xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 0a8314e06cf69b7a9c73de807ee5890329301224)
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_futex.h>
79 #include <compat/linux/linux_ioctl.h>
80 #include <compat/linux/linux_mib.h>
81 #include <compat/linux/linux_misc.h>
82 #include <compat/linux/linux_signal.h>
83 #include <compat/linux/linux_sysproto.h>
84 #include <compat/linux/linux_util.h>
85 #include <compat/linux/linux_vdso.h>
86 
87 MODULE_VERSION(linux64, 1);
88 
89 const char *linux_kplatform;
90 static int linux_szsigcode;
91 static vm_object_t linux_shared_page_obj;
92 static char *linux_shared_page_mapping;
93 extern char _binary_linux_locore_o_start;
94 extern char _binary_linux_locore_o_end;
95 
96 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
97 
98 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
99 
100 static register_t * linux_copyout_strings(struct image_params *imgp);
101 static int	linux_fixup_elf(register_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 		    u_long 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 	sa->narg = sa->callp->sy_narg;
202 
203 	td->td_retval[0] = 0;
204 	return (0);
205 }
206 
207 static void
208 linux_set_syscall_retval(struct thread *td, int error)
209 {
210 	struct trapframe *frame = td->td_frame;
211 
212 	/*
213 	 * On Linux only %rcx and %r11 values are not preserved across
214 	 * the syscall.  So, do not clobber %rdx and %r10.
215 	 */
216 	td->td_retval[1] = frame->tf_rdx;
217 	if (error != EJUSTRETURN)
218 		frame->tf_r10 = frame->tf_rcx;
219 
220 	cpu_set_syscall_retval(td, error);
221 
222 	 /* Restore all registers. */
223 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
224 }
225 
226 static int
227 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
228 {
229 	Elf_Auxargs *args;
230 	Elf_Auxinfo *argarray, *pos;
231 	Elf_Addr *auxbase, *base;
232 	struct ps_strings *arginfo;
233 	struct proc *p;
234 	int error, issetugid;
235 
236 	p = imgp->proc;
237 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
238 
239 	KASSERT(curthread->td_proc == imgp->proc,
240 	    ("unsafe linux_fixup_elf(), should be curproc"));
241 	base = (Elf64_Addr *)*stack_base;
242 	args = (Elf64_Auxargs *)imgp->auxargs;
243 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
244 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
245 	    M_WAITOK | M_ZERO);
246 
247 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
248 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
249 	    imgp->proc->p_sysent->sv_shared_page_base);
250 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
251 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
252 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
253 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
254 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
255 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
256 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
257 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
258 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
259 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
260 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
261 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
262 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
263 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
264 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
265 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
266 	if (imgp->execpathp != 0)
267 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
268 	if (args->execfd != -1)
269 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
270 	AUXARGS_ENTRY(pos, AT_NULL, 0);
271 	free(imgp->auxargs, M_TEMP);
272 	imgp->auxargs = NULL;
273 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
274 
275 	error = copyout(argarray, auxbase, sizeof(*argarray) * LINUX_AT_COUNT);
276 	free(argarray, M_TEMP);
277 	if (error != 0)
278 		return (error);
279 
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 	if (imgp->auxargs) {
332 		/*
333 		 * Allocate room on the stack for the ELF auxargs
334 		 * array.  It has LINUX_AT_COUNT entries.
335 		 */
336 		vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf64_Auxinfo),
337 		    sizeof(*vectp));
338 	}
339 
340 	/*
341 	 * Allocate room for the argv[] and env vectors including the
342 	 * terminating NULL pointers.
343 	 */
344 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
345 
346 	/* vectp also becomes our initial stack base. */
347 	stack_base = (register_t *)vectp;
348 
349 	stringp = imgp->args->begin_argv;
350 	argc = imgp->args->argc;
351 	envc = imgp->args->envc;
352 
353 	/* Copy out strings - arguments and environment. */
354 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
355 
356 	/* Fill in "ps_strings" struct for ps, w, etc. */
357 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
358 	suword(&arginfo->ps_nargvstr, argc);
359 
360 	/* Fill in argument portion of vector table. */
361 	for (; argc > 0; --argc) {
362 		suword(vectp++, (long)(intptr_t)destp);
363 		while (*stringp++ != 0)
364 			destp++;
365 		destp++;
366 	}
367 
368 	/* A null vector table pointer separates the argp's from the envp's. */
369 	suword(vectp++, 0);
370 
371 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
372 	suword(&arginfo->ps_nenvstr, envc);
373 
374 	/* Fill in environment portion of vector table. */
375 	for (; envc > 0; --envc) {
376 		suword(vectp++, (long)(intptr_t)destp);
377 		while (*stringp++ != 0)
378 			destp++;
379 		destp++;
380 	}
381 
382 	/* The end of the vector table is a null pointer. */
383 	suword(vectp, 0);
384 	return (stack_base);
385 }
386 
387 /*
388  * Reset registers to default values on exec.
389  */
390 static void
391 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
392 {
393 	struct trapframe *regs;
394 	struct pcb *pcb;
395 	register_t saved_rflags;
396 
397 	regs = td->td_frame;
398 	pcb = td->td_pcb;
399 
400 	if (td->td_proc->p_md.md_ldt != NULL)
401 		user_ldt_free(td);
402 
403 	pcb->pcb_fsbase = 0;
404 	pcb->pcb_gsbase = 0;
405 	clear_pcb_flags(pcb, PCB_32BIT);
406 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
407 	set_pcb_flags(pcb, PCB_FULL_IRET);
408 
409 	saved_rflags = regs->tf_rflags & PSL_T;
410 	bzero((char *)regs, sizeof(struct trapframe));
411 	regs->tf_rip = imgp->entry_addr;
412 	regs->tf_rsp = stack;
413 	regs->tf_rflags = PSL_USER | saved_rflags;
414 	regs->tf_ss = _udatasel;
415 	regs->tf_cs = _ucodesel;
416 	regs->tf_ds = _udatasel;
417 	regs->tf_es = _udatasel;
418 	regs->tf_fs = _ufssel;
419 	regs->tf_gs = _ugssel;
420 	regs->tf_flags = TF_HASSEGS;
421 
422 	/*
423 	 * Reset the hardware debug registers if they were in use.
424 	 * They won't have any meaning for the newly exec'd process.
425 	 */
426 	if (pcb->pcb_flags & PCB_DBREGS) {
427 		pcb->pcb_dr0 = 0;
428 		pcb->pcb_dr1 = 0;
429 		pcb->pcb_dr2 = 0;
430 		pcb->pcb_dr3 = 0;
431 		pcb->pcb_dr6 = 0;
432 		pcb->pcb_dr7 = 0;
433 		if (pcb == curpcb) {
434 			/*
435 			 * Clear the debug registers on the running
436 			 * CPU, otherwise they will end up affecting
437 			 * the next process we switch to.
438 			 */
439 			reset_dbregs();
440 		}
441 		clear_pcb_flags(pcb, PCB_DBREGS);
442 	}
443 
444 	/*
445 	 * Drop the FP state if we hold it, so that the process gets a
446 	 * clean FP state if it uses the FPU again.
447 	 */
448 	fpstate_drop(td);
449 }
450 
451 /*
452  * Copied from amd64/amd64/machdep.c
453  *
454  * XXX fpu state need? don't think so
455  */
456 int
457 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
458 {
459 	struct proc *p;
460 	struct l_ucontext uc;
461 	struct l_sigcontext *context;
462 	struct trapframe *regs;
463 	unsigned long rflags;
464 	int error;
465 	ksiginfo_t ksi;
466 
467 	regs = td->td_frame;
468 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
469 	if (error != 0)
470 		return (error);
471 
472 	p = td->td_proc;
473 	context = &uc.uc_mcontext;
474 	rflags = context->sc_rflags;
475 
476 	/*
477 	 * Don't allow users to change privileged or reserved flags.
478 	 */
479 	/*
480 	 * XXX do allow users to change the privileged flag PSL_RF.
481 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
482 	 * should sometimes set it there too.  tf_rflags is kept in
483 	 * the signal context during signal handling and there is no
484 	 * other place to remember it, so the PSL_RF bit may be
485 	 * corrupted by the signal handler without us knowing.
486 	 * Corruption of the PSL_RF bit at worst causes one more or
487 	 * one less debugger trap, so allowing it is fairly harmless.
488 	 */
489 
490 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
491 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
492 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
493 		return (EINVAL);
494 	}
495 
496 	/*
497 	 * Don't allow users to load a valid privileged %cs.  Let the
498 	 * hardware check for invalid selectors, excess privilege in
499 	 * other selectors, invalid %eip's and invalid %esp's.
500 	 */
501 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
502 	if (!CS_SECURE(context->sc_cs)) {
503 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
504 		ksiginfo_init_trap(&ksi);
505 		ksi.ksi_signo = SIGBUS;
506 		ksi.ksi_code = BUS_OBJERR;
507 		ksi.ksi_trapno = T_PROTFLT;
508 		ksi.ksi_addr = (void *)regs->tf_rip;
509 		trapsignal(td, &ksi);
510 		return (EINVAL);
511 	}
512 
513 	PROC_LOCK(p);
514 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
515 	SIG_CANTMASK(td->td_sigmask);
516 	signotify(td);
517 	PROC_UNLOCK(p);
518 
519 	regs->tf_rdi    = context->sc_rdi;
520 	regs->tf_rsi    = context->sc_rsi;
521 	regs->tf_rdx    = context->sc_rdx;
522 	regs->tf_rbp    = context->sc_rbp;
523 	regs->tf_rbx    = context->sc_rbx;
524 	regs->tf_rcx    = context->sc_rcx;
525 	regs->tf_rax    = context->sc_rax;
526 	regs->tf_rip    = context->sc_rip;
527 	regs->tf_rsp    = context->sc_rsp;
528 	regs->tf_r8     = context->sc_r8;
529 	regs->tf_r9     = context->sc_r9;
530 	regs->tf_r10    = context->sc_r10;
531 	regs->tf_r11    = context->sc_r11;
532 	regs->tf_r12    = context->sc_r12;
533 	regs->tf_r13    = context->sc_r13;
534 	regs->tf_r14    = context->sc_r14;
535 	regs->tf_r15    = context->sc_r15;
536 	regs->tf_cs     = context->sc_cs;
537 	regs->tf_err    = context->sc_err;
538 	regs->tf_rflags = rflags;
539 
540 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
541 	return (EJUSTRETURN);
542 }
543 
544 /*
545  * copied from amd64/amd64/machdep.c
546  *
547  * Send an interrupt to process.
548  */
549 static void
550 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
551 {
552 	struct l_rt_sigframe sf, *sfp;
553 	struct proc *p;
554 	struct thread *td;
555 	struct sigacts *psp;
556 	caddr_t sp;
557 	struct trapframe *regs;
558 	int sig, code;
559 	int oonstack;
560 
561 	td = curthread;
562 	p = td->td_proc;
563 	PROC_LOCK_ASSERT(p, MA_OWNED);
564 	sig = ksi->ksi_signo;
565 	psp = p->p_sigacts;
566 	code = ksi->ksi_code;
567 	mtx_assert(&psp->ps_mtx, MA_OWNED);
568 	regs = td->td_frame;
569 	oonstack = sigonstack(regs->tf_rsp);
570 
571 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
572 	    catcher, sig, mask, code);
573 
574 	/* Allocate space for the signal handler context. */
575 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
576 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
577 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
578 		    sizeof(struct l_rt_sigframe);
579 	} else
580 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
581 	/* Align to 16 bytes. */
582 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
583 	mtx_unlock(&psp->ps_mtx);
584 
585 	/* Translate the signal. */
586 	sig = bsd_to_linux_signal(sig);
587 
588 	/* Save user context. */
589 	bzero(&sf, sizeof(sf));
590 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
591 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
592 
593 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
594 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
595 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
596 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
597 	PROC_UNLOCK(p);
598 
599 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
600 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
601 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
602 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
603 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
604 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
605 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
606 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
607 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
608 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
609 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
610 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
611 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
612 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
613 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
614 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
615 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
616 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
617 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
618 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
619 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
620 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
621 
622 	/* Build the argument list for the signal handler. */
623 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
624 	regs->tf_rax = 0;
625 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
626 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
627 
628 	sf.sf_handler = catcher;
629 	/* Fill in POSIX parts. */
630 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
631 
632 	/* Copy the sigframe out to the user's stack. */
633 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
634 		PROC_LOCK(p);
635 		sigexit(td, SIGILL);
636 	}
637 
638 	regs->tf_rsp = (long)sfp;
639 	regs->tf_rip = linux_rt_sigcode;
640 	regs->tf_rflags &= ~(PSL_T | PSL_D);
641 	regs->tf_cs = _ucodesel;
642 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
643 	PROC_LOCK(p);
644 	mtx_lock(&psp->ps_mtx);
645 }
646 
647 #define	LINUX_VSYSCALL_START		(-10UL << 20)
648 #define	LINUX_VSYSCALL_SZ		1024
649 
650 const unsigned long linux_vsyscall_vector[] = {
651 	LINUX_SYS_gettimeofday,
652 	LINUX_SYS_linux_time,
653 				/* getcpu not implemented */
654 };
655 
656 static int
657 linux_vsyscall(struct thread *td)
658 {
659 	struct trapframe *frame;
660 	uint64_t retqaddr;
661 	int code, traced;
662 	int error;
663 
664 	frame = td->td_frame;
665 
666 	/* Check %rip for vsyscall area. */
667 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
668 		return (EINVAL);
669 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
670 		return (EINVAL);
671 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
672 	if (code >= nitems(linux_vsyscall_vector))
673 		return (EINVAL);
674 
675 	/*
676 	 * vsyscall called as callq *(%rax), so we must
677 	 * use return address from %rsp and also fixup %rsp.
678 	 */
679 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
680 	if (error)
681 		return (error);
682 
683 	frame->tf_rip = retqaddr;
684 	frame->tf_rax = linux_vsyscall_vector[code];
685 	frame->tf_rsp += 8;
686 
687 	traced = (frame->tf_flags & PSL_T);
688 
689 	amd64_syscall(td, traced);
690 
691 	return (0);
692 }
693 
694 struct sysentvec elf_linux_sysvec = {
695 	.sv_size	= LINUX_SYS_MAXSYSCALL,
696 	.sv_table	= linux_sysent,
697 	.sv_errsize	= ELAST + 1,
698 	.sv_errtbl	= linux_errtbl,
699 	.sv_transtrap	= linux_translate_traps,
700 	.sv_fixup	= linux_fixup_elf,
701 	.sv_sendsig	= linux_rt_sendsig,
702 	.sv_sigcode	= &_binary_linux_locore_o_start,
703 	.sv_szsigcode	= &linux_szsigcode,
704 	.sv_name	= "Linux ELF64",
705 	.sv_coredump	= elf64_coredump,
706 	.sv_imgact_try	= linux_exec_imgact_try,
707 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
708 	.sv_minuser	= VM_MIN_ADDRESS,
709 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
710 	.sv_usrstack	= USRSTACK,
711 	.sv_psstrings	= PS_STRINGS,
712 	.sv_stackprot	= VM_PROT_ALL,
713 	.sv_copyout_strings = linux_copyout_strings,
714 	.sv_setregs	= linux_exec_setregs,
715 	.sv_fixlimit	= NULL,
716 	.sv_maxssiz	= NULL,
717 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
718 	.sv_set_syscall_retval = linux_set_syscall_retval,
719 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
720 	.sv_syscallnames = NULL,
721 	.sv_shared_page_base = SHAREDPAGE,
722 	.sv_shared_page_len = PAGE_SIZE,
723 	.sv_schedtail	= linux_schedtail,
724 	.sv_thread_detach = linux_thread_detach,
725 	.sv_trap	= linux_vsyscall,
726 };
727 
728 static void
729 linux_vdso_install(void *param)
730 {
731 
732 	amd64_lower_shared_page(&elf_linux_sysvec);
733 
734 	linux_szsigcode = (&_binary_linux_locore_o_end -
735 	    &_binary_linux_locore_o_start);
736 
737 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
738 		panic("Linux invalid vdso size\n");
739 
740 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
741 
742 	linux_shared_page_obj = __elfN(linux_shared_page_init)
743 	    (&linux_shared_page_mapping);
744 
745 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
746 
747 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
748 	    linux_szsigcode);
749 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
750 
751 	linux_kplatform = linux_shared_page_mapping +
752 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
753 }
754 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
755     linux_vdso_install, NULL);
756 
757 static void
758 linux_vdso_deinstall(void *param)
759 {
760 
761 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
762 }
763 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
764     linux_vdso_deinstall, NULL);
765 
766 static char GNULINUX_ABI_VENDOR[] = "GNU";
767 static int GNULINUX_ABI_DESC = 0;
768 
769 static bool
770 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
771 {
772 	const Elf32_Word *desc;
773 	uintptr_t p;
774 
775 	p = (uintptr_t)(note + 1);
776 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
777 
778 	desc = (const Elf32_Word *)p;
779 	if (desc[0] != GNULINUX_ABI_DESC)
780 		return (false);
781 
782 	/*
783 	 * For Linux we encode osrel using the Linux convention of
784 	 * 	(version << 16) | (major << 8) | (minor)
785 	 * See macro in linux_mib.h
786 	 */
787 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
788 
789 	return (true);
790 }
791 
792 static Elf_Brandnote linux64_brandnote = {
793 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
794 	.hdr.n_descsz	= 16,
795 	.hdr.n_type	= 1,
796 	.vendor		= GNULINUX_ABI_VENDOR,
797 	.flags		= BN_TRANSLATE_OSREL,
798 	.trans_osrel	= linux_trans_osrel
799 };
800 
801 static Elf64_Brandinfo linux_glibc2brand = {
802 	.brand		= ELFOSABI_LINUX,
803 	.machine	= EM_X86_64,
804 	.compat_3_brand	= "Linux",
805 	.emul_path	= "/compat/linux",
806 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
807 	.sysvec		= &elf_linux_sysvec,
808 	.interp_newpath	= NULL,
809 	.brand_note	= &linux64_brandnote,
810 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
811 };
812 
813 static Elf64_Brandinfo linux_glibc2brandshort = {
814 	.brand		= ELFOSABI_LINUX,
815 	.machine	= EM_X86_64,
816 	.compat_3_brand	= "Linux",
817 	.emul_path	= "/compat/linux",
818 	.interp_path	= "/lib64/ld-linux.so.2",
819 	.sysvec		= &elf_linux_sysvec,
820 	.interp_newpath	= NULL,
821 	.brand_note	= &linux64_brandnote,
822 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
823 };
824 
825 static Elf64_Brandinfo linux_muslbrand = {
826 	.brand		= ELFOSABI_LINUX,
827 	.machine	= EM_X86_64,
828 	.compat_3_brand	= "Linux",
829 	.emul_path	= "/compat/linux",
830 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
831 	.sysvec		= &elf_linux_sysvec,
832 	.interp_newpath	= NULL,
833 	.brand_note	= &linux64_brandnote,
834 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
835 };
836 
837 Elf64_Brandinfo *linux_brandlist[] = {
838 	&linux_glibc2brand,
839 	&linux_glibc2brandshort,
840 	&linux_muslbrand,
841 	NULL
842 };
843 
844 static int
845 linux64_elf_modevent(module_t mod, int type, void *data)
846 {
847 	Elf64_Brandinfo **brandinfo;
848 	int error;
849 	struct linux_ioctl_handler **lihp;
850 
851 	error = 0;
852 
853 	switch(type) {
854 	case MOD_LOAD:
855 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
856 		     ++brandinfo)
857 			if (elf64_insert_brand_entry(*brandinfo) < 0)
858 				error = EINVAL;
859 		if (error == 0) {
860 			SET_FOREACH(lihp, linux_ioctl_handler_set)
861 				linux_ioctl_register_handler(*lihp);
862 			LIST_INIT(&futex_list);
863 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
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 			mtx_destroy(&futex_mtx);
885 			if (bootverbose)
886 				printf("Linux ELF exec handler removed\n");
887 		} else
888 			printf("Could not deinstall ELF interpreter entry\n");
889 		break;
890 	default:
891 		return (EOPNOTSUPP);
892 	}
893 	return (error);
894 }
895 
896 static moduledata_t linux64_elf_mod = {
897 	"linux64elf",
898 	linux64_elf_modevent,
899 	0
900 };
901 
902 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
903 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
904 FEATURE(linux64, "Linux 64bit support");
905