xref: /freebsd-13.1/sys/amd64/linux/linux_sysvec.c (revision 5caa67fa)
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 register_t * linux_copyout_strings(struct image_params *imgp);
100 static int	linux_fixup_elf(register_t **stack_base,
101 		    struct image_params *iparams);
102 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
103 static void	linux_vdso_install(void *param);
104 static void	linux_vdso_deinstall(void *param);
105 static void	linux_set_syscall_retval(struct thread *td, int error);
106 static int	linux_fetch_syscall_args(struct thread *td);
107 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
108 		    u_long stack);
109 static int	linux_vsyscall(struct thread *td);
110 
111 #define LINUX_T_UNKNOWN  255
112 static int _bsd_to_linux_trapcode[] = {
113 	LINUX_T_UNKNOWN,	/* 0 */
114 	6,			/* 1  T_PRIVINFLT */
115 	LINUX_T_UNKNOWN,	/* 2 */
116 	3,			/* 3  T_BPTFLT */
117 	LINUX_T_UNKNOWN,	/* 4 */
118 	LINUX_T_UNKNOWN,	/* 5 */
119 	16,			/* 6  T_ARITHTRAP */
120 	254,			/* 7  T_ASTFLT */
121 	LINUX_T_UNKNOWN,	/* 8 */
122 	13,			/* 9  T_PROTFLT */
123 	1,			/* 10 T_TRCTRAP */
124 	LINUX_T_UNKNOWN,	/* 11 */
125 	14,			/* 12 T_PAGEFLT */
126 	LINUX_T_UNKNOWN,	/* 13 */
127 	17,			/* 14 T_ALIGNFLT */
128 	LINUX_T_UNKNOWN,	/* 15 */
129 	LINUX_T_UNKNOWN,	/* 16 */
130 	LINUX_T_UNKNOWN,	/* 17 */
131 	0,			/* 18 T_DIVIDE */
132 	2,			/* 19 T_NMI */
133 	4,			/* 20 T_OFLOW */
134 	5,			/* 21 T_BOUND */
135 	7,			/* 22 T_DNA */
136 	8,			/* 23 T_DOUBLEFLT */
137 	9,			/* 24 T_FPOPFLT */
138 	10,			/* 25 T_TSSFLT */
139 	11,			/* 26 T_SEGNPFLT */
140 	12,			/* 27 T_STKFLT */
141 	18,			/* 28 T_MCHK */
142 	19,			/* 29 T_XMMFLT */
143 	15			/* 30 T_RESERVED */
144 };
145 #define bsd_to_linux_trapcode(code) \
146     ((code)<nitems(_bsd_to_linux_trapcode)? \
147      _bsd_to_linux_trapcode[(code)]: \
148      LINUX_T_UNKNOWN)
149 
150 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
151 LINUX_VDSO_SYM_CHAR(linux_platform);
152 
153 /*
154  * If FreeBSD & Linux have a difference of opinion about what a trap
155  * means, deal with it here.
156  *
157  * MPSAFE
158  */
159 static int
160 linux_translate_traps(int signal, int trap_code)
161 {
162 
163 	if (signal != SIGBUS)
164 		return (signal);
165 	switch (trap_code) {
166 	case T_PROTFLT:
167 	case T_TSSFLT:
168 	case T_DOUBLEFLT:
169 	case T_PAGEFLT:
170 		return (SIGSEGV);
171 	default:
172 		return (signal);
173 	}
174 }
175 
176 static int
177 linux_fetch_syscall_args(struct thread *td)
178 {
179 	struct proc *p;
180 	struct trapframe *frame;
181 	struct syscall_args *sa;
182 
183 	p = td->td_proc;
184 	frame = td->td_frame;
185 	sa = &td->td_sa;
186 
187 	sa->args[0] = frame->tf_rdi;
188 	sa->args[1] = frame->tf_rsi;
189 	sa->args[2] = frame->tf_rdx;
190 	sa->args[3] = frame->tf_rcx;
191 	sa->args[4] = frame->tf_r8;
192 	sa->args[5] = frame->tf_r9;
193 	sa->code = frame->tf_rax;
194 
195 	if (sa->code >= p->p_sysent->sv_size)
196 		/* nosys */
197 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
198 	else
199 		sa->callp = &p->p_sysent->sv_table[sa->code];
200 	sa->narg = sa->callp->sy_narg;
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 	 /* Restore all registers. */
222 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
223 }
224 
225 static void
226 linux_copyout_auxargs(struct image_params *imgp, u_long *base)
227 {
228 	Elf_Auxargs *args;
229 	Elf_Auxinfo *argarray, *pos;
230 	u_long auxlen;
231 	struct proc *p;
232 	int issetugid;
233 
234 	p = imgp->proc;
235 	args = (Elf64_Auxargs *)imgp->auxargs;
236 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
237 	    M_WAITOK | M_ZERO);
238 
239 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
240 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
241 	    imgp->proc->p_sysent->sv_shared_page_base);
242 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
243 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
244 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
245 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
246 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
247 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
248 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
249 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
250 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
251 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
252 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
253 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
254 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
255 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
256 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
257 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
258 	if (imgp->execpathp != 0)
259 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
260 	if (args->execfd != -1)
261 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
262 	AUXARGS_ENTRY(pos, AT_NULL, 0);
263 
264 	free(imgp->auxargs, M_TEMP);
265 	imgp->auxargs = NULL;
266 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
267 
268 	auxlen = sizeof(*argarray) * (pos - argarray);
269 	*base -= auxlen;
270 	copyout(argarray, (void *)*base, auxlen);
271 	free(argarray, M_TEMP);
272 }
273 
274 static int
275 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
276 {
277 	Elf_Addr *base;
278 
279 	base = (Elf64_Addr *)*stack_base;
280 	base--;
281 	if (suword(base, (uint64_t)imgp->args->argc) == -1)
282 		return (EFAULT);
283 
284 	*stack_base = (register_t *)base;
285 	return (0);
286 }
287 
288 /*
289  * Copy strings out to the new process address space, constructing new arg
290  * and env vector tables. Return a pointer to the base so that it can be used
291  * as the initial stack pointer.
292  */
293 static register_t *
294 linux_copyout_strings(struct image_params *imgp)
295 {
296 	int argc, envc;
297 	char **vectp;
298 	char *stringp, *destp;
299 	register_t *stack_base;
300 	struct ps_strings *arginfo;
301 	char canary[LINUX_AT_RANDOM_LEN];
302 	size_t execpath_len;
303 	struct proc *p;
304 
305 	/* Calculate string base and vector table pointers. */
306 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
307 		execpath_len = strlen(imgp->execpath) + 1;
308 	else
309 		execpath_len = 0;
310 
311 	p = imgp->proc;
312 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
313 	destp = (caddr_t)arginfo - SPARE_USRSPACE -
314 	    roundup(sizeof(canary), sizeof(char *)) -
315 	    roundup(execpath_len, sizeof(char *)) -
316 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
317 
318 	if (execpath_len != 0) {
319 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
320 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
321 	}
322 
323 	/* Prepare the canary for SSP. */
324 	arc4rand(canary, sizeof(canary), 0);
325 	imgp->canary = (uintptr_t)arginfo -
326 	    roundup(execpath_len, sizeof(char *)) -
327 	    roundup(sizeof(canary), sizeof(char *));
328 	copyout(canary, (void *)imgp->canary, sizeof(canary));
329 
330 	vectp = (char **)destp;
331 
332 	/*
333 	 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
334 	 * One "long argc" will be prepended later.
335 	 */
336 	vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
337 
338 	if (imgp->auxargs)
339 		imgp->sysent->sv_copyout_auxargs(imgp, (u_long *)&vectp);
340 
341 	/*
342 	 * Allocate room for the argv[] and env vectors including the
343 	 * terminating NULL pointers.
344 	 */
345 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
346 
347 	/* vectp also becomes our initial stack base. */
348 	stack_base = (register_t *)vectp;
349 
350 	stringp = imgp->args->begin_argv;
351 	argc = imgp->args->argc;
352 	envc = imgp->args->envc;
353 
354 	/* Copy out strings - arguments and environment. */
355 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
356 
357 	/* Fill in "ps_strings" struct for ps, w, etc. */
358 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
359 	suword(&arginfo->ps_nargvstr, argc);
360 
361 	/* Fill in argument portion of vector table. */
362 	for (; argc > 0; --argc) {
363 		suword(vectp++, (long)(intptr_t)destp);
364 		while (*stringp++ != 0)
365 			destp++;
366 		destp++;
367 	}
368 
369 	/* A null vector table pointer separates the argp's from the envp's. */
370 	suword(vectp++, 0);
371 
372 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
373 	suword(&arginfo->ps_nenvstr, envc);
374 
375 	/* Fill in environment portion of vector table. */
376 	for (; envc > 0; --envc) {
377 		suword(vectp++, (long)(intptr_t)destp);
378 		while (*stringp++ != 0)
379 			destp++;
380 		destp++;
381 	}
382 
383 	/* The end of the vector table is a null pointer. */
384 	suword(vectp, 0);
385 	return (stack_base);
386 }
387 
388 /*
389  * Reset registers to default values on exec.
390  */
391 static void
392 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
393 {
394 	struct trapframe *regs;
395 	struct pcb *pcb;
396 	register_t saved_rflags;
397 
398 	regs = td->td_frame;
399 	pcb = td->td_pcb;
400 
401 	if (td->td_proc->p_md.md_ldt != NULL)
402 		user_ldt_free(td);
403 
404 	pcb->pcb_fsbase = 0;
405 	pcb->pcb_gsbase = 0;
406 	clear_pcb_flags(pcb, PCB_32BIT);
407 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
408 	set_pcb_flags(pcb, PCB_FULL_IRET);
409 
410 	saved_rflags = regs->tf_rflags & PSL_T;
411 	bzero((char *)regs, sizeof(struct trapframe));
412 	regs->tf_rip = imgp->entry_addr;
413 	regs->tf_rsp = stack;
414 	regs->tf_rflags = PSL_USER | saved_rflags;
415 	regs->tf_ss = _udatasel;
416 	regs->tf_cs = _ucodesel;
417 	regs->tf_ds = _udatasel;
418 	regs->tf_es = _udatasel;
419 	regs->tf_fs = _ufssel;
420 	regs->tf_gs = _ugssel;
421 	regs->tf_flags = TF_HASSEGS;
422 
423 	/*
424 	 * Reset the hardware debug registers if they were in use.
425 	 * They won't have any meaning for the newly exec'd process.
426 	 */
427 	if (pcb->pcb_flags & PCB_DBREGS) {
428 		pcb->pcb_dr0 = 0;
429 		pcb->pcb_dr1 = 0;
430 		pcb->pcb_dr2 = 0;
431 		pcb->pcb_dr3 = 0;
432 		pcb->pcb_dr6 = 0;
433 		pcb->pcb_dr7 = 0;
434 		if (pcb == curpcb) {
435 			/*
436 			 * Clear the debug registers on the running
437 			 * CPU, otherwise they will end up affecting
438 			 * the next process we switch to.
439 			 */
440 			reset_dbregs();
441 		}
442 		clear_pcb_flags(pcb, PCB_DBREGS);
443 	}
444 
445 	/*
446 	 * Drop the FP state if we hold it, so that the process gets a
447 	 * clean FP state if it uses the FPU again.
448 	 */
449 	fpstate_drop(td);
450 }
451 
452 /*
453  * Copied from amd64/amd64/machdep.c
454  *
455  * XXX fpu state need? don't think so
456  */
457 int
458 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
459 {
460 	struct proc *p;
461 	struct l_ucontext uc;
462 	struct l_sigcontext *context;
463 	struct trapframe *regs;
464 	unsigned long rflags;
465 	int error;
466 	ksiginfo_t ksi;
467 
468 	regs = td->td_frame;
469 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
470 	if (error != 0)
471 		return (error);
472 
473 	p = td->td_proc;
474 	context = &uc.uc_mcontext;
475 	rflags = context->sc_rflags;
476 
477 	/*
478 	 * Don't allow users to change privileged or reserved flags.
479 	 */
480 	/*
481 	 * XXX do allow users to change the privileged flag PSL_RF.
482 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
483 	 * should sometimes set it there too.  tf_rflags is kept in
484 	 * the signal context during signal handling and there is no
485 	 * other place to remember it, so the PSL_RF bit may be
486 	 * corrupted by the signal handler without us knowing.
487 	 * Corruption of the PSL_RF bit at worst causes one more or
488 	 * one less debugger trap, so allowing it is fairly harmless.
489 	 */
490 
491 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
492 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
493 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
494 		return (EINVAL);
495 	}
496 
497 	/*
498 	 * Don't allow users to load a valid privileged %cs.  Let the
499 	 * hardware check for invalid selectors, excess privilege in
500 	 * other selectors, invalid %eip's and invalid %esp's.
501 	 */
502 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
503 	if (!CS_SECURE(context->sc_cs)) {
504 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
505 		ksiginfo_init_trap(&ksi);
506 		ksi.ksi_signo = SIGBUS;
507 		ksi.ksi_code = BUS_OBJERR;
508 		ksi.ksi_trapno = T_PROTFLT;
509 		ksi.ksi_addr = (void *)regs->tf_rip;
510 		trapsignal(td, &ksi);
511 		return (EINVAL);
512 	}
513 
514 	PROC_LOCK(p);
515 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
516 	SIG_CANTMASK(td->td_sigmask);
517 	signotify(td);
518 	PROC_UNLOCK(p);
519 
520 	regs->tf_rdi    = context->sc_rdi;
521 	regs->tf_rsi    = context->sc_rsi;
522 	regs->tf_rdx    = context->sc_rdx;
523 	regs->tf_rbp    = context->sc_rbp;
524 	regs->tf_rbx    = context->sc_rbx;
525 	regs->tf_rcx    = context->sc_rcx;
526 	regs->tf_rax    = context->sc_rax;
527 	regs->tf_rip    = context->sc_rip;
528 	regs->tf_rsp    = context->sc_rsp;
529 	regs->tf_r8     = context->sc_r8;
530 	regs->tf_r9     = context->sc_r9;
531 	regs->tf_r10    = context->sc_r10;
532 	regs->tf_r11    = context->sc_r11;
533 	regs->tf_r12    = context->sc_r12;
534 	regs->tf_r13    = context->sc_r13;
535 	regs->tf_r14    = context->sc_r14;
536 	regs->tf_r15    = context->sc_r15;
537 	regs->tf_cs     = context->sc_cs;
538 	regs->tf_err    = context->sc_err;
539 	regs->tf_rflags = rflags;
540 
541 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
542 	return (EJUSTRETURN);
543 }
544 
545 /*
546  * copied from amd64/amd64/machdep.c
547  *
548  * Send an interrupt to process.
549  */
550 static void
551 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
552 {
553 	struct l_rt_sigframe sf, *sfp;
554 	struct proc *p;
555 	struct thread *td;
556 	struct sigacts *psp;
557 	caddr_t sp;
558 	struct trapframe *regs;
559 	int sig, code;
560 	int oonstack;
561 
562 	td = curthread;
563 	p = td->td_proc;
564 	PROC_LOCK_ASSERT(p, MA_OWNED);
565 	sig = ksi->ksi_signo;
566 	psp = p->p_sigacts;
567 	code = ksi->ksi_code;
568 	mtx_assert(&psp->ps_mtx, MA_OWNED);
569 	regs = td->td_frame;
570 	oonstack = sigonstack(regs->tf_rsp);
571 
572 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
573 	    catcher, sig, mask, code);
574 
575 	/* Allocate space for the signal handler context. */
576 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
577 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
578 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
579 		    sizeof(struct l_rt_sigframe);
580 	} else
581 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
582 	/* Align to 16 bytes. */
583 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
584 	mtx_unlock(&psp->ps_mtx);
585 
586 	/* Translate the signal. */
587 	sig = bsd_to_linux_signal(sig);
588 
589 	/* Save user context. */
590 	bzero(&sf, sizeof(sf));
591 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
592 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
593 
594 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
595 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
596 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
597 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
598 	PROC_UNLOCK(p);
599 
600 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
601 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
602 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
603 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
604 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
605 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
606 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
607 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
608 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
609 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
610 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
611 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
612 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
613 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
614 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
615 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
616 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
617 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
618 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
619 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
620 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
621 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
622 
623 	/* Build the argument list for the signal handler. */
624 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
625 	regs->tf_rax = 0;
626 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
627 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
628 
629 	sf.sf_handler = catcher;
630 	/* Fill in POSIX parts. */
631 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
632 
633 	/* Copy the sigframe out to the user's stack. */
634 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
635 		PROC_LOCK(p);
636 		sigexit(td, SIGILL);
637 	}
638 
639 	regs->tf_rsp = (long)sfp;
640 	regs->tf_rip = linux_rt_sigcode;
641 	regs->tf_rflags &= ~(PSL_T | PSL_D);
642 	regs->tf_cs = _ucodesel;
643 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
644 	PROC_LOCK(p);
645 	mtx_lock(&psp->ps_mtx);
646 }
647 
648 #define	LINUX_VSYSCALL_START		(-10UL << 20)
649 #define	LINUX_VSYSCALL_SZ		1024
650 
651 const unsigned long linux_vsyscall_vector[] = {
652 	LINUX_SYS_gettimeofday,
653 	LINUX_SYS_linux_time,
654 				/* getcpu not implemented */
655 };
656 
657 static int
658 linux_vsyscall(struct thread *td)
659 {
660 	struct trapframe *frame;
661 	uint64_t retqaddr;
662 	int code, traced;
663 	int error;
664 
665 	frame = td->td_frame;
666 
667 	/* Check %rip for vsyscall area. */
668 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
669 		return (EINVAL);
670 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
671 		return (EINVAL);
672 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
673 	if (code >= nitems(linux_vsyscall_vector))
674 		return (EINVAL);
675 
676 	/*
677 	 * vsyscall called as callq *(%rax), so we must
678 	 * use return address from %rsp and also fixup %rsp.
679 	 */
680 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
681 	if (error)
682 		return (error);
683 
684 	frame->tf_rip = retqaddr;
685 	frame->tf_rax = linux_vsyscall_vector[code];
686 	frame->tf_rsp += 8;
687 
688 	traced = (frame->tf_flags & PSL_T);
689 
690 	amd64_syscall(td, traced);
691 
692 	return (0);
693 }
694 
695 struct sysentvec elf_linux_sysvec = {
696 	.sv_size	= LINUX_SYS_MAXSYSCALL,
697 	.sv_table	= linux_sysent,
698 	.sv_errsize	= ELAST + 1,
699 	.sv_errtbl	= linux_errtbl,
700 	.sv_transtrap	= linux_translate_traps,
701 	.sv_fixup	= linux_fixup_elf,
702 	.sv_sendsig	= linux_rt_sendsig,
703 	.sv_sigcode	= &_binary_linux_locore_o_start,
704 	.sv_szsigcode	= &linux_szsigcode,
705 	.sv_name	= "Linux ELF64",
706 	.sv_coredump	= elf64_coredump,
707 	.sv_imgact_try	= linux_exec_imgact_try,
708 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
709 	.sv_minuser	= VM_MIN_ADDRESS,
710 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
711 	.sv_usrstack	= USRSTACK,
712 	.sv_psstrings	= PS_STRINGS,
713 	.sv_stackprot	= VM_PROT_ALL,
714 	.sv_copyout_auxargs = linux_copyout_auxargs,
715 	.sv_copyout_strings = linux_copyout_strings,
716 	.sv_setregs	= linux_exec_setregs,
717 	.sv_fixlimit	= NULL,
718 	.sv_maxssiz	= NULL,
719 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
720 	.sv_set_syscall_retval = linux_set_syscall_retval,
721 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
722 	.sv_syscallnames = NULL,
723 	.sv_shared_page_base = SHAREDPAGE,
724 	.sv_shared_page_len = PAGE_SIZE,
725 	.sv_schedtail	= linux_schedtail,
726 	.sv_thread_detach = linux_thread_detach,
727 	.sv_trap	= linux_vsyscall,
728 };
729 
730 static void
731 linux_vdso_install(void *param)
732 {
733 
734 	amd64_lower_shared_page(&elf_linux_sysvec);
735 
736 	linux_szsigcode = (&_binary_linux_locore_o_end -
737 	    &_binary_linux_locore_o_start);
738 
739 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
740 		panic("Linux invalid vdso size\n");
741 
742 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
743 
744 	linux_shared_page_obj = __elfN(linux_shared_page_init)
745 	    (&linux_shared_page_mapping);
746 
747 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
748 
749 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
750 	    linux_szsigcode);
751 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
752 
753 	linux_kplatform = linux_shared_page_mapping +
754 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
755 }
756 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
757     linux_vdso_install, NULL);
758 
759 static void
760 linux_vdso_deinstall(void *param)
761 {
762 
763 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
764 }
765 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
766     linux_vdso_deinstall, NULL);
767 
768 static char GNULINUX_ABI_VENDOR[] = "GNU";
769 static int GNULINUX_ABI_DESC = 0;
770 
771 static bool
772 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
773 {
774 	const Elf32_Word *desc;
775 	uintptr_t p;
776 
777 	p = (uintptr_t)(note + 1);
778 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
779 
780 	desc = (const Elf32_Word *)p;
781 	if (desc[0] != GNULINUX_ABI_DESC)
782 		return (false);
783 
784 	/*
785 	 * For Linux we encode osrel using the Linux convention of
786 	 * 	(version << 16) | (major << 8) | (minor)
787 	 * See macro in linux_mib.h
788 	 */
789 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
790 
791 	return (true);
792 }
793 
794 static Elf_Brandnote linux64_brandnote = {
795 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
796 	.hdr.n_descsz	= 16,
797 	.hdr.n_type	= 1,
798 	.vendor		= GNULINUX_ABI_VENDOR,
799 	.flags		= BN_TRANSLATE_OSREL,
800 	.trans_osrel	= linux_trans_osrel
801 };
802 
803 static Elf64_Brandinfo linux_glibc2brand = {
804 	.brand		= ELFOSABI_LINUX,
805 	.machine	= EM_X86_64,
806 	.compat_3_brand	= "Linux",
807 	.emul_path	= "/compat/linux",
808 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
809 	.sysvec		= &elf_linux_sysvec,
810 	.interp_newpath	= NULL,
811 	.brand_note	= &linux64_brandnote,
812 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
813 };
814 
815 static Elf64_Brandinfo linux_glibc2brandshort = {
816 	.brand		= ELFOSABI_LINUX,
817 	.machine	= EM_X86_64,
818 	.compat_3_brand	= "Linux",
819 	.emul_path	= "/compat/linux",
820 	.interp_path	= "/lib64/ld-linux.so.2",
821 	.sysvec		= &elf_linux_sysvec,
822 	.interp_newpath	= NULL,
823 	.brand_note	= &linux64_brandnote,
824 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
825 };
826 
827 static Elf64_Brandinfo linux_muslbrand = {
828 	.brand		= ELFOSABI_LINUX,
829 	.machine	= EM_X86_64,
830 	.compat_3_brand	= "Linux",
831 	.emul_path	= "/compat/linux",
832 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
833 	.sysvec		= &elf_linux_sysvec,
834 	.interp_newpath	= NULL,
835 	.brand_note	= &linux64_brandnote,
836 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
837 };
838 
839 Elf64_Brandinfo *linux_brandlist[] = {
840 	&linux_glibc2brand,
841 	&linux_glibc2brandshort,
842 	&linux_muslbrand,
843 	NULL
844 };
845 
846 static int
847 linux64_elf_modevent(module_t mod, int type, void *data)
848 {
849 	Elf64_Brandinfo **brandinfo;
850 	int error;
851 	struct linux_ioctl_handler **lihp;
852 
853 	error = 0;
854 
855 	switch(type) {
856 	case MOD_LOAD:
857 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
858 		     ++brandinfo)
859 			if (elf64_insert_brand_entry(*brandinfo) < 0)
860 				error = EINVAL;
861 		if (error == 0) {
862 			SET_FOREACH(lihp, linux_ioctl_handler_set)
863 				linux_ioctl_register_handler(*lihp);
864 			stclohz = (stathz ? stathz : hz);
865 			if (bootverbose)
866 				printf("Linux x86-64 ELF exec handler installed\n");
867 		} else
868 			printf("cannot insert Linux x86-64 ELF brand handler\n");
869 		break;
870 	case MOD_UNLOAD:
871 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
872 		     ++brandinfo)
873 			if (elf64_brand_inuse(*brandinfo))
874 				error = EBUSY;
875 		if (error == 0) {
876 			for (brandinfo = &linux_brandlist[0];
877 			     *brandinfo != NULL; ++brandinfo)
878 				if (elf64_remove_brand_entry(*brandinfo) < 0)
879 					error = EINVAL;
880 		}
881 		if (error == 0) {
882 			SET_FOREACH(lihp, linux_ioctl_handler_set)
883 				linux_ioctl_unregister_handler(*lihp);
884 			if (bootverbose)
885 				printf("Linux ELF exec handler removed\n");
886 		} else
887 			printf("Could not deinstall ELF interpreter entry\n");
888 		break;
889 	default:
890 		return (EOPNOTSUPP);
891 	}
892 	return (error);
893 }
894 
895 static moduledata_t linux64_elf_mod = {
896 	"linux64elf",
897 	linux64_elf_modevent,
898 	0
899 };
900 
901 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
902 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
903 FEATURE(linux64, "Linux 64bit support");
904