xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 25ecdc7d52770caf1c9b44b5ec11f468f6b636f3)
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 #define	__ELF_WORD_SIZE	64
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/exec.h>
42 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/kernel.h>
46 #include <sys/ktr.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/module.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/resourcevar.h>
53 #include <sys/signalvar.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysent.h>
57 #include <sys/sysproto.h>
58 #include <sys/vnode.h>
59 #include <sys/eventhandler.h>
60 
61 #include <vm/vm.h>
62 #include <vm/pmap.h>
63 #include <vm/vm_extern.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_param.h>
68 
69 #include <machine/cpu.h>
70 #include <machine/md_var.h>
71 #include <machine/pcb.h>
72 #include <machine/specialreg.h>
73 #include <machine/trap.h>
74 
75 #include <amd64/linux/linux.h>
76 #include <amd64/linux/linux_proto.h>
77 #include <compat/linux/linux_emul.h>
78 #include <compat/linux/linux_ioctl.h>
79 #include <compat/linux/linux_mib.h>
80 #include <compat/linux/linux_misc.h>
81 #include <compat/linux/linux_signal.h>
82 #include <compat/linux/linux_sysproto.h>
83 #include <compat/linux/linux_util.h>
84 #include <compat/linux/linux_vdso.h>
85 
86 MODULE_VERSION(linux64, 1);
87 
88 const char *linux_kplatform;
89 static int linux_szsigcode;
90 static vm_object_t linux_shared_page_obj;
91 static char *linux_shared_page_mapping;
92 extern char _binary_linux_locore_o_start;
93 extern char _binary_linux_locore_o_end;
94 
95 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
96 
97 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
98 
99 static int	linux_copyout_strings(struct image_params *imgp,
100 		    uintptr_t *stack_base);
101 static int	linux_fixup_elf(uintptr_t *stack_base,
102 		    struct image_params *iparams);
103 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
104 static void	linux_vdso_install(void *param);
105 static void	linux_vdso_deinstall(void *param);
106 static void	linux_set_syscall_retval(struct thread *td, int error);
107 static int	linux_fetch_syscall_args(struct thread *td);
108 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
109 		    uintptr_t stack);
110 static int	linux_vsyscall(struct thread *td);
111 
112 #define LINUX_T_UNKNOWN  255
113 static int _bsd_to_linux_trapcode[] = {
114 	LINUX_T_UNKNOWN,	/* 0 */
115 	6,			/* 1  T_PRIVINFLT */
116 	LINUX_T_UNKNOWN,	/* 2 */
117 	3,			/* 3  T_BPTFLT */
118 	LINUX_T_UNKNOWN,	/* 4 */
119 	LINUX_T_UNKNOWN,	/* 5 */
120 	16,			/* 6  T_ARITHTRAP */
121 	254,			/* 7  T_ASTFLT */
122 	LINUX_T_UNKNOWN,	/* 8 */
123 	13,			/* 9  T_PROTFLT */
124 	1,			/* 10 T_TRCTRAP */
125 	LINUX_T_UNKNOWN,	/* 11 */
126 	14,			/* 12 T_PAGEFLT */
127 	LINUX_T_UNKNOWN,	/* 13 */
128 	17,			/* 14 T_ALIGNFLT */
129 	LINUX_T_UNKNOWN,	/* 15 */
130 	LINUX_T_UNKNOWN,	/* 16 */
131 	LINUX_T_UNKNOWN,	/* 17 */
132 	0,			/* 18 T_DIVIDE */
133 	2,			/* 19 T_NMI */
134 	4,			/* 20 T_OFLOW */
135 	5,			/* 21 T_BOUND */
136 	7,			/* 22 T_DNA */
137 	8,			/* 23 T_DOUBLEFLT */
138 	9,			/* 24 T_FPOPFLT */
139 	10,			/* 25 T_TSSFLT */
140 	11,			/* 26 T_SEGNPFLT */
141 	12,			/* 27 T_STKFLT */
142 	18,			/* 28 T_MCHK */
143 	19,			/* 29 T_XMMFLT */
144 	15			/* 30 T_RESERVED */
145 };
146 #define bsd_to_linux_trapcode(code) \
147     ((code)<nitems(_bsd_to_linux_trapcode)? \
148      _bsd_to_linux_trapcode[(code)]: \
149      LINUX_T_UNKNOWN)
150 
151 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
152 LINUX_VDSO_SYM_CHAR(linux_platform);
153 
154 /*
155  * If FreeBSD & Linux have a difference of opinion about what a trap
156  * means, deal with it here.
157  *
158  * MPSAFE
159  */
160 static int
161 linux_translate_traps(int signal, int trap_code)
162 {
163 
164 	if (signal != SIGBUS)
165 		return (signal);
166 	switch (trap_code) {
167 	case T_PROTFLT:
168 	case T_TSSFLT:
169 	case T_DOUBLEFLT:
170 	case T_PAGEFLT:
171 		return (SIGSEGV);
172 	default:
173 		return (signal);
174 	}
175 }
176 
177 static int
178 linux_fetch_syscall_args(struct thread *td)
179 {
180 	struct proc *p;
181 	struct trapframe *frame;
182 	struct syscall_args *sa;
183 
184 	p = td->td_proc;
185 	frame = td->td_frame;
186 	sa = &td->td_sa;
187 
188 	sa->args[0] = frame->tf_rdi;
189 	sa->args[1] = frame->tf_rsi;
190 	sa->args[2] = frame->tf_rdx;
191 	sa->args[3] = frame->tf_rcx;
192 	sa->args[4] = frame->tf_r8;
193 	sa->args[5] = frame->tf_r9;
194 	sa->code = frame->tf_rax;
195 
196 	if (sa->code >= p->p_sysent->sv_size)
197 		/* nosys */
198 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
199 	else
200 		sa->callp = &p->p_sysent->sv_table[sa->code];
201 
202 	td->td_retval[0] = 0;
203 	return (0);
204 }
205 
206 static void
207 linux_set_syscall_retval(struct thread *td, int error)
208 {
209 	struct trapframe *frame;
210 
211 	frame = td->td_frame;
212 
213 	switch (error) {
214 	case 0:
215 		frame->tf_rax = td->td_retval[0];
216  		frame->tf_r10 = frame->tf_rcx;
217 		break;
218 
219 	case ERESTART:
220 		/*
221 		 * Reconstruct pc, we know that 'syscall' is 2 bytes,
222 		 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
223 		 * We saved this in tf_err.
224 		 *
225 		 */
226 		frame->tf_rip -= frame->tf_err;
227 		frame->tf_r10 = frame->tf_rcx;
228 		break;
229 
230 	case EJUSTRETURN:
231 		break;
232 
233 	default:
234 		frame->tf_rax = bsd_to_linux_errno(error);
235 		frame->tf_r10 = frame->tf_rcx;
236 		break;
237 	}
238 
239 	/*
240 	 * Differently from FreeBSD native ABI, on Linux only %rcx
241 	 * and %r11 values are not preserved across the syscall.
242 	 * Require full context restore to get all registers except
243 	 * those two restored at return to usermode.
244 	 *
245 	 * XXX: Would be great to be able to avoid PCB_FULL_IRET
246 	 *      for the error == 0 case.
247 	 */
248 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
249 }
250 
251 static int
252 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
253 {
254 	Elf_Auxargs *args;
255 	Elf_Auxinfo *argarray, *pos;
256 	struct proc *p;
257 	int error, issetugid;
258 
259 	p = imgp->proc;
260 	args = (Elf64_Auxargs *)imgp->auxargs;
261 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
262 	    M_WAITOK | M_ZERO);
263 
264 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
265 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
266 	    imgp->proc->p_sysent->sv_shared_page_base);
267 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
268 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
269 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
270 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
271 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
272 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
273 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
274 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
275 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
276 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
277 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
278 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
279 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
280 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
281 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
282 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
283 	if (imgp->execpathp != 0)
284 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
285 	if (args->execfd != -1)
286 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
287 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
288 	AUXARGS_ENTRY(pos, AT_NULL, 0);
289 
290 	free(imgp->auxargs, M_TEMP);
291 	imgp->auxargs = NULL;
292 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
293 
294 	error = copyout(argarray, (void *)base,
295 	    sizeof(*argarray) * LINUX_AT_COUNT);
296 	free(argarray, M_TEMP);
297 	return (error);
298 }
299 
300 static int
301 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
302 {
303 	Elf_Addr *base;
304 
305 	base = (Elf64_Addr *)*stack_base;
306 	base--;
307 	if (suword(base, (uint64_t)imgp->args->argc) == -1)
308 		return (EFAULT);
309 
310 	*stack_base = (uintptr_t)base;
311 	return (0);
312 }
313 
314 /*
315  * Copy strings out to the new process address space, constructing new arg
316  * and env vector tables. Return a pointer to the base so that it can be used
317  * as the initial stack pointer.
318  */
319 static int
320 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
321 {
322 	int argc, envc, error;
323 	char **vectp;
324 	char *stringp;
325 	uintptr_t destp, ustringp;
326 	struct ps_strings *arginfo;
327 	char canary[LINUX_AT_RANDOM_LEN];
328 	size_t execpath_len;
329 	struct proc *p;
330 
331 	/* Calculate string base and vector table pointers. */
332 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
333 		execpath_len = strlen(imgp->execpath) + 1;
334 	else
335 		execpath_len = 0;
336 
337 	p = imgp->proc;
338 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
339 	destp = (uintptr_t)arginfo;
340 
341 	if (execpath_len != 0) {
342 		destp -= execpath_len;
343 		destp = rounddown2(destp, sizeof(void *));
344 		imgp->execpathp = (void *)destp;
345 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
346 		if (error != 0)
347 			return (error);
348 	}
349 
350 	/* Prepare the canary for SSP. */
351 	arc4rand(canary, sizeof(canary), 0);
352 	destp -= roundup(sizeof(canary), sizeof(void *));
353 	imgp->canary = (void *)destp;
354 	error = copyout(canary, imgp->canary, sizeof(canary));
355 	if (error != 0)
356 		return (error);
357 
358 	/* Allocate room for the argument and environment strings. */
359 	destp -= ARG_MAX - imgp->args->stringspace;
360 	destp = rounddown2(destp, sizeof(void *));
361 	ustringp = destp;
362 
363 	if (imgp->auxargs) {
364 		/*
365 		 * Allocate room on the stack for the ELF auxargs
366 		 * array.  It has LINUX_AT_COUNT entries.
367 		 */
368 		destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
369 		destp = rounddown2(destp, sizeof(void *));
370 	}
371 
372 	vectp = (char **)destp;
373 
374 	/*
375 	 * Allocate room for the argv[] and env vectors including the
376 	 * terminating NULL pointers.
377 	 */
378 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
379 
380 	/*
381 	 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
382 	 * One "long argc" will be prepended later.
383 	 */
384 	vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
385 
386 	/* vectp also becomes our initial stack base. */
387 	*stack_base = (uintptr_t)vectp;
388 
389 	stringp = imgp->args->begin_argv;
390 	argc = imgp->args->argc;
391 	envc = imgp->args->envc;
392 
393 	/* Copy out strings - arguments and environment. */
394 	error = copyout(stringp, (void *)ustringp,
395 	    ARG_MAX - imgp->args->stringspace);
396 	if (error != 0)
397 		return (error);
398 
399 	/* Fill in "ps_strings" struct for ps, w, etc. */
400 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
401 	    suword(&arginfo->ps_nargvstr, argc) != 0)
402 		return (EFAULT);
403 
404 	/* Fill in argument portion of vector table. */
405 	for (; argc > 0; --argc) {
406 		if (suword(vectp++, ustringp) != 0)
407 			return (EFAULT);
408 		while (*stringp++ != 0)
409 			ustringp++;
410 		ustringp++;
411 	}
412 
413 	/* A null vector table pointer separates the argp's from the envp's. */
414 	if (suword(vectp++, 0) != 0)
415 		return (EFAULT);
416 
417 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
418 	    suword(&arginfo->ps_nenvstr, envc) != 0)
419 		return (EFAULT);
420 
421 	/* Fill in environment portion of vector table. */
422 	for (; envc > 0; --envc) {
423 		if (suword(vectp++, ustringp) != 0)
424 			return (EFAULT);
425 		while (*stringp++ != 0)
426 			ustringp++;
427 		ustringp++;
428 	}
429 
430 	/* The end of the vector table is a null pointer. */
431 	if (suword(vectp, 0) != 0)
432 		return (EFAULT);
433 
434 	if (imgp->auxargs) {
435 		vectp++;
436 		error = imgp->sysent->sv_copyout_auxargs(imgp,
437 		    (uintptr_t)vectp);
438 		if (error != 0)
439 			return (error);
440 	}
441 
442 	return (0);
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,
450     uintptr_t stack)
451 {
452 	struct trapframe *regs;
453 	struct pcb *pcb;
454 	register_t saved_rflags;
455 
456 	regs = td->td_frame;
457 	pcb = td->td_pcb;
458 
459 	if (td->td_proc->p_md.md_ldt != NULL)
460 		user_ldt_free(td);
461 
462 	pcb->pcb_fsbase = 0;
463 	pcb->pcb_gsbase = 0;
464 	clear_pcb_flags(pcb, PCB_32BIT);
465 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
466 	set_pcb_flags(pcb, PCB_FULL_IRET);
467 
468 	saved_rflags = regs->tf_rflags & PSL_T;
469 	bzero((char *)regs, sizeof(struct trapframe));
470 	regs->tf_rip = imgp->entry_addr;
471 	regs->tf_rsp = stack;
472 	regs->tf_rflags = PSL_USER | saved_rflags;
473 	regs->tf_ss = _udatasel;
474 	regs->tf_cs = _ucodesel;
475 	regs->tf_ds = _udatasel;
476 	regs->tf_es = _udatasel;
477 	regs->tf_fs = _ufssel;
478 	regs->tf_gs = _ugssel;
479 	regs->tf_flags = TF_HASSEGS;
480 
481 	x86_clear_dbregs(pcb);
482 
483 	/*
484 	 * Drop the FP state if we hold it, so that the process gets a
485 	 * clean FP state if it uses the FPU again.
486 	 */
487 	fpstate_drop(td);
488 }
489 
490 /*
491  * Copied from amd64/amd64/machdep.c
492  *
493  * XXX fpu state need? don't think so
494  */
495 int
496 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
497 {
498 	struct proc *p;
499 	struct l_ucontext uc;
500 	struct l_sigcontext *context;
501 	struct trapframe *regs;
502 	unsigned long rflags;
503 	int error;
504 	ksiginfo_t ksi;
505 
506 	regs = td->td_frame;
507 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
508 	if (error != 0)
509 		return (error);
510 
511 	p = td->td_proc;
512 	context = &uc.uc_mcontext;
513 	rflags = context->sc_rflags;
514 
515 	/*
516 	 * Don't allow users to change privileged or reserved flags.
517 	 */
518 	/*
519 	 * XXX do allow users to change the privileged flag PSL_RF.
520 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
521 	 * should sometimes set it there too.  tf_rflags is kept in
522 	 * the signal context during signal handling and there is no
523 	 * other place to remember it, so the PSL_RF bit may be
524 	 * corrupted by the signal handler without us knowing.
525 	 * Corruption of the PSL_RF bit at worst causes one more or
526 	 * one less debugger trap, so allowing it is fairly harmless.
527 	 */
528 
529 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
530 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
531 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
532 		return (EINVAL);
533 	}
534 
535 	/*
536 	 * Don't allow users to load a valid privileged %cs.  Let the
537 	 * hardware check for invalid selectors, excess privilege in
538 	 * other selectors, invalid %eip's and invalid %esp's.
539 	 */
540 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
541 	if (!CS_SECURE(context->sc_cs)) {
542 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
543 		ksiginfo_init_trap(&ksi);
544 		ksi.ksi_signo = SIGBUS;
545 		ksi.ksi_code = BUS_OBJERR;
546 		ksi.ksi_trapno = T_PROTFLT;
547 		ksi.ksi_addr = (void *)regs->tf_rip;
548 		trapsignal(td, &ksi);
549 		return (EINVAL);
550 	}
551 
552 	PROC_LOCK(p);
553 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
554 	SIG_CANTMASK(td->td_sigmask);
555 	signotify(td);
556 	PROC_UNLOCK(p);
557 
558 	regs->tf_rdi    = context->sc_rdi;
559 	regs->tf_rsi    = context->sc_rsi;
560 	regs->tf_rdx    = context->sc_rdx;
561 	regs->tf_rbp    = context->sc_rbp;
562 	regs->tf_rbx    = context->sc_rbx;
563 	regs->tf_rcx    = context->sc_rcx;
564 	regs->tf_rax    = context->sc_rax;
565 	regs->tf_rip    = context->sc_rip;
566 	regs->tf_rsp    = context->sc_rsp;
567 	regs->tf_r8     = context->sc_r8;
568 	regs->tf_r9     = context->sc_r9;
569 	regs->tf_r10    = context->sc_r10;
570 	regs->tf_r11    = context->sc_r11;
571 	regs->tf_r12    = context->sc_r12;
572 	regs->tf_r13    = context->sc_r13;
573 	regs->tf_r14    = context->sc_r14;
574 	regs->tf_r15    = context->sc_r15;
575 	regs->tf_cs     = context->sc_cs;
576 	regs->tf_err    = context->sc_err;
577 	regs->tf_rflags = rflags;
578 
579 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
580 	return (EJUSTRETURN);
581 }
582 
583 /*
584  * copied from amd64/amd64/machdep.c
585  *
586  * Send an interrupt to process.
587  */
588 static void
589 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
590 {
591 	struct l_rt_sigframe sf, *sfp;
592 	struct proc *p;
593 	struct thread *td;
594 	struct sigacts *psp;
595 	caddr_t sp;
596 	struct trapframe *regs;
597 	int sig, code;
598 	int oonstack;
599 
600 	td = curthread;
601 	p = td->td_proc;
602 	PROC_LOCK_ASSERT(p, MA_OWNED);
603 	sig = ksi->ksi_signo;
604 	psp = p->p_sigacts;
605 	code = ksi->ksi_code;
606 	mtx_assert(&psp->ps_mtx, MA_OWNED);
607 	regs = td->td_frame;
608 	oonstack = sigonstack(regs->tf_rsp);
609 
610 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
611 	    catcher, sig, mask, code);
612 
613 	/* Allocate space for the signal handler context. */
614 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
615 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
616 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
617 		    sizeof(struct l_rt_sigframe);
618 	} else
619 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
620 	/* Align to 16 bytes. */
621 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
622 	mtx_unlock(&psp->ps_mtx);
623 
624 	/* Translate the signal. */
625 	sig = bsd_to_linux_signal(sig);
626 
627 	/* Save user context. */
628 	bzero(&sf, sizeof(sf));
629 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
630 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
631 
632 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
633 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
634 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
635 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
636 	PROC_UNLOCK(p);
637 
638 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
639 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
640 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
641 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
642 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
643 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
644 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
645 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
646 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
647 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
648 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
649 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
650 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
651 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
652 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
653 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
654 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
655 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
656 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
657 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
658 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
659 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
660 
661 	/* Build the argument list for the signal handler. */
662 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
663 	regs->tf_rax = 0;
664 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
665 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
666 
667 	sf.sf_handler = catcher;
668 	/* Fill in POSIX parts. */
669 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
670 
671 	/* Copy the sigframe out to the user's stack. */
672 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
673 		PROC_LOCK(p);
674 		sigexit(td, SIGILL);
675 	}
676 
677 	regs->tf_rsp = (long)sfp;
678 	regs->tf_rip = linux_rt_sigcode;
679 	regs->tf_rflags &= ~(PSL_T | PSL_D);
680 	regs->tf_cs = _ucodesel;
681 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
682 	PROC_LOCK(p);
683 	mtx_lock(&psp->ps_mtx);
684 }
685 
686 #define	LINUX_VSYSCALL_START		(-10UL << 20)
687 #define	LINUX_VSYSCALL_SZ		1024
688 
689 const unsigned long linux_vsyscall_vector[] = {
690 	LINUX_SYS_gettimeofday,
691 	LINUX_SYS_linux_time,
692 	LINUX_SYS_linux_getcpu,
693 };
694 
695 static int
696 linux_vsyscall(struct thread *td)
697 {
698 	struct trapframe *frame;
699 	uint64_t retqaddr;
700 	int code, traced;
701 	int error;
702 
703 	frame = td->td_frame;
704 
705 	/* Check %rip for vsyscall area. */
706 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
707 		return (EINVAL);
708 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
709 		return (EINVAL);
710 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
711 	if (code >= nitems(linux_vsyscall_vector))
712 		return (EINVAL);
713 
714 	/*
715 	 * vsyscall called as callq *(%rax), so we must
716 	 * use return address from %rsp and also fixup %rsp.
717 	 */
718 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
719 	if (error)
720 		return (error);
721 
722 	frame->tf_rip = retqaddr;
723 	frame->tf_rax = linux_vsyscall_vector[code];
724 	frame->tf_rsp += 8;
725 
726 	traced = (frame->tf_flags & PSL_T);
727 
728 	amd64_syscall(td, traced);
729 
730 	return (0);
731 }
732 
733 struct sysentvec elf_linux_sysvec = {
734 	.sv_size	= LINUX_SYS_MAXSYSCALL,
735 	.sv_table	= linux_sysent,
736 	.sv_transtrap	= linux_translate_traps,
737 	.sv_fixup	= linux_fixup_elf,
738 	.sv_sendsig	= linux_rt_sendsig,
739 	.sv_sigcode	= &_binary_linux_locore_o_start,
740 	.sv_szsigcode	= &linux_szsigcode,
741 	.sv_name	= "Linux ELF64",
742 	.sv_coredump	= elf64_coredump,
743 	.sv_imgact_try	= linux_exec_imgact_try,
744 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
745 	.sv_minuser	= VM_MIN_ADDRESS,
746 	.sv_maxuser	= VM_MAXUSER_ADDRESS_LA48,
747 	.sv_usrstack	= USRSTACK_LA48,
748 	.sv_psstrings	= PS_STRINGS_LA48,
749 	.sv_stackprot	= VM_PROT_ALL,
750 	.sv_copyout_auxargs = linux_copyout_auxargs,
751 	.sv_copyout_strings = linux_copyout_strings,
752 	.sv_setregs	= linux_exec_setregs,
753 	.sv_fixlimit	= NULL,
754 	.sv_maxssiz	= NULL,
755 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
756 	.sv_set_syscall_retval = linux_set_syscall_retval,
757 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
758 	.sv_syscallnames = NULL,
759 	.sv_shared_page_base = SHAREDPAGE_LA48,
760 	.sv_shared_page_len = PAGE_SIZE,
761 	.sv_schedtail	= linux_schedtail,
762 	.sv_thread_detach = linux_thread_detach,
763 	.sv_trap	= linux_vsyscall,
764 	.sv_onexec	= linux_on_exec,
765 	.sv_onexit	= linux_on_exit,
766 	.sv_ontdexit	= linux_thread_dtor,
767 };
768 
769 static void
770 linux_vdso_install(void *param)
771 {
772 
773 	amd64_lower_shared_page(&elf_linux_sysvec);
774 
775 	linux_szsigcode = (&_binary_linux_locore_o_end -
776 	    &_binary_linux_locore_o_start);
777 
778 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
779 		panic("Linux invalid vdso size\n");
780 
781 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
782 
783 	linux_shared_page_obj = __elfN(linux_shared_page_init)
784 	    (&linux_shared_page_mapping);
785 
786 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
787 
788 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
789 	    linux_szsigcode);
790 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
791 
792 	linux_kplatform = linux_shared_page_mapping +
793 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
794 }
795 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
796     linux_vdso_install, NULL);
797 
798 static void
799 linux_vdso_deinstall(void *param)
800 {
801 
802 	__elfN(linux_shared_page_fini)(linux_shared_page_obj,
803 	    linux_shared_page_mapping);
804 }
805 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
806     linux_vdso_deinstall, NULL);
807 
808 static char GNULINUX_ABI_VENDOR[] = "GNU";
809 static int GNULINUX_ABI_DESC = 0;
810 
811 static bool
812 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
813 {
814 	const Elf32_Word *desc;
815 	uintptr_t p;
816 
817 	p = (uintptr_t)(note + 1);
818 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
819 
820 	desc = (const Elf32_Word *)p;
821 	if (desc[0] != GNULINUX_ABI_DESC)
822 		return (false);
823 
824 	/*
825 	 * For Linux we encode osrel using the Linux convention of
826 	 * 	(version << 16) | (major << 8) | (minor)
827 	 * See macro in linux_mib.h
828 	 */
829 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
830 
831 	return (true);
832 }
833 
834 static Elf_Brandnote linux64_brandnote = {
835 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
836 	.hdr.n_descsz	= 16,
837 	.hdr.n_type	= 1,
838 	.vendor		= GNULINUX_ABI_VENDOR,
839 	.flags		= BN_TRANSLATE_OSREL,
840 	.trans_osrel	= linux_trans_osrel
841 };
842 
843 static Elf64_Brandinfo linux_glibc2brand = {
844 	.brand		= ELFOSABI_LINUX,
845 	.machine	= EM_X86_64,
846 	.compat_3_brand	= "Linux",
847 	.emul_path	= linux_emul_path,
848 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
849 	.sysvec		= &elf_linux_sysvec,
850 	.interp_newpath	= NULL,
851 	.brand_note	= &linux64_brandnote,
852 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
853 };
854 
855 static Elf64_Brandinfo linux_glibc2brandshort = {
856 	.brand		= ELFOSABI_LINUX,
857 	.machine	= EM_X86_64,
858 	.compat_3_brand	= "Linux",
859 	.emul_path	= linux_emul_path,
860 	.interp_path	= "/lib64/ld-linux.so.2",
861 	.sysvec		= &elf_linux_sysvec,
862 	.interp_newpath	= NULL,
863 	.brand_note	= &linux64_brandnote,
864 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
865 };
866 
867 static Elf64_Brandinfo linux_muslbrand = {
868 	.brand		= ELFOSABI_LINUX,
869 	.machine	= EM_X86_64,
870 	.compat_3_brand	= "Linux",
871 	.emul_path	= linux_emul_path,
872 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
873 	.sysvec		= &elf_linux_sysvec,
874 	.interp_newpath	= NULL,
875 	.brand_note	= &linux64_brandnote,
876 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
877 };
878 
879 Elf64_Brandinfo *linux_brandlist[] = {
880 	&linux_glibc2brand,
881 	&linux_glibc2brandshort,
882 	&linux_muslbrand,
883 	NULL
884 };
885 
886 static int
887 linux64_elf_modevent(module_t mod, int type, void *data)
888 {
889 	Elf64_Brandinfo **brandinfo;
890 	int error;
891 	struct linux_ioctl_handler **lihp;
892 
893 	error = 0;
894 
895 	switch(type) {
896 	case MOD_LOAD:
897 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
898 		     ++brandinfo)
899 			if (elf64_insert_brand_entry(*brandinfo) < 0)
900 				error = EINVAL;
901 		if (error == 0) {
902 			SET_FOREACH(lihp, linux_ioctl_handler_set)
903 				linux_ioctl_register_handler(*lihp);
904 			stclohz = (stathz ? stathz : hz);
905 			if (bootverbose)
906 				printf("Linux x86-64 ELF exec handler installed\n");
907 		} else
908 			printf("cannot insert Linux x86-64 ELF brand handler\n");
909 		break;
910 	case MOD_UNLOAD:
911 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
912 		     ++brandinfo)
913 			if (elf64_brand_inuse(*brandinfo))
914 				error = EBUSY;
915 		if (error == 0) {
916 			for (brandinfo = &linux_brandlist[0];
917 			     *brandinfo != NULL; ++brandinfo)
918 				if (elf64_remove_brand_entry(*brandinfo) < 0)
919 					error = EINVAL;
920 		}
921 		if (error == 0) {
922 			SET_FOREACH(lihp, linux_ioctl_handler_set)
923 				linux_ioctl_unregister_handler(*lihp);
924 			if (bootverbose)
925 				printf("Linux ELF exec handler removed\n");
926 		} else
927 			printf("Could not deinstall ELF interpreter entry\n");
928 		break;
929 	default:
930 		return (EOPNOTSUPP);
931 	}
932 	return (error);
933 }
934 
935 static moduledata_t linux64_elf_mod = {
936 	"linux64elf",
937 	linux64_elf_modevent,
938 	0
939 };
940 
941 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
942 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
943 FEATURE(linux64, "Linux 64bit support");
944