xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 752c1d14e398faab8e6f19f99ee29df87ecb05e1)
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 = td->td_frame;
210 
211 	/*
212 	 * On Linux only %rcx and %r11 values are not preserved across
213 	 * the syscall.  So, do not clobber %rdx and %r10.
214 	 */
215 	td->td_retval[1] = frame->tf_rdx;
216 	if (error != EJUSTRETURN)
217 		frame->tf_r10 = frame->tf_rcx;
218 
219 	cpu_set_syscall_retval(td, error);
220 
221 	if (__predict_false(error != 0)) {
222 		if (error != ERESTART && error != EJUSTRETURN)
223 			frame->tf_rax = linux_to_bsd_errno(error);
224 	}
225 
226 	 /* Restore all registers. */
227 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
228 }
229 
230 static int
231 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
232 {
233 	Elf_Auxargs *args;
234 	Elf_Auxinfo *argarray, *pos;
235 	struct proc *p;
236 	int error, issetugid;
237 
238 	p = imgp->proc;
239 	args = (Elf64_Auxargs *)imgp->auxargs;
240 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
241 	    M_WAITOK | M_ZERO);
242 
243 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
244 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
245 	    imgp->proc->p_sysent->sv_shared_page_base);
246 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
247 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
248 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
249 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
250 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
251 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
252 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
253 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
254 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
255 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
256 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
257 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
258 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
259 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
260 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
261 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
262 	if (imgp->execpathp != 0)
263 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
264 	if (args->execfd != -1)
265 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
266 	AUXARGS_ENTRY(pos, AT_NULL, 0);
267 
268 	free(imgp->auxargs, M_TEMP);
269 	imgp->auxargs = NULL;
270 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
271 
272 	error = copyout(argarray, (void *)base,
273 	    sizeof(*argarray) * LINUX_AT_COUNT);
274 	free(argarray, M_TEMP);
275 	return (error);
276 }
277 
278 static int
279 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
280 {
281 	Elf_Addr *base;
282 
283 	base = (Elf64_Addr *)*stack_base;
284 	base--;
285 	if (suword(base, (uint64_t)imgp->args->argc) == -1)
286 		return (EFAULT);
287 
288 	*stack_base = (uintptr_t)base;
289 	return (0);
290 }
291 
292 /*
293  * Copy strings out to the new process address space, constructing new arg
294  * and env vector tables. Return a pointer to the base so that it can be used
295  * as the initial stack pointer.
296  */
297 static int
298 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
299 {
300 	int argc, envc, error;
301 	char **vectp;
302 	char *stringp;
303 	uintptr_t destp, ustringp;
304 	struct ps_strings *arginfo;
305 	char canary[LINUX_AT_RANDOM_LEN];
306 	size_t execpath_len;
307 	struct proc *p;
308 
309 	/* Calculate string base and vector table pointers. */
310 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
311 		execpath_len = strlen(imgp->execpath) + 1;
312 	else
313 		execpath_len = 0;
314 
315 	p = imgp->proc;
316 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
317 	destp = (uintptr_t)arginfo;
318 
319 	if (execpath_len != 0) {
320 		destp -= execpath_len;
321 		destp = rounddown2(destp, sizeof(void *));
322 		imgp->execpathp = (void *)destp;
323 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
324 		if (error != 0)
325 			return (error);
326 	}
327 
328 	/* Prepare the canary for SSP. */
329 	arc4rand(canary, sizeof(canary), 0);
330 	destp -= roundup(sizeof(canary), sizeof(void *));
331 	imgp->canary = (void *)destp;
332 	error = copyout(canary, imgp->canary, sizeof(canary));
333 	if (error != 0)
334 		return (error);
335 
336 	/* Allocate room for the argument and environment strings. */
337 	destp -= ARG_MAX - imgp->args->stringspace;
338 	destp = rounddown2(destp, sizeof(void *));
339 	ustringp = destp;
340 
341 	if (imgp->auxargs) {
342 		/*
343 		 * Allocate room on the stack for the ELF auxargs
344 		 * array.  It has LINUX_AT_COUNT entries.
345 		 */
346 		destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
347 		destp = rounddown2(destp, sizeof(void *));
348 	}
349 
350 	vectp = (char **)destp;
351 
352 	/*
353 	 * Allocate room for the argv[] and env vectors including the
354 	 * terminating NULL pointers.
355 	 */
356 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
357 
358 	/*
359 	 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
360 	 * One "long argc" will be prepended later.
361 	 */
362 	vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
363 
364 	/* vectp also becomes our initial stack base. */
365 	*stack_base = (uintptr_t)vectp;
366 
367 	stringp = imgp->args->begin_argv;
368 	argc = imgp->args->argc;
369 	envc = imgp->args->envc;
370 
371 	/* Copy out strings - arguments and environment. */
372 	error = copyout(stringp, (void *)ustringp,
373 	    ARG_MAX - imgp->args->stringspace);
374 	if (error != 0)
375 		return (error);
376 
377 	/* Fill in "ps_strings" struct for ps, w, etc. */
378 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
379 	    suword(&arginfo->ps_nargvstr, argc) != 0)
380 		return (EFAULT);
381 
382 	/* Fill in argument portion of vector table. */
383 	for (; argc > 0; --argc) {
384 		if (suword(vectp++, ustringp) != 0)
385 			return (EFAULT);
386 		while (*stringp++ != 0)
387 			ustringp++;
388 		ustringp++;
389 	}
390 
391 	/* A null vector table pointer separates the argp's from the envp's. */
392 	if (suword(vectp++, 0) != 0)
393 		return (EFAULT);
394 
395 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
396 	    suword(&arginfo->ps_nenvstr, envc) != 0)
397 		return (EFAULT);
398 
399 	/* Fill in environment portion of vector table. */
400 	for (; envc > 0; --envc) {
401 		if (suword(vectp++, ustringp) != 0)
402 			return (EFAULT);
403 		while (*stringp++ != 0)
404 			ustringp++;
405 		ustringp++;
406 	}
407 
408 	/* The end of the vector table is a null pointer. */
409 	if (suword(vectp, 0) != 0)
410 		return (EFAULT);
411 
412 	if (imgp->auxargs) {
413 		vectp++;
414 		error = imgp->sysent->sv_copyout_auxargs(imgp,
415 		    (uintptr_t)vectp);
416 		if (error != 0)
417 			return (error);
418 	}
419 
420 	return (0);
421 }
422 
423 /*
424  * Reset registers to default values on exec.
425  */
426 static void
427 linux_exec_setregs(struct thread *td, struct image_params *imgp,
428     uintptr_t stack)
429 {
430 	struct trapframe *regs;
431 	struct pcb *pcb;
432 	register_t saved_rflags;
433 
434 	regs = td->td_frame;
435 	pcb = td->td_pcb;
436 
437 	if (td->td_proc->p_md.md_ldt != NULL)
438 		user_ldt_free(td);
439 
440 	pcb->pcb_fsbase = 0;
441 	pcb->pcb_gsbase = 0;
442 	clear_pcb_flags(pcb, PCB_32BIT);
443 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
444 	set_pcb_flags(pcb, PCB_FULL_IRET);
445 
446 	saved_rflags = regs->tf_rflags & PSL_T;
447 	bzero((char *)regs, sizeof(struct trapframe));
448 	regs->tf_rip = imgp->entry_addr;
449 	regs->tf_rsp = stack;
450 	regs->tf_rflags = PSL_USER | saved_rflags;
451 	regs->tf_ss = _udatasel;
452 	regs->tf_cs = _ucodesel;
453 	regs->tf_ds = _udatasel;
454 	regs->tf_es = _udatasel;
455 	regs->tf_fs = _ufssel;
456 	regs->tf_gs = _ugssel;
457 	regs->tf_flags = TF_HASSEGS;
458 
459 	/*
460 	 * Reset the hardware debug registers if they were in use.
461 	 * They won't have any meaning for the newly exec'd process.
462 	 */
463 	if (pcb->pcb_flags & PCB_DBREGS) {
464 		pcb->pcb_dr0 = 0;
465 		pcb->pcb_dr1 = 0;
466 		pcb->pcb_dr2 = 0;
467 		pcb->pcb_dr3 = 0;
468 		pcb->pcb_dr6 = 0;
469 		pcb->pcb_dr7 = 0;
470 		if (pcb == curpcb) {
471 			/*
472 			 * Clear the debug registers on the running
473 			 * CPU, otherwise they will end up affecting
474 			 * the next process we switch to.
475 			 */
476 			reset_dbregs();
477 		}
478 		clear_pcb_flags(pcb, PCB_DBREGS);
479 	}
480 
481 	/*
482 	 * Drop the FP state if we hold it, so that the process gets a
483 	 * clean FP state if it uses the FPU again.
484 	 */
485 	fpstate_drop(td);
486 }
487 
488 /*
489  * Copied from amd64/amd64/machdep.c
490  *
491  * XXX fpu state need? don't think so
492  */
493 int
494 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
495 {
496 	struct proc *p;
497 	struct l_ucontext uc;
498 	struct l_sigcontext *context;
499 	struct trapframe *regs;
500 	unsigned long rflags;
501 	int error;
502 	ksiginfo_t ksi;
503 
504 	regs = td->td_frame;
505 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
506 	if (error != 0)
507 		return (error);
508 
509 	p = td->td_proc;
510 	context = &uc.uc_mcontext;
511 	rflags = context->sc_rflags;
512 
513 	/*
514 	 * Don't allow users to change privileged or reserved flags.
515 	 */
516 	/*
517 	 * XXX do allow users to change the privileged flag PSL_RF.
518 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
519 	 * should sometimes set it there too.  tf_rflags is kept in
520 	 * the signal context during signal handling and there is no
521 	 * other place to remember it, so the PSL_RF bit may be
522 	 * corrupted by the signal handler without us knowing.
523 	 * Corruption of the PSL_RF bit at worst causes one more or
524 	 * one less debugger trap, so allowing it is fairly harmless.
525 	 */
526 
527 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
528 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
529 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
530 		return (EINVAL);
531 	}
532 
533 	/*
534 	 * Don't allow users to load a valid privileged %cs.  Let the
535 	 * hardware check for invalid selectors, excess privilege in
536 	 * other selectors, invalid %eip's and invalid %esp's.
537 	 */
538 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
539 	if (!CS_SECURE(context->sc_cs)) {
540 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
541 		ksiginfo_init_trap(&ksi);
542 		ksi.ksi_signo = SIGBUS;
543 		ksi.ksi_code = BUS_OBJERR;
544 		ksi.ksi_trapno = T_PROTFLT;
545 		ksi.ksi_addr = (void *)regs->tf_rip;
546 		trapsignal(td, &ksi);
547 		return (EINVAL);
548 	}
549 
550 	PROC_LOCK(p);
551 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
552 	SIG_CANTMASK(td->td_sigmask);
553 	signotify(td);
554 	PROC_UNLOCK(p);
555 
556 	regs->tf_rdi    = context->sc_rdi;
557 	regs->tf_rsi    = context->sc_rsi;
558 	regs->tf_rdx    = context->sc_rdx;
559 	regs->tf_rbp    = context->sc_rbp;
560 	regs->tf_rbx    = context->sc_rbx;
561 	regs->tf_rcx    = context->sc_rcx;
562 	regs->tf_rax    = context->sc_rax;
563 	regs->tf_rip    = context->sc_rip;
564 	regs->tf_rsp    = context->sc_rsp;
565 	regs->tf_r8     = context->sc_r8;
566 	regs->tf_r9     = context->sc_r9;
567 	regs->tf_r10    = context->sc_r10;
568 	regs->tf_r11    = context->sc_r11;
569 	regs->tf_r12    = context->sc_r12;
570 	regs->tf_r13    = context->sc_r13;
571 	regs->tf_r14    = context->sc_r14;
572 	regs->tf_r15    = context->sc_r15;
573 	regs->tf_cs     = context->sc_cs;
574 	regs->tf_err    = context->sc_err;
575 	regs->tf_rflags = rflags;
576 
577 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
578 	return (EJUSTRETURN);
579 }
580 
581 /*
582  * copied from amd64/amd64/machdep.c
583  *
584  * Send an interrupt to process.
585  */
586 static void
587 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
588 {
589 	struct l_rt_sigframe sf, *sfp;
590 	struct proc *p;
591 	struct thread *td;
592 	struct sigacts *psp;
593 	caddr_t sp;
594 	struct trapframe *regs;
595 	int sig, code;
596 	int oonstack;
597 
598 	td = curthread;
599 	p = td->td_proc;
600 	PROC_LOCK_ASSERT(p, MA_OWNED);
601 	sig = ksi->ksi_signo;
602 	psp = p->p_sigacts;
603 	code = ksi->ksi_code;
604 	mtx_assert(&psp->ps_mtx, MA_OWNED);
605 	regs = td->td_frame;
606 	oonstack = sigonstack(regs->tf_rsp);
607 
608 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
609 	    catcher, sig, mask, code);
610 
611 	/* Allocate space for the signal handler context. */
612 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
613 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
614 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
615 		    sizeof(struct l_rt_sigframe);
616 	} else
617 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
618 	/* Align to 16 bytes. */
619 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
620 	mtx_unlock(&psp->ps_mtx);
621 
622 	/* Translate the signal. */
623 	sig = bsd_to_linux_signal(sig);
624 
625 	/* Save user context. */
626 	bzero(&sf, sizeof(sf));
627 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
628 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
629 
630 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
631 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
632 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
633 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
634 	PROC_UNLOCK(p);
635 
636 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
637 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
638 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
639 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
640 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
641 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
642 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
643 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
644 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
645 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
646 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
647 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
648 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
649 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
650 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
651 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
652 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
653 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
654 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
655 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
656 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
657 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
658 
659 	/* Build the argument list for the signal handler. */
660 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
661 	regs->tf_rax = 0;
662 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
663 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
664 
665 	sf.sf_handler = catcher;
666 	/* Fill in POSIX parts. */
667 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
668 
669 	/* Copy the sigframe out to the user's stack. */
670 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
671 		PROC_LOCK(p);
672 		sigexit(td, SIGILL);
673 	}
674 
675 	regs->tf_rsp = (long)sfp;
676 	regs->tf_rip = linux_rt_sigcode;
677 	regs->tf_rflags &= ~(PSL_T | PSL_D);
678 	regs->tf_cs = _ucodesel;
679 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
680 	PROC_LOCK(p);
681 	mtx_lock(&psp->ps_mtx);
682 }
683 
684 #define	LINUX_VSYSCALL_START		(-10UL << 20)
685 #define	LINUX_VSYSCALL_SZ		1024
686 
687 const unsigned long linux_vsyscall_vector[] = {
688 	LINUX_SYS_gettimeofday,
689 	LINUX_SYS_linux_time,
690 	LINUX_SYS_linux_getcpu,
691 };
692 
693 static int
694 linux_vsyscall(struct thread *td)
695 {
696 	struct trapframe *frame;
697 	uint64_t retqaddr;
698 	int code, traced;
699 	int error;
700 
701 	frame = td->td_frame;
702 
703 	/* Check %rip for vsyscall area. */
704 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
705 		return (EINVAL);
706 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
707 		return (EINVAL);
708 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
709 	if (code >= nitems(linux_vsyscall_vector))
710 		return (EINVAL);
711 
712 	/*
713 	 * vsyscall called as callq *(%rax), so we must
714 	 * use return address from %rsp and also fixup %rsp.
715 	 */
716 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
717 	if (error)
718 		return (error);
719 
720 	frame->tf_rip = retqaddr;
721 	frame->tf_rax = linux_vsyscall_vector[code];
722 	frame->tf_rsp += 8;
723 
724 	traced = (frame->tf_flags & PSL_T);
725 
726 	amd64_syscall(td, traced);
727 
728 	return (0);
729 }
730 
731 struct sysentvec elf_linux_sysvec = {
732 	.sv_size	= LINUX_SYS_MAXSYSCALL,
733 	.sv_table	= linux_sysent,
734 	.sv_transtrap	= linux_translate_traps,
735 	.sv_fixup	= linux_fixup_elf,
736 	.sv_sendsig	= linux_rt_sendsig,
737 	.sv_sigcode	= &_binary_linux_locore_o_start,
738 	.sv_szsigcode	= &linux_szsigcode,
739 	.sv_name	= "Linux ELF64",
740 	.sv_coredump	= elf64_coredump,
741 	.sv_imgact_try	= linux_exec_imgact_try,
742 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
743 	.sv_minuser	= VM_MIN_ADDRESS,
744 	.sv_maxuser	= VM_MAXUSER_ADDRESS_LA48,
745 	.sv_usrstack	= USRSTACK_LA48,
746 	.sv_psstrings	= PS_STRINGS_LA48,
747 	.sv_stackprot	= VM_PROT_ALL,
748 	.sv_copyout_auxargs = linux_copyout_auxargs,
749 	.sv_copyout_strings = linux_copyout_strings,
750 	.sv_setregs	= linux_exec_setregs,
751 	.sv_fixlimit	= NULL,
752 	.sv_maxssiz	= NULL,
753 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
754 	.sv_set_syscall_retval = linux_set_syscall_retval,
755 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
756 	.sv_syscallnames = NULL,
757 	.sv_shared_page_base = SHAREDPAGE_LA48,
758 	.sv_shared_page_len = PAGE_SIZE,
759 	.sv_schedtail	= linux_schedtail,
760 	.sv_thread_detach = linux_thread_detach,
761 	.sv_trap	= linux_vsyscall,
762 };
763 
764 static void
765 linux_vdso_install(void *param)
766 {
767 
768 	amd64_lower_shared_page(&elf_linux_sysvec);
769 
770 	linux_szsigcode = (&_binary_linux_locore_o_end -
771 	    &_binary_linux_locore_o_start);
772 
773 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
774 		panic("Linux invalid vdso size\n");
775 
776 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
777 
778 	linux_shared_page_obj = __elfN(linux_shared_page_init)
779 	    (&linux_shared_page_mapping);
780 
781 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
782 
783 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
784 	    linux_szsigcode);
785 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
786 
787 	linux_kplatform = linux_shared_page_mapping +
788 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
789 }
790 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
791     linux_vdso_install, NULL);
792 
793 static void
794 linux_vdso_deinstall(void *param)
795 {
796 
797 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
798 }
799 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
800     linux_vdso_deinstall, NULL);
801 
802 static char GNULINUX_ABI_VENDOR[] = "GNU";
803 static int GNULINUX_ABI_DESC = 0;
804 
805 static bool
806 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
807 {
808 	const Elf32_Word *desc;
809 	uintptr_t p;
810 
811 	p = (uintptr_t)(note + 1);
812 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
813 
814 	desc = (const Elf32_Word *)p;
815 	if (desc[0] != GNULINUX_ABI_DESC)
816 		return (false);
817 
818 	/*
819 	 * For Linux we encode osrel using the Linux convention of
820 	 * 	(version << 16) | (major << 8) | (minor)
821 	 * See macro in linux_mib.h
822 	 */
823 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
824 
825 	return (true);
826 }
827 
828 static Elf_Brandnote linux64_brandnote = {
829 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
830 	.hdr.n_descsz	= 16,
831 	.hdr.n_type	= 1,
832 	.vendor		= GNULINUX_ABI_VENDOR,
833 	.flags		= BN_TRANSLATE_OSREL,
834 	.trans_osrel	= linux_trans_osrel
835 };
836 
837 static Elf64_Brandinfo linux_glibc2brand = {
838 	.brand		= ELFOSABI_LINUX,
839 	.machine	= EM_X86_64,
840 	.compat_3_brand	= "Linux",
841 	.emul_path	= linux_emul_path,
842 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
843 	.sysvec		= &elf_linux_sysvec,
844 	.interp_newpath	= NULL,
845 	.brand_note	= &linux64_brandnote,
846 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
847 };
848 
849 static Elf64_Brandinfo linux_glibc2brandshort = {
850 	.brand		= ELFOSABI_LINUX,
851 	.machine	= EM_X86_64,
852 	.compat_3_brand	= "Linux",
853 	.emul_path	= linux_emul_path,
854 	.interp_path	= "/lib64/ld-linux.so.2",
855 	.sysvec		= &elf_linux_sysvec,
856 	.interp_newpath	= NULL,
857 	.brand_note	= &linux64_brandnote,
858 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
859 };
860 
861 static Elf64_Brandinfo linux_muslbrand = {
862 	.brand		= ELFOSABI_LINUX,
863 	.machine	= EM_X86_64,
864 	.compat_3_brand	= "Linux",
865 	.emul_path	= linux_emul_path,
866 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
867 	.sysvec		= &elf_linux_sysvec,
868 	.interp_newpath	= NULL,
869 	.brand_note	= &linux64_brandnote,
870 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
871 };
872 
873 Elf64_Brandinfo *linux_brandlist[] = {
874 	&linux_glibc2brand,
875 	&linux_glibc2brandshort,
876 	&linux_muslbrand,
877 	NULL
878 };
879 
880 static int
881 linux64_elf_modevent(module_t mod, int type, void *data)
882 {
883 	Elf64_Brandinfo **brandinfo;
884 	int error;
885 	struct linux_ioctl_handler **lihp;
886 
887 	error = 0;
888 
889 	switch(type) {
890 	case MOD_LOAD:
891 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
892 		     ++brandinfo)
893 			if (elf64_insert_brand_entry(*brandinfo) < 0)
894 				error = EINVAL;
895 		if (error == 0) {
896 			SET_FOREACH(lihp, linux_ioctl_handler_set)
897 				linux_ioctl_register_handler(*lihp);
898 			stclohz = (stathz ? stathz : hz);
899 			if (bootverbose)
900 				printf("Linux x86-64 ELF exec handler installed\n");
901 		} else
902 			printf("cannot insert Linux x86-64 ELF brand handler\n");
903 		break;
904 	case MOD_UNLOAD:
905 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
906 		     ++brandinfo)
907 			if (elf64_brand_inuse(*brandinfo))
908 				error = EBUSY;
909 		if (error == 0) {
910 			for (brandinfo = &linux_brandlist[0];
911 			     *brandinfo != NULL; ++brandinfo)
912 				if (elf64_remove_brand_entry(*brandinfo) < 0)
913 					error = EINVAL;
914 		}
915 		if (error == 0) {
916 			SET_FOREACH(lihp, linux_ioctl_handler_set)
917 				linux_ioctl_unregister_handler(*lihp);
918 			if (bootverbose)
919 				printf("Linux ELF exec handler removed\n");
920 		} else
921 			printf("Could not deinstall ELF interpreter entry\n");
922 		break;
923 	default:
924 		return (EOPNOTSUPP);
925 	}
926 	return (error);
927 }
928 
929 static moduledata_t linux64_elf_mod = {
930 	"linux64elf",
931 	linux64_elf_modevent,
932 	0
933 };
934 
935 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
936 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
937 FEATURE(linux64, "Linux 64bit support");
938