xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 118063fb70c3f74f16678dc7ddbc8528c483a528)
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);
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, -61, -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)
221 {
222 	struct proc *p;
223 	struct trapframe *frame;
224 	struct syscall_args *sa;
225 
226 	p = td->td_proc;
227 	frame = td->td_frame;
228 	sa = &td->td_sa;
229 
230 	sa->args[0] = frame->tf_rdi;
231 	sa->args[1] = frame->tf_rsi;
232 	sa->args[2] = frame->tf_rdx;
233 	sa->args[3] = frame->tf_rcx;
234 	sa->args[4] = frame->tf_r8;
235 	sa->args[5] = frame->tf_r9;
236 	sa->code = frame->tf_rax;
237 
238 	if (sa->code >= p->p_sysent->sv_size)
239 		/* nosys */
240 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
241 	else
242 		sa->callp = &p->p_sysent->sv_table[sa->code];
243 	sa->narg = sa->callp->sy_narg;
244 
245 	td->td_retval[0] = 0;
246 	return (0);
247 }
248 
249 static void
250 linux_set_syscall_retval(struct thread *td, int error)
251 {
252 	struct trapframe *frame = td->td_frame;
253 
254 	/*
255 	 * On Linux only %rcx and %r11 values are not preserved across
256 	 * the syscall.
257 	 * So, do not clobber %rdx and %r10
258 	 */
259 	td->td_retval[1] = frame->tf_rdx;
260 	frame->tf_r10 = frame->tf_rcx;
261 
262 	cpu_set_syscall_retval(td, error);
263 
264 	 /* Restore all registers. */
265 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
266 }
267 
268 static int
269 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
270 {
271 	Elf_Auxargs *args;
272 	Elf_Addr *base;
273 	Elf_Addr *pos;
274 	struct ps_strings *arginfo;
275 	struct proc *p;
276 	int issetugid;
277 
278 	p = imgp->proc;
279 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
280 
281 	KASSERT(curthread->td_proc == imgp->proc,
282 	    ("unsafe elf_linux_fixup(), should be curproc"));
283 	base = (Elf64_Addr *)*stack_base;
284 	args = (Elf64_Auxargs *)imgp->auxargs;
285 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
286 
287 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
288 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
289 	    imgp->proc->p_sysent->sv_shared_page_base);
290 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
291 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
292 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
293 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
294 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
295 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
296 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
297 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
298 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
299 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
300 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
301 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
302 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
303 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
304 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
305 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
306 	if (imgp->execpathp != 0)
307 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
308 	if (args->execfd != -1)
309 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
310 	AUXARGS_ENTRY(pos, AT_NULL, 0);
311 	free(imgp->auxargs, M_TEMP);
312 	imgp->auxargs = NULL;
313 
314 	base--;
315 	suword(base, (uint64_t)imgp->args->argc);
316 
317 	*stack_base = (register_t *)base;
318 	return (0);
319 }
320 
321 /*
322  * Copy strings out to the new process address space, constructing new arg
323  * and env vector tables. Return a pointer to the base so that it can be used
324  * as the initial stack pointer.
325  */
326 static register_t *
327 linux_copyout_strings(struct image_params *imgp)
328 {
329 	int argc, envc;
330 	char **vectp;
331 	char *stringp, *destp;
332 	register_t *stack_base;
333 	struct ps_strings *arginfo;
334 	char canary[LINUX_AT_RANDOM_LEN];
335 	size_t execpath_len;
336 	struct proc *p;
337 
338 	/*
339 	 * Calculate string base and vector table pointers.
340 	 */
341 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
342 		execpath_len = strlen(imgp->execpath) + 1;
343 	else
344 		execpath_len = 0;
345 
346 	p = imgp->proc;
347 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
348 	destp =	(caddr_t)arginfo - SPARE_USRSPACE -
349 	    roundup(sizeof(canary), sizeof(char *)) -
350 	    roundup(execpath_len, sizeof(char *)) -
351 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
352 
353 	if (execpath_len != 0) {
354 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
355 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
356 	}
357 
358 	/*
359 	 * Prepare the canary for SSP.
360 	 */
361 	arc4rand(canary, sizeof(canary), 0);
362 	imgp->canary = (uintptr_t)arginfo -
363 	    roundup(execpath_len, sizeof(char *)) -
364 	    roundup(sizeof(canary), sizeof(char *));
365 	copyout(canary, (void *)imgp->canary, sizeof(canary));
366 
367 	/*
368 	 * If we have a valid auxargs ptr, prepare some room
369 	 * on the stack.
370 	 */
371 	if (imgp->auxargs) {
372 		/*
373 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
374 		 * lower compatibility.
375 		 */
376 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
377 		    (LINUX_AT_COUNT * 2);
378 
379 		/*
380 		 * The '+ 2' is for the null pointers at the end of each of
381 		 * the arg and env vector sets,and imgp->auxarg_size is room
382 		 * for argument of Runtime loader.
383 		 */
384 		vectp = (char **)(destp - (imgp->args->argc +
385 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
386 
387 	} else {
388 		/*
389 		 * The '+ 2' is for the null pointers at the end of each of
390 		 * the arg and env vector sets
391 		 */
392 		vectp = (char **)(destp - (imgp->args->argc +
393 		    imgp->args->envc + 2) * sizeof(char *));
394 	}
395 
396 	/*
397 	 * vectp also becomes our initial stack base
398 	 */
399 	stack_base = (register_t *)vectp;
400 
401 	stringp = imgp->args->begin_argv;
402 	argc = imgp->args->argc;
403 	envc = imgp->args->envc;
404 
405 	/*
406 	 * Copy out strings - arguments and environment.
407 	 */
408 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
409 
410 	/*
411 	 * Fill in "ps_strings" struct for ps, w, etc.
412 	 */
413 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
414 	suword(&arginfo->ps_nargvstr, argc);
415 
416 	/*
417 	 * Fill in argument portion of vector table.
418 	 */
419 	for (; argc > 0; --argc) {
420 		suword(vectp++, (long)(intptr_t)destp);
421 		while (*stringp++ != 0)
422 			destp++;
423 		destp++;
424 	}
425 
426 	/* a null vector table pointer separates the argp's from the envp's */
427 	suword(vectp++, 0);
428 
429 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
430 	suword(&arginfo->ps_nenvstr, envc);
431 
432 	/*
433 	 * Fill in environment portion of vector table.
434 	 */
435 	for (; envc > 0; --envc) {
436 		suword(vectp++, (long)(intptr_t)destp);
437 		while (*stringp++ != 0)
438 			destp++;
439 		destp++;
440 	}
441 
442 	/* end of vector table is a null pointer */
443 	suword(vectp, 0);
444 	return (stack_base);
445 }
446 
447 /*
448  * Reset registers to default values on exec.
449  */
450 static void
451 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
452 {
453 	struct trapframe *regs = td->td_frame;
454 	struct pcb *pcb = td->td_pcb;
455 
456 	if (td->td_proc->p_md.md_ldt != NULL)
457 		user_ldt_free(td);
458 
459 	pcb->pcb_fsbase = 0;
460 	pcb->pcb_gsbase = 0;
461 	clear_pcb_flags(pcb, PCB_32BIT);
462 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
463 	set_pcb_flags(pcb, PCB_FULL_IRET);
464 
465 	bzero((char *)regs, sizeof(struct trapframe));
466 	regs->tf_rip = imgp->entry_addr;
467 	regs->tf_rsp = stack;
468 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
469 	regs->tf_ss = _udatasel;
470 	regs->tf_cs = _ucodesel;
471 	regs->tf_ds = _udatasel;
472 	regs->tf_es = _udatasel;
473 	regs->tf_fs = _ufssel;
474 	regs->tf_gs = _ugssel;
475 	regs->tf_flags = TF_HASSEGS;
476 
477 	/*
478 	 * Reset the hardware debug registers if they were in use.
479 	 * They won't have any meaning for the newly exec'd process.
480 	 */
481 	if (pcb->pcb_flags & PCB_DBREGS) {
482 		pcb->pcb_dr0 = 0;
483 		pcb->pcb_dr1 = 0;
484 		pcb->pcb_dr2 = 0;
485 		pcb->pcb_dr3 = 0;
486 		pcb->pcb_dr6 = 0;
487 		pcb->pcb_dr7 = 0;
488 		if (pcb == curpcb) {
489 			/*
490 			 * Clear the debug registers on the running
491 			 * CPU, otherwise they will end up affecting
492 			 * the next process we switch to.
493 			 */
494 			reset_dbregs();
495 		}
496 		clear_pcb_flags(pcb, PCB_DBREGS);
497 	}
498 
499 	/*
500 	 * Drop the FP state if we hold it, so that the process gets a
501 	 * clean FP state if it uses the FPU again.
502 	 */
503 	fpstate_drop(td);
504 }
505 
506 /*
507  * Copied from amd64/amd64/machdep.c
508  *
509  * XXX fpu state need? don't think so
510  */
511 int
512 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
513 {
514 	struct proc *p;
515 	struct l_ucontext uc;
516 	struct l_sigcontext *context;
517 	struct trapframe *regs;
518 	unsigned long rflags;
519 	int error;
520 	ksiginfo_t ksi;
521 
522 	regs = td->td_frame;
523 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
524 	if (error != 0)
525 		return (error);
526 
527 	p = td->td_proc;
528 	context = &uc.uc_mcontext;
529 	rflags = context->sc_rflags;
530 
531 	/*
532 	 * Don't allow users to change privileged or reserved flags.
533 	 */
534 	/*
535 	 * XXX do allow users to change the privileged flag PSL_RF.
536 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
537 	 * should sometimes set it there too.  tf_rflags is kept in
538 	 * the signal context during signal handling and there is no
539 	 * other place to remember it, so the PSL_RF bit may be
540 	 * corrupted by the signal handler without us knowing.
541 	 * Corruption of the PSL_RF bit at worst causes one more or
542 	 * one less debugger trap, so allowing it is fairly harmless.
543 	 */
544 
545 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
546 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
547 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
548 		return (EINVAL);
549 	}
550 
551 	/*
552 	 * Don't allow users to load a valid privileged %cs.  Let the
553 	 * hardware check for invalid selectors, excess privilege in
554 	 * other selectors, invalid %eip's and invalid %esp's.
555 	 */
556 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
557 	if (!CS_SECURE(context->sc_cs)) {
558 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
559 		ksiginfo_init_trap(&ksi);
560 		ksi.ksi_signo = SIGBUS;
561 		ksi.ksi_code = BUS_OBJERR;
562 		ksi.ksi_trapno = T_PROTFLT;
563 		ksi.ksi_addr = (void *)regs->tf_rip;
564 		trapsignal(td, &ksi);
565 		return (EINVAL);
566 	}
567 
568 	PROC_LOCK(p);
569 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
570 	SIG_CANTMASK(td->td_sigmask);
571 	signotify(td);
572 	PROC_UNLOCK(p);
573 
574 	regs->tf_rdi    = context->sc_rdi;
575 	regs->tf_rsi    = context->sc_rsi;
576 	regs->tf_rdx    = context->sc_rdx;
577 	regs->tf_rbp    = context->sc_rbp;
578 	regs->tf_rbx    = context->sc_rbx;
579 	regs->tf_rcx    = context->sc_rcx;
580 	regs->tf_rax    = context->sc_rax;
581 	regs->tf_rip    = context->sc_rip;
582 	regs->tf_rsp    = context->sc_rsp;
583 	regs->tf_r8     = context->sc_r8;
584 	regs->tf_r9     = context->sc_r9;
585 	regs->tf_r10    = context->sc_r10;
586 	regs->tf_r11    = context->sc_r11;
587 	regs->tf_r12    = context->sc_r12;
588 	regs->tf_r13    = context->sc_r13;
589 	regs->tf_r14    = context->sc_r14;
590 	regs->tf_r15    = context->sc_r15;
591 	regs->tf_cs     = context->sc_cs;
592 	regs->tf_err    = context->sc_err;
593 	regs->tf_rflags = rflags;
594 
595 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
596 	return (EJUSTRETURN);
597 }
598 
599 /*
600  * copied from amd64/amd64/machdep.c
601  *
602  * Send an interrupt to process.
603  */
604 static void
605 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
606 {
607 	struct l_rt_sigframe sf, *sfp;
608 	struct proc *p;
609 	struct thread *td;
610 	struct sigacts *psp;
611 	caddr_t sp;
612 	struct trapframe *regs;
613 	int sig, code;
614 	int oonstack;
615 
616 	td = curthread;
617 	p = td->td_proc;
618 	PROC_LOCK_ASSERT(p, MA_OWNED);
619 	sig = ksi->ksi_signo;
620 	psp = p->p_sigacts;
621 	code = ksi->ksi_code;
622 	mtx_assert(&psp->ps_mtx, MA_OWNED);
623 	regs = td->td_frame;
624 	oonstack = sigonstack(regs->tf_rsp);
625 
626 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
627 	    catcher, sig, mask, code);
628 
629 	/* Allocate space for the signal handler context. */
630 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
631 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
632 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
633 		    sizeof(struct l_rt_sigframe);
634 	} else
635 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
636 	/* Align to 16 bytes. */
637 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
638 	mtx_unlock(&psp->ps_mtx);
639 
640 	/* Translate the signal. */
641 	sig = bsd_to_linux_signal(sig);
642 
643 	/* Save user context. */
644 	bzero(&sf, sizeof(sf));
645 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
646 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
647 
648 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
649 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
650 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
651 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
652 	PROC_UNLOCK(p);
653 
654 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
655 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
656 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
657 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
658 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
659 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
660 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
661 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
662 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
663 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
664 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
665 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
666 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
667 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
668 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
669 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
670 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
671 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
672 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
673 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
674 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
675 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
676 
677 	/* Build the argument list for the signal handler. */
678 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
679 	regs->tf_rax = 0;
680 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
681 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
682 
683 	sf.sf_handler = catcher;
684 	/* Fill in POSIX parts */
685 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
686 
687 	/*
688 	 * Copy the sigframe out to the user's stack.
689 	 */
690 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
691 #ifdef DEBUG
692 		printf("process %ld has trashed its stack\n", (long)p->p_pid);
693 #endif
694 		PROC_LOCK(p);
695 		sigexit(td, SIGILL);
696 	}
697 
698 	regs->tf_rsp = (long)sfp;
699 	regs->tf_rip = linux_rt_sigcode;
700 	regs->tf_rflags &= ~(PSL_T | PSL_D);
701 	regs->tf_cs = _ucodesel;
702 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
703 	PROC_LOCK(p);
704 	mtx_lock(&psp->ps_mtx);
705 }
706 
707 /*
708  * If a Linux binary is exec'ing something, try this image activator
709  * first.  We override standard shell script execution in order to
710  * be able to modify the interpreter path.  We only do this if a Linux
711  * binary is doing the exec, so we do not create an EXEC module for it.
712  */
713 static int exec_linux_imgact_try(struct image_params *iparams);
714 
715 static int
716 exec_linux_imgact_try(struct image_params *imgp)
717 {
718 	const char *head = (const char *)imgp->image_header;
719 	char *rpath;
720 	int error = -1;
721 
722 	/*
723 	 * The interpreter for shell scripts run from a Linux binary needs
724 	 * to be located in /compat/linux if possible in order to recursively
725 	 * maintain Linux path emulation.
726 	 */
727 	if (((const short *)head)[0] == SHELLMAGIC) {
728 		/*
729 		 * Run our normal shell image activator.  If it succeeds
730 		 * attempt to use the alternate path for the interpreter.
731 		 * If an alternate path is found, use our stringspace
732 		 * to store it.
733 		 */
734 		if ((error = exec_shell_imgact(imgp)) == 0) {
735 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
736 			    imgp->interpreter_name, UIO_SYSSPACE,
737 			    &rpath, 0, AT_FDCWD);
738 			if (rpath != NULL)
739 				imgp->args->fname_buf =
740 				    imgp->interpreter_name = rpath;
741 		}
742 	}
743 	return (error);
744 }
745 
746 #define	LINUX_VSYSCALL_START		(-10UL << 20)
747 #define	LINUX_VSYSCALL_SZ		1024
748 
749 const unsigned long linux_vsyscall_vector[] = {
750 	LINUX_SYS_gettimeofday,
751 	LINUX_SYS_linux_time,
752 				/* getcpu not implemented */
753 };
754 
755 static int
756 linux_vsyscall(struct thread *td)
757 {
758 	struct trapframe *frame;
759 	uint64_t retqaddr;
760 	int code, traced;
761 	int error;
762 
763 	frame = td->td_frame;
764 
765 	/* Check %rip for vsyscall area */
766 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
767 		return (EINVAL);
768 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
769 		return (EINVAL);
770 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
771 	if (code >= nitems(linux_vsyscall_vector))
772 		return (EINVAL);
773 
774 	/*
775 	 * vsyscall called as callq *(%rax), so we must
776 	 * use return address from %rsp and also fixup %rsp
777 	 */
778 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
779 	if (error)
780 		return (error);
781 
782 	frame->tf_rip = retqaddr;
783 	frame->tf_rax = linux_vsyscall_vector[code];
784 	frame->tf_rsp += 8;
785 
786 	traced = (frame->tf_flags & PSL_T);
787 
788 	amd64_syscall(td, traced);
789 
790 	return (0);
791 }
792 
793 struct sysentvec elf_linux_sysvec = {
794 	.sv_size	= LINUX_SYS_MAXSYSCALL,
795 	.sv_table	= linux_sysent,
796 	.sv_mask	= 0,
797 	.sv_errsize	= ELAST + 1,
798 	.sv_errtbl	= bsd_to_linux_errno,
799 	.sv_transtrap	= translate_traps,
800 	.sv_fixup	= elf_linux_fixup,
801 	.sv_sendsig	= linux_rt_sendsig,
802 	.sv_sigcode	= &_binary_linux_locore_o_start,
803 	.sv_szsigcode	= &linux_szsigcode,
804 	.sv_name	= "Linux ELF64",
805 	.sv_coredump	= elf64_coredump,
806 	.sv_imgact_try	= exec_linux_imgact_try,
807 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
808 	.sv_pagesize	= PAGE_SIZE,
809 	.sv_minuser	= VM_MIN_ADDRESS,
810 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
811 	.sv_usrstack	= USRSTACK,
812 	.sv_psstrings	= PS_STRINGS,
813 	.sv_stackprot	= VM_PROT_ALL,
814 	.sv_copyout_strings = linux_copyout_strings,
815 	.sv_setregs	= linux_exec_setregs,
816 	.sv_fixlimit	= NULL,
817 	.sv_maxssiz	= NULL,
818 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
819 	.sv_set_syscall_retval = linux_set_syscall_retval,
820 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
821 	.sv_syscallnames = NULL,
822 	.sv_shared_page_base = SHAREDPAGE,
823 	.sv_shared_page_len = PAGE_SIZE,
824 	.sv_schedtail	= linux_schedtail,
825 	.sv_thread_detach = linux_thread_detach,
826 	.sv_trap	= linux_vsyscall,
827 };
828 
829 static void
830 linux_vdso_install(void *param)
831 {
832 
833 	amd64_lower_shared_page(&elf_linux_sysvec);
834 
835 	linux_szsigcode = (&_binary_linux_locore_o_end -
836 	    &_binary_linux_locore_o_start);
837 
838 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
839 		panic("Linux invalid vdso size\n");
840 
841 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
842 
843 	linux_shared_page_obj = __elfN(linux_shared_page_init)
844 	    (&linux_shared_page_mapping);
845 
846 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
847 
848 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
849 	    linux_szsigcode);
850 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
851 
852 	linux_kplatform = linux_shared_page_mapping +
853 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
854 }
855 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
856     (sysinit_cfunc_t)linux_vdso_install, NULL);
857 
858 static void
859 linux_vdso_deinstall(void *param)
860 {
861 
862 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
863 };
864 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
865     (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
866 
867 static char GNULINUX_ABI_VENDOR[] = "GNU";
868 static int GNULINUX_ABI_DESC = 0;
869 
870 static boolean_t
871 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
872 {
873 	const Elf32_Word *desc;
874 	uintptr_t p;
875 
876 	p = (uintptr_t)(note + 1);
877 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
878 
879 	desc = (const Elf32_Word *)p;
880 	if (desc[0] != GNULINUX_ABI_DESC)
881 		return (FALSE);
882 
883 	/*
884 	 * For Linux we encode osrel as follows (see linux_mib.c):
885 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
886 	 */
887 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
888 
889 	return (TRUE);
890 }
891 
892 static Elf_Brandnote linux64_brandnote = {
893 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
894 	.hdr.n_descsz	= 16,
895 	.hdr.n_type	= 1,
896 	.vendor		= GNULINUX_ABI_VENDOR,
897 	.flags		= BN_TRANSLATE_OSREL,
898 	.trans_osrel	= linux_trans_osrel
899 };
900 
901 static Elf64_Brandinfo linux_glibc2brand = {
902 	.brand		= ELFOSABI_LINUX,
903 	.machine	= EM_X86_64,
904 	.compat_3_brand	= "Linux",
905 	.emul_path	= "/compat/linux",
906 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
907 	.sysvec		= &elf_linux_sysvec,
908 	.interp_newpath	= NULL,
909 	.brand_note	= &linux64_brandnote,
910 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
911 };
912 
913 static Elf64_Brandinfo linux_glibc2brandshort = {
914 	.brand		= ELFOSABI_LINUX,
915 	.machine	= EM_X86_64,
916 	.compat_3_brand	= "Linux",
917 	.emul_path	= "/compat/linux",
918 	.interp_path	= "/lib64/ld-linux.so.2",
919 	.sysvec		= &elf_linux_sysvec,
920 	.interp_newpath	= NULL,
921 	.brand_note	= &linux64_brandnote,
922 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
923 };
924 
925 static Elf64_Brandinfo linux_muslbrand = {
926 	.brand		= ELFOSABI_LINUX,
927 	.machine	= EM_X86_64,
928 	.compat_3_brand	= "Linux",
929 	.emul_path	= "/compat/linux",
930 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
931 	.sysvec		= &elf_linux_sysvec,
932 	.interp_newpath	= NULL,
933 	.brand_note	= &linux64_brandnote,
934 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
935 };
936 
937 Elf64_Brandinfo *linux_brandlist[] = {
938 	&linux_glibc2brand,
939 	&linux_glibc2brandshort,
940 	&linux_muslbrand,
941 	NULL
942 };
943 
944 static int
945 linux64_elf_modevent(module_t mod, int type, void *data)
946 {
947 	Elf64_Brandinfo **brandinfo;
948 	int error;
949 	struct linux_ioctl_handler **lihp;
950 
951 	error = 0;
952 
953 	switch(type) {
954 	case MOD_LOAD:
955 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
956 		     ++brandinfo)
957 			if (elf64_insert_brand_entry(*brandinfo) < 0)
958 				error = EINVAL;
959 		if (error == 0) {
960 			SET_FOREACH(lihp, linux_ioctl_handler_set)
961 				linux_ioctl_register_handler(*lihp);
962 			LIST_INIT(&futex_list);
963 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
964 			stclohz = (stathz ? stathz : hz);
965 			if (bootverbose)
966 				printf("Linux x86-64 ELF exec handler installed\n");
967 		} else
968 			printf("cannot insert Linux x86-64 ELF brand handler\n");
969 		break;
970 	case MOD_UNLOAD:
971 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
972 		     ++brandinfo)
973 			if (elf64_brand_inuse(*brandinfo))
974 				error = EBUSY;
975 		if (error == 0) {
976 			for (brandinfo = &linux_brandlist[0];
977 			     *brandinfo != NULL; ++brandinfo)
978 				if (elf64_remove_brand_entry(*brandinfo) < 0)
979 					error = EINVAL;
980 		}
981 		if (error == 0) {
982 			SET_FOREACH(lihp, linux_ioctl_handler_set)
983 				linux_ioctl_unregister_handler(*lihp);
984 			mtx_destroy(&futex_mtx);
985 			if (bootverbose)
986 				printf("Linux ELF exec handler removed\n");
987 		} else
988 			printf("Could not deinstall ELF interpreter entry\n");
989 		break;
990 	default:
991 		return (EOPNOTSUPP);
992 	}
993 	return (error);
994 }
995 
996 static moduledata_t linux64_elf_mod = {
997 	"linux64elf",
998 	linux64_elf_modevent,
999 	0
1000 };
1001 
1002 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1003 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
1004 FEATURE(linux64, "Linux 64bit support");
1005