xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 0e33efe4e4b5d24e2d416938af8bc6e6e4160ec8)
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	linux_fixup_elf(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 int	linux_exec_imgact_try(struct image_params *iparams);
130 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
131 		    u_long stack);
132 static int	linux_vsyscall(struct thread *td);
133 
134 #define LINUX_T_UNKNOWN  255
135 static int _bsd_to_linux_trapcode[] = {
136 	LINUX_T_UNKNOWN,	/* 0 */
137 	6,			/* 1  T_PRIVINFLT */
138 	LINUX_T_UNKNOWN,	/* 2 */
139 	3,			/* 3  T_BPTFLT */
140 	LINUX_T_UNKNOWN,	/* 4 */
141 	LINUX_T_UNKNOWN,	/* 5 */
142 	16,			/* 6  T_ARITHTRAP */
143 	254,			/* 7  T_ASTFLT */
144 	LINUX_T_UNKNOWN,	/* 8 */
145 	13,			/* 9  T_PROTFLT */
146 	1,			/* 10 T_TRCTRAP */
147 	LINUX_T_UNKNOWN,	/* 11 */
148 	14,			/* 12 T_PAGEFLT */
149 	LINUX_T_UNKNOWN,	/* 13 */
150 	17,			/* 14 T_ALIGNFLT */
151 	LINUX_T_UNKNOWN,	/* 15 */
152 	LINUX_T_UNKNOWN,	/* 16 */
153 	LINUX_T_UNKNOWN,	/* 17 */
154 	0,			/* 18 T_DIVIDE */
155 	2,			/* 19 T_NMI */
156 	4,			/* 20 T_OFLOW */
157 	5,			/* 21 T_BOUND */
158 	7,			/* 22 T_DNA */
159 	8,			/* 23 T_DOUBLEFLT */
160 	9,			/* 24 T_FPOPFLT */
161 	10,			/* 25 T_TSSFLT */
162 	11,			/* 26 T_SEGNPFLT */
163 	12,			/* 27 T_STKFLT */
164 	18,			/* 28 T_MCHK */
165 	19,			/* 29 T_XMMFLT */
166 	15			/* 30 T_RESERVED */
167 };
168 #define bsd_to_linux_trapcode(code) \
169     ((code)<nitems(_bsd_to_linux_trapcode)? \
170      _bsd_to_linux_trapcode[(code)]: \
171      LINUX_T_UNKNOWN)
172 
173 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
174 LINUX_VDSO_SYM_CHAR(linux_platform);
175 
176 /*
177  * If FreeBSD & Linux have a difference of opinion about what a trap
178  * means, deal with it here.
179  *
180  * MPSAFE
181  */
182 static int
183 linux_translate_traps(int signal, int trap_code)
184 {
185 
186 	if (signal != SIGBUS)
187 		return signal;
188 	switch (trap_code) {
189 	case T_PROTFLT:
190 	case T_TSSFLT:
191 	case T_DOUBLEFLT:
192 	case T_PAGEFLT:
193 		return SIGSEGV;
194 	default:
195 		return signal;
196 	}
197 }
198 
199 static int
200 linux_fetch_syscall_args(struct thread *td)
201 {
202 	struct proc *p;
203 	struct trapframe *frame;
204 	struct syscall_args *sa;
205 
206 	p = td->td_proc;
207 	frame = td->td_frame;
208 	sa = &td->td_sa;
209 
210 	sa->args[0] = frame->tf_rdi;
211 	sa->args[1] = frame->tf_rsi;
212 	sa->args[2] = frame->tf_rdx;
213 	sa->args[3] = frame->tf_rcx;
214 	sa->args[4] = frame->tf_r8;
215 	sa->args[5] = frame->tf_r9;
216 	sa->code = frame->tf_rax;
217 
218 	if (sa->code >= p->p_sysent->sv_size)
219 		/* nosys */
220 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
221 	else
222 		sa->callp = &p->p_sysent->sv_table[sa->code];
223 	sa->narg = sa->callp->sy_narg;
224 
225 	td->td_retval[0] = 0;
226 	return (0);
227 }
228 
229 static void
230 linux_set_syscall_retval(struct thread *td, int error)
231 {
232 	struct trapframe *frame = td->td_frame;
233 
234 	/*
235 	 * On Linux only %rcx and %r11 values are not preserved across
236 	 * the syscall.  So, do not clobber %rdx and %r10.
237 	 */
238 	td->td_retval[1] = frame->tf_rdx;
239 	frame->tf_r10 = frame->tf_rcx;
240 
241 	cpu_set_syscall_retval(td, error);
242 
243 	 /* Restore all registers. */
244 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
245 }
246 
247 static int
248 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
249 {
250 	Elf_Auxargs *args;
251 	Elf_Addr *base;
252 	Elf_Addr *pos;
253 	struct ps_strings *arginfo;
254 	struct proc *p;
255 	int issetugid;
256 
257 	p = imgp->proc;
258 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
259 
260 	KASSERT(curthread->td_proc == imgp->proc,
261 	    ("unsafe linux_fixup_elf(), should be curproc"));
262 	base = (Elf64_Addr *)*stack_base;
263 	args = (Elf64_Auxargs *)imgp->auxargs;
264 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
265 
266 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
267 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
268 	    imgp->proc->p_sysent->sv_shared_page_base);
269 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
270 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
271 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
272 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
273 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
274 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
275 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
276 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
277 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
278 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
279 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
280 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
281 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
282 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
283 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
284 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
285 	if (imgp->execpathp != 0)
286 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
287 	if (args->execfd != -1)
288 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
289 	AUXARGS_ENTRY(pos, AT_NULL, 0);
290 	free(imgp->auxargs, M_TEMP);
291 	imgp->auxargs = NULL;
292 
293 	base--;
294 	suword(base, (uint64_t)imgp->args->argc);
295 
296 	*stack_base = (register_t *)base;
297 	return (0);
298 }
299 
300 /*
301  * Copy strings out to the new process address space, constructing new arg
302  * and env vector tables. Return a pointer to the base so that it can be used
303  * as the initial stack pointer.
304  */
305 static register_t *
306 linux_copyout_strings(struct image_params *imgp)
307 {
308 	int argc, envc;
309 	char **vectp;
310 	char *stringp, *destp;
311 	register_t *stack_base;
312 	struct ps_strings *arginfo;
313 	char canary[LINUX_AT_RANDOM_LEN];
314 	size_t execpath_len;
315 	struct proc *p;
316 
317 	/* Calculate string base and vector table pointers. */
318 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
319 		execpath_len = strlen(imgp->execpath) + 1;
320 	else
321 		execpath_len = 0;
322 
323 	p = imgp->proc;
324 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
325 	destp = (caddr_t)arginfo - SPARE_USRSPACE -
326 	    roundup(sizeof(canary), sizeof(char *)) -
327 	    roundup(execpath_len, sizeof(char *)) -
328 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
329 
330 	if (execpath_len != 0) {
331 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
332 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
333 	}
334 
335 	/* Prepare the canary for SSP. */
336 	arc4rand(canary, sizeof(canary), 0);
337 	imgp->canary = (uintptr_t)arginfo -
338 	    roundup(execpath_len, sizeof(char *)) -
339 	    roundup(sizeof(canary), sizeof(char *));
340 	copyout(canary, (void *)imgp->canary, sizeof(canary));
341 
342 	/* If we have a valid auxargs ptr, prepare some room on the stack. */
343 	if (imgp->auxargs) {
344 		/*
345 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
346 		 * lower compatibility.
347 		 */
348 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
349 		    (LINUX_AT_COUNT * 2);
350 
351 		/*
352 		 * The '+ 2' is for the null pointers at the end of each of
353 		 * the arg and env vector sets,and imgp->auxarg_size is room
354 		 * for argument of Runtime loader.
355 		 */
356 		vectp = (char **)(destp - (imgp->args->argc +
357 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
358 
359 	} else {
360 		/*
361 		 * The '+ 2' is for the null pointers at the end of each of
362 		 * the arg and env vector sets
363 		 */
364 		vectp = (char **)(destp - (imgp->args->argc +
365 		    imgp->args->envc + 2) * sizeof(char *));
366 	}
367 
368 	/* vectp also becomes our initial stack base. */
369 	stack_base = (register_t *)vectp;
370 
371 	stringp = imgp->args->begin_argv;
372 	argc = imgp->args->argc;
373 	envc = imgp->args->envc;
374 
375 	/* Copy out strings - arguments and environment. */
376 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
377 
378 	/* Fill in "ps_strings" struct for ps, w, etc. */
379 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
380 	suword(&arginfo->ps_nargvstr, argc);
381 
382 	/* Fill in argument portion of vector table. */
383 	for (; argc > 0; --argc) {
384 		suword(vectp++, (long)(intptr_t)destp);
385 		while (*stringp++ != 0)
386 			destp++;
387 		destp++;
388 	}
389 
390 	/* A null vector table pointer separates the argp's from the envp's. */
391 	suword(vectp++, 0);
392 
393 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
394 	suword(&arginfo->ps_nenvstr, envc);
395 
396 	/* Fill in environment portion of vector table. */
397 	for (; envc > 0; --envc) {
398 		suword(vectp++, (long)(intptr_t)destp);
399 		while (*stringp++ != 0)
400 			destp++;
401 		destp++;
402 	}
403 
404 	/* The end of the vector table is a null pointer. */
405 	suword(vectp, 0);
406 	return (stack_base);
407 }
408 
409 /*
410  * Reset registers to default values on exec.
411  */
412 static void
413 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
414 {
415 	struct trapframe *regs = td->td_frame;
416 	struct pcb *pcb = td->td_pcb;
417 
418 	if (td->td_proc->p_md.md_ldt != NULL)
419 		user_ldt_free(td);
420 
421 	pcb->pcb_fsbase = 0;
422 	pcb->pcb_gsbase = 0;
423 	clear_pcb_flags(pcb, PCB_32BIT);
424 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
425 	set_pcb_flags(pcb, PCB_FULL_IRET);
426 
427 	bzero((char *)regs, sizeof(struct trapframe));
428 	regs->tf_rip = imgp->entry_addr;
429 	regs->tf_rsp = stack;
430 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
431 	regs->tf_ss = _udatasel;
432 	regs->tf_cs = _ucodesel;
433 	regs->tf_ds = _udatasel;
434 	regs->tf_es = _udatasel;
435 	regs->tf_fs = _ufssel;
436 	regs->tf_gs = _ugssel;
437 	regs->tf_flags = TF_HASSEGS;
438 
439 	/*
440 	 * Reset the hardware debug registers if they were in use.
441 	 * They won't have any meaning for the newly exec'd process.
442 	 */
443 	if (pcb->pcb_flags & PCB_DBREGS) {
444 		pcb->pcb_dr0 = 0;
445 		pcb->pcb_dr1 = 0;
446 		pcb->pcb_dr2 = 0;
447 		pcb->pcb_dr3 = 0;
448 		pcb->pcb_dr6 = 0;
449 		pcb->pcb_dr7 = 0;
450 		if (pcb == curpcb) {
451 			/*
452 			 * Clear the debug registers on the running
453 			 * CPU, otherwise they will end up affecting
454 			 * the next process we switch to.
455 			 */
456 			reset_dbregs();
457 		}
458 		clear_pcb_flags(pcb, PCB_DBREGS);
459 	}
460 
461 	/*
462 	 * Drop the FP state if we hold it, so that the process gets a
463 	 * clean FP state if it uses the FPU again.
464 	 */
465 	fpstate_drop(td);
466 }
467 
468 /*
469  * Copied from amd64/amd64/machdep.c
470  *
471  * XXX fpu state need? don't think so
472  */
473 int
474 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
475 {
476 	struct proc *p;
477 	struct l_ucontext uc;
478 	struct l_sigcontext *context;
479 	struct trapframe *regs;
480 	unsigned long rflags;
481 	int error;
482 	ksiginfo_t ksi;
483 
484 	regs = td->td_frame;
485 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
486 	if (error != 0)
487 		return (error);
488 
489 	p = td->td_proc;
490 	context = &uc.uc_mcontext;
491 	rflags = context->sc_rflags;
492 
493 	/*
494 	 * Don't allow users to change privileged or reserved flags.
495 	 */
496 	/*
497 	 * XXX do allow users to change the privileged flag PSL_RF.
498 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
499 	 * should sometimes set it there too.  tf_rflags is kept in
500 	 * the signal context during signal handling and there is no
501 	 * other place to remember it, so the PSL_RF bit may be
502 	 * corrupted by the signal handler without us knowing.
503 	 * Corruption of the PSL_RF bit at worst causes one more or
504 	 * one less debugger trap, so allowing it is fairly harmless.
505 	 */
506 
507 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
508 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
509 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
510 		return (EINVAL);
511 	}
512 
513 	/*
514 	 * Don't allow users to load a valid privileged %cs.  Let the
515 	 * hardware check for invalid selectors, excess privilege in
516 	 * other selectors, invalid %eip's and invalid %esp's.
517 	 */
518 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
519 	if (!CS_SECURE(context->sc_cs)) {
520 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
521 		ksiginfo_init_trap(&ksi);
522 		ksi.ksi_signo = SIGBUS;
523 		ksi.ksi_code = BUS_OBJERR;
524 		ksi.ksi_trapno = T_PROTFLT;
525 		ksi.ksi_addr = (void *)regs->tf_rip;
526 		trapsignal(td, &ksi);
527 		return (EINVAL);
528 	}
529 
530 	PROC_LOCK(p);
531 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
532 	SIG_CANTMASK(td->td_sigmask);
533 	signotify(td);
534 	PROC_UNLOCK(p);
535 
536 	regs->tf_rdi    = context->sc_rdi;
537 	regs->tf_rsi    = context->sc_rsi;
538 	regs->tf_rdx    = context->sc_rdx;
539 	regs->tf_rbp    = context->sc_rbp;
540 	regs->tf_rbx    = context->sc_rbx;
541 	regs->tf_rcx    = context->sc_rcx;
542 	regs->tf_rax    = context->sc_rax;
543 	regs->tf_rip    = context->sc_rip;
544 	regs->tf_rsp    = context->sc_rsp;
545 	regs->tf_r8     = context->sc_r8;
546 	regs->tf_r9     = context->sc_r9;
547 	regs->tf_r10    = context->sc_r10;
548 	regs->tf_r11    = context->sc_r11;
549 	regs->tf_r12    = context->sc_r12;
550 	regs->tf_r13    = context->sc_r13;
551 	regs->tf_r14    = context->sc_r14;
552 	regs->tf_r15    = context->sc_r15;
553 	regs->tf_cs     = context->sc_cs;
554 	regs->tf_err    = context->sc_err;
555 	regs->tf_rflags = rflags;
556 
557 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
558 	return (EJUSTRETURN);
559 }
560 
561 /*
562  * copied from amd64/amd64/machdep.c
563  *
564  * Send an interrupt to process.
565  */
566 static void
567 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
568 {
569 	struct l_rt_sigframe sf, *sfp;
570 	struct proc *p;
571 	struct thread *td;
572 	struct sigacts *psp;
573 	caddr_t sp;
574 	struct trapframe *regs;
575 	int sig, code;
576 	int oonstack;
577 
578 	td = curthread;
579 	p = td->td_proc;
580 	PROC_LOCK_ASSERT(p, MA_OWNED);
581 	sig = ksi->ksi_signo;
582 	psp = p->p_sigacts;
583 	code = ksi->ksi_code;
584 	mtx_assert(&psp->ps_mtx, MA_OWNED);
585 	regs = td->td_frame;
586 	oonstack = sigonstack(regs->tf_rsp);
587 
588 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
589 	    catcher, sig, mask, code);
590 
591 	/* Allocate space for the signal handler context. */
592 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
593 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
594 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
595 		    sizeof(struct l_rt_sigframe);
596 	} else
597 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
598 	/* Align to 16 bytes. */
599 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
600 	mtx_unlock(&psp->ps_mtx);
601 
602 	/* Translate the signal. */
603 	sig = bsd_to_linux_signal(sig);
604 
605 	/* Save user context. */
606 	bzero(&sf, sizeof(sf));
607 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
608 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
609 
610 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
611 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
612 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
613 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
614 	PROC_UNLOCK(p);
615 
616 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
617 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
618 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
619 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
620 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
621 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
622 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
623 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
624 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
625 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
626 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
627 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
628 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
629 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
630 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
631 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
632 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
633 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
634 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
635 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
636 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
637 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
638 
639 	/* Build the argument list for the signal handler. */
640 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
641 	regs->tf_rax = 0;
642 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
643 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
644 
645 	sf.sf_handler = catcher;
646 	/* Fill in POSIX parts. */
647 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
648 
649 	/* Copy the sigframe out to the user's stack. */
650 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
651 #ifdef DEBUG
652 		printf("process %ld has trashed its stack\n", (long)p->p_pid);
653 #endif
654 		PROC_LOCK(p);
655 		sigexit(td, SIGILL);
656 	}
657 
658 	regs->tf_rsp = (long)sfp;
659 	regs->tf_rip = linux_rt_sigcode;
660 	regs->tf_rflags &= ~(PSL_T | PSL_D);
661 	regs->tf_cs = _ucodesel;
662 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
663 	PROC_LOCK(p);
664 	mtx_lock(&psp->ps_mtx);
665 }
666 
667 /*
668  * If a Linux binary is exec'ing something, try this image activator
669  * first.  We override standard shell script execution in order to
670  * be able to modify the interpreter path.  We only do this if a Linux
671  * binary is doing the exec, so we do not create an EXEC module for it.
672  */
673 static int
674 linux_exec_imgact_try(struct image_params *imgp)
675 {
676 	const char *head = (const char *)imgp->image_header;
677 	char *rpath;
678 	int error = -1;
679 
680 	/*
681 	 * The interpreter for shell scripts run from a Linux binary needs
682 	 * to be located in /compat/linux if possible in order to recursively
683 	 * maintain Linux path emulation.
684 	 */
685 	if (((const short *)head)[0] == SHELLMAGIC) {
686 		/*
687 		 * Run our normal shell image activator.  If it succeeds then
688 		 * attempt to use the alternate path for the interpreter.  If
689 		 * an alternate path is found, use our stringspace to store it.
690 		 */
691 		if ((error = exec_shell_imgact(imgp)) == 0) {
692 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
693 			    imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
694 			    AT_FDCWD);
695 			if (rpath != NULL)
696 				imgp->args->fname_buf =
697 				    imgp->interpreter_name = rpath;
698 		}
699 	}
700 	return (error);
701 }
702 
703 #define	LINUX_VSYSCALL_START		(-10UL << 20)
704 #define	LINUX_VSYSCALL_SZ		1024
705 
706 const unsigned long linux_vsyscall_vector[] = {
707 	LINUX_SYS_gettimeofday,
708 	LINUX_SYS_linux_time,
709 				/* getcpu not implemented */
710 };
711 
712 static int
713 linux_vsyscall(struct thread *td)
714 {
715 	struct trapframe *frame;
716 	uint64_t retqaddr;
717 	int code, traced;
718 	int error;
719 
720 	frame = td->td_frame;
721 
722 	/* Check %rip for vsyscall area. */
723 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
724 		return (EINVAL);
725 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
726 		return (EINVAL);
727 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
728 	if (code >= nitems(linux_vsyscall_vector))
729 		return (EINVAL);
730 
731 	/*
732 	 * vsyscall called as callq *(%rax), so we must
733 	 * use return address from %rsp and also fixup %rsp.
734 	 */
735 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
736 	if (error)
737 		return (error);
738 
739 	frame->tf_rip = retqaddr;
740 	frame->tf_rax = linux_vsyscall_vector[code];
741 	frame->tf_rsp += 8;
742 
743 	traced = (frame->tf_flags & PSL_T);
744 
745 	amd64_syscall(td, traced);
746 
747 	return (0);
748 }
749 
750 struct sysentvec elf_linux_sysvec = {
751 	.sv_size	= LINUX_SYS_MAXSYSCALL,
752 	.sv_table	= linux_sysent,
753 	.sv_mask	= 0,
754 	.sv_errsize	= ELAST + 1,
755 	.sv_errtbl	= bsd_to_linux_errno_generic,
756 	.sv_transtrap	= linux_translate_traps,
757 	.sv_fixup	= linux_fixup_elf,
758 	.sv_sendsig	= linux_rt_sendsig,
759 	.sv_sigcode	= &_binary_linux_locore_o_start,
760 	.sv_szsigcode	= &linux_szsigcode,
761 	.sv_name	= "Linux ELF64",
762 	.sv_coredump	= elf64_coredump,
763 	.sv_imgact_try	= linux_exec_imgact_try,
764 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
765 	.sv_pagesize	= PAGE_SIZE,
766 	.sv_minuser	= VM_MIN_ADDRESS,
767 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
768 	.sv_usrstack	= USRSTACK,
769 	.sv_psstrings	= PS_STRINGS,
770 	.sv_stackprot	= VM_PROT_ALL,
771 	.sv_copyout_strings = linux_copyout_strings,
772 	.sv_setregs	= linux_exec_setregs,
773 	.sv_fixlimit	= NULL,
774 	.sv_maxssiz	= NULL,
775 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
776 	.sv_set_syscall_retval = linux_set_syscall_retval,
777 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
778 	.sv_syscallnames = NULL,
779 	.sv_shared_page_base = SHAREDPAGE,
780 	.sv_shared_page_len = PAGE_SIZE,
781 	.sv_schedtail	= linux_schedtail,
782 	.sv_thread_detach = linux_thread_detach,
783 	.sv_trap	= linux_vsyscall,
784 };
785 
786 static void
787 linux_vdso_install(void *param)
788 {
789 
790 	amd64_lower_shared_page(&elf_linux_sysvec);
791 
792 	linux_szsigcode = (&_binary_linux_locore_o_end -
793 	    &_binary_linux_locore_o_start);
794 
795 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
796 		panic("Linux invalid vdso size\n");
797 
798 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
799 
800 	linux_shared_page_obj = __elfN(linux_shared_page_init)
801 	    (&linux_shared_page_mapping);
802 
803 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
804 
805 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
806 	    linux_szsigcode);
807 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
808 
809 	linux_kplatform = linux_shared_page_mapping +
810 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
811 }
812 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
813     (sysinit_cfunc_t)linux_vdso_install, NULL);
814 
815 static void
816 linux_vdso_deinstall(void *param)
817 {
818 
819 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
820 }
821 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
822     (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
823 
824 static char GNULINUX_ABI_VENDOR[] = "GNU";
825 static int GNULINUX_ABI_DESC = 0;
826 
827 static bool
828 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
829 {
830 	const Elf32_Word *desc;
831 	uintptr_t p;
832 
833 	p = (uintptr_t)(note + 1);
834 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
835 
836 	desc = (const Elf32_Word *)p;
837 	if (desc[0] != GNULINUX_ABI_DESC)
838 		return (false);
839 
840 	/*
841 	 * For Linux we encode osrel as follows (see linux_mib.c):
842 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
843 	 */
844 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
845 
846 	return (true);
847 }
848 
849 static Elf_Brandnote linux64_brandnote = {
850 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
851 	.hdr.n_descsz	= 16,
852 	.hdr.n_type	= 1,
853 	.vendor		= GNULINUX_ABI_VENDOR,
854 	.flags		= BN_TRANSLATE_OSREL,
855 	.trans_osrel	= linux_trans_osrel
856 };
857 
858 static Elf64_Brandinfo linux_glibc2brand = {
859 	.brand		= ELFOSABI_LINUX,
860 	.machine	= EM_X86_64,
861 	.compat_3_brand	= "Linux",
862 	.emul_path	= "/compat/linux",
863 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
864 	.sysvec		= &elf_linux_sysvec,
865 	.interp_newpath	= NULL,
866 	.brand_note	= &linux64_brandnote,
867 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
868 };
869 
870 static Elf64_Brandinfo linux_glibc2brandshort = {
871 	.brand		= ELFOSABI_LINUX,
872 	.machine	= EM_X86_64,
873 	.compat_3_brand	= "Linux",
874 	.emul_path	= "/compat/linux",
875 	.interp_path	= "/lib64/ld-linux.so.2",
876 	.sysvec		= &elf_linux_sysvec,
877 	.interp_newpath	= NULL,
878 	.brand_note	= &linux64_brandnote,
879 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
880 };
881 
882 static Elf64_Brandinfo linux_muslbrand = {
883 	.brand		= ELFOSABI_LINUX,
884 	.machine	= EM_X86_64,
885 	.compat_3_brand	= "Linux",
886 	.emul_path	= "/compat/linux",
887 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
888 	.sysvec		= &elf_linux_sysvec,
889 	.interp_newpath	= NULL,
890 	.brand_note	= &linux64_brandnote,
891 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
892 };
893 
894 Elf64_Brandinfo *linux_brandlist[] = {
895 	&linux_glibc2brand,
896 	&linux_glibc2brandshort,
897 	&linux_muslbrand,
898 	NULL
899 };
900 
901 static int
902 linux64_elf_modevent(module_t mod, int type, void *data)
903 {
904 	Elf64_Brandinfo **brandinfo;
905 	int error;
906 	struct linux_ioctl_handler **lihp;
907 
908 	error = 0;
909 
910 	switch(type) {
911 	case MOD_LOAD:
912 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
913 		     ++brandinfo)
914 			if (elf64_insert_brand_entry(*brandinfo) < 0)
915 				error = EINVAL;
916 		if (error == 0) {
917 			SET_FOREACH(lihp, linux_ioctl_handler_set)
918 				linux_ioctl_register_handler(*lihp);
919 			LIST_INIT(&futex_list);
920 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
921 			stclohz = (stathz ? stathz : hz);
922 			if (bootverbose)
923 				printf("Linux x86-64 ELF exec handler installed\n");
924 		} else
925 			printf("cannot insert Linux x86-64 ELF brand handler\n");
926 		break;
927 	case MOD_UNLOAD:
928 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
929 		     ++brandinfo)
930 			if (elf64_brand_inuse(*brandinfo))
931 				error = EBUSY;
932 		if (error == 0) {
933 			for (brandinfo = &linux_brandlist[0];
934 			     *brandinfo != NULL; ++brandinfo)
935 				if (elf64_remove_brand_entry(*brandinfo) < 0)
936 					error = EINVAL;
937 		}
938 		if (error == 0) {
939 			SET_FOREACH(lihp, linux_ioctl_handler_set)
940 				linux_ioctl_unregister_handler(*lihp);
941 			mtx_destroy(&futex_mtx);
942 			if (bootverbose)
943 				printf("Linux ELF exec handler removed\n");
944 		} else
945 			printf("Could not deinstall ELF interpreter entry\n");
946 		break;
947 	default:
948 		return (EOPNOTSUPP);
949 	}
950 	return (error);
951 }
952 
953 static moduledata_t linux64_elf_mod = {
954 	"linux64elf",
955 	linux64_elf_modevent,
956 	0
957 };
958 
959 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
960 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
961 FEATURE(linux64, "Linux 64bit support");
962