xref: /freebsd-14.2/sys/amd64/linux/linux_sysvec.c (revision 052d3c12)
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 #include "opt_compat.h"
38 
39 #define	__ELF_WORD_SIZE	64
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/exec.h>
44 #include <sys/fcntl.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/ktr.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/module.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/resourcevar.h>
55 #include <sys/signalvar.h>
56 #include <sys/sysctl.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
59 #include <sys/sysproto.h>
60 #include <sys/vnode.h>
61 #include <sys/eventhandler.h>
62 
63 #include <vm/vm.h>
64 #include <vm/pmap.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_param.h>
70 
71 #include <machine/cpu.h>
72 #include <machine/md_var.h>
73 #include <machine/pcb.h>
74 #include <machine/specialreg.h>
75 
76 #include <amd64/linux/linux.h>
77 #include <amd64/linux/linux_proto.h>
78 #include <compat/linux/linux_emul.h>
79 #include <compat/linux/linux_futex.h>
80 #include <compat/linux/linux_ioctl.h>
81 #include <compat/linux/linux_mib.h>
82 #include <compat/linux/linux_misc.h>
83 #include <compat/linux/linux_signal.h>
84 #include <compat/linux/linux_sysproto.h>
85 #include <compat/linux/linux_util.h>
86 #include <compat/linux/linux_vdso.h>
87 
88 MODULE_VERSION(linux64, 1);
89 
90 #if BYTE_ORDER == LITTLE_ENDIAN
91 #define SHELLMAGIC      0x2123 /* #! */
92 #else
93 #define SHELLMAGIC      0x2321
94 #endif
95 
96 #if defined(DEBUG)
97 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
98 	    CTLTYPE_STRING | CTLFLAG_RW,
99 	    0, 0, linux_sysctl_debug, "A",
100 	    "Linux 64 debugging control");
101 #endif
102 
103 /*
104  * Allow the sendsig functions to use the ldebug() facility even though they
105  * are not syscalls themselves.  Map them to syscall 0.  This is slightly less
106  * bogus than using ldebug(sigreturn).
107  */
108 #define	LINUX_SYS_linux_rt_sendsig	0
109 
110 const char *linux_kplatform;
111 static int linux_szsigcode;
112 static vm_object_t linux_shared_page_obj;
113 static char *linux_shared_page_mapping;
114 extern char _binary_linux_locore_o_start;
115 extern char _binary_linux_locore_o_end;
116 
117 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
118 
119 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
120 
121 static register_t * linux_copyout_strings(struct image_params *imgp);
122 static int	elf_linux_fixup(register_t **stack_base,
123 		    struct image_params *iparams);
124 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
125 static void	linux_vdso_install(void *param);
126 static void	linux_vdso_deinstall(void *param);
127 static void	linux_set_syscall_retval(struct thread *td, int error);
128 static int	linux_fetch_syscall_args(struct thread *td);
129 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
130 		    u_long stack);
131 static int	linux_vsyscall(struct thread *td);
132 
133 /*
134  * Linux syscalls return negative errno's, we do positive and map them
135  * Reference:
136  *   FreeBSD: src/sys/sys/errno.h
137  *   Linux:   linux-2.6.17.8/include/asm-generic/errno-base.h
138  *            linux-2.6.17.8/include/asm-generic/errno.h
139  */
140 static int bsd_to_linux_errno[ELAST + 1] = {
141 	-0,  -1,  -2,  -3,  -4,  -5,  -6,  -7,  -8,  -9,
142 	-10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
143 	-20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
144 	-30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
145 	-90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
146 	-100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
147 	-110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
148 	-116, -66,  -6,  -6,  -6,  -6,  -6, -37, -38,  -9,
149 	  -6,  -6, -43, -42, -75,-125, -84, -61, -16, -74,
150 	 -72, -67, -71
151 };
152 
153 #define LINUX_T_UNKNOWN  255
154 static int _bsd_to_linux_trapcode[] = {
155 	LINUX_T_UNKNOWN,	/* 0 */
156 	6,			/* 1  T_PRIVINFLT */
157 	LINUX_T_UNKNOWN,	/* 2 */
158 	3,			/* 3  T_BPTFLT */
159 	LINUX_T_UNKNOWN,	/* 4 */
160 	LINUX_T_UNKNOWN,	/* 5 */
161 	16,			/* 6  T_ARITHTRAP */
162 	254,			/* 7  T_ASTFLT */
163 	LINUX_T_UNKNOWN,	/* 8 */
164 	13,			/* 9  T_PROTFLT */
165 	1,			/* 10 T_TRCTRAP */
166 	LINUX_T_UNKNOWN,	/* 11 */
167 	14,			/* 12 T_PAGEFLT */
168 	LINUX_T_UNKNOWN,	/* 13 */
169 	17,			/* 14 T_ALIGNFLT */
170 	LINUX_T_UNKNOWN,	/* 15 */
171 	LINUX_T_UNKNOWN,	/* 16 */
172 	LINUX_T_UNKNOWN,	/* 17 */
173 	0,			/* 18 T_DIVIDE */
174 	2,			/* 19 T_NMI */
175 	4,			/* 20 T_OFLOW */
176 	5,			/* 21 T_BOUND */
177 	7,			/* 22 T_DNA */
178 	8,			/* 23 T_DOUBLEFLT */
179 	9,			/* 24 T_FPOPFLT */
180 	10,			/* 25 T_TSSFLT */
181 	11,			/* 26 T_SEGNPFLT */
182 	12,			/* 27 T_STKFLT */
183 	18,			/* 28 T_MCHK */
184 	19,			/* 29 T_XMMFLT */
185 	15			/* 30 T_RESERVED */
186 };
187 #define bsd_to_linux_trapcode(code) \
188     ((code)<nitems(_bsd_to_linux_trapcode)? \
189      _bsd_to_linux_trapcode[(code)]: \
190      LINUX_T_UNKNOWN)
191 
192 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
193 LINUX_VDSO_SYM_CHAR(linux_platform);
194 
195 /*
196  * If FreeBSD & Linux have a difference of opinion about what a trap
197  * means, deal with it here.
198  *
199  * MPSAFE
200  */
201 static int
202 translate_traps(int signal, int trap_code)
203 {
204 
205 	if (signal != SIGBUS)
206 		return signal;
207 	switch (trap_code) {
208 	case T_PROTFLT:
209 	case T_TSSFLT:
210 	case T_DOUBLEFLT:
211 	case T_PAGEFLT:
212 		return SIGSEGV;
213 	default:
214 		return signal;
215 	}
216 }
217 
218 static int
219 linux_fetch_syscall_args(struct thread *td)
220 {
221 	struct proc *p;
222 	struct trapframe *frame;
223 	struct syscall_args *sa;
224 
225 	p = td->td_proc;
226 	frame = td->td_frame;
227 	sa = &td->td_sa;
228 
229 	sa->args[0] = frame->tf_rdi;
230 	sa->args[1] = frame->tf_rsi;
231 	sa->args[2] = frame->tf_rdx;
232 	sa->args[3] = frame->tf_rcx;
233 	sa->args[4] = frame->tf_r8;
234 	sa->args[5] = frame->tf_r9;
235 	sa->code = frame->tf_rax;
236 
237 	if (sa->code >= p->p_sysent->sv_size)
238 		/* nosys */
239 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
240 	else
241 		sa->callp = &p->p_sysent->sv_table[sa->code];
242 	sa->narg = sa->callp->sy_narg;
243 
244 	td->td_retval[0] = 0;
245 	return (0);
246 }
247 
248 static void
249 linux_set_syscall_retval(struct thread *td, int error)
250 {
251 	struct trapframe *frame = td->td_frame;
252 
253 	/*
254 	 * On Linux only %rcx and %r11 values are not preserved across
255 	 * the syscall.  So, do not clobber %rdx and %r10.
256 	 */
257 	td->td_retval[1] = frame->tf_rdx;
258 	frame->tf_r10 = frame->tf_rcx;
259 
260 	cpu_set_syscall_retval(td, error);
261 
262 	 /* Restore all registers. */
263 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
264 }
265 
266 static int
267 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
268 {
269 	Elf_Auxargs *args;
270 	Elf_Addr *base;
271 	Elf_Addr *pos;
272 	struct ps_strings *arginfo;
273 	struct proc *p;
274 	int issetugid;
275 
276 	p = imgp->proc;
277 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
278 
279 	KASSERT(curthread->td_proc == imgp->proc,
280 	    ("unsafe elf_linux_fixup(), should be curproc"));
281 	base = (Elf64_Addr *)*stack_base;
282 	args = (Elf64_Auxargs *)imgp->auxargs;
283 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
284 
285 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
286 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
287 	    imgp->proc->p_sysent->sv_shared_page_base);
288 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
289 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
290 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
291 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
292 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
293 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
294 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
295 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
296 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
297 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
298 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
299 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
300 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
301 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
302 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
303 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
304 	if (imgp->execpathp != 0)
305 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
306 	if (args->execfd != -1)
307 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
308 	AUXARGS_ENTRY(pos, AT_NULL, 0);
309 	free(imgp->auxargs, M_TEMP);
310 	imgp->auxargs = NULL;
311 
312 	base--;
313 	suword(base, (uint64_t)imgp->args->argc);
314 
315 	*stack_base = (register_t *)base;
316 	return (0);
317 }
318 
319 /*
320  * Copy strings out to the new process address space, constructing new arg
321  * and env vector tables. Return a pointer to the base so that it can be used
322  * as the initial stack pointer.
323  */
324 static register_t *
325 linux_copyout_strings(struct image_params *imgp)
326 {
327 	int argc, envc;
328 	char **vectp;
329 	char *stringp, *destp;
330 	register_t *stack_base;
331 	struct ps_strings *arginfo;
332 	char canary[LINUX_AT_RANDOM_LEN];
333 	size_t execpath_len;
334 	struct proc *p;
335 
336 	/* Calculate string base and vector table pointers. */
337 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
338 		execpath_len = strlen(imgp->execpath) + 1;
339 	else
340 		execpath_len = 0;
341 
342 	p = imgp->proc;
343 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
344 	destp = (caddr_t)arginfo - SPARE_USRSPACE -
345 	    roundup(sizeof(canary), sizeof(char *)) -
346 	    roundup(execpath_len, sizeof(char *)) -
347 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
348 
349 	if (execpath_len != 0) {
350 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
351 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
352 	}
353 
354 	/* Prepare the canary for SSP. */
355 	arc4rand(canary, sizeof(canary), 0);
356 	imgp->canary = (uintptr_t)arginfo -
357 	    roundup(execpath_len, sizeof(char *)) -
358 	    roundup(sizeof(canary), sizeof(char *));
359 	copyout(canary, (void *)imgp->canary, sizeof(canary));
360 
361 	/* If we have a valid auxargs ptr, prepare some room on the stack. */
362 	if (imgp->auxargs) {
363 		/*
364 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
365 		 * lower compatibility.
366 		 */
367 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
368 		    (LINUX_AT_COUNT * 2);
369 
370 		/*
371 		 * The '+ 2' is for the null pointers at the end of each of
372 		 * the arg and env vector sets,and imgp->auxarg_size is room
373 		 * for argument of Runtime loader.
374 		 */
375 		vectp = (char **)(destp - (imgp->args->argc +
376 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
377 
378 	} else {
379 		/*
380 		 * The '+ 2' is for the null pointers at the end of each of
381 		 * the arg and env vector sets
382 		 */
383 		vectp = (char **)(destp - (imgp->args->argc +
384 		    imgp->args->envc + 2) * sizeof(char *));
385 	}
386 
387 	/* vectp also becomes our initial stack base. */
388 	stack_base = (register_t *)vectp;
389 
390 	stringp = imgp->args->begin_argv;
391 	argc = imgp->args->argc;
392 	envc = imgp->args->envc;
393 
394 	/* Copy out strings - arguments and environment. */
395 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
396 
397 	/* Fill in "ps_strings" struct for ps, w, etc. */
398 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
399 	suword(&arginfo->ps_nargvstr, argc);
400 
401 	/* Fill in argument portion of vector table. */
402 	for (; argc > 0; --argc) {
403 		suword(vectp++, (long)(intptr_t)destp);
404 		while (*stringp++ != 0)
405 			destp++;
406 		destp++;
407 	}
408 
409 	/* A null vector table pointer separates the argp's from the envp's. */
410 	suword(vectp++, 0);
411 
412 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
413 	suword(&arginfo->ps_nenvstr, envc);
414 
415 	/* Fill in environment portion of vector table. */
416 	for (; envc > 0; --envc) {
417 		suword(vectp++, (long)(intptr_t)destp);
418 		while (*stringp++ != 0)
419 			destp++;
420 		destp++;
421 	}
422 
423 	/* The end of the vector table is a null pointer. */
424 	suword(vectp, 0);
425 	return (stack_base);
426 }
427 
428 /*
429  * Reset registers to default values on exec.
430  */
431 static void
432 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
433 {
434 	struct trapframe *regs = td->td_frame;
435 	struct pcb *pcb = td->td_pcb;
436 
437 	if (td->td_proc->p_md.md_ldt != NULL)
438 		user_ldt_free(td);
439 
440 	pcb->pcb_fsbase = 0;
441 	pcb->pcb_gsbase = 0;
442 	clear_pcb_flags(pcb, PCB_32BIT);
443 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
444 	set_pcb_flags(pcb, PCB_FULL_IRET);
445 
446 	bzero((char *)regs, sizeof(struct trapframe));
447 	regs->tf_rip = imgp->entry_addr;
448 	regs->tf_rsp = stack;
449 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
450 	regs->tf_ss = _udatasel;
451 	regs->tf_cs = _ucodesel;
452 	regs->tf_ds = _udatasel;
453 	regs->tf_es = _udatasel;
454 	regs->tf_fs = _ufssel;
455 	regs->tf_gs = _ugssel;
456 	regs->tf_flags = TF_HASSEGS;
457 
458 	/*
459 	 * Reset the hardware debug registers if they were in use.
460 	 * They won't have any meaning for the newly exec'd process.
461 	 */
462 	if (pcb->pcb_flags & PCB_DBREGS) {
463 		pcb->pcb_dr0 = 0;
464 		pcb->pcb_dr1 = 0;
465 		pcb->pcb_dr2 = 0;
466 		pcb->pcb_dr3 = 0;
467 		pcb->pcb_dr6 = 0;
468 		pcb->pcb_dr7 = 0;
469 		if (pcb == curpcb) {
470 			/*
471 			 * Clear the debug registers on the running
472 			 * CPU, otherwise they will end up affecting
473 			 * the next process we switch to.
474 			 */
475 			reset_dbregs();
476 		}
477 		clear_pcb_flags(pcb, PCB_DBREGS);
478 	}
479 
480 	/*
481 	 * Drop the FP state if we hold it, so that the process gets a
482 	 * clean FP state if it uses the FPU again.
483 	 */
484 	fpstate_drop(td);
485 }
486 
487 /*
488  * Copied from amd64/amd64/machdep.c
489  *
490  * XXX fpu state need? don't think so
491  */
492 int
493 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
494 {
495 	struct proc *p;
496 	struct l_ucontext uc;
497 	struct l_sigcontext *context;
498 	struct trapframe *regs;
499 	unsigned long rflags;
500 	int error;
501 	ksiginfo_t ksi;
502 
503 	regs = td->td_frame;
504 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
505 	if (error != 0)
506 		return (error);
507 
508 	p = td->td_proc;
509 	context = &uc.uc_mcontext;
510 	rflags = context->sc_rflags;
511 
512 	/*
513 	 * Don't allow users to change privileged or reserved flags.
514 	 */
515 	/*
516 	 * XXX do allow users to change the privileged flag PSL_RF.
517 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
518 	 * should sometimes set it there too.  tf_rflags is kept in
519 	 * the signal context during signal handling and there is no
520 	 * other place to remember it, so the PSL_RF bit may be
521 	 * corrupted by the signal handler without us knowing.
522 	 * Corruption of the PSL_RF bit at worst causes one more or
523 	 * one less debugger trap, so allowing it is fairly harmless.
524 	 */
525 
526 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
527 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
528 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
529 		return (EINVAL);
530 	}
531 
532 	/*
533 	 * Don't allow users to load a valid privileged %cs.  Let the
534 	 * hardware check for invalid selectors, excess privilege in
535 	 * other selectors, invalid %eip's and invalid %esp's.
536 	 */
537 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
538 	if (!CS_SECURE(context->sc_cs)) {
539 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
540 		ksiginfo_init_trap(&ksi);
541 		ksi.ksi_signo = SIGBUS;
542 		ksi.ksi_code = BUS_OBJERR;
543 		ksi.ksi_trapno = T_PROTFLT;
544 		ksi.ksi_addr = (void *)regs->tf_rip;
545 		trapsignal(td, &ksi);
546 		return (EINVAL);
547 	}
548 
549 	PROC_LOCK(p);
550 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
551 	SIG_CANTMASK(td->td_sigmask);
552 	signotify(td);
553 	PROC_UNLOCK(p);
554 
555 	regs->tf_rdi    = context->sc_rdi;
556 	regs->tf_rsi    = context->sc_rsi;
557 	regs->tf_rdx    = context->sc_rdx;
558 	regs->tf_rbp    = context->sc_rbp;
559 	regs->tf_rbx    = context->sc_rbx;
560 	regs->tf_rcx    = context->sc_rcx;
561 	regs->tf_rax    = context->sc_rax;
562 	regs->tf_rip    = context->sc_rip;
563 	regs->tf_rsp    = context->sc_rsp;
564 	regs->tf_r8     = context->sc_r8;
565 	regs->tf_r9     = context->sc_r9;
566 	regs->tf_r10    = context->sc_r10;
567 	regs->tf_r11    = context->sc_r11;
568 	regs->tf_r12    = context->sc_r12;
569 	regs->tf_r13    = context->sc_r13;
570 	regs->tf_r14    = context->sc_r14;
571 	regs->tf_r15    = context->sc_r15;
572 	regs->tf_cs     = context->sc_cs;
573 	regs->tf_err    = context->sc_err;
574 	regs->tf_rflags = rflags;
575 
576 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
577 	return (EJUSTRETURN);
578 }
579 
580 /*
581  * copied from amd64/amd64/machdep.c
582  *
583  * Send an interrupt to process.
584  */
585 static void
586 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
587 {
588 	struct l_rt_sigframe sf, *sfp;
589 	struct proc *p;
590 	struct thread *td;
591 	struct sigacts *psp;
592 	caddr_t sp;
593 	struct trapframe *regs;
594 	int sig, code;
595 	int oonstack;
596 
597 	td = curthread;
598 	p = td->td_proc;
599 	PROC_LOCK_ASSERT(p, MA_OWNED);
600 	sig = ksi->ksi_signo;
601 	psp = p->p_sigacts;
602 	code = ksi->ksi_code;
603 	mtx_assert(&psp->ps_mtx, MA_OWNED);
604 	regs = td->td_frame;
605 	oonstack = sigonstack(regs->tf_rsp);
606 
607 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
608 	    catcher, sig, mask, code);
609 
610 	/* Allocate space for the signal handler context. */
611 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
612 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
613 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
614 		    sizeof(struct l_rt_sigframe);
615 	} else
616 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
617 	/* Align to 16 bytes. */
618 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
619 	mtx_unlock(&psp->ps_mtx);
620 
621 	/* Translate the signal. */
622 	sig = bsd_to_linux_signal(sig);
623 
624 	/* Save user context. */
625 	bzero(&sf, sizeof(sf));
626 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
627 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
628 
629 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
630 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
631 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
632 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
633 	PROC_UNLOCK(p);
634 
635 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
636 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
637 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
638 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
639 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
640 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
641 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
642 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
643 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
644 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
645 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
646 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
647 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
648 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
649 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
650 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
651 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
652 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
653 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
654 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
655 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
656 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
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 	sf.sf_handler = catcher;
665 	/* Fill in POSIX parts. */
666 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
667 
668 	/* Copy the sigframe out to the user's stack. */
669 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
670 #ifdef DEBUG
671 		printf("process %ld has trashed its stack\n", (long)p->p_pid);
672 #endif
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 /*
687  * If a Linux binary is exec'ing something, try this image activator
688  * first.  We override standard shell script execution in order to
689  * be able to modify the interpreter path.  We only do this if a Linux
690  * binary is doing the exec, so we do not create an EXEC module for it.
691  */
692 static int exec_linux_imgact_try(struct image_params *iparams);
693 
694 static int
695 exec_linux_imgact_try(struct image_params *imgp)
696 {
697 	const char *head = (const char *)imgp->image_header;
698 	char *rpath;
699 	int error = -1;
700 
701 	/*
702 	 * The interpreter for shell scripts run from a Linux binary needs
703 	 * to be located in /compat/linux if possible in order to recursively
704 	 * maintain Linux path emulation.
705 	 */
706 	if (((const short *)head)[0] == SHELLMAGIC) {
707 		/*
708 		 * Run our normal shell image activator.  If it succeeds
709 		 * attempt to use the alternate path for the interpreter.
710 		 * If an alternate path is found, use our stringspace
711 		 * to store it.
712 		 */
713 		if ((error = exec_shell_imgact(imgp)) == 0) {
714 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
715 			    imgp->interpreter_name, UIO_SYSSPACE,
716 			    &rpath, 0, AT_FDCWD);
717 			if (rpath != NULL)
718 				imgp->args->fname_buf =
719 				    imgp->interpreter_name = rpath;
720 		}
721 	}
722 	return (error);
723 }
724 
725 #define	LINUX_VSYSCALL_START		(-10UL << 20)
726 #define	LINUX_VSYSCALL_SZ		1024
727 
728 const unsigned long linux_vsyscall_vector[] = {
729 	LINUX_SYS_gettimeofday,
730 	LINUX_SYS_linux_time,
731 				/* getcpu not implemented */
732 };
733 
734 static int
735 linux_vsyscall(struct thread *td)
736 {
737 	struct trapframe *frame;
738 	uint64_t retqaddr;
739 	int code, traced;
740 	int error;
741 
742 	frame = td->td_frame;
743 
744 	/* Check %rip for vsyscall area. */
745 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
746 		return (EINVAL);
747 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
748 		return (EINVAL);
749 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
750 	if (code >= nitems(linux_vsyscall_vector))
751 		return (EINVAL);
752 
753 	/*
754 	 * vsyscall called as callq *(%rax), so we must
755 	 * use return address from %rsp and also fixup %rsp.
756 	 */
757 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
758 	if (error)
759 		return (error);
760 
761 	frame->tf_rip = retqaddr;
762 	frame->tf_rax = linux_vsyscall_vector[code];
763 	frame->tf_rsp += 8;
764 
765 	traced = (frame->tf_flags & PSL_T);
766 
767 	amd64_syscall(td, traced);
768 
769 	return (0);
770 }
771 
772 struct sysentvec elf_linux_sysvec = {
773 	.sv_size	= LINUX_SYS_MAXSYSCALL,
774 	.sv_table	= linux_sysent,
775 	.sv_mask	= 0,
776 	.sv_errsize	= ELAST + 1,
777 	.sv_errtbl	= bsd_to_linux_errno,
778 	.sv_transtrap	= translate_traps,
779 	.sv_fixup	= elf_linux_fixup,
780 	.sv_sendsig	= linux_rt_sendsig,
781 	.sv_sigcode	= &_binary_linux_locore_o_start,
782 	.sv_szsigcode	= &linux_szsigcode,
783 	.sv_name	= "Linux ELF64",
784 	.sv_coredump	= elf64_coredump,
785 	.sv_imgact_try	= exec_linux_imgact_try,
786 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
787 	.sv_pagesize	= PAGE_SIZE,
788 	.sv_minuser	= VM_MIN_ADDRESS,
789 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
790 	.sv_usrstack	= USRSTACK,
791 	.sv_psstrings	= PS_STRINGS,
792 	.sv_stackprot	= VM_PROT_ALL,
793 	.sv_copyout_strings = linux_copyout_strings,
794 	.sv_setregs	= linux_exec_setregs,
795 	.sv_fixlimit	= NULL,
796 	.sv_maxssiz	= NULL,
797 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
798 	.sv_set_syscall_retval = linux_set_syscall_retval,
799 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
800 	.sv_syscallnames = NULL,
801 	.sv_shared_page_base = SHAREDPAGE,
802 	.sv_shared_page_len = PAGE_SIZE,
803 	.sv_schedtail	= linux_schedtail,
804 	.sv_thread_detach = linux_thread_detach,
805 	.sv_trap	= linux_vsyscall,
806 };
807 
808 static void
809 linux_vdso_install(void *param)
810 {
811 
812 	amd64_lower_shared_page(&elf_linux_sysvec);
813 
814 	linux_szsigcode = (&_binary_linux_locore_o_end -
815 	    &_binary_linux_locore_o_start);
816 
817 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
818 		panic("Linux invalid vdso size\n");
819 
820 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
821 
822 	linux_shared_page_obj = __elfN(linux_shared_page_init)
823 	    (&linux_shared_page_mapping);
824 
825 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
826 
827 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
828 	    linux_szsigcode);
829 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
830 
831 	linux_kplatform = linux_shared_page_mapping +
832 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
833 }
834 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
835     (sysinit_cfunc_t)linux_vdso_install, NULL);
836 
837 static void
838 linux_vdso_deinstall(void *param)
839 {
840 
841 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
842 };
843 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
844     (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
845 
846 static char GNULINUX_ABI_VENDOR[] = "GNU";
847 static int GNULINUX_ABI_DESC = 0;
848 
849 static bool
850 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
851 {
852 	const Elf32_Word *desc;
853 	uintptr_t p;
854 
855 	p = (uintptr_t)(note + 1);
856 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
857 
858 	desc = (const Elf32_Word *)p;
859 	if (desc[0] != GNULINUX_ABI_DESC)
860 		return (false);
861 
862 	/*
863 	 * For Linux we encode osrel as follows (see linux_mib.c):
864 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
865 	 */
866 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
867 
868 	return (true);
869 }
870 
871 static Elf_Brandnote linux64_brandnote = {
872 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
873 	.hdr.n_descsz	= 16,
874 	.hdr.n_type	= 1,
875 	.vendor		= GNULINUX_ABI_VENDOR,
876 	.flags		= BN_TRANSLATE_OSREL,
877 	.trans_osrel	= linux_trans_osrel
878 };
879 
880 static Elf64_Brandinfo linux_glibc2brand = {
881 	.brand		= ELFOSABI_LINUX,
882 	.machine	= EM_X86_64,
883 	.compat_3_brand	= "Linux",
884 	.emul_path	= "/compat/linux",
885 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
886 	.sysvec		= &elf_linux_sysvec,
887 	.interp_newpath	= NULL,
888 	.brand_note	= &linux64_brandnote,
889 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
890 };
891 
892 static Elf64_Brandinfo linux_glibc2brandshort = {
893 	.brand		= ELFOSABI_LINUX,
894 	.machine	= EM_X86_64,
895 	.compat_3_brand	= "Linux",
896 	.emul_path	= "/compat/linux",
897 	.interp_path	= "/lib64/ld-linux.so.2",
898 	.sysvec		= &elf_linux_sysvec,
899 	.interp_newpath	= NULL,
900 	.brand_note	= &linux64_brandnote,
901 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
902 };
903 
904 static Elf64_Brandinfo linux_muslbrand = {
905 	.brand		= ELFOSABI_LINUX,
906 	.machine	= EM_X86_64,
907 	.compat_3_brand	= "Linux",
908 	.emul_path	= "/compat/linux",
909 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
910 	.sysvec		= &elf_linux_sysvec,
911 	.interp_newpath	= NULL,
912 	.brand_note	= &linux64_brandnote,
913 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
914 };
915 
916 Elf64_Brandinfo *linux_brandlist[] = {
917 	&linux_glibc2brand,
918 	&linux_glibc2brandshort,
919 	&linux_muslbrand,
920 	NULL
921 };
922 
923 static int
924 linux64_elf_modevent(module_t mod, int type, void *data)
925 {
926 	Elf64_Brandinfo **brandinfo;
927 	int error;
928 	struct linux_ioctl_handler **lihp;
929 
930 	error = 0;
931 
932 	switch(type) {
933 	case MOD_LOAD:
934 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
935 		     ++brandinfo)
936 			if (elf64_insert_brand_entry(*brandinfo) < 0)
937 				error = EINVAL;
938 		if (error == 0) {
939 			SET_FOREACH(lihp, linux_ioctl_handler_set)
940 				linux_ioctl_register_handler(*lihp);
941 			LIST_INIT(&futex_list);
942 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
943 			stclohz = (stathz ? stathz : hz);
944 			if (bootverbose)
945 				printf("Linux x86-64 ELF exec handler installed\n");
946 		} else
947 			printf("cannot insert Linux x86-64 ELF brand handler\n");
948 		break;
949 	case MOD_UNLOAD:
950 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
951 		     ++brandinfo)
952 			if (elf64_brand_inuse(*brandinfo))
953 				error = EBUSY;
954 		if (error == 0) {
955 			for (brandinfo = &linux_brandlist[0];
956 			     *brandinfo != NULL; ++brandinfo)
957 				if (elf64_remove_brand_entry(*brandinfo) < 0)
958 					error = EINVAL;
959 		}
960 		if (error == 0) {
961 			SET_FOREACH(lihp, linux_ioctl_handler_set)
962 				linux_ioctl_unregister_handler(*lihp);
963 			mtx_destroy(&futex_mtx);
964 			if (bootverbose)
965 				printf("Linux ELF exec handler removed\n");
966 		} else
967 			printf("Could not deinstall ELF interpreter entry\n");
968 		break;
969 	default:
970 		return (EOPNOTSUPP);
971 	}
972 	return (error);
973 }
974 
975 static moduledata_t linux64_elf_mod = {
976 	"linux64elf",
977 	linux64_elf_modevent,
978 	0
979 };
980 
981 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
982 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
983 FEATURE(linux64, "Linux 64bit support");
984