xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 2ff91c175eca50b7d0d9da6b31eae4109c034137)
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 #if defined(DEBUG)
90 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
91 	    CTLTYPE_STRING | CTLFLAG_RW,
92 	    0, 0, linux_sysctl_debug, "A",
93 	    "Linux 64 debugging control");
94 #endif
95 
96 /*
97  * Allow the sendsig functions to use the ldebug() facility even though they
98  * are not syscalls themselves.  Map them to syscall 0.  This is slightly less
99  * bogus than using ldebug(sigreturn).
100  */
101 #define	LINUX_SYS_linux_rt_sendsig	0
102 
103 const char *linux_kplatform;
104 static int linux_szsigcode;
105 static vm_object_t linux_shared_page_obj;
106 static char *linux_shared_page_mapping;
107 extern char _binary_linux_locore_o_start;
108 extern char _binary_linux_locore_o_end;
109 
110 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
111 
112 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
113 
114 static register_t * linux_copyout_strings(struct image_params *imgp);
115 static int	linux_fixup_elf(register_t **stack_base,
116 		    struct image_params *iparams);
117 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
118 static void	linux_vdso_install(void *param);
119 static void	linux_vdso_deinstall(void *param);
120 static void	linux_set_syscall_retval(struct thread *td, int error);
121 static int	linux_fetch_syscall_args(struct thread *td);
122 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
123 		    u_long stack);
124 static int	linux_vsyscall(struct thread *td);
125 
126 #define LINUX_T_UNKNOWN  255
127 static int _bsd_to_linux_trapcode[] = {
128 	LINUX_T_UNKNOWN,	/* 0 */
129 	6,			/* 1  T_PRIVINFLT */
130 	LINUX_T_UNKNOWN,	/* 2 */
131 	3,			/* 3  T_BPTFLT */
132 	LINUX_T_UNKNOWN,	/* 4 */
133 	LINUX_T_UNKNOWN,	/* 5 */
134 	16,			/* 6  T_ARITHTRAP */
135 	254,			/* 7  T_ASTFLT */
136 	LINUX_T_UNKNOWN,	/* 8 */
137 	13,			/* 9  T_PROTFLT */
138 	1,			/* 10 T_TRCTRAP */
139 	LINUX_T_UNKNOWN,	/* 11 */
140 	14,			/* 12 T_PAGEFLT */
141 	LINUX_T_UNKNOWN,	/* 13 */
142 	17,			/* 14 T_ALIGNFLT */
143 	LINUX_T_UNKNOWN,	/* 15 */
144 	LINUX_T_UNKNOWN,	/* 16 */
145 	LINUX_T_UNKNOWN,	/* 17 */
146 	0,			/* 18 T_DIVIDE */
147 	2,			/* 19 T_NMI */
148 	4,			/* 20 T_OFLOW */
149 	5,			/* 21 T_BOUND */
150 	7,			/* 22 T_DNA */
151 	8,			/* 23 T_DOUBLEFLT */
152 	9,			/* 24 T_FPOPFLT */
153 	10,			/* 25 T_TSSFLT */
154 	11,			/* 26 T_SEGNPFLT */
155 	12,			/* 27 T_STKFLT */
156 	18,			/* 28 T_MCHK */
157 	19,			/* 29 T_XMMFLT */
158 	15			/* 30 T_RESERVED */
159 };
160 #define bsd_to_linux_trapcode(code) \
161     ((code)<nitems(_bsd_to_linux_trapcode)? \
162      _bsd_to_linux_trapcode[(code)]: \
163      LINUX_T_UNKNOWN)
164 
165 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
166 LINUX_VDSO_SYM_CHAR(linux_platform);
167 
168 /*
169  * If FreeBSD & Linux have a difference of opinion about what a trap
170  * means, deal with it here.
171  *
172  * MPSAFE
173  */
174 static int
175 linux_translate_traps(int signal, int trap_code)
176 {
177 
178 	if (signal != SIGBUS)
179 		return (signal);
180 	switch (trap_code) {
181 	case T_PROTFLT:
182 	case T_TSSFLT:
183 	case T_DOUBLEFLT:
184 	case T_PAGEFLT:
185 		return (SIGSEGV);
186 	default:
187 		return (signal);
188 	}
189 }
190 
191 static int
192 linux_fetch_syscall_args(struct thread *td)
193 {
194 	struct proc *p;
195 	struct trapframe *frame;
196 	struct syscall_args *sa;
197 
198 	p = td->td_proc;
199 	frame = td->td_frame;
200 	sa = &td->td_sa;
201 
202 	sa->args[0] = frame->tf_rdi;
203 	sa->args[1] = frame->tf_rsi;
204 	sa->args[2] = frame->tf_rdx;
205 	sa->args[3] = frame->tf_rcx;
206 	sa->args[4] = frame->tf_r8;
207 	sa->args[5] = frame->tf_r9;
208 	sa->code = frame->tf_rax;
209 
210 	if (sa->code >= p->p_sysent->sv_size)
211 		/* nosys */
212 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
213 	else
214 		sa->callp = &p->p_sysent->sv_table[sa->code];
215 	sa->narg = sa->callp->sy_narg;
216 
217 	td->td_retval[0] = 0;
218 	return (0);
219 }
220 
221 static void
222 linux_set_syscall_retval(struct thread *td, int error)
223 {
224 	struct trapframe *frame = td->td_frame;
225 
226 	/*
227 	 * On Linux only %rcx and %r11 values are not preserved across
228 	 * the syscall.  So, do not clobber %rdx and %r10.
229 	 */
230 	td->td_retval[1] = frame->tf_rdx;
231 	frame->tf_r10 = frame->tf_rcx;
232 
233 	cpu_set_syscall_retval(td, error);
234 
235 	 /* Restore all registers. */
236 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
237 }
238 
239 static int
240 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
241 {
242 	Elf_Auxargs *args;
243 	Elf_Addr *base;
244 	Elf_Addr *pos;
245 	struct ps_strings *arginfo;
246 	struct proc *p;
247 	int issetugid;
248 
249 	p = imgp->proc;
250 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
251 
252 	KASSERT(curthread->td_proc == imgp->proc,
253 	    ("unsafe linux_fixup_elf(), should be curproc"));
254 	base = (Elf64_Addr *)*stack_base;
255 	args = (Elf64_Auxargs *)imgp->auxargs;
256 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
257 
258 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
259 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
260 	    imgp->proc->p_sysent->sv_shared_page_base);
261 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
262 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
263 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
264 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
265 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
266 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
267 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
268 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
269 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
270 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
271 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
272 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
273 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
274 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
275 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
276 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
277 	if (imgp->execpathp != 0)
278 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
279 	if (args->execfd != -1)
280 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
281 	AUXARGS_ENTRY(pos, AT_NULL, 0);
282 	free(imgp->auxargs, M_TEMP);
283 	imgp->auxargs = NULL;
284 
285 	base--;
286 	suword(base, (uint64_t)imgp->args->argc);
287 
288 	*stack_base = (register_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 register_t *
298 linux_copyout_strings(struct image_params *imgp)
299 {
300 	int argc, envc;
301 	char **vectp;
302 	char *stringp, *destp;
303 	register_t *stack_base;
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 = (caddr_t)arginfo - SPARE_USRSPACE -
318 	    roundup(sizeof(canary), sizeof(char *)) -
319 	    roundup(execpath_len, sizeof(char *)) -
320 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
321 
322 	if (execpath_len != 0) {
323 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
324 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
325 	}
326 
327 	/* Prepare the canary for SSP. */
328 	arc4rand(canary, sizeof(canary), 0);
329 	imgp->canary = (uintptr_t)arginfo -
330 	    roundup(execpath_len, sizeof(char *)) -
331 	    roundup(sizeof(canary), sizeof(char *));
332 	copyout(canary, (void *)imgp->canary, sizeof(canary));
333 
334 	/* If we have a valid auxargs ptr, prepare some room on the stack. */
335 	if (imgp->auxargs) {
336 		/*
337 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
338 		 * lower compatibility.
339 		 */
340 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
341 		    (LINUX_AT_COUNT * 2);
342 
343 		/*
344 		 * The '+ 2' is for the null pointers at the end of each of
345 		 * the arg and env vector sets,and imgp->auxarg_size is room
346 		 * for argument of Runtime loader.
347 		 */
348 		vectp = (char **)(destp - (imgp->args->argc +
349 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
350 
351 	} else {
352 		/*
353 		 * The '+ 2' is for the null pointers at the end of each of
354 		 * the arg and env vector sets
355 		 */
356 		vectp = (char **)(destp - (imgp->args->argc +
357 		    imgp->args->envc + 2) * sizeof(char *));
358 	}
359 
360 	/* vectp also becomes our initial stack base. */
361 	stack_base = (register_t *)vectp;
362 
363 	stringp = imgp->args->begin_argv;
364 	argc = imgp->args->argc;
365 	envc = imgp->args->envc;
366 
367 	/* Copy out strings - arguments and environment. */
368 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
369 
370 	/* Fill in "ps_strings" struct for ps, w, etc. */
371 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
372 	suword(&arginfo->ps_nargvstr, argc);
373 
374 	/* Fill in argument portion of vector table. */
375 	for (; argc > 0; --argc) {
376 		suword(vectp++, (long)(intptr_t)destp);
377 		while (*stringp++ != 0)
378 			destp++;
379 		destp++;
380 	}
381 
382 	/* A null vector table pointer separates the argp's from the envp's. */
383 	suword(vectp++, 0);
384 
385 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
386 	suword(&arginfo->ps_nenvstr, envc);
387 
388 	/* Fill in environment portion of vector table. */
389 	for (; envc > 0; --envc) {
390 		suword(vectp++, (long)(intptr_t)destp);
391 		while (*stringp++ != 0)
392 			destp++;
393 		destp++;
394 	}
395 
396 	/* The end of the vector table is a null pointer. */
397 	suword(vectp, 0);
398 	return (stack_base);
399 }
400 
401 /*
402  * Reset registers to default values on exec.
403  */
404 static void
405 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
406 {
407 	struct trapframe *regs = td->td_frame;
408 	struct pcb *pcb = td->td_pcb;
409 
410 	if (td->td_proc->p_md.md_ldt != NULL)
411 		user_ldt_free(td);
412 
413 	pcb->pcb_fsbase = 0;
414 	pcb->pcb_gsbase = 0;
415 	clear_pcb_flags(pcb, PCB_32BIT);
416 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
417 	set_pcb_flags(pcb, PCB_FULL_IRET);
418 
419 	bzero((char *)regs, sizeof(struct trapframe));
420 	regs->tf_rip = imgp->entry_addr;
421 	regs->tf_rsp = stack;
422 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
423 	regs->tf_ss = _udatasel;
424 	regs->tf_cs = _ucodesel;
425 	regs->tf_ds = _udatasel;
426 	regs->tf_es = _udatasel;
427 	regs->tf_fs = _ufssel;
428 	regs->tf_gs = _ugssel;
429 	regs->tf_flags = TF_HASSEGS;
430 
431 	/*
432 	 * Reset the hardware debug registers if they were in use.
433 	 * They won't have any meaning for the newly exec'd process.
434 	 */
435 	if (pcb->pcb_flags & PCB_DBREGS) {
436 		pcb->pcb_dr0 = 0;
437 		pcb->pcb_dr1 = 0;
438 		pcb->pcb_dr2 = 0;
439 		pcb->pcb_dr3 = 0;
440 		pcb->pcb_dr6 = 0;
441 		pcb->pcb_dr7 = 0;
442 		if (pcb == curpcb) {
443 			/*
444 			 * Clear the debug registers on the running
445 			 * CPU, otherwise they will end up affecting
446 			 * the next process we switch to.
447 			 */
448 			reset_dbregs();
449 		}
450 		clear_pcb_flags(pcb, PCB_DBREGS);
451 	}
452 
453 	/*
454 	 * Drop the FP state if we hold it, so that the process gets a
455 	 * clean FP state if it uses the FPU again.
456 	 */
457 	fpstate_drop(td);
458 }
459 
460 /*
461  * Copied from amd64/amd64/machdep.c
462  *
463  * XXX fpu state need? don't think so
464  */
465 int
466 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
467 {
468 	struct proc *p;
469 	struct l_ucontext uc;
470 	struct l_sigcontext *context;
471 	struct trapframe *regs;
472 	unsigned long rflags;
473 	int error;
474 	ksiginfo_t ksi;
475 
476 	regs = td->td_frame;
477 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
478 	if (error != 0)
479 		return (error);
480 
481 	p = td->td_proc;
482 	context = &uc.uc_mcontext;
483 	rflags = context->sc_rflags;
484 
485 	/*
486 	 * Don't allow users to change privileged or reserved flags.
487 	 */
488 	/*
489 	 * XXX do allow users to change the privileged flag PSL_RF.
490 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
491 	 * should sometimes set it there too.  tf_rflags is kept in
492 	 * the signal context during signal handling and there is no
493 	 * other place to remember it, so the PSL_RF bit may be
494 	 * corrupted by the signal handler without us knowing.
495 	 * Corruption of the PSL_RF bit at worst causes one more or
496 	 * one less debugger trap, so allowing it is fairly harmless.
497 	 */
498 
499 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
500 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
501 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
502 		return (EINVAL);
503 	}
504 
505 	/*
506 	 * Don't allow users to load a valid privileged %cs.  Let the
507 	 * hardware check for invalid selectors, excess privilege in
508 	 * other selectors, invalid %eip's and invalid %esp's.
509 	 */
510 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
511 	if (!CS_SECURE(context->sc_cs)) {
512 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
513 		ksiginfo_init_trap(&ksi);
514 		ksi.ksi_signo = SIGBUS;
515 		ksi.ksi_code = BUS_OBJERR;
516 		ksi.ksi_trapno = T_PROTFLT;
517 		ksi.ksi_addr = (void *)regs->tf_rip;
518 		trapsignal(td, &ksi);
519 		return (EINVAL);
520 	}
521 
522 	PROC_LOCK(p);
523 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
524 	SIG_CANTMASK(td->td_sigmask);
525 	signotify(td);
526 	PROC_UNLOCK(p);
527 
528 	regs->tf_rdi    = context->sc_rdi;
529 	regs->tf_rsi    = context->sc_rsi;
530 	regs->tf_rdx    = context->sc_rdx;
531 	regs->tf_rbp    = context->sc_rbp;
532 	regs->tf_rbx    = context->sc_rbx;
533 	regs->tf_rcx    = context->sc_rcx;
534 	regs->tf_rax    = context->sc_rax;
535 	regs->tf_rip    = context->sc_rip;
536 	regs->tf_rsp    = context->sc_rsp;
537 	regs->tf_r8     = context->sc_r8;
538 	regs->tf_r9     = context->sc_r9;
539 	regs->tf_r10    = context->sc_r10;
540 	regs->tf_r11    = context->sc_r11;
541 	regs->tf_r12    = context->sc_r12;
542 	regs->tf_r13    = context->sc_r13;
543 	regs->tf_r14    = context->sc_r14;
544 	regs->tf_r15    = context->sc_r15;
545 	regs->tf_cs     = context->sc_cs;
546 	regs->tf_err    = context->sc_err;
547 	regs->tf_rflags = rflags;
548 
549 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
550 	return (EJUSTRETURN);
551 }
552 
553 /*
554  * copied from amd64/amd64/machdep.c
555  *
556  * Send an interrupt to process.
557  */
558 static void
559 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
560 {
561 	struct l_rt_sigframe sf, *sfp;
562 	struct proc *p;
563 	struct thread *td;
564 	struct sigacts *psp;
565 	caddr_t sp;
566 	struct trapframe *regs;
567 	int sig, code;
568 	int oonstack;
569 
570 	td = curthread;
571 	p = td->td_proc;
572 	PROC_LOCK_ASSERT(p, MA_OWNED);
573 	sig = ksi->ksi_signo;
574 	psp = p->p_sigacts;
575 	code = ksi->ksi_code;
576 	mtx_assert(&psp->ps_mtx, MA_OWNED);
577 	regs = td->td_frame;
578 	oonstack = sigonstack(regs->tf_rsp);
579 
580 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
581 	    catcher, sig, mask, code);
582 
583 	/* Allocate space for the signal handler context. */
584 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
585 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
586 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
587 		    sizeof(struct l_rt_sigframe);
588 	} else
589 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
590 	/* Align to 16 bytes. */
591 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
592 	mtx_unlock(&psp->ps_mtx);
593 
594 	/* Translate the signal. */
595 	sig = bsd_to_linux_signal(sig);
596 
597 	/* Save user context. */
598 	bzero(&sf, sizeof(sf));
599 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
600 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
601 
602 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
603 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
604 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
605 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
606 	PROC_UNLOCK(p);
607 
608 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
609 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
610 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
611 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
612 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
613 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
614 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
615 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
616 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
617 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
618 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
619 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
620 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
621 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
622 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
623 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
624 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
625 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
626 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
627 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
628 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
629 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
630 
631 	/* Build the argument list for the signal handler. */
632 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
633 	regs->tf_rax = 0;
634 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
635 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
636 
637 	sf.sf_handler = catcher;
638 	/* Fill in POSIX parts. */
639 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
640 
641 	/* Copy the sigframe out to the user's stack. */
642 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
643 #ifdef DEBUG
644 		printf("process %ld has trashed its stack\n", (long)p->p_pid);
645 #endif
646 		PROC_LOCK(p);
647 		sigexit(td, SIGILL);
648 	}
649 
650 	regs->tf_rsp = (long)sfp;
651 	regs->tf_rip = linux_rt_sigcode;
652 	regs->tf_rflags &= ~(PSL_T | PSL_D);
653 	regs->tf_cs = _ucodesel;
654 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
655 	PROC_LOCK(p);
656 	mtx_lock(&psp->ps_mtx);
657 }
658 
659 #define	LINUX_VSYSCALL_START		(-10UL << 20)
660 #define	LINUX_VSYSCALL_SZ		1024
661 
662 const unsigned long linux_vsyscall_vector[] = {
663 	LINUX_SYS_gettimeofday,
664 	LINUX_SYS_linux_time,
665 				/* getcpu not implemented */
666 };
667 
668 static int
669 linux_vsyscall(struct thread *td)
670 {
671 	struct trapframe *frame;
672 	uint64_t retqaddr;
673 	int code, traced;
674 	int error;
675 
676 	frame = td->td_frame;
677 
678 	/* Check %rip for vsyscall area. */
679 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
680 		return (EINVAL);
681 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
682 		return (EINVAL);
683 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
684 	if (code >= nitems(linux_vsyscall_vector))
685 		return (EINVAL);
686 
687 	/*
688 	 * vsyscall called as callq *(%rax), so we must
689 	 * use return address from %rsp and also fixup %rsp.
690 	 */
691 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
692 	if (error)
693 		return (error);
694 
695 	frame->tf_rip = retqaddr;
696 	frame->tf_rax = linux_vsyscall_vector[code];
697 	frame->tf_rsp += 8;
698 
699 	traced = (frame->tf_flags & PSL_T);
700 
701 	amd64_syscall(td, traced);
702 
703 	return (0);
704 }
705 
706 struct sysentvec elf_linux_sysvec = {
707 	.sv_size	= LINUX_SYS_MAXSYSCALL,
708 	.sv_table	= linux_sysent,
709 	.sv_mask	= 0,
710 	.sv_errsize	= ELAST + 1,
711 	.sv_errtbl	= linux_errtbl,
712 	.sv_transtrap	= linux_translate_traps,
713 	.sv_fixup	= linux_fixup_elf,
714 	.sv_sendsig	= linux_rt_sendsig,
715 	.sv_sigcode	= &_binary_linux_locore_o_start,
716 	.sv_szsigcode	= &linux_szsigcode,
717 	.sv_name	= "Linux ELF64",
718 	.sv_coredump	= elf64_coredump,
719 	.sv_imgact_try	= linux_exec_imgact_try,
720 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
721 	.sv_pagesize	= PAGE_SIZE,
722 	.sv_minuser	= VM_MIN_ADDRESS,
723 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
724 	.sv_usrstack	= USRSTACK,
725 	.sv_psstrings	= PS_STRINGS,
726 	.sv_stackprot	= VM_PROT_ALL,
727 	.sv_copyout_strings = linux_copyout_strings,
728 	.sv_setregs	= linux_exec_setregs,
729 	.sv_fixlimit	= NULL,
730 	.sv_maxssiz	= NULL,
731 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
732 	.sv_set_syscall_retval = linux_set_syscall_retval,
733 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
734 	.sv_syscallnames = NULL,
735 	.sv_shared_page_base = SHAREDPAGE,
736 	.sv_shared_page_len = PAGE_SIZE,
737 	.sv_schedtail	= linux_schedtail,
738 	.sv_thread_detach = linux_thread_detach,
739 	.sv_trap	= linux_vsyscall,
740 };
741 
742 static void
743 linux_vdso_install(void *param)
744 {
745 
746 	amd64_lower_shared_page(&elf_linux_sysvec);
747 
748 	linux_szsigcode = (&_binary_linux_locore_o_end -
749 	    &_binary_linux_locore_o_start);
750 
751 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
752 		panic("Linux invalid vdso size\n");
753 
754 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
755 
756 	linux_shared_page_obj = __elfN(linux_shared_page_init)
757 	    (&linux_shared_page_mapping);
758 
759 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
760 
761 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
762 	    linux_szsigcode);
763 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
764 
765 	linux_kplatform = linux_shared_page_mapping +
766 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
767 }
768 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
769     linux_vdso_install, NULL);
770 
771 static void
772 linux_vdso_deinstall(void *param)
773 {
774 
775 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
776 }
777 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
778     linux_vdso_deinstall, NULL);
779 
780 static char GNULINUX_ABI_VENDOR[] = "GNU";
781 static int GNULINUX_ABI_DESC = 0;
782 
783 static bool
784 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
785 {
786 	const Elf32_Word *desc;
787 	uintptr_t p;
788 
789 	p = (uintptr_t)(note + 1);
790 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
791 
792 	desc = (const Elf32_Word *)p;
793 	if (desc[0] != GNULINUX_ABI_DESC)
794 		return (false);
795 
796 	/*
797 	 * For Linux we encode osrel as follows (see linux_mib.c):
798 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
799 	 */
800 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
801 
802 	return (true);
803 }
804 
805 static Elf_Brandnote linux64_brandnote = {
806 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
807 	.hdr.n_descsz	= 16,
808 	.hdr.n_type	= 1,
809 	.vendor		= GNULINUX_ABI_VENDOR,
810 	.flags		= BN_TRANSLATE_OSREL,
811 	.trans_osrel	= linux_trans_osrel
812 };
813 
814 static Elf64_Brandinfo linux_glibc2brand = {
815 	.brand		= ELFOSABI_LINUX,
816 	.machine	= EM_X86_64,
817 	.compat_3_brand	= "Linux",
818 	.emul_path	= "/compat/linux",
819 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
820 	.sysvec		= &elf_linux_sysvec,
821 	.interp_newpath	= NULL,
822 	.brand_note	= &linux64_brandnote,
823 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
824 };
825 
826 static Elf64_Brandinfo linux_glibc2brandshort = {
827 	.brand		= ELFOSABI_LINUX,
828 	.machine	= EM_X86_64,
829 	.compat_3_brand	= "Linux",
830 	.emul_path	= "/compat/linux",
831 	.interp_path	= "/lib64/ld-linux.so.2",
832 	.sysvec		= &elf_linux_sysvec,
833 	.interp_newpath	= NULL,
834 	.brand_note	= &linux64_brandnote,
835 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
836 };
837 
838 static Elf64_Brandinfo linux_muslbrand = {
839 	.brand		= ELFOSABI_LINUX,
840 	.machine	= EM_X86_64,
841 	.compat_3_brand	= "Linux",
842 	.emul_path	= "/compat/linux",
843 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
844 	.sysvec		= &elf_linux_sysvec,
845 	.interp_newpath	= NULL,
846 	.brand_note	= &linux64_brandnote,
847 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
848 };
849 
850 Elf64_Brandinfo *linux_brandlist[] = {
851 	&linux_glibc2brand,
852 	&linux_glibc2brandshort,
853 	&linux_muslbrand,
854 	NULL
855 };
856 
857 static int
858 linux64_elf_modevent(module_t mod, int type, void *data)
859 {
860 	Elf64_Brandinfo **brandinfo;
861 	int error;
862 	struct linux_ioctl_handler **lihp;
863 
864 	error = 0;
865 
866 	switch(type) {
867 	case MOD_LOAD:
868 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
869 		     ++brandinfo)
870 			if (elf64_insert_brand_entry(*brandinfo) < 0)
871 				error = EINVAL;
872 		if (error == 0) {
873 			SET_FOREACH(lihp, linux_ioctl_handler_set)
874 				linux_ioctl_register_handler(*lihp);
875 			LIST_INIT(&futex_list);
876 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
877 			stclohz = (stathz ? stathz : hz);
878 			if (bootverbose)
879 				printf("Linux x86-64 ELF exec handler installed\n");
880 		} else
881 			printf("cannot insert Linux x86-64 ELF brand handler\n");
882 		break;
883 	case MOD_UNLOAD:
884 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
885 		     ++brandinfo)
886 			if (elf64_brand_inuse(*brandinfo))
887 				error = EBUSY;
888 		if (error == 0) {
889 			for (brandinfo = &linux_brandlist[0];
890 			     *brandinfo != NULL; ++brandinfo)
891 				if (elf64_remove_brand_entry(*brandinfo) < 0)
892 					error = EINVAL;
893 		}
894 		if (error == 0) {
895 			SET_FOREACH(lihp, linux_ioctl_handler_set)
896 				linux_ioctl_unregister_handler(*lihp);
897 			mtx_destroy(&futex_mtx);
898 			if (bootverbose)
899 				printf("Linux ELF exec handler removed\n");
900 		} else
901 			printf("Could not deinstall ELF interpreter entry\n");
902 		break;
903 	default:
904 		return (EOPNOTSUPP);
905 	}
906 	return (error);
907 }
908 
909 static moduledata_t linux64_elf_mod = {
910 	"linux64elf",
911 	linux64_elf_modevent,
912 	0
913 };
914 
915 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
916 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
917 FEATURE(linux64, "Linux 64bit support");
918