/* 27 T_STKFLT */
184 	18,			/* 28 T_MCHK */
185 	19,			/* 29 T_XMMFLT */
186 	15			/* 30 T_RESERVED */
187 };
188 #define bsd_to_linux_trapcode(code) \
189     ((code)<sizeof(_bsd_to_linux_trapcode)/sizeof(*_bsd_to_linux_trapcode)? \
190      _bsd_to_linux_trapcode[(code)]: \
191      LINUX_T_UNKNOWN)
192 
193 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
194 LINUX_VDSO_SYM_CHAR(linux_platform);
195 
196 /*
197  * If FreeBSD & Linux have a difference of opinion about what a trap
198  * means, deal with it here.
199  *
200  * MPSAFE
201  */
202 static int
203 translate_traps(int signal, int trap_code)
204 {
205 
206 	if (signal != SIGBUS)
207 		return signal;
208 	switch (trap_code) {
209 	case T_PROTFLT:
210 	case T_TSSFLT:
211 	case T_DOUBLEFLT:
212 	case T_PAGEFLT:
213 		return SIGSEGV;
214 	default:
215 		return signal;
216 	}
217 }
218 
219 static int
220 linux_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
221 {
222 	struct proc *p;
223 	struct trapframe *frame;
224 
225 	p = td->td_proc;
226 	frame = td->td_frame;
227 
228 	sa->args[0] = frame->tf_rdi;
229 	sa->args[1] = frame->tf_rsi;
230 	sa->args[2] = frame->tf_rdx;
231 	sa->args[3] = frame->tf_rcx;
232 	sa->args[4] = frame->tf_r8;
233 	sa->args[5] = frame->tf_r9;
234 	sa->code = frame->tf_rax;
235 
236 	if (sa->code >= p->p_sysent->sv_size)
237 		/* nosys */
238 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
239 	else
240 		sa->callp = &p->p_sysent->sv_table[sa->code];
241 	sa->narg = sa->callp->sy_narg;
242 
243 	td->td_retval[0] = 0;
244 	return (0);
245 }
246 
247 static void
248 linux_set_syscall_retval(struct thread *td, int error)
249 {
250 	struct trapframe *frame = td->td_frame;
251 
252 	/*
253 	 * On Linux only %rcx and %r11 values are not preserved across
254 	 * the syscall.
255 	 * So, do not clobber %rdx and %r10
256 	 */
257 	td->td_retval[1] = frame->tf_rdx;
258 	frame->tf_r10 = frame->tf_rcx;
259 
260 	cpu_set_syscall_retval(td, error);
261 
262 	 /* Restore all registers. */
263 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
264 }
265 
266 static int
267 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
268 {
269 	Elf_Auxargs *args;
270 	Elf_Addr *base;
271 	Elf_Addr *pos;
272 	struct ps_strings *arginfo;
273 	struct proc *p;
274 
275 	p = imgp->proc;
276 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
277 
278 	KASSERT(curthread->td_proc == imgp->proc,
279 	    ("unsafe elf_linux_fixup(), should be curproc"));
280 	base = (Elf64_Addr *)*stack_base;
281 	args = (Elf64_Auxargs *)imgp->auxargs;
282 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
283 
284 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
285 	    imgp->proc->p_sysent->sv_shared_page_base);
286 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
287 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
288 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
289 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
290 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
291 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
292 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
293 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
294 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
295 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
296 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
297 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
298 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
299 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, 0);
300 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
301 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
302 	if (imgp->execpathp != 0)
303 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
304 	if (args->execfd != -1)
305 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
306 	AUXARGS_ENTRY(pos, AT_NULL, 0);
307 	free(imgp->auxargs, M_TEMP);
308 	imgp->auxargs = NULL;
309 
310 	base--;
311 	suword(base, (uint64_t)imgp->args->argc);
312 
313 	*stack_base = (register_t *)base;
314 	return (0);
315 }
316 
317 /*
318  * Copy strings out to the new process address space, constructing new arg
319  * and env vector tables. Return a pointer to the base so that it can be used
320  * as the initial stack pointer.
321  */
322 static register_t *
323 linux_copyout_strings(struct image_params *imgp)
324 {
325 	int argc, envc;
326 	char **vectp;
327 	char *stringp, *destp;
328 	register_t *stack_base;
329 	struct ps_strings *arginfo;
330 	char canary[LINUX_AT_RANDOM_LEN];
331 	size_t execpath_len;
332 	struct proc *p;
333 
334 	/*
335 	 * Calculate string base and vector table pointers.
336 	 */
337 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
338 		execpath_len = strlen(imgp->execpath) + 1;
339 	else
340 		execpath_len = 0;
341 
342 	p = imgp->proc;
343 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
344 	destp =	(caddr_t)arginfo - SPARE_USRSPACE -
345 	    roundup(sizeof(canary), sizeof(char *)) -
346 	    roundup(execpath_len, sizeof(char *)) -
347 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
348 
349 	if (execpath_len != 0) {
350 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
351 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
352 	}
353 
354 	/*
355 	 * Prepare the canary for SSP.
356 	 */
357 	arc4rand(canary, sizeof(canary), 0);
358 	imgp->canary = (uintptr_t)arginfo -
359 	    roundup(execpath_len, sizeof(char *)) -
360 	    roundup(sizeof(canary), sizeof(char *));
361 	copyout(canary, (void *)imgp->canary, sizeof(canary));
362 
363 	/*
364 	 * If we have a valid auxargs ptr, prepare some room
365 	 * on the stack.
366 	 */
367 	if (imgp->auxargs) {
368 		/*
369 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
370 		 * lower compatibility.
371 		 */
372 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
373 		    (LINUX_AT_COUNT * 2);
374 
375 		/*
376 		 * The '+ 2' is for the null pointers at the end of each of
377 		 * the arg and env vector sets,and imgp->auxarg_size is room
378 		 * for argument of Runtime loader.
379 		 */
380 		vectp = (char **)(destp - (imgp->args->argc +
381 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
382 
383 	} else {
384 		/*
385 		 * The '+ 2' is for the null pointers at the end of each of
386 		 * the arg and env vector sets
387 		 */
388 		vectp = (char **)(destp - (imgp->args->argc +
389 		    imgp->args->envc + 2) * sizeof(char *));
390 	}
391 
392 	/*
393 	 * vectp also becomes our initial stack base
394 	 */
395 	stack_base = (register_t *)vectp;
396 
397 	stringp = imgp->args->begin_argv;
398 	argc = imgp->args->argc;
399 	envc = imgp->args->envc;
400 
401 	/*
402 	 * Copy out strings - arguments and environment.
403 	 */
404 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
405 
406 	/*
407 	 * Fill in "ps_strings" struct for ps, w, etc.
408 	 */
409 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
410 	suword(&arginfo->ps_nargvstr, argc);
411 
412 	/*
413 	 * Fill in argument portion of vector table.
414 	 */
415 	for (; argc > 0; --argc) {
416 		suword(vectp++, (long)(intptr_t)destp);
417 		while (*stringp++ != 0)
418 			destp++;
419 		destp++;
420 	}
421 
422 	/* a null vector table pointer separates the argp's from the envp's */
423 	suword(vectp++, 0);
424 
425 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
426 	suword(&arginfo->ps_nenvstr, envc);
427 
428 	/*
429 	 * Fill in environment portion of vector table.
430 	 */
431 	for (; envc > 0; --envc) {
432 		suword(vectp++, (long)(intptr_t)destp);
433 		while (*stringp++ != 0)
434 			destp++;
435 		destp++;
436 	}
437 
438 	/* end of vector table is a null pointer */
439 	suword(vectp, 0);
440 	return (stack_base);
441 }
442 
443 /*
444  * Reset registers to default values on exec.
445  */
446 static void
447 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
448 {
449 	struct trapframe *regs = td->td_frame;
450 	struct pcb *pcb = td->td_pcb;
451 
452 	mtx_lock(&dt_lock);
453 	if (td->td_proc->p_md.md_ldt != NULL)
454 		user_ldt_free(td);
455 	else
456 		mtx_unlock(&dt_lock);
457 
458 	pcb->pcb_fsbase = 0;
459 	pcb->pcb_gsbase = 0;
460 	clear_pcb_flags(pcb, PCB_32BIT);
461 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
462 	set_pcb_flags(pcb, PCB_FULL_IRET);
463 
464 	bzero((char *)regs, sizeof(struct trapframe));
465 	regs->tf_rip = imgp->entry_addr;
466 	regs->tf_rsp = stack;
467 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
468 	regs->tf_ss = _udatasel;
469 	regs->tf_cs = _ucodesel;
470 	regs->tf_ds = _udatasel;
471 	regs->tf_es = _udatasel;
472 	regs->tf_fs = _ufssel;
473 	regs->tf_gs = _ugssel;
474 	regs->tf_flags = TF_HASSEGS;
475 
476 	/*
477 	 * Reset the hardware debug registers if they were in use.
478 	 * They won't have any meaning for the newly exec'd process.
479 	 */
480 	if (pcb->pcb_flags & PCB_DBREGS) {
481 		pcb->pcb_dr0 = 0;
482 		pcb->pcb_dr1 = 0;
483 		pcb->pcb_dr2 = 0;
484 		pcb->pcb_dr3 = 0;
485 		pcb->pcb_dr6 = 0;
486 		pcb->pcb_dr7 = 0;
487 		if (pcb == curpcb) {
488 			/*
489 			 * Clear the debug registers on the running
490 			 * CPU, otherwise they will end up affecting
491 			 * the next process we switch to.
492 			 */
493 			reset_dbregs();
494 		}
495 		clear_pcb_flags(pcb, PCB_DBREGS);
496 	}
497 
498 	/*
499 	 * Drop the FP state if we hold it, so that the process gets a
500 	 * clean FP state if it uses the FPU again.
501 	 */
502 	fpstate_drop(td);
503 }
504 
505 /*
506  * Copied from amd64/amd64/machdep.c
507  *
508  * XXX fpu state need? don't think so
509  */
510 int
511 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
512 {
513 	struct proc *p;
514 	struct l_ucontext uc;
515 	struct l_sigcontext *context;
516 	struct trapframe *regs;
517 	unsigned long rflags;
518 	int error;
519 	ksiginfo_t ksi;
520 
521 	regs = td->td_frame;
522 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
523 	if (error != 0)
524 		return (error);
525 
526 	p = td->td_proc;
527 	context = &uc.uc_mcontext;
528 	rflags = context->sc_rflags;
529 
530 	/*
531 	 * Don't allow users to change privileged or reserved flags.
532 	 */
533 	/*
534 	 * XXX do allow users to change the privileged flag PSL_RF.
535 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
536 	 * should sometimes set it there too.  tf_rflags is kept in
537 	 * the signal context during signal handling and there is no
538 	 * other place to remember it, so the PSL_RF bit may be
539 	 * corrupted by the signal handler without us knowing.
540 	 * Corruption of the PSL_RF bit at worst causes one more or
541 	 * one less debugger trap, so allowing it is fairly harmless.
542 	 */
543 
544 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
545 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
546 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
547 		return (EINVAL);
548 	}
549 
550 	/*
551 	 * Don't allow users to load a valid privileged %cs.  Let the
552 	 * hardware check for invalid selectors, excess privilege in
553 	 * other selectors, invalid %eip's and invalid %esp's.
554 	 */
555 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
556 	if (!CS_SECURE(context->sc_cs)) {
557 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
558 		ksiginfo_init_trap(&ksi);
559 		ksi.ksi_signo = SIGBUS;
560 		ksi.ksi_code = BUS_OBJERR;
561 		ksi.ksi_trapno = T_PROTFLT;
562 		ksi.ksi_addr = (void *)regs->tf_rip;
563 		trapsignal(td, &ksi);
564 		return (EINVAL);
565 	}
566 
567 	PROC_LOCK(p);
568 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
569 	SIG_CANTMASK(td->td_sigmask);
570 	signotify(td);
571 	PROC_UNLOCK(p);
572 
573 	regs->tf_rdi    = context->sc_rdi;
574 	regs->tf_rsi    = context->sc_rsi;
575 	regs->tf_rdx    = context->sc_rdx;
576 	regs->tf_rbp    = context->sc_rbp;
577 	regs->tf_rbx    = context->sc_rbx;
578 	regs->tf_rcx    = context->sc_rcx;
579 	regs->tf_rax    = context->sc_rax;
580 	regs->tf_rip    = context->sc_rip;
581 	regs->tf_rsp    = context->sc_rsp;
582 	regs->tf_r8     = context->sc_r8;
583 	regs->tf_r9     = context->sc_r9;
584 	regs->tf_r10    = context->sc_r10;
585 	regs->tf_r11    = context->sc_r11;
586 	regs->tf_r12    = context->sc_r12;
587 	regs->tf_r13    = context->sc_r13;
588 	regs->tf_r14    = context->sc_r14;
589 	regs->tf_r15    = context->sc_r15;
590 	regs->tf_cs     = context->sc_cs;
591 	regs->tf_err    = context->sc_err;
592 	regs->tf_rflags = rflags;
593 
594 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
595 	return (EJUSTRETURN);
596 }
597 
598 /*
599  * copied from amd64/amd64/machdep.c
600  *
601  * Send an interrupt to process.
602  */
603 static void
604 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
605 {
606 	struct l_rt_sigframe sf, *sfp;
607 	struct proc *p;
608 	struct thread *td;
609 	struct sigacts *psp;
610 	caddr_t sp;
611 	struct trapframe *regs;
612 	int sig, code;
613 	int oonstack;
614 
615 	td = curthread;
616 	p = td->td_proc;
617 	PROC_LOCK_ASSERT(p, MA_OWNED);
618 	sig = ksi->ksi_signo;
619 	psp = p->p_sigacts;
620 	code = ksi->ksi_code;
621 	mtx_assert(&psp->ps_mtx, MA_OWNED);
622 	regs = td->td_frame;
623 	oonstack = sigonstack(regs->tf_rsp);
624 
625 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
626 	    catcher, sig, mask, code);
627 
628 	/* Allocate space for the signal handler context. */
629 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
630 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
631 		sp = td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
632 		    sizeof(struct l_rt_sigframe);
633 	} else
634 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
635 	/* Align to 16 bytes. */
636 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
637 	mtx_unlock(&psp->ps_mtx);
638 
639 	/* Translate the signal. */
640 	sig = bsd_to_linux_signal(sig);
641 
642 	/* Save user context. */
643 	bzero(&sf, sizeof(sf));
644 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
645 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
646 
647 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
648 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
649 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
650 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
651 	PROC_UNLOCK(p);
652 
653 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
654 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
655 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
656 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
657 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
658 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
659 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
660 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
661 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
662 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
663 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
664 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
665 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
666 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
667 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
668 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
669 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
670 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
671 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
672 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
673 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
674 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
675 
676 	/* Build the argument list for the signal handler. */
677 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
678 	regs->tf_rax = 0;
679 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
680 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
681 
682 	sf.sf_handler = catcher;
683 	/* Fill in POSIX parts */
684 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
685 
686 	/*
687 	 * Copy the sigframe out to the user's stack.
688 	 */
689 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
690 #ifdef DEBUG
691 		printf("process %ld has trashed its stack\n", (long)p->p_pid);
692 #endif
693 		PROC_LOCK(p);
694 		sigexit(td, SIGILL);
695 	}
696 
697 	regs->tf_rsp = (long)sfp;
698 	regs->tf_rip = linux_rt_sigcode;
699 	regs->tf_rflags &= ~(PSL_T | PSL_D);
700 	regs->tf_cs = _ucodesel;
701 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
702 	PROC_LOCK(p);
703 	mtx_lock(&psp->ps_mtx);
704 }
705 
706 /*
707  * If a linux binary is exec'ing something, try this image activator
708  * first.  We override standard shell script execution in order to
709  * be able to modify the interpreter path.  We only do this if a linux
710  * binary is doing the exec, so we do not create an EXEC module for it.
711  */
712 static int exec_linux_imgact_try(struct image_params *iparams);
713 
714 static int
715 exec_linux_imgact_try(struct image_params *imgp)
716 {
717 	const char *head = (const char *)imgp->image_header;
718 	char *rpath;
719 	int error = -1, len;
720 
721 	/*
722 	 * The interpreter for shell scripts run from a linux binary needs
723 	 * to be located in /compat/linux if possible in order to recursively
724 	 * maintain linux path emulation.
725 	 */
726 	if (((const short *)head)[0] == SHELLMAGIC) {
727 		/*
728 		 * Run our normal shell image activator.  If it succeeds
729 		 * attempt to use the alternate path for the interpreter.
730 		 * If an alternate path is found, use our stringspace
731 		 * to store it.
732 		 */
733 		if ((error = exec_shell_imgact(imgp)) == 0) {
734 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
735 			    imgp->interpreter_name, UIO_SYSSPACE,
736 			    &rpath, 0, AT_FDCWD);
737 			if (rpath != NULL) {
738 				len = strlen(rpath) + 1;
739 
740 				if (len <= MAXSHELLCMDLEN)
741 					memcpy(imgp->interpreter_name,
742 					    rpath, len);
743 				free(rpath, M_TEMP);
744 			}
745 		}
746 	}
747 	return(error);
748 }
749 
750 #define	LINUX_VSYSCALL_START		(-10UL << 20)
751 #define	LINUX_VSYSCALL_SZ		1024
752 
753 const unsigned long linux_vsyscall_vector[] = {
754 	LINUX_SYS_gettimeofday,
755 	LINUX_SYS_linux_time,
756 				/* getcpu not implemented */
757 };
758 
759 static int
760 linux_vsyscall(struct thread *td)
761 {
762 	struct trapframe *frame;
763 	uint64_t retqaddr;
764 	int code, traced;
765 	int error;
766 
767 	frame = td->td_frame;
768 
769 	/* Check %rip for vsyscall area */
770 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
771 		return (EINVAL);
772 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
773 		return (EINVAL);
774 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
775 	if (code >= nitems(linux_vsyscall_vector))
776 		return (EINVAL);
777 
778 	/*
779 	 * vsyscall called as callq *(%rax), so we must
780 	 * use return address from %rsp and also fixup %rsp
781 	 */
782 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
783 	if (error)
784 		return (error);
785 
786 	frame->tf_rip = retqaddr;
787 	frame->tf_rax = linux_vsyscall_vector[code];
788 	frame->tf_rsp += 8;
789 
790 	traced = (frame->tf_flags & PSL_T);
791 
792 	amd64_syscall(td, traced);
793 
794 	return (0);
795 }
796 
797 struct sysentvec elf_linux_sysvec = {
798 	.sv_size	= LINUX_SYS_MAXSYSCALL,
799 	.sv_table	= linux_sysent,
800 	.sv_mask	= 0,
801 	.sv_errsize	= ELAST + 1,
802 	.sv_errtbl	= bsd_to_linux_errno,
803 	.sv_transtrap	= translate_traps,
804 	.sv_fixup	= elf_linux_fixup,
805 	.sv_sendsig	= linux_rt_sendsig,
806 	.sv_sigcode	= &_binary_linux_locore_o_start,
807 	.sv_szsigcode	= &linux_szsigcode,
808 	.sv_name	= "Linux ELF64",
809 	.sv_coredump	= elf64_coredump,
810 	.sv_imgact_try	= exec_linux_imgact_try,
811 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
812 	.sv_pagesize	= PAGE_SIZE,
813 	.sv_minuser	= VM_MIN_ADDRESS,
814 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
815 	.sv_usrstack	= USRSTACK,
816 	.sv_psstrings	= PS_STRINGS,
817 	.sv_stackprot	= VM_PROT_ALL,
818 	.sv_copyout_strings = linux_copyout_strings,
819 	.sv_setregs	= linux_exec_setregs,
820 	.sv_fixlimit	= NULL,
821 	.sv_maxssiz	= NULL,
822 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
823 	.sv_set_syscall_retval = linux_set_syscall_retval,
824 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
825 	.sv_syscallnames = NULL,
826 	.sv_shared_page_base = SHAREDPAGE,
827 	.sv_shared_page_len = PAGE_SIZE,
828 	.sv_schedtail	= linux_schedtail,
829 	.sv_thread_detach = linux_thread_detach,
830 	.sv_trap	= linux_vsyscall,
831 };
832 
833 static void
834 linux_vdso_install(void *param)
835 {
836 
837 	linux_szsigcode = (&_binary_linux_locore_o_end -
838 	    &_binary_linux_locore_o_start);
839 
840 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
841 		panic("Linux invalid vdso size\n");
842 
843 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
844 
845 	linux_shared_page_obj = __elfN(linux_shared_page_init)
846 	    (&linux_shared_page_mapping);
847 
848 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec, SHAREDPAGE);
849 
850 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
851 	    linux_szsigcode);
852 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
853 
854 	linux_kplatform = linux_shared_page_mapping +
855 	    (linux_platform - (caddr_t)SHAREDPAGE);
856 }
857 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
858     (sysinit_cfunc_t)linux_vdso_install, NULL);
859 
860 static void
861 linux_vdso_deinstall(void *param)
862 {
863 
864 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
865 };
866 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
867     (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
868 
869 static char GNULINUX_ABI_VENDOR[] = "GNU";
870 static int GNULINUX_ABI_DESC = 0;
871 
872 static boolean_t
873 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
874 {
875 	const Elf32_Word *desc;
876 	uintptr_t p;
877 
878 	p = (uintptr_t)(note + 1);
879 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
880 
881 	desc = (const Elf32_Word *)p;
882 	if (desc[0] != GNULINUX_ABI_DESC)
883 		return (FALSE);
884 
885 	/*
886 	 * For linux we encode osrel as follows (see linux_mib.c):
887 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
888 	 */
889 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
890 
891 	return (TRUE);
892 }
893 
894 static Elf_Brandnote linux64_brandnote = {
895 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
896 	.hdr.n_descsz	= 16,
897 	.hdr.n_type	= 1,
898 	.vendor		= GNULINUX_ABI_VENDOR,
899 	.flags		= BN_TRANSLATE_OSREL,
900 	.trans_osrel	= linux_trans_osrel
901 };
902 
903 static Elf64_Brandinfo linux_glibc2brand = {
904 	.brand		= ELFOSABI_LINUX,
905 	.machine	= EM_X86_64,
906 	.compat_3_brand	= "Linux",
907 	.emul_path	= "/compat/linux",
908 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
909 	.sysvec		= &elf_linux_sysvec,
910 	.interp_newpath	= NULL,
911 	.brand_note	= &linux64_brandnote,
912 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
913 };
914 
915 static Elf64_Brandinfo linux_glibc2brandshort = {
916 	.brand		= ELFOSABI_LINUX,
917 	.machine	= EM_X86_64,
918 	.compat_3_brand	= "Linux",
919 	.emul_path	= "/compat/linux",
920 	.interp_path	= "/lib64/ld-linux.so.2",
921 	.sysvec		= &elf_linux_sysvec,
922 	.interp_newpath	= NULL,
923 	.brand_note	= &linux64_brandnote,
924 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
925 };
926 
927 Elf64_Brandinfo *linux_brandlist[] = {
928 	&linux_glibc2brand,
929 	&linux_glibc2brandshort,
930 	NULL
931 };
932 
933 static int
934 linux64_elf_modevent(module_t mod, int type, void *data)
935 {
936 	Elf64_Brandinfo **brandinfo;
937 	int error;
938 	struct linux_ioctl_handler **lihp;
939 
940 	error = 0;
941 
942 	switch(type) {
943 	case MOD_LOAD:
944 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
945 		     ++brandinfo)
946 			if (elf64_insert_brand_entry(*brandinfo) < 0)
947 				error = EINVAL;
948 		if (error == 0) {
949 			SET_FOREACH(lihp, linux_ioctl_handler_set)
950 				linux_ioctl_register_handler(*lihp);
951 			LIST_INIT(&futex_list);
952 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
953 			stclohz = (stathz ? stathz : hz);
954 			if (bootverbose)
955 				printf("Linux x86-64 ELF exec handler installed\n");
956 		} else
957 			printf("cannot insert Linux x86-64 ELF brand handler\n");
958 		break;
959 	case MOD_UNLOAD:
960 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
961 		     ++brandinfo)
962 			if (elf64_brand_inuse(*brandinfo))
963 				error = EBUSY;
964 		if (error == 0) {
965 			for (brandinfo = &linux_brandlist[0];
966 			     *brandinfo != NULL; ++brandinfo)
967 				if (elf64_remove_brand_entry(*brandinfo) < 0)
968 					error = EINVAL;
969 		}
970 		if (error == 0) {
971 			SET_FOREACH(lihp, linux_ioctl_handler_set)
972 				linux_ioctl_unregister_handler(*lihp);
973 			mtx_destroy(&futex_mtx);
974 			if (bootverbose)
975 				printf("Linux ELF exec handler removed\n");
976 		} else
977 			printf("Could not deinstall ELF interpreter entry\n");
978 		break;
979 	default:
980 		return (EOPNOTSUPP);
981 	}
982 	return (error);
983 }
984 
985 static moduledata_t linux64_elf_mod = {
986 	"linux64elf",
987 	linux64_elf_modevent,
988 	0
989 };
990 
991 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
992 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
993