xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 193d9e768ba63fcfb187cfd17f461f7d41345048)
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)<nitems(_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 	int issetugid;
275 
276 	p = imgp->proc;
277 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
278 
279 	KASSERT(curthread->td_proc == imgp->proc,
280 	    ("unsafe elf_linux_fixup(), should be curproc"));
281 	base = (Elf64_Addr *)*stack_base;
282 	args = (Elf64_Auxargs *)imgp->auxargs;
283 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
284 
285 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
286 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
287 	    imgp->proc->p_sysent->sv_shared_page_base);
288 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
289 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
290 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
291 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
292 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
293 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
294 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
295 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
296 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
297 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
298 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
299 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
300 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
301 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
302 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
303 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
304 	if (imgp->execpathp != 0)
305 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
306 	if (args->execfd != -1)
307 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
308 	AUXARGS_ENTRY(pos, AT_NULL, 0);
309 	free(imgp->auxargs, M_TEMP);
310 	imgp->auxargs = NULL;
311 
312 	base--;
313 	suword(base, (uint64_t)imgp->args->argc);
314 
315 	*stack_base = (register_t *)base;
316 	return (0);
317 }
318 
319 /*
320  * Copy strings out to the new process address space, constructing new arg
321  * and env vector tables. Return a pointer to the base so that it can be used
322  * as the initial stack pointer.
323  */
324 static register_t *
325 linux_copyout_strings(struct image_params *imgp)
326 {
327 	int argc, envc;
328 	char **vectp;
329 	char *stringp, *destp;
330 	register_t *stack_base;
331 	struct ps_strings *arginfo;
332 	char canary[LINUX_AT_RANDOM_LEN];
333 	size_t execpath_len;
334 	struct proc *p;
335 
336 	/*
337 	 * Calculate string base and vector table pointers.
338 	 */
339 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
340 		execpath_len = strlen(imgp->execpath) + 1;
341 	else
342 		execpath_len = 0;
343 
344 	p = imgp->proc;
345 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
346 	destp =	(caddr_t)arginfo - SPARE_USRSPACE -
347 	    roundup(sizeof(canary), sizeof(char *)) -
348 	    roundup(execpath_len, sizeof(char *)) -
349 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
350 
351 	if (execpath_len != 0) {
352 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
353 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
354 	}
355 
356 	/*
357 	 * Prepare the canary for SSP.
358 	 */
359 	arc4rand(canary, sizeof(canary), 0);
360 	imgp->canary = (uintptr_t)arginfo -
361 	    roundup(execpath_len, sizeof(char *)) -
362 	    roundup(sizeof(canary), sizeof(char *));
363 	copyout(canary, (void *)imgp->canary, sizeof(canary));
364 
365 	/*
366 	 * If we have a valid auxargs ptr, prepare some room
367 	 * on the stack.
368 	 */
369 	if (imgp->auxargs) {
370 		/*
371 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
372 		 * lower compatibility.
373 		 */
374 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
375 		    (LINUX_AT_COUNT * 2);
376 
377 		/*
378 		 * The '+ 2' is for the null pointers at the end of each of
379 		 * the arg and env vector sets,and imgp->auxarg_size is room
380 		 * for argument of Runtime loader.
381 		 */
382 		vectp = (char **)(destp - (imgp->args->argc +
383 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
384 
385 	} else {
386 		/*
387 		 * The '+ 2' is for the null pointers at the end of each of
388 		 * the arg and env vector sets
389 		 */
390 		vectp = (char **)(destp - (imgp->args->argc +
391 		    imgp->args->envc + 2) * sizeof(char *));
392 	}
393 
394 	/*
395 	 * vectp also becomes our initial stack base
396 	 */
397 	stack_base = (register_t *)vectp;
398 
399 	stringp = imgp->args->begin_argv;
400 	argc = imgp->args->argc;
401 	envc = imgp->args->envc;
402 
403 	/*
404 	 * Copy out strings - arguments and environment.
405 	 */
406 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
407 
408 	/*
409 	 * Fill in "ps_strings" struct for ps, w, etc.
410 	 */
411 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
412 	suword(&arginfo->ps_nargvstr, argc);
413 
414 	/*
415 	 * Fill in argument portion of vector table.
416 	 */
417 	for (; argc > 0; --argc) {
418 		suword(vectp++, (long)(intptr_t)destp);
419 		while (*stringp++ != 0)
420 			destp++;
421 		destp++;
422 	}
423 
424 	/* a null vector table pointer separates the argp's from the envp's */
425 	suword(vectp++, 0);
426 
427 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
428 	suword(&arginfo->ps_nenvstr, envc);
429 
430 	/*
431 	 * Fill in environment portion of vector table.
432 	 */
433 	for (; envc > 0; --envc) {
434 		suword(vectp++, (long)(intptr_t)destp);
435 		while (*stringp++ != 0)
436 			destp++;
437 		destp++;
438 	}
439 
440 	/* end of vector table is a null pointer */
441 	suword(vectp, 0);
442 	return (stack_base);
443 }
444 
445 /*
446  * Reset registers to default values on exec.
447  */
448 static void
449 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
450 {
451 	struct trapframe *regs = td->td_frame;
452 	struct pcb *pcb = td->td_pcb;
453 
454 	mtx_lock(&dt_lock);
455 	if (td->td_proc->p_md.md_ldt != NULL)
456 		user_ldt_free(td);
457 	else
458 		mtx_unlock(&dt_lock);
459 
460 	pcb->pcb_fsbase = 0;
461 	pcb->pcb_gsbase = 0;
462 	clear_pcb_flags(pcb, PCB_32BIT);
463 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
464 	set_pcb_flags(pcb, PCB_FULL_IRET);
465 
466 	bzero((char *)regs, sizeof(struct trapframe));
467 	regs->tf_rip = imgp->entry_addr;
468 	regs->tf_rsp = stack;
469 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
470 	regs->tf_ss = _udatasel;
471 	regs->tf_cs = _ucodesel;
472 	regs->tf_ds = _udatasel;
473 	regs->tf_es = _udatasel;
474 	regs->tf_fs = _ufssel;
475 	regs->tf_gs = _ugssel;
476 	regs->tf_flags = TF_HASSEGS;
477 
478 	/*
479 	 * Reset the hardware debug registers if they were in use.
480 	 * They won't have any meaning for the newly exec'd process.
481 	 */
482 	if (pcb->pcb_flags & PCB_DBREGS) {
483 		pcb->pcb_dr0 = 0;
484 		pcb->pcb_dr1 = 0;
485 		pcb->pcb_dr2 = 0;
486 		pcb->pcb_dr3 = 0;
487 		pcb->pcb_dr6 = 0;
488 		pcb->pcb_dr7 = 0;
489 		if (pcb == curpcb) {
490 			/*
491 			 * Clear the debug registers on the running
492 			 * CPU, otherwise they will end up affecting
493 			 * the next process we switch to.
494 			 */
495 			reset_dbregs();
496 		}
497 		clear_pcb_flags(pcb, PCB_DBREGS);
498 	}
499 
500 	/*
501 	 * Drop the FP state if we hold it, so that the process gets a
502 	 * clean FP state if it uses the FPU again.
503 	 */
504 	fpstate_drop(td);
505 }
506 
507 /*
508  * Copied from amd64/amd64/machdep.c
509  *
510  * XXX fpu state need? don't think so
511  */
512 int
513 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
514 {
515 	struct proc *p;
516 	struct l_ucontext uc;
517 	struct l_sigcontext *context;
518 	struct trapframe *regs;
519 	unsigned long rflags;
520 	int error;
521 	ksiginfo_t ksi;
522 
523 	regs = td->td_frame;
524 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
525 	if (error != 0)
526 		return (error);
527 
528 	p = td->td_proc;
529 	context = &uc.uc_mcontext;
530 	rflags = context->sc_rflags;
531 
532 	/*
533 	 * Don't allow users to change privileged or reserved flags.
534 	 */
535 	/*
536 	 * XXX do allow users to change the privileged flag PSL_RF.
537 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
538 	 * should sometimes set it there too.  tf_rflags is kept in
539 	 * the signal context during signal handling and there is no
540 	 * other place to remember it, so the PSL_RF bit may be
541 	 * corrupted by the signal handler without us knowing.
542 	 * Corruption of the PSL_RF bit at worst causes one more or
543 	 * one less debugger trap, so allowing it is fairly harmless.
544 	 */
545 
546 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
547 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
548 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
549 		return (EINVAL);
550 	}
551 
552 	/*
553 	 * Don't allow users to load a valid privileged %cs.  Let the
554 	 * hardware check for invalid selectors, excess privilege in
555 	 * other selectors, invalid %eip's and invalid %esp's.
556 	 */
557 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
558 	if (!CS_SECURE(context->sc_cs)) {
559 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
560 		ksiginfo_init_trap(&ksi);
561 		ksi.ksi_signo = SIGBUS;
562 		ksi.ksi_code = BUS_OBJERR;
563 		ksi.ksi_trapno = T_PROTFLT;
564 		ksi.ksi_addr = (void *)regs->tf_rip;
565 		trapsignal(td, &ksi);
566 		return (EINVAL);
567 	}
568 
569 	PROC_LOCK(p);
570 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
571 	SIG_CANTMASK(td->td_sigmask);
572 	signotify(td);
573 	PROC_UNLOCK(p);
574 
575 	regs->tf_rdi    = context->sc_rdi;
576 	regs->tf_rsi    = context->sc_rsi;
577 	regs->tf_rdx    = context->sc_rdx;
578 	regs->tf_rbp    = context->sc_rbp;
579 	regs->tf_rbx    = context->sc_rbx;
580 	regs->tf_rcx    = context->sc_rcx;
581 	regs->tf_rax    = context->sc_rax;
582 	regs->tf_rip    = context->sc_rip;
583 	regs->tf_rsp    = context->sc_rsp;
584 	regs->tf_r8     = context->sc_r8;
585 	regs->tf_r9     = context->sc_r9;
586 	regs->tf_r10    = context->sc_r10;
587 	regs->tf_r11    = context->sc_r11;
588 	regs->tf_r12    = context->sc_r12;
589 	regs->tf_r13    = context->sc_r13;
590 	regs->tf_r14    = context->sc_r14;
591 	regs->tf_r15    = context->sc_r15;
592 	regs->tf_cs     = context->sc_cs;
593 	regs->tf_err    = context->sc_err;
594 	regs->tf_rflags = rflags;
595 
596 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
597 	return (EJUSTRETURN);
598 }
599 
600 /*
601  * copied from amd64/amd64/machdep.c
602  *
603  * Send an interrupt to process.
604  */
605 static void
606 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
607 {
608 	struct l_rt_sigframe sf, *sfp;
609 	struct proc *p;
610 	struct thread *td;
611 	struct sigacts *psp;
612 	caddr_t sp;
613 	struct trapframe *regs;
614 	int sig, code;
615 	int oonstack;
616 
617 	td = curthread;
618 	p = td->td_proc;
619 	PROC_LOCK_ASSERT(p, MA_OWNED);
620 	sig = ksi->ksi_signo;
621 	psp = p->p_sigacts;
622 	code = ksi->ksi_code;
623 	mtx_assert(&psp->ps_mtx, MA_OWNED);
624 	regs = td->td_frame;
625 	oonstack = sigonstack(regs->tf_rsp);
626 
627 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
628 	    catcher, sig, mask, code);
629 
630 	/* Allocate space for the signal handler context. */
631 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
632 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
633 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
634 		    sizeof(struct l_rt_sigframe);
635 	} else
636 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
637 	/* Align to 16 bytes. */
638 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
639 	mtx_unlock(&psp->ps_mtx);
640 
641 	/* Translate the signal. */
642 	sig = bsd_to_linux_signal(sig);
643 
644 	/* Save user context. */
645 	bzero(&sf, sizeof(sf));
646 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
647 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
648 
649 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
650 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
651 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
652 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
653 	PROC_UNLOCK(p);
654 
655 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
656 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
657 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
658 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
659 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
660 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
661 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
662 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
663 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
664 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
665 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
666 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
667 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
668 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
669 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
670 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
671 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
672 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
673 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
674 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
675 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
676 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
677 
678 	/* Build the argument list for the signal handler. */
679 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
680 	regs->tf_rax = 0;
681 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
682 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
683 
684 	sf.sf_handler = catcher;
685 	/* Fill in POSIX parts */
686 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
687 
688 	/*
689 	 * Copy the sigframe out to the user's stack.
690 	 */
691 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
692 #ifdef DEBUG
693 		printf("process %ld has trashed its stack\n", (long)p->p_pid);
694 #endif
695 		PROC_LOCK(p);
696 		sigexit(td, SIGILL);
697 	}
698 
699 	regs->tf_rsp = (long)sfp;
700 	regs->tf_rip = linux_rt_sigcode;
701 	regs->tf_rflags &= ~(PSL_T | PSL_D);
702 	regs->tf_cs = _ucodesel;
703 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
704 	PROC_LOCK(p);
705 	mtx_lock(&psp->ps_mtx);
706 }
707 
708 /*
709  * If a linux binary is exec'ing something, try this image activator
710  * first.  We override standard shell script execution in order to
711  * be able to modify the interpreter path.  We only do this if a linux
712  * binary is doing the exec, so we do not create an EXEC module for it.
713  */
714 static int exec_linux_imgact_try(struct image_params *iparams);
715 
716 static int
717 exec_linux_imgact_try(struct image_params *imgp)
718 {
719 	const char *head = (const char *)imgp->image_header;
720 	char *rpath;
721 	int error = -1;
722 
723 	/*
724 	 * The interpreter for shell scripts run from a linux binary needs
725 	 * to be located in /compat/linux if possible in order to recursively
726 	 * maintain linux path emulation.
727 	 */
728 	if (((const short *)head)[0] == SHELLMAGIC) {
729 		/*
730 		 * Run our normal shell image activator.  If it succeeds
731 		 * attempt to use the alternate path for the interpreter.
732 		 * If an alternate path is found, use our stringspace
733 		 * to store it.
734 		 */
735 		if ((error = exec_shell_imgact(imgp)) == 0) {
736 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
737 			    imgp->interpreter_name, UIO_SYSSPACE,
738 			    &rpath, 0, AT_FDCWD);
739 			if (rpath != NULL)
740 				imgp->args->fname_buf =
741 				    imgp->interpreter_name = rpath;
742 		}
743 	}
744 	return (error);
745 }
746 
747 #define	LINUX_VSYSCALL_START		(-10UL << 20)
748 #define	LINUX_VSYSCALL_SZ		1024
749 
750 const unsigned long linux_vsyscall_vector[] = {
751 	LINUX_SYS_gettimeofday,
752 	LINUX_SYS_linux_time,
753 				/* getcpu not implemented */
754 };
755 
756 static int
757 linux_vsyscall(struct thread *td)
758 {
759 	struct trapframe *frame;
760 	uint64_t retqaddr;
761 	int code, traced;
762 	int error;
763 
764 	frame = td->td_frame;
765 
766 	/* Check %rip for vsyscall area */
767 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
768 		return (EINVAL);
769 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
770 		return (EINVAL);
771 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
772 	if (code >= nitems(linux_vsyscall_vector))
773 		return (EINVAL);
774 
775 	/*
776 	 * vsyscall called as callq *(%rax), so we must
777 	 * use return address from %rsp and also fixup %rsp
778 	 */
779 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
780 	if (error)
781 		return (error);
782 
783 	frame->tf_rip = retqaddr;
784 	frame->tf_rax = linux_vsyscall_vector[code];
785 	frame->tf_rsp += 8;
786 
787 	traced = (frame->tf_flags & PSL_T);
788 
789 	amd64_syscall(td, traced);
790 
791 	return (0);
792 }
793 
794 struct sysentvec elf_linux_sysvec = {
795 	.sv_size	= LINUX_SYS_MAXSYSCALL,
796 	.sv_table	= linux_sysent,
797 	.sv_mask	= 0,
798 	.sv_errsize	= ELAST + 1,
799 	.sv_errtbl	= bsd_to_linux_errno,
800 	.sv_transtrap	= translate_traps,
801 	.sv_fixup	= elf_linux_fixup,
802 	.sv_sendsig	= linux_rt_sendsig,
803 	.sv_sigcode	= &_binary_linux_locore_o_start,
804 	.sv_szsigcode	= &linux_szsigcode,
805 	.sv_name	= "Linux ELF64",
806 	.sv_coredump	= elf64_coredump,
807 	.sv_imgact_try	= exec_linux_imgact_try,
808 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
809 	.sv_pagesize	= PAGE_SIZE,
810 	.sv_minuser	= VM_MIN_ADDRESS,
811 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
812 	.sv_usrstack	= USRSTACK,
813 	.sv_psstrings	= PS_STRINGS,
814 	.sv_stackprot	= VM_PROT_ALL,
815 	.sv_copyout_strings = linux_copyout_strings,
816 	.sv_setregs	= linux_exec_setregs,
817 	.sv_fixlimit	= NULL,
818 	.sv_maxssiz	= NULL,
819 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
820 	.sv_set_syscall_retval = linux_set_syscall_retval,
821 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
822 	.sv_syscallnames = NULL,
823 	.sv_shared_page_base = SHAREDPAGE,
824 	.sv_shared_page_len = PAGE_SIZE,
825 	.sv_schedtail	= linux_schedtail,
826 	.sv_thread_detach = linux_thread_detach,
827 	.sv_trap	= linux_vsyscall,
828 };
829 
830 static void
831 linux_vdso_install(void *param)
832 {
833 
834 	linux_szsigcode = (&_binary_linux_locore_o_end -
835 	    &_binary_linux_locore_o_start);
836 
837 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
838 		panic("Linux invalid vdso size\n");
839 
840 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
841 
842 	linux_shared_page_obj = __elfN(linux_shared_page_init)
843 	    (&linux_shared_page_mapping);
844 
845 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec, SHAREDPAGE);
846 
847 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
848 	    linux_szsigcode);
849 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
850 
851 	linux_kplatform = linux_shared_page_mapping +
852 	    (linux_platform - (caddr_t)SHAREDPAGE);
853 }
854 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
855     (sysinit_cfunc_t)linux_vdso_install, NULL);
856 
857 static void
858 linux_vdso_deinstall(void *param)
859 {
860 
861 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
862 };
863 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
864     (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
865 
866 static char GNULINUX_ABI_VENDOR[] = "GNU";
867 static int GNULINUX_ABI_DESC = 0;
868 
869 static boolean_t
870 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
871 {
872 	const Elf32_Word *desc;
873 	uintptr_t p;
874 
875 	p = (uintptr_t)(note + 1);
876 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
877 
878 	desc = (const Elf32_Word *)p;
879 	if (desc[0] != GNULINUX_ABI_DESC)
880 		return (FALSE);
881 
882 	/*
883 	 * For linux we encode osrel as follows (see linux_mib.c):
884 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
885 	 */
886 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
887 
888 	return (TRUE);
889 }
890 
891 static Elf_Brandnote linux64_brandnote = {
892 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
893 	.hdr.n_descsz	= 16,
894 	.hdr.n_type	= 1,
895 	.vendor		= GNULINUX_ABI_VENDOR,
896 	.flags		= BN_TRANSLATE_OSREL,
897 	.trans_osrel	= linux_trans_osrel
898 };
899 
900 static Elf64_Brandinfo linux_glibc2brand = {
901 	.brand		= ELFOSABI_LINUX,
902 	.machine	= EM_X86_64,
903 	.compat_3_brand	= "Linux",
904 	.emul_path	= "/compat/linux",
905 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
906 	.sysvec		= &elf_linux_sysvec,
907 	.interp_newpath	= NULL,
908 	.brand_note	= &linux64_brandnote,
909 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
910 };
911 
912 static Elf64_Brandinfo linux_glibc2brandshort = {
913 	.brand		= ELFOSABI_LINUX,
914 	.machine	= EM_X86_64,
915 	.compat_3_brand	= "Linux",
916 	.emul_path	= "/compat/linux",
917 	.interp_path	= "/lib64/ld-linux.so.2",
918 	.sysvec		= &elf_linux_sysvec,
919 	.interp_newpath	= NULL,
920 	.brand_note	= &linux64_brandnote,
921 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
922 };
923 
924 Elf64_Brandinfo *linux_brandlist[] = {
925 	&linux_glibc2brand,
926 	&linux_glibc2brandshort,
927 	NULL
928 };
929 
930 static int
931 linux64_elf_modevent(module_t mod, int type, void *data)
932 {
933 	Elf64_Brandinfo **brandinfo;
934 	int error;
935 	struct linux_ioctl_handler **lihp;
936 
937 	error = 0;
938 
939 	switch(type) {
940 	case MOD_LOAD:
941 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
942 		     ++brandinfo)
943 			if (elf64_insert_brand_entry(*brandinfo) < 0)
944 				error = EINVAL;
945 		if (error == 0) {
946 			SET_FOREACH(lihp, linux_ioctl_handler_set)
947 				linux_ioctl_register_handler(*lihp);
948 			LIST_INIT(&futex_list);
949 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
950 			stclohz = (stathz ? stathz : hz);
951 			if (bootverbose)
952 				printf("Linux x86-64 ELF exec handler installed\n");
953 		} else
954 			printf("cannot insert Linux x86-64 ELF brand handler\n");
955 		break;
956 	case MOD_UNLOAD:
957 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
958 		     ++brandinfo)
959 			if (elf64_brand_inuse(*brandinfo))
960 				error = EBUSY;
961 		if (error == 0) {
962 			for (brandinfo = &linux_brandlist[0];
963 			     *brandinfo != NULL; ++brandinfo)
964 				if (elf64_remove_brand_entry(*brandinfo) < 0)
965 					error = EINVAL;
966 		}
967 		if (error == 0) {
968 			SET_FOREACH(lihp, linux_ioctl_handler_set)
969 				linux_ioctl_unregister_handler(*lihp);
970 			mtx_destroy(&futex_mtx);
971 			if (bootverbose)
972 				printf("Linux ELF exec handler removed\n");
973 		} else
974 			printf("Could not deinstall ELF interpreter entry\n");
975 		break;
976 	default:
977 		return (EOPNOTSUPP);
978 	}
979 	return (error);
980 }
981 
982 static moduledata_t linux64_elf_mod = {
983 	"linux64elf",
984 	linux64_elf_modevent,
985 	0
986 };
987 
988 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
989 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
990 FEATURE(linux64, "Linux 64bit support");
991