xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 2617128a21bfb1b1e3065e54d46f9982d37c0bf8)
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 static int linux_szsigcode;
89 static vm_object_t linux_shared_page_obj;
90 static char *linux_shared_page_mapping;
91 extern char _binary_linux_locore_o_start;
92 extern char _binary_linux_locore_o_end;
93 
94 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
95 
96 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
97 
98 static int	linux_copyout_strings(struct image_params *imgp,
99 		    uintptr_t *stack_base);
100 static int	linux_fixup_elf(uintptr_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 		    uintptr_t 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 
201 	td->td_retval[0] = 0;
202 	return (0);
203 }
204 
205 static void
206 linux_set_syscall_retval(struct thread *td, int error)
207 {
208 	struct trapframe *frame;
209 
210 	frame = td->td_frame;
211 
212 	switch (error) {
213 	case 0:
214 		frame->tf_rax = td->td_retval[0];
215  		frame->tf_r10 = frame->tf_rcx;
216 		break;
217 
218 	case ERESTART:
219 		/*
220 		 * Reconstruct pc, we know that 'syscall' is 2 bytes,
221 		 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
222 		 * We saved this in tf_err.
223 		 *
224 		 */
225 		frame->tf_rip -= frame->tf_err;
226 		frame->tf_r10 = frame->tf_rcx;
227 		break;
228 
229 	case EJUSTRETURN:
230 		break;
231 
232 	default:
233 		frame->tf_rax = bsd_to_linux_errno(error);
234 		frame->tf_r10 = frame->tf_rcx;
235 		break;
236 	}
237 
238 	/*
239 	 * Differently from FreeBSD native ABI, on Linux only %rcx
240 	 * and %r11 values are not preserved across the syscall.
241 	 * Require full context restore to get all registers except
242 	 * those two restored at return to usermode.
243 	 *
244 	 * XXX: Would be great to be able to avoid PCB_FULL_IRET
245 	 *      for the error == 0 case.
246 	 */
247 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
248 }
249 
250 static int
251 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
252 {
253 	Elf_Auxargs *args;
254 	Elf_Auxinfo *argarray, *pos;
255 	struct proc *p;
256 	int error, issetugid;
257 
258 	p = imgp->proc;
259 	args = (Elf64_Auxargs *)imgp->auxargs;
260 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
261 	    M_WAITOK | M_ZERO);
262 
263 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
264 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
265 	    imgp->proc->p_sysent->sv_shared_page_base);
266 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
267 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
268 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
269 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
270 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
271 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
272 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
273 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
274 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
275 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
276 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
277 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
278 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
279 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
280 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
281 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
282 	if (imgp->execpathp != 0)
283 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
284 	if (args->execfd != -1)
285 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
286 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
287 	AUXARGS_ENTRY(pos, AT_NULL, 0);
288 
289 	free(imgp->auxargs, M_TEMP);
290 	imgp->auxargs = NULL;
291 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
292 
293 	error = copyout(argarray, (void *)base,
294 	    sizeof(*argarray) * LINUX_AT_COUNT);
295 	free(argarray, M_TEMP);
296 	return (error);
297 }
298 
299 static int
300 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
301 {
302 	Elf_Addr *base;
303 
304 	base = (Elf64_Addr *)*stack_base;
305 	base--;
306 	if (suword(base, (uint64_t)imgp->args->argc) == -1)
307 		return (EFAULT);
308 
309 	*stack_base = (uintptr_t)base;
310 	return (0);
311 }
312 
313 /*
314  * Copy strings out to the new process address space, constructing new arg
315  * and env vector tables. Return a pointer to the base so that it can be used
316  * as the initial stack pointer.
317  */
318 static int
319 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
320 {
321 	int argc, envc, error;
322 	char **vectp;
323 	char *stringp;
324 	uintptr_t destp, ustringp;
325 	struct ps_strings *arginfo;
326 	char canary[LINUX_AT_RANDOM_LEN];
327 	size_t execpath_len;
328 	struct proc *p;
329 
330 	/* Calculate string base and vector table pointers. */
331 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
332 		execpath_len = strlen(imgp->execpath) + 1;
333 	else
334 		execpath_len = 0;
335 
336 	p = imgp->proc;
337 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
338 	destp = (uintptr_t)arginfo;
339 
340 	if (execpath_len != 0) {
341 		destp -= execpath_len;
342 		destp = rounddown2(destp, sizeof(void *));
343 		imgp->execpathp = (void *)destp;
344 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
345 		if (error != 0)
346 			return (error);
347 	}
348 
349 	/* Prepare the canary for SSP. */
350 	arc4rand(canary, sizeof(canary), 0);
351 	destp -= roundup(sizeof(canary), sizeof(void *));
352 	imgp->canary = (void *)destp;
353 	error = copyout(canary, imgp->canary, sizeof(canary));
354 	if (error != 0)
355 		return (error);
356 
357 	/* Allocate room for the argument and environment strings. */
358 	destp -= ARG_MAX - imgp->args->stringspace;
359 	destp = rounddown2(destp, sizeof(void *));
360 	ustringp = destp;
361 
362 	if (imgp->auxargs) {
363 		/*
364 		 * Allocate room on the stack for the ELF auxargs
365 		 * array.  It has LINUX_AT_COUNT entries.
366 		 */
367 		destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
368 		destp = rounddown2(destp, sizeof(void *));
369 	}
370 
371 	vectp = (char **)destp;
372 
373 	/*
374 	 * Allocate room for the argv[] and env vectors including the
375 	 * terminating NULL pointers.
376 	 */
377 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
378 
379 	/*
380 	 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
381 	 * One "long argc" will be prepended later.
382 	 */
383 	vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
384 
385 	/* vectp also becomes our initial stack base. */
386 	*stack_base = (uintptr_t)vectp;
387 
388 	stringp = imgp->args->begin_argv;
389 	argc = imgp->args->argc;
390 	envc = imgp->args->envc;
391 
392 	/* Copy out strings - arguments and environment. */
393 	error = copyout(stringp, (void *)ustringp,
394 	    ARG_MAX - imgp->args->stringspace);
395 	if (error != 0)
396 		return (error);
397 
398 	/* Fill in "ps_strings" struct for ps, w, etc. */
399 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
400 	    suword(&arginfo->ps_nargvstr, argc) != 0)
401 		return (EFAULT);
402 
403 	/* Fill in argument portion of vector table. */
404 	for (; argc > 0; --argc) {
405 		if (suword(vectp++, ustringp) != 0)
406 			return (EFAULT);
407 		while (*stringp++ != 0)
408 			ustringp++;
409 		ustringp++;
410 	}
411 
412 	/* A null vector table pointer separates the argp's from the envp's. */
413 	if (suword(vectp++, 0) != 0)
414 		return (EFAULT);
415 
416 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
417 	    suword(&arginfo->ps_nenvstr, envc) != 0)
418 		return (EFAULT);
419 
420 	/* Fill in environment portion of vector table. */
421 	for (; envc > 0; --envc) {
422 		if (suword(vectp++, ustringp) != 0)
423 			return (EFAULT);
424 		while (*stringp++ != 0)
425 			ustringp++;
426 		ustringp++;
427 	}
428 
429 	/* The end of the vector table is a null pointer. */
430 	if (suword(vectp, 0) != 0)
431 		return (EFAULT);
432 
433 	if (imgp->auxargs) {
434 		vectp++;
435 		error = imgp->sysent->sv_copyout_auxargs(imgp,
436 		    (uintptr_t)vectp);
437 		if (error != 0)
438 			return (error);
439 	}
440 
441 	return (0);
442 }
443 
444 /*
445  * Reset registers to default values on exec.
446  */
447 static void
448 linux_exec_setregs(struct thread *td, struct image_params *imgp,
449     uintptr_t stack)
450 {
451 	struct trapframe *regs;
452 	struct pcb *pcb;
453 	register_t saved_rflags;
454 
455 	regs = td->td_frame;
456 	pcb = td->td_pcb;
457 
458 	if (td->td_proc->p_md.md_ldt != NULL)
459 		user_ldt_free(td);
460 
461 	pcb->pcb_fsbase = 0;
462 	pcb->pcb_gsbase = 0;
463 	clear_pcb_flags(pcb, PCB_32BIT);
464 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
465 	set_pcb_flags(pcb, PCB_FULL_IRET);
466 
467 	saved_rflags = regs->tf_rflags & PSL_T;
468 	bzero((char *)regs, sizeof(struct trapframe));
469 	regs->tf_rip = imgp->entry_addr;
470 	regs->tf_rsp = stack;
471 	regs->tf_rflags = PSL_USER | saved_rflags;
472 	regs->tf_ss = _udatasel;
473 	regs->tf_cs = _ucodesel;
474 	regs->tf_ds = _udatasel;
475 	regs->tf_es = _udatasel;
476 	regs->tf_fs = _ufssel;
477 	regs->tf_gs = _ugssel;
478 	regs->tf_flags = TF_HASSEGS;
479 
480 	x86_clear_dbregs(pcb);
481 
482 	/*
483 	 * Drop the FP state if we hold it, so that the process gets a
484 	 * clean FP state if it uses the FPU again.
485 	 */
486 	fpstate_drop(td);
487 }
488 
489 /*
490  * Copied from amd64/amd64/machdep.c
491  *
492  * XXX fpu state need? don't think so
493  */
494 int
495 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
496 {
497 	struct proc *p;
498 	struct l_ucontext uc;
499 	struct l_sigcontext *context;
500 	struct trapframe *regs;
501 	unsigned long rflags;
502 	int error;
503 	ksiginfo_t ksi;
504 
505 	regs = td->td_frame;
506 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
507 	if (error != 0)
508 		return (error);
509 
510 	p = td->td_proc;
511 	context = &uc.uc_mcontext;
512 	rflags = context->sc_rflags;
513 
514 	/*
515 	 * Don't allow users to change privileged or reserved flags.
516 	 */
517 	/*
518 	 * XXX do allow users to change the privileged flag PSL_RF.
519 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
520 	 * should sometimes set it there too.  tf_rflags is kept in
521 	 * the signal context during signal handling and there is no
522 	 * other place to remember it, so the PSL_RF bit may be
523 	 * corrupted by the signal handler without us knowing.
524 	 * Corruption of the PSL_RF bit at worst causes one more or
525 	 * one less debugger trap, so allowing it is fairly harmless.
526 	 */
527 
528 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
529 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
530 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
531 		return (EINVAL);
532 	}
533 
534 	/*
535 	 * Don't allow users to load a valid privileged %cs.  Let the
536 	 * hardware check for invalid selectors, excess privilege in
537 	 * other selectors, invalid %eip's and invalid %esp's.
538 	 */
539 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
540 	if (!CS_SECURE(context->sc_cs)) {
541 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
542 		ksiginfo_init_trap(&ksi);
543 		ksi.ksi_signo = SIGBUS;
544 		ksi.ksi_code = BUS_OBJERR;
545 		ksi.ksi_trapno = T_PROTFLT;
546 		ksi.ksi_addr = (void *)regs->tf_rip;
547 		trapsignal(td, &ksi);
548 		return (EINVAL);
549 	}
550 
551 	PROC_LOCK(p);
552 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
553 	SIG_CANTMASK(td->td_sigmask);
554 	signotify(td);
555 	PROC_UNLOCK(p);
556 
557 	regs->tf_rdi    = context->sc_rdi;
558 	regs->tf_rsi    = context->sc_rsi;
559 	regs->tf_rdx    = context->sc_rdx;
560 	regs->tf_rbp    = context->sc_rbp;
561 	regs->tf_rbx    = context->sc_rbx;
562 	regs->tf_rcx    = context->sc_rcx;
563 	regs->tf_rax    = context->sc_rax;
564 	regs->tf_rip    = context->sc_rip;
565 	regs->tf_rsp    = context->sc_rsp;
566 	regs->tf_r8     = context->sc_r8;
567 	regs->tf_r9     = context->sc_r9;
568 	regs->tf_r10    = context->sc_r10;
569 	regs->tf_r11    = context->sc_r11;
570 	regs->tf_r12    = context->sc_r12;
571 	regs->tf_r13    = context->sc_r13;
572 	regs->tf_r14    = context->sc_r14;
573 	regs->tf_r15    = context->sc_r15;
574 	regs->tf_cs     = context->sc_cs;
575 	regs->tf_err    = context->sc_err;
576 	regs->tf_rflags = rflags;
577 
578 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
579 	return (EJUSTRETURN);
580 }
581 
582 /*
583  * copied from amd64/amd64/machdep.c
584  *
585  * Send an interrupt to process.
586  */
587 static void
588 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
589 {
590 	struct l_rt_sigframe sf, *sfp;
591 	struct proc *p;
592 	struct thread *td;
593 	struct sigacts *psp;
594 	caddr_t sp;
595 	struct trapframe *regs;
596 	int sig, code;
597 	int oonstack;
598 
599 	td = curthread;
600 	p = td->td_proc;
601 	PROC_LOCK_ASSERT(p, MA_OWNED);
602 	sig = ksi->ksi_signo;
603 	psp = p->p_sigacts;
604 	code = ksi->ksi_code;
605 	mtx_assert(&psp->ps_mtx, MA_OWNED);
606 	regs = td->td_frame;
607 	oonstack = sigonstack(regs->tf_rsp);
608 
609 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
610 	    catcher, sig, mask, code);
611 
612 	/* Translate the signal. */
613 	sig = bsd_to_linux_signal(sig);
614 
615 	/* Save user context. */
616 	bzero(&sf, sizeof(sf));
617 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
618 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
619 
620 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
621 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
622 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
623 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
624 
625 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
626 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
627 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
628 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
629 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
630 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
631 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
632 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
633 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
634 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
635 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
636 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
637 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
638 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
639 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
640 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
641 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
642 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
643 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
644 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
645 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
646 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
647 
648 	/* Allocate space for the signal handler context. */
649 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
650 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
651 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
652 		    sizeof(struct l_rt_sigframe);
653 	} else
654 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
655 	/* Align to 16 bytes. */
656 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
657 
658 	/* Build the argument list for the signal handler. */
659 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
660 	regs->tf_rax = 0;
661 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
662 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
663 
664 	/* Fill in POSIX parts. */
665 	siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig);
666 	sf.sf_handler = catcher;
667 
668 	mtx_unlock(&psp->ps_mtx);
669 	PROC_UNLOCK(p);
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_elf_core_osabi = ELFOSABI_NONE,
744 	.sv_elf_core_abi_vendor = FREEBSD_ABI_VENDOR,
745 	.sv_elf_core_prepare_notes = elf64_prepare_notes,
746 	.sv_imgact_try	= linux_exec_imgact_try,
747 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
748 	.sv_minuser	= VM_MIN_ADDRESS,
749 	.sv_maxuser	= VM_MAXUSER_ADDRESS_LA48,
750 	.sv_usrstack	= USRSTACK_LA48,
751 	.sv_psstrings	= PS_STRINGS_LA48,
752 	.sv_stackprot	= VM_PROT_ALL,
753 	.sv_copyout_auxargs = linux_copyout_auxargs,
754 	.sv_copyout_strings = linux_copyout_strings,
755 	.sv_setregs	= linux_exec_setregs,
756 	.sv_fixlimit	= NULL,
757 	.sv_maxssiz	= NULL,
758 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
759 	    SV_SIG_WAITNDQ,
760 	.sv_set_syscall_retval = linux_set_syscall_retval,
761 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
762 	.sv_syscallnames = NULL,
763 	.sv_shared_page_base = SHAREDPAGE_LA48,
764 	.sv_shared_page_len = PAGE_SIZE,
765 	.sv_schedtail	= linux_schedtail,
766 	.sv_thread_detach = linux_thread_detach,
767 	.sv_trap	= linux_vsyscall,
768 	.sv_onexec	= linux_on_exec,
769 	.sv_onexit	= linux_on_exit,
770 	.sv_ontdexit	= linux_thread_dtor,
771 	.sv_setid_allowed = &linux_setid_allowed_query,
772 };
773 
774 static void
775 linux_vdso_install(void *param)
776 {
777 
778 	amd64_lower_shared_page(&elf_linux_sysvec);
779 
780 	linux_szsigcode = (&_binary_linux_locore_o_end -
781 	    &_binary_linux_locore_o_start);
782 
783 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
784 		panic("Linux invalid vdso size\n");
785 
786 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
787 
788 	linux_shared_page_obj = __elfN(linux_shared_page_init)
789 	    (&linux_shared_page_mapping);
790 
791 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
792 
793 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
794 	    linux_szsigcode);
795 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
796 }
797 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
798     linux_vdso_install, NULL);
799 
800 static void
801 linux_vdso_deinstall(void *param)
802 {
803 
804 	__elfN(linux_shared_page_fini)(linux_shared_page_obj,
805 	    linux_shared_page_mapping);
806 }
807 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
808     linux_vdso_deinstall, NULL);
809 
810 static char GNULINUX_ABI_VENDOR[] = "GNU";
811 static int GNULINUX_ABI_DESC = 0;
812 
813 static bool
814 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
815 {
816 	const Elf32_Word *desc;
817 	uintptr_t p;
818 
819 	p = (uintptr_t)(note + 1);
820 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
821 
822 	desc = (const Elf32_Word *)p;
823 	if (desc[0] != GNULINUX_ABI_DESC)
824 		return (false);
825 
826 	/*
827 	 * For Linux we encode osrel using the Linux convention of
828 	 * 	(version << 16) | (major << 8) | (minor)
829 	 * See macro in linux_mib.h
830 	 */
831 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
832 
833 	return (true);
834 }
835 
836 static Elf_Brandnote linux64_brandnote = {
837 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
838 	.hdr.n_descsz	= 16,
839 	.hdr.n_type	= 1,
840 	.vendor		= GNULINUX_ABI_VENDOR,
841 	.flags		= BN_TRANSLATE_OSREL,
842 	.trans_osrel	= linux_trans_osrel
843 };
844 
845 static Elf64_Brandinfo linux_glibc2brand = {
846 	.brand		= ELFOSABI_LINUX,
847 	.machine	= EM_X86_64,
848 	.compat_3_brand	= "Linux",
849 	.emul_path	= linux_emul_path,
850 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
851 	.sysvec		= &elf_linux_sysvec,
852 	.interp_newpath	= NULL,
853 	.brand_note	= &linux64_brandnote,
854 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
855 };
856 
857 static Elf64_Brandinfo linux_glibc2brandshort = {
858 	.brand		= ELFOSABI_LINUX,
859 	.machine	= EM_X86_64,
860 	.compat_3_brand	= "Linux",
861 	.emul_path	= linux_emul_path,
862 	.interp_path	= "/lib64/ld-linux.so.2",
863 	.sysvec		= &elf_linux_sysvec,
864 	.interp_newpath	= NULL,
865 	.brand_note	= &linux64_brandnote,
866 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
867 };
868 
869 static Elf64_Brandinfo linux_muslbrand = {
870 	.brand		= ELFOSABI_LINUX,
871 	.machine	= EM_X86_64,
872 	.compat_3_brand	= "Linux",
873 	.emul_path	= linux_emul_path,
874 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
875 	.sysvec		= &elf_linux_sysvec,
876 	.interp_newpath	= NULL,
877 	.brand_note	= &linux64_brandnote,
878 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
879 };
880 
881 Elf64_Brandinfo *linux_brandlist[] = {
882 	&linux_glibc2brand,
883 	&linux_glibc2brandshort,
884 	&linux_muslbrand,
885 	NULL
886 };
887 
888 static int
889 linux64_elf_modevent(module_t mod, int type, void *data)
890 {
891 	Elf64_Brandinfo **brandinfo;
892 	int error;
893 	struct linux_ioctl_handler **lihp;
894 
895 	error = 0;
896 
897 	switch(type) {
898 	case MOD_LOAD:
899 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
900 		     ++brandinfo)
901 			if (elf64_insert_brand_entry(*brandinfo) < 0)
902 				error = EINVAL;
903 		if (error == 0) {
904 			SET_FOREACH(lihp, linux_ioctl_handler_set)
905 				linux_ioctl_register_handler(*lihp);
906 			stclohz = (stathz ? stathz : hz);
907 			if (bootverbose)
908 				printf("Linux x86-64 ELF exec handler installed\n");
909 		} else
910 			printf("cannot insert Linux x86-64 ELF brand handler\n");
911 		break;
912 	case MOD_UNLOAD:
913 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
914 		     ++brandinfo)
915 			if (elf64_brand_inuse(*brandinfo))
916 				error = EBUSY;
917 		if (error == 0) {
918 			for (brandinfo = &linux_brandlist[0];
919 			     *brandinfo != NULL; ++brandinfo)
920 				if (elf64_remove_brand_entry(*brandinfo) < 0)
921 					error = EINVAL;
922 		}
923 		if (error == 0) {
924 			SET_FOREACH(lihp, linux_ioctl_handler_set)
925 				linux_ioctl_unregister_handler(*lihp);
926 			if (bootverbose)
927 				printf("Linux ELF exec handler removed\n");
928 		} else
929 			printf("Could not deinstall ELF interpreter entry\n");
930 		break;
931 	default:
932 		return (EOPNOTSUPP);
933 	}
934 	return (error);
935 }
936 
937 static moduledata_t linux64_elf_mod = {
938 	"linux64elf",
939 	linux64_elf_modevent,
940 	0
941 };
942 
943 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
944 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
945 FEATURE(linux64, "Linux 64bit support");
946