xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 48c779cdecb5f803e5fe5d761987e976ca9609db)
1 /*-
2  * Copyright (c) 2013 Dmitry Chagin
3  * Copyright (c) 2004 Tim J. Robbins
4  * Copyright (c) 2003 Peter Wemm
5  * Copyright (c) 2002 Doug Rabson
6  * Copyright (c) 1998-1999 Andrew Gallatin
7  * Copyright (c) 1994-1996 Søren Schmidt
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #define	__ELF_WORD_SIZE	64
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/exec.h>
42 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/kernel.h>
46 #include <sys/ktr.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/module.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/resourcevar.h>
53 #include <sys/signalvar.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysent.h>
57 #include <sys/sysproto.h>
58 #include <sys/vnode.h>
59 #include <sys/eventhandler.h>
60 
61 #include <vm/vm.h>
62 #include <vm/pmap.h>
63 #include <vm/vm_extern.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_param.h>
68 
69 #include <machine/cpu.h>
70 #include <machine/md_var.h>
71 #include <machine/pcb.h>
72 #include <machine/specialreg.h>
73 #include <machine/trap.h>
74 
75 #include <amd64/linux/linux.h>
76 #include <amd64/linux/linux_proto.h>
77 #include <compat/linux/linux_emul.h>
78 #include <compat/linux/linux_ioctl.h>
79 #include <compat/linux/linux_mib.h>
80 #include <compat/linux/linux_misc.h>
81 #include <compat/linux/linux_signal.h>
82 #include <compat/linux/linux_sysproto.h>
83 #include <compat/linux/linux_util.h>
84 #include <compat/linux/linux_vdso.h>
85 
86 MODULE_VERSION(linux64, 1);
87 
88 const char *linux_kplatform;
89 static int linux_szsigcode;
90 static vm_object_t linux_shared_page_obj;
91 static char *linux_shared_page_mapping;
92 extern char _binary_linux_locore_o_start;
93 extern char _binary_linux_locore_o_end;
94 
95 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
96 
97 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
98 
99 static register_t * linux_copyout_strings(struct image_params *imgp);
100 static int	linux_fixup_elf(register_t **stack_base,
101 		    struct image_params *iparams);
102 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
103 static void	linux_vdso_install(void *param);
104 static void	linux_vdso_deinstall(void *param);
105 static void	linux_set_syscall_retval(struct thread *td, int error);
106 static int	linux_fetch_syscall_args(struct thread *td);
107 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
108 		    u_long stack);
109 static int	linux_vsyscall(struct thread *td);
110 
111 #define LINUX_T_UNKNOWN  255
112 static int _bsd_to_linux_trapcode[] = {
113 	LINUX_T_UNKNOWN,	/* 0 */
114 	6,			/* 1  T_PRIVINFLT */
115 	LINUX_T_UNKNOWN,	/* 2 */
116 	3,			/* 3  T_BPTFLT */
117 	LINUX_T_UNKNOWN,	/* 4 */
118 	LINUX_T_UNKNOWN,	/* 5 */
119 	16,			/* 6  T_ARITHTRAP */
120 	254,			/* 7  T_ASTFLT */
121 	LINUX_T_UNKNOWN,	/* 8 */
122 	13,			/* 9  T_PROTFLT */
123 	1,			/* 10 T_TRCTRAP */
124 	LINUX_T_UNKNOWN,	/* 11 */
125 	14,			/* 12 T_PAGEFLT */
126 	LINUX_T_UNKNOWN,	/* 13 */
127 	17,			/* 14 T_ALIGNFLT */
128 	LINUX_T_UNKNOWN,	/* 15 */
129 	LINUX_T_UNKNOWN,	/* 16 */
130 	LINUX_T_UNKNOWN,	/* 17 */
131 	0,			/* 18 T_DIVIDE */
132 	2,			/* 19 T_NMI */
133 	4,			/* 20 T_OFLOW */
134 	5,			/* 21 T_BOUND */
135 	7,			/* 22 T_DNA */
136 	8,			/* 23 T_DOUBLEFLT */
137 	9,			/* 24 T_FPOPFLT */
138 	10,			/* 25 T_TSSFLT */
139 	11,			/* 26 T_SEGNPFLT */
140 	12,			/* 27 T_STKFLT */
141 	18,			/* 28 T_MCHK */
142 	19,			/* 29 T_XMMFLT */
143 	15			/* 30 T_RESERVED */
144 };
145 #define bsd_to_linux_trapcode(code) \
146     ((code)<nitems(_bsd_to_linux_trapcode)? \
147      _bsd_to_linux_trapcode[(code)]: \
148      LINUX_T_UNKNOWN)
149 
150 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
151 LINUX_VDSO_SYM_CHAR(linux_platform);
152 
153 /*
154  * If FreeBSD & Linux have a difference of opinion about what a trap
155  * means, deal with it here.
156  *
157  * MPSAFE
158  */
159 static int
160 linux_translate_traps(int signal, int trap_code)
161 {
162 
163 	if (signal != SIGBUS)
164 		return (signal);
165 	switch (trap_code) {
166 	case T_PROTFLT:
167 	case T_TSSFLT:
168 	case T_DOUBLEFLT:
169 	case T_PAGEFLT:
170 		return (SIGSEGV);
171 	default:
172 		return (signal);
173 	}
174 }
175 
176 static int
177 linux_fetch_syscall_args(struct thread *td)
178 {
179 	struct proc *p;
180 	struct trapframe *frame;
181 	struct syscall_args *sa;
182 
183 	p = td->td_proc;
184 	frame = td->td_frame;
185 	sa = &td->td_sa;
186 
187 	sa->args[0] = frame->tf_rdi;
188 	sa->args[1] = frame->tf_rsi;
189 	sa->args[2] = frame->tf_rdx;
190 	sa->args[3] = frame->tf_rcx;
191 	sa->args[4] = frame->tf_r8;
192 	sa->args[5] = frame->tf_r9;
193 	sa->code = frame->tf_rax;
194 
195 	if (sa->code >= p->p_sysent->sv_size)
196 		/* nosys */
197 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
198 	else
199 		sa->callp = &p->p_sysent->sv_table[sa->code];
200 	sa->narg = sa->callp->sy_narg;
201 
202 	td->td_retval[0] = 0;
203 	return (0);
204 }
205 
206 static void
207 linux_set_syscall_retval(struct thread *td, int error)
208 {
209 	struct trapframe *frame = td->td_frame;
210 
211 	/*
212 	 * On Linux only %rcx and %r11 values are not preserved across
213 	 * the syscall.  So, do not clobber %rdx and %r10.
214 	 */
215 	td->td_retval[1] = frame->tf_rdx;
216 	if (error != EJUSTRETURN)
217 		frame->tf_r10 = frame->tf_rcx;
218 
219 	cpu_set_syscall_retval(td, error);
220 
221 	 /* Restore all registers. */
222 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
223 }
224 
225 static int
226 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
227 {
228 	Elf_Auxargs *args;
229 	Elf_Auxinfo *argarray, *pos;
230 	Elf_Addr *auxbase, *base;
231 	struct ps_strings *arginfo;
232 	struct proc *p;
233 	int error, issetugid;
234 
235 	p = imgp->proc;
236 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
237 
238 	KASSERT(curthread->td_proc == imgp->proc,
239 	    ("unsafe linux_fixup_elf(), should be curproc"));
240 	base = (Elf64_Addr *)*stack_base;
241 	args = (Elf64_Auxargs *)imgp->auxargs;
242 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
243 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
244 	    M_WAITOK | M_ZERO);
245 
246 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
247 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
248 	    imgp->proc->p_sysent->sv_shared_page_base);
249 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
250 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
251 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
252 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
253 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
254 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
255 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
256 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
257 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
258 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
259 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
260 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
261 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
262 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
263 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
264 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
265 	if (imgp->execpathp != 0)
266 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
267 	if (args->execfd != -1)
268 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
269 	AUXARGS_ENTRY(pos, AT_NULL, 0);
270 	free(imgp->auxargs, M_TEMP);
271 	imgp->auxargs = NULL;
272 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
273 
274 	error = copyout(argarray, auxbase, sizeof(*argarray) * LINUX_AT_COUNT);
275 	free(argarray, M_TEMP);
276 	if (error != 0)
277 		return (error);
278 
279 	base--;
280 	if (suword(base, (uint64_t)imgp->args->argc) == -1)
281 		return (EFAULT);
282 
283 	*stack_base = (register_t *)base;
284 	return (0);
285 }
286 
287 /*
288  * Copy strings out to the new process address space, constructing new arg
289  * and env vector tables. Return a pointer to the base so that it can be used
290  * as the initial stack pointer.
291  */
292 static register_t *
293 linux_copyout_strings(struct image_params *imgp)
294 {
295 	int argc, envc;
296 	char **vectp;
297 	char *stringp, *destp;
298 	register_t *stack_base;
299 	struct ps_strings *arginfo;
300 	char canary[LINUX_AT_RANDOM_LEN];
301 	size_t execpath_len;
302 	struct proc *p;
303 
304 	/* Calculate string base and vector table pointers. */
305 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
306 		execpath_len = strlen(imgp->execpath) + 1;
307 	else
308 		execpath_len = 0;
309 
310 	p = imgp->proc;
311 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
312 	destp = (caddr_t)arginfo - SPARE_USRSPACE -
313 	    roundup(sizeof(canary), sizeof(char *)) -
314 	    roundup(execpath_len, sizeof(char *)) -
315 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
316 
317 	if (execpath_len != 0) {
318 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
319 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
320 	}
321 
322 	/* Prepare the canary for SSP. */
323 	arc4rand(canary, sizeof(canary), 0);
324 	imgp->canary = (uintptr_t)arginfo -
325 	    roundup(execpath_len, sizeof(char *)) -
326 	    roundup(sizeof(canary), sizeof(char *));
327 	copyout(canary, (void *)imgp->canary, sizeof(canary));
328 
329 	vectp = (char **)destp;
330 	if (imgp->auxargs) {
331 		/*
332 		 * Allocate room on the stack for the ELF auxargs
333 		 * array.  It has LINUX_AT_COUNT entries.
334 		 */
335 		vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf64_Auxinfo),
336 		    sizeof(*vectp));
337 	}
338 
339 	/*
340 	 * Allocate room for the argv[] and env vectors including the
341 	 * terminating NULL pointers.
342 	 */
343 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
344 
345 	/*
346 	 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
347 	 * One "long argc" will be prepended later.
348 	 */
349 	vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
350 
351 	/* vectp also becomes our initial stack base. */
352 	stack_base = (register_t *)vectp;
353 
354 	stringp = imgp->args->begin_argv;
355 	argc = imgp->args->argc;
356 	envc = imgp->args->envc;
357 
358 	/* Copy out strings - arguments and environment. */
359 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
360 
361 	/* Fill in "ps_strings" struct for ps, w, etc. */
362 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
363 	suword(&arginfo->ps_nargvstr, argc);
364 
365 	/* Fill in argument portion of vector table. */
366 	for (; argc > 0; --argc) {
367 		suword(vectp++, (long)(intptr_t)destp);
368 		while (*stringp++ != 0)
369 			destp++;
370 		destp++;
371 	}
372 
373 	/* A null vector table pointer separates the argp's from the envp's. */
374 	suword(vectp++, 0);
375 
376 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
377 	suword(&arginfo->ps_nenvstr, envc);
378 
379 	/* Fill in environment portion of vector table. */
380 	for (; envc > 0; --envc) {
381 		suword(vectp++, (long)(intptr_t)destp);
382 		while (*stringp++ != 0)
383 			destp++;
384 		destp++;
385 	}
386 
387 	/* The end of the vector table is a null pointer. */
388 	suword(vectp, 0);
389 	return (stack_base);
390 }
391 
392 /*
393  * Reset registers to default values on exec.
394  */
395 static void
396 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
397 {
398 	struct trapframe *regs;
399 	struct pcb *pcb;
400 	register_t saved_rflags;
401 
402 	regs = td->td_frame;
403 	pcb = td->td_pcb;
404 
405 	if (td->td_proc->p_md.md_ldt != NULL)
406 		user_ldt_free(td);
407 
408 	pcb->pcb_fsbase = 0;
409 	pcb->pcb_gsbase = 0;
410 	clear_pcb_flags(pcb, PCB_32BIT);
411 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
412 	set_pcb_flags(pcb, PCB_FULL_IRET);
413 
414 	saved_rflags = regs->tf_rflags & PSL_T;
415 	bzero((char *)regs, sizeof(struct trapframe));
416 	regs->tf_rip = imgp->entry_addr;
417 	regs->tf_rsp = stack;
418 	regs->tf_rflags = PSL_USER | saved_rflags;
419 	regs->tf_ss = _udatasel;
420 	regs->tf_cs = _ucodesel;
421 	regs->tf_ds = _udatasel;
422 	regs->tf_es = _udatasel;
423 	regs->tf_fs = _ufssel;
424 	regs->tf_gs = _ugssel;
425 	regs->tf_flags = TF_HASSEGS;
426 
427 	/*
428 	 * Reset the hardware debug registers if they were in use.
429 	 * They won't have any meaning for the newly exec'd process.
430 	 */
431 	if (pcb->pcb_flags & PCB_DBREGS) {
432 		pcb->pcb_dr0 = 0;
433 		pcb->pcb_dr1 = 0;
434 		pcb->pcb_dr2 = 0;
435 		pcb->pcb_dr3 = 0;
436 		pcb->pcb_dr6 = 0;
437 		pcb->pcb_dr7 = 0;
438 		if (pcb == curpcb) {
439 			/*
440 			 * Clear the debug registers on the running
441 			 * CPU, otherwise they will end up affecting
442 			 * the next process we switch to.
443 			 */
444 			reset_dbregs();
445 		}
446 		clear_pcb_flags(pcb, PCB_DBREGS);
447 	}
448 
449 	/*
450 	 * Drop the FP state if we hold it, so that the process gets a
451 	 * clean FP state if it uses the FPU again.
452 	 */
453 	fpstate_drop(td);
454 }
455 
456 /*
457  * Copied from amd64/amd64/machdep.c
458  *
459  * XXX fpu state need? don't think so
460  */
461 int
462 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
463 {
464 	struct proc *p;
465 	struct l_ucontext uc;
466 	struct l_sigcontext *context;
467 	struct trapframe *regs;
468 	unsigned long rflags;
469 	int error;
470 	ksiginfo_t ksi;
471 
472 	regs = td->td_frame;
473 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
474 	if (error != 0)
475 		return (error);
476 
477 	p = td->td_proc;
478 	context = &uc.uc_mcontext;
479 	rflags = context->sc_rflags;
480 
481 	/*
482 	 * Don't allow users to change privileged or reserved flags.
483 	 */
484 	/*
485 	 * XXX do allow users to change the privileged flag PSL_RF.
486 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
487 	 * should sometimes set it there too.  tf_rflags is kept in
488 	 * the signal context during signal handling and there is no
489 	 * other place to remember it, so the PSL_RF bit may be
490 	 * corrupted by the signal handler without us knowing.
491 	 * Corruption of the PSL_RF bit at worst causes one more or
492 	 * one less debugger trap, so allowing it is fairly harmless.
493 	 */
494 
495 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
496 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
497 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
498 		return (EINVAL);
499 	}
500 
501 	/*
502 	 * Don't allow users to load a valid privileged %cs.  Let the
503 	 * hardware check for invalid selectors, excess privilege in
504 	 * other selectors, invalid %eip's and invalid %esp's.
505 	 */
506 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
507 	if (!CS_SECURE(context->sc_cs)) {
508 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
509 		ksiginfo_init_trap(&ksi);
510 		ksi.ksi_signo = SIGBUS;
511 		ksi.ksi_code = BUS_OBJERR;
512 		ksi.ksi_trapno = T_PROTFLT;
513 		ksi.ksi_addr = (void *)regs->tf_rip;
514 		trapsignal(td, &ksi);
515 		return (EINVAL);
516 	}
517 
518 	PROC_LOCK(p);
519 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
520 	SIG_CANTMASK(td->td_sigmask);
521 	signotify(td);
522 	PROC_UNLOCK(p);
523 
524 	regs->tf_rdi    = context->sc_rdi;
525 	regs->tf_rsi    = context->sc_rsi;
526 	regs->tf_rdx    = context->sc_rdx;
527 	regs->tf_rbp    = context->sc_rbp;
528 	regs->tf_rbx    = context->sc_rbx;
529 	regs->tf_rcx    = context->sc_rcx;
530 	regs->tf_rax    = context->sc_rax;
531 	regs->tf_rip    = context->sc_rip;
532 	regs->tf_rsp    = context->sc_rsp;
533 	regs->tf_r8     = context->sc_r8;
534 	regs->tf_r9     = context->sc_r9;
535 	regs->tf_r10    = context->sc_r10;
536 	regs->tf_r11    = context->sc_r11;
537 	regs->tf_r12    = context->sc_r12;
538 	regs->tf_r13    = context->sc_r13;
539 	regs->tf_r14    = context->sc_r14;
540 	regs->tf_r15    = context->sc_r15;
541 	regs->tf_cs     = context->sc_cs;
542 	regs->tf_err    = context->sc_err;
543 	regs->tf_rflags = rflags;
544 
545 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
546 	return (EJUSTRETURN);
547 }
548 
549 /*
550  * copied from amd64/amd64/machdep.c
551  *
552  * Send an interrupt to process.
553  */
554 static void
555 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
556 {
557 	struct l_rt_sigframe sf, *sfp;
558 	struct proc *p;
559 	struct thread *td;
560 	struct sigacts *psp;
561 	caddr_t sp;
562 	struct trapframe *regs;
563 	int sig, code;
564 	int oonstack;
565 
566 	td = curthread;
567 	p = td->td_proc;
568 	PROC_LOCK_ASSERT(p, MA_OWNED);
569 	sig = ksi->ksi_signo;
570 	psp = p->p_sigacts;
571 	code = ksi->ksi_code;
572 	mtx_assert(&psp->ps_mtx, MA_OWNED);
573 	regs = td->td_frame;
574 	oonstack = sigonstack(regs->tf_rsp);
575 
576 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
577 	    catcher, sig, mask, code);
578 
579 	/* Allocate space for the signal handler context. */
580 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
581 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
582 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
583 		    sizeof(struct l_rt_sigframe);
584 	} else
585 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
586 	/* Align to 16 bytes. */
587 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
588 	mtx_unlock(&psp->ps_mtx);
589 
590 	/* Translate the signal. */
591 	sig = bsd_to_linux_signal(sig);
592 
593 	/* Save user context. */
594 	bzero(&sf, sizeof(sf));
595 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
596 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
597 
598 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
599 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
600 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
601 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
602 	PROC_UNLOCK(p);
603 
604 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
605 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
606 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
607 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
608 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
609 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
610 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
611 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
612 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
613 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
614 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
615 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
616 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
617 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
618 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
619 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
620 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
621 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
622 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
623 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
624 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
625 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
626 
627 	/* Build the argument list for the signal handler. */
628 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
629 	regs->tf_rax = 0;
630 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
631 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
632 
633 	sf.sf_handler = catcher;
634 	/* Fill in POSIX parts. */
635 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
636 
637 	/* Copy the sigframe out to the user's stack. */
638 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
639 		PROC_LOCK(p);
640 		sigexit(td, SIGILL);
641 	}
642 
643 	regs->tf_rsp = (long)sfp;
644 	regs->tf_rip = linux_rt_sigcode;
645 	regs->tf_rflags &= ~(PSL_T | PSL_D);
646 	regs->tf_cs = _ucodesel;
647 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
648 	PROC_LOCK(p);
649 	mtx_lock(&psp->ps_mtx);
650 }
651 
652 #define	LINUX_VSYSCALL_START		(-10UL << 20)
653 #define	LINUX_VSYSCALL_SZ		1024
654 
655 const unsigned long linux_vsyscall_vector[] = {
656 	LINUX_SYS_gettimeofday,
657 	LINUX_SYS_linux_time,
658 				/* getcpu not implemented */
659 };
660 
661 static int
662 linux_vsyscall(struct thread *td)
663 {
664 	struct trapframe *frame;
665 	uint64_t retqaddr;
666 	int code, traced;
667 	int error;
668 
669 	frame = td->td_frame;
670 
671 	/* Check %rip for vsyscall area. */
672 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
673 		return (EINVAL);
674 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
675 		return (EINVAL);
676 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
677 	if (code >= nitems(linux_vsyscall_vector))
678 		return (EINVAL);
679 
680 	/*
681 	 * vsyscall called as callq *(%rax), so we must
682 	 * use return address from %rsp and also fixup %rsp.
683 	 */
684 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
685 	if (error)
686 		return (error);
687 
688 	frame->tf_rip = retqaddr;
689 	frame->tf_rax = linux_vsyscall_vector[code];
690 	frame->tf_rsp += 8;
691 
692 	traced = (frame->tf_flags & PSL_T);
693 
694 	amd64_syscall(td, traced);
695 
696 	return (0);
697 }
698 
699 struct sysentvec elf_linux_sysvec = {
700 	.sv_size	= LINUX_SYS_MAXSYSCALL,
701 	.sv_table	= linux_sysent,
702 	.sv_errsize	= ELAST + 1,
703 	.sv_errtbl	= linux_errtbl,
704 	.sv_transtrap	= linux_translate_traps,
705 	.sv_fixup	= linux_fixup_elf,
706 	.sv_sendsig	= linux_rt_sendsig,
707 	.sv_sigcode	= &_binary_linux_locore_o_start,
708 	.sv_szsigcode	= &linux_szsigcode,
709 	.sv_name	= "Linux ELF64",
710 	.sv_coredump	= elf64_coredump,
711 	.sv_imgact_try	= linux_exec_imgact_try,
712 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
713 	.sv_minuser	= VM_MIN_ADDRESS,
714 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
715 	.sv_usrstack	= USRSTACK,
716 	.sv_psstrings	= PS_STRINGS,
717 	.sv_stackprot	= VM_PROT_ALL,
718 	.sv_copyout_strings = linux_copyout_strings,
719 	.sv_setregs	= linux_exec_setregs,
720 	.sv_fixlimit	= NULL,
721 	.sv_maxssiz	= NULL,
722 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
723 	.sv_set_syscall_retval = linux_set_syscall_retval,
724 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
725 	.sv_syscallnames = NULL,
726 	.sv_shared_page_base = SHAREDPAGE,
727 	.sv_shared_page_len = PAGE_SIZE,
728 	.sv_schedtail	= linux_schedtail,
729 	.sv_thread_detach = linux_thread_detach,
730 	.sv_trap	= linux_vsyscall,
731 };
732 
733 static void
734 linux_vdso_install(void *param)
735 {
736 
737 	amd64_lower_shared_page(&elf_linux_sysvec);
738 
739 	linux_szsigcode = (&_binary_linux_locore_o_end -
740 	    &_binary_linux_locore_o_start);
741 
742 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
743 		panic("Linux invalid vdso size\n");
744 
745 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
746 
747 	linux_shared_page_obj = __elfN(linux_shared_page_init)
748 	    (&linux_shared_page_mapping);
749 
750 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
751 
752 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
753 	    linux_szsigcode);
754 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
755 
756 	linux_kplatform = linux_shared_page_mapping +
757 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
758 }
759 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
760     linux_vdso_install, NULL);
761 
762 static void
763 linux_vdso_deinstall(void *param)
764 {
765 
766 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
767 }
768 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
769     linux_vdso_deinstall, NULL);
770 
771 static char GNULINUX_ABI_VENDOR[] = "GNU";
772 static int GNULINUX_ABI_DESC = 0;
773 
774 static bool
775 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
776 {
777 	const Elf32_Word *desc;
778 	uintptr_t p;
779 
780 	p = (uintptr_t)(note + 1);
781 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
782 
783 	desc = (const Elf32_Word *)p;
784 	if (desc[0] != GNULINUX_ABI_DESC)
785 		return (false);
786 
787 	/*
788 	 * For Linux we encode osrel using the Linux convention of
789 	 * 	(version << 16) | (major << 8) | (minor)
790 	 * See macro in linux_mib.h
791 	 */
792 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
793 
794 	return (true);
795 }
796 
797 static Elf_Brandnote linux64_brandnote = {
798 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
799 	.hdr.n_descsz	= 16,
800 	.hdr.n_type	= 1,
801 	.vendor		= GNULINUX_ABI_VENDOR,
802 	.flags		= BN_TRANSLATE_OSREL,
803 	.trans_osrel	= linux_trans_osrel
804 };
805 
806 static Elf64_Brandinfo linux_glibc2brand = {
807 	.brand		= ELFOSABI_LINUX,
808 	.machine	= EM_X86_64,
809 	.compat_3_brand	= "Linux",
810 	.emul_path	= "/compat/linux",
811 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
812 	.sysvec		= &elf_linux_sysvec,
813 	.interp_newpath	= NULL,
814 	.brand_note	= &linux64_brandnote,
815 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
816 };
817 
818 static Elf64_Brandinfo linux_glibc2brandshort = {
819 	.brand		= ELFOSABI_LINUX,
820 	.machine	= EM_X86_64,
821 	.compat_3_brand	= "Linux",
822 	.emul_path	= "/compat/linux",
823 	.interp_path	= "/lib64/ld-linux.so.2",
824 	.sysvec		= &elf_linux_sysvec,
825 	.interp_newpath	= NULL,
826 	.brand_note	= &linux64_brandnote,
827 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
828 };
829 
830 static Elf64_Brandinfo linux_muslbrand = {
831 	.brand		= ELFOSABI_LINUX,
832 	.machine	= EM_X86_64,
833 	.compat_3_brand	= "Linux",
834 	.emul_path	= "/compat/linux",
835 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
836 	.sysvec		= &elf_linux_sysvec,
837 	.interp_newpath	= NULL,
838 	.brand_note	= &linux64_brandnote,
839 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
840 };
841 
842 Elf64_Brandinfo *linux_brandlist[] = {
843 	&linux_glibc2brand,
844 	&linux_glibc2brandshort,
845 	&linux_muslbrand,
846 	NULL
847 };
848 
849 static int
850 linux64_elf_modevent(module_t mod, int type, void *data)
851 {
852 	Elf64_Brandinfo **brandinfo;
853 	int error;
854 	struct linux_ioctl_handler **lihp;
855 
856 	error = 0;
857 
858 	switch(type) {
859 	case MOD_LOAD:
860 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
861 		     ++brandinfo)
862 			if (elf64_insert_brand_entry(*brandinfo) < 0)
863 				error = EINVAL;
864 		if (error == 0) {
865 			SET_FOREACH(lihp, linux_ioctl_handler_set)
866 				linux_ioctl_register_handler(*lihp);
867 			stclohz = (stathz ? stathz : hz);
868 			if (bootverbose)
869 				printf("Linux x86-64 ELF exec handler installed\n");
870 		} else
871 			printf("cannot insert Linux x86-64 ELF brand handler\n");
872 		break;
873 	case MOD_UNLOAD:
874 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
875 		     ++brandinfo)
876 			if (elf64_brand_inuse(*brandinfo))
877 				error = EBUSY;
878 		if (error == 0) {
879 			for (brandinfo = &linux_brandlist[0];
880 			     *brandinfo != NULL; ++brandinfo)
881 				if (elf64_remove_brand_entry(*brandinfo) < 0)
882 					error = EINVAL;
883 		}
884 		if (error == 0) {
885 			SET_FOREACH(lihp, linux_ioctl_handler_set)
886 				linux_ioctl_unregister_handler(*lihp);
887 			if (bootverbose)
888 				printf("Linux ELF exec handler removed\n");
889 		} else
890 			printf("Could not deinstall ELF interpreter entry\n");
891 		break;
892 	default:
893 		return (EOPNOTSUPP);
894 	}
895 	return (error);
896 }
897 
898 static moduledata_t linux64_elf_mod = {
899 	"linux64elf",
900 	linux64_elf_modevent,
901 	0
902 };
903 
904 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
905 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
906 FEATURE(linux64, "Linux 64bit support");
907