xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision e9ac41698b2f322d55ccf9da50a3596edb2c1800)
1 /*-
2  * Copyright (c) 2004 Tim J. Robbins
3  * Copyright (c) 2003 Peter Wemm
4  * Copyright (c) 2002 Doug Rabson
5  * Copyright (c) 1998-1999 Andrew Gallatin
6  * Copyright (c) 1994-1996 Søren Schmidt
7  * All rights reserved.
8  * Copyright (c) 2013, 2021 Dmitry Chagin <dchagin@FreeBSD.org>
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 #define	__ELF_WORD_SIZE	64
35 
36 #include <sys/param.h>
37 #include <sys/exec.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_elf.h>
40 #include <sys/kernel.h>
41 #include <sys/ktr.h>
42 #include <sys/lock.h>
43 #include <sys/module.h>
44 #include <sys/mutex.h>
45 #include <sys/proc.h>
46 #include <sys/stddef.h>
47 #include <sys/syscallsubr.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
50 
51 #include <vm/pmap.h>
52 #include <vm/vm.h>
53 #include <vm/vm_param.h>
54 
55 #include <machine/md_var.h>
56 #include <machine/trap.h>
57 
58 #include <x86/linux/linux_x86.h>
59 #include <amd64/linux/linux.h>
60 #include <amd64/linux/linux_proto.h>
61 #include <compat/linux/linux_elf.h>
62 #include <compat/linux/linux_emul.h>
63 #include <compat/linux/linux_fork.h>
64 #include <compat/linux/linux_ioctl.h>
65 #include <compat/linux/linux_mib.h>
66 #include <compat/linux/linux_misc.h>
67 #include <compat/linux/linux_signal.h>
68 #include <compat/linux/linux_util.h>
69 #include <compat/linux/linux_vdso.h>
70 
71 #include <x86/linux/linux_x86_sigframe.h>
72 
73 _Static_assert(sizeof(struct l_fpstate) ==
74     sizeof(__typeof(((mcontext_t *)0)->mc_fpstate)),
75     "fxsave area size incorrect");
76 
77 MODULE_VERSION(linux64, 1);
78 
79 #define	LINUX_VDSOPAGE_SIZE	PAGE_SIZE * 2
80 #define	LINUX_VDSOPAGE_LA48	(VM_MAXUSER_ADDRESS_LA48 - \
81 				    LINUX_VDSOPAGE_SIZE)
82 #define	LINUX_SHAREDPAGE_LA48	(LINUX_VDSOPAGE_LA48 - PAGE_SIZE)
83 				/*
84 				 * PAGE_SIZE - the size
85 				 * of the native SHAREDPAGE
86 				 */
87 #define	LINUX_USRSTACK_LA48	LINUX_SHAREDPAGE_LA48
88 #define	LINUX_PS_STRINGS_LA48	(LINUX_USRSTACK_LA48 - \
89 				    sizeof(struct ps_strings))
90 
91 static int linux_szsigcode;
92 static vm_object_t linux_vdso_obj;
93 static char *linux_vdso_mapping;
94 extern char _binary_linux_vdso_so_o_start;
95 extern char _binary_linux_vdso_so_o_end;
96 static vm_offset_t linux_vdso_base;
97 
98 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
99 extern const char *linux_syscallnames[];
100 
101 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
102 
103 static void	linux_vdso_install(const void *param);
104 static void	linux_vdso_deinstall(const void *param);
105 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
106 static void	linux_set_syscall_retval(struct thread *td, int error);
107 static int	linux_fetch_syscall_args(struct thread *td);
108 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
109 		    uintptr_t stack);
110 static void	linux_exec_sysvec_init(void *param);
111 static int	linux_on_exec_vmspace(struct proc *p,
112 		    struct image_params *imgp);
113 static void	linux_set_fork_retval(struct thread *td);
114 static int	linux_vsyscall(struct thread *td);
115 
116 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
117 LINUX_VDSO_SYM_CHAR(linux_platform);
118 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
119 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
120 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
121 
122 /*
123  * According to the Intel x86 ISA 64-bit syscall
124  * saves %rip to %rcx and rflags to %r11. Registers on syscall entry:
125  * %rax  system call number
126  * %rcx  return address
127  * %r11  saved rflags
128  * %rdi  arg1
129  * %rsi  arg2
130  * %rdx  arg3
131  * %r10  arg4
132  * %r8   arg5
133  * %r9   arg6
134  *
135  * Then FreeBSD fast_syscall() move registers:
136  * %rcx -> trapframe.tf_rip
137  * %r10 -> trapframe.tf_rcx
138  */
139 static int
140 linux_fetch_syscall_args(struct thread *td)
141 {
142 	struct proc *p;
143 	struct trapframe *frame;
144 	struct syscall_args *sa;
145 
146 	p = td->td_proc;
147 	frame = td->td_frame;
148 	sa = &td->td_sa;
149 
150 	sa->args[0] = frame->tf_rdi;
151 	sa->args[1] = frame->tf_rsi;
152 	sa->args[2] = frame->tf_rdx;
153 	sa->args[3] = frame->tf_rcx;
154 	sa->args[4] = frame->tf_r8;
155 	sa->args[5] = frame->tf_r9;
156 	sa->code = frame->tf_rax;
157 	sa->original_code = sa->code;
158 
159 	if (sa->code >= p->p_sysent->sv_size)
160 		/* nosys */
161 		sa->callp = &nosys_sysent;
162 	else
163 		sa->callp = &p->p_sysent->sv_table[sa->code];
164 
165 	/* Restore r10 earlier to avoid doing this multiply times. */
166 	frame->tf_r10 = frame->tf_rcx;
167 	/* Restore %rcx for machine context. */
168 	frame->tf_rcx = frame->tf_rip;
169 
170 	td->td_retval[0] = 0;
171 	return (0);
172 }
173 
174 static void
175 linux_set_syscall_retval(struct thread *td, int error)
176 {
177 	struct trapframe *frame;
178 
179 	frame = td->td_frame;
180 
181 	switch (error) {
182 	case 0:
183 		frame->tf_rax = td->td_retval[0];
184 		break;
185 
186 	case ERESTART:
187 		/*
188 		 * Reconstruct pc, we know that 'syscall' is 2 bytes,
189 		 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
190 		 * We saved this in tf_err.
191 		 *
192 		 */
193 		frame->tf_rip -= frame->tf_err;
194 		break;
195 
196 	case EJUSTRETURN:
197 		break;
198 
199 	default:
200 		frame->tf_rax = bsd_to_linux_errno(error);
201 		break;
202 	}
203 
204 	/*
205 	 * Differently from FreeBSD native ABI, on Linux only %rcx
206 	 * and %r11 values are not preserved across the syscall.
207 	 * Require full context restore to get all registers except
208 	 * those two restored at return to usermode.
209 	 */
210 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
211 }
212 
213 static void
214 linux_set_fork_retval(struct thread *td)
215 {
216 	struct trapframe *frame = td->td_frame;
217 
218 	frame->tf_rax = 0;
219 }
220 
221 void
222 linux64_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos)
223 {
224 
225 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
226 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature);
227 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2());
228 	AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform));
229 }
230 
231 /*
232  * Reset registers to default values on exec.
233  */
234 static void
235 linux_exec_setregs(struct thread *td, struct image_params *imgp,
236     uintptr_t stack)
237 {
238 	struct trapframe *regs;
239 	struct pcb *pcb;
240 	register_t saved_rflags;
241 
242 	regs = td->td_frame;
243 	pcb = td->td_pcb;
244 
245 	if (td->td_proc->p_md.md_ldt != NULL)
246 		user_ldt_free(td);
247 
248 	pcb->pcb_fsbase = 0;
249 	pcb->pcb_gsbase = 0;
250 	clear_pcb_flags(pcb, PCB_32BIT);
251 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
252 	set_pcb_flags(pcb, PCB_FULL_IRET);
253 
254 	saved_rflags = regs->tf_rflags & PSL_T;
255 	bzero((char *)regs, sizeof(struct trapframe));
256 	regs->tf_rip = imgp->entry_addr;
257 	regs->tf_rsp = stack;
258 	regs->tf_rflags = PSL_USER | saved_rflags;
259 	regs->tf_ss = _udatasel;
260 	regs->tf_cs = _ucodesel;
261 	regs->tf_ds = _udatasel;
262 	regs->tf_es = _udatasel;
263 	regs->tf_fs = _ufssel;
264 	regs->tf_gs = _ugssel;
265 	regs->tf_flags = TF_HASSEGS;
266 
267 	x86_clear_dbregs(pcb);
268 
269 	/*
270 	 * Drop the FP state if we hold it, so that the process gets a
271 	 * clean FP state if it uses the FPU again.
272 	 */
273 	fpstate_drop(td);
274 }
275 
276 static int
277 linux_fxrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc)
278 {
279 	struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0];
280 	int error;
281 
282 	error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate));
283 	if (error != 0)
284 		return (error);
285 	bzero(&fp->sv_pad[0], sizeof(fp->sv_pad));
286 	return (set_fpcontext(td, mcp, NULL, 0));
287 }
288 
289 static int
290 linux_xrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc)
291 {
292 	struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0];
293 	char *xfpustate;
294 	struct proc *p;
295 	uint32_t magic2;
296 	int error;
297 
298 	p = td->td_proc;
299 	mcp->mc_xfpustate_len = cpu_max_ext_state_size - sizeof(struct savefpu);
300 
301 	/* Legacy region of an xsave area. */
302 	error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate));
303 	if (error != 0)
304 		return (error);
305 	bzero(&fp->sv_pad[0], sizeof(fp->sv_pad));
306 
307 	/* Extended region of an xsave area. */
308 	sc->sc_fpstate += sizeof(mcp->mc_fpstate);
309 	xfpustate = (char *)fpu_save_area_alloc();
310 	error = copyin(PTRIN(sc->sc_fpstate), xfpustate, mcp->mc_xfpustate_len);
311 	if (error != 0) {
312 		fpu_save_area_free((struct savefpu *)xfpustate);
313 		uprintf("pid %d (%s): linux xrstor failed\n", p->p_pid,
314 		    td->td_name);
315 		return (error);
316 	}
317 
318 	/* Linux specific end of xsave area marker. */
319 	sc->sc_fpstate += mcp->mc_xfpustate_len;
320 	error = copyin(PTRIN(sc->sc_fpstate), &magic2, LINUX_FP_XSTATE_MAGIC2_SIZE);
321 	if (error != 0 || magic2 != LINUX_FP_XSTATE_MAGIC2) {
322 		fpu_save_area_free((struct savefpu *)xfpustate);
323 		uprintf("pid %d (%s): sigreturn magic2 0x%x error %d\n",
324 		    p->p_pid, td->td_name, magic2, error);
325 		return (error);
326 	}
327 
328 	error = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len);
329 	fpu_save_area_free((struct savefpu *)xfpustate);
330 	if (error != 0) {
331 		uprintf("pid %d (%s): sigreturn set_fpcontext error %d\n",
332 		    p->p_pid, td->td_name, error);
333 	}
334 	return (error);
335 }
336 
337 static int
338 linux_copyin_fpstate(struct thread *td, struct l_ucontext *uc)
339 {
340 	mcontext_t mc;
341 
342 	bzero(&mc, sizeof(mc));
343 	mc.mc_ownedfp = _MC_FPOWNED_FPU;
344 	mc.mc_fpformat = _MC_FPFMT_XMM;
345 
346 	if ((uc->uc_flags & LINUX_UC_FP_XSTATE) != 0)
347 		return (linux_xrstor(td, &mc, &uc->uc_mcontext));
348 	else
349 		return (linux_fxrstor(td, &mc, &uc->uc_mcontext));
350 }
351 
352 /*
353  * Copied from amd64/amd64/machdep.c
354  */
355 int
356 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
357 {
358 	struct proc *p;
359 	struct l_rt_sigframe sf;
360 	struct l_sigcontext *context;
361 	struct trapframe *regs;
362 	unsigned long rflags;
363 	sigset_t bmask;
364 	int error;
365 	ksiginfo_t ksi;
366 
367 	regs = td->td_frame;
368 	error = copyin((void *)regs->tf_rbx, &sf, sizeof(sf));
369 	if (error != 0)
370 		return (error);
371 
372 	p = td->td_proc;
373 	context = &sf.sf_uc.uc_mcontext;
374 	rflags = context->sc_rflags;
375 
376 	/*
377 	 * Don't allow users to change privileged or reserved flags.
378 	 */
379 	/*
380 	 * XXX do allow users to change the privileged flag PSL_RF.
381 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
382 	 * should sometimes set it there too.  tf_rflags is kept in
383 	 * the signal context during signal handling and there is no
384 	 * other place to remember it, so the PSL_RF bit may be
385 	 * corrupted by the signal handler without us knowing.
386 	 * Corruption of the PSL_RF bit at worst causes one more or
387 	 * one less debugger trap, so allowing it is fairly harmless.
388 	 */
389 	if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
390 		uprintf("pid %d comm %s linux mangled rflags %#lx\n",
391 		    p->p_pid, p->p_comm, rflags);
392 		return (EINVAL);
393 	}
394 
395 	/*
396 	 * Don't allow users to load a valid privileged %cs.  Let the
397 	 * hardware check for invalid selectors, excess privilege in
398 	 * other selectors, invalid %eip's and invalid %esp's.
399 	 */
400 	if (!CS_SECURE(context->sc_cs)) {
401 		uprintf("pid %d comm %s linux mangled cs %#x\n",
402 		    p->p_pid, p->p_comm, context->sc_cs);
403 		ksiginfo_init_trap(&ksi);
404 		ksi.ksi_signo = SIGBUS;
405 		ksi.ksi_code = BUS_OBJERR;
406 		ksi.ksi_trapno = T_PROTFLT;
407 		ksi.ksi_addr = (void *)regs->tf_rip;
408 		trapsignal(td, &ksi);
409 		return (EINVAL);
410 	}
411 
412 	linux_to_bsd_sigset(&sf.sf_uc.uc_sigmask, &bmask);
413 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
414 
415 	regs->tf_rdi    = context->sc_rdi;
416 	regs->tf_rsi    = context->sc_rsi;
417 	regs->tf_rdx    = context->sc_rdx;
418 	regs->tf_rbp    = context->sc_rbp;
419 	regs->tf_rbx    = context->sc_rbx;
420 	regs->tf_rcx    = context->sc_rcx;
421 	regs->tf_rax    = context->sc_rax;
422 	regs->tf_rip    = context->sc_rip;
423 	regs->tf_rsp    = context->sc_rsp;
424 	regs->tf_r8     = context->sc_r8;
425 	regs->tf_r9     = context->sc_r9;
426 	regs->tf_r10    = context->sc_r10;
427 	regs->tf_r11    = context->sc_r11;
428 	regs->tf_r12    = context->sc_r12;
429 	regs->tf_r13    = context->sc_r13;
430 	regs->tf_r14    = context->sc_r14;
431 	regs->tf_r15    = context->sc_r15;
432 	regs->tf_cs     = context->sc_cs;
433 	regs->tf_err    = context->sc_err;
434 	regs->tf_rflags = rflags;
435 
436 	error = linux_copyin_fpstate(td, &sf.sf_uc);
437 	if (error != 0) {
438 		uprintf("pid %d comm %s linux can't restore fpu state %d\n",
439 		    p->p_pid, p->p_comm, error);
440 		return (error);
441 	}
442 
443 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
444 	return (EJUSTRETURN);
445 }
446 
447 static int
448 linux_fxsave(mcontext_t *mcp, void *ufp)
449 {
450 	struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0];
451 
452 	bzero(&fx->reserved2[0], sizeof(fx->reserved2));
453 	return (copyout(fx, ufp, sizeof(*fx)));
454 }
455 
456 static int
457 linux_xsave(mcontext_t *mcp, char *xfpusave, char *ufp)
458 {
459 	struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0];
460 	uint32_t magic2;
461 	int error;
462 
463 	/* Legacy region of an xsave area. */
464 	fx->sw_reserved.magic1 = LINUX_FP_XSTATE_MAGIC1;
465 	fx->sw_reserved.xstate_size = mcp->mc_xfpustate_len + sizeof(*fx);
466 	fx->sw_reserved.extended_size = fx->sw_reserved.xstate_size +
467 	    LINUX_FP_XSTATE_MAGIC2_SIZE;
468 	fx->sw_reserved.xfeatures = xsave_mask;
469 
470 	error = copyout(fx, ufp, sizeof(*fx));
471 	if (error != 0)
472 		return (error);
473 	ufp += sizeof(*fx);
474 
475 	/* Extended region of an xsave area. */
476 	error = copyout(xfpusave, ufp, mcp->mc_xfpustate_len);
477 	if (error != 0)
478 		return (error);
479 
480 	/* Linux specific end of xsave area marker. */
481 	ufp += mcp->mc_xfpustate_len;
482 	magic2 = LINUX_FP_XSTATE_MAGIC2;
483 	return (copyout(&magic2, ufp, LINUX_FP_XSTATE_MAGIC2_SIZE));
484 }
485 
486 static int
487 linux_copyout_fpstate(struct thread *td, struct l_ucontext *uc, char **sp)
488 {
489 	size_t xfpusave_len;
490 	char *xfpusave;
491 	mcontext_t mc;
492 	char *ufp = *sp;
493 
494 	get_fpcontext(td, &mc, &xfpusave, &xfpusave_len);
495 	KASSERT(mc.mc_fpformat != _MC_FPFMT_NODEV, ("fpu not present"));
496 
497 	/* Room for fxsave area. */
498 	ufp -= sizeof(struct l_fpstate);
499 	if (xfpusave != NULL) {
500 		/* Room for xsave area. */
501 		ufp -= (xfpusave_len + LINUX_FP_XSTATE_MAGIC2_SIZE);
502 		uc->uc_flags |= LINUX_UC_FP_XSTATE;
503 	}
504 	*sp = ufp = (char *)((unsigned long)ufp & ~0x3Ful);
505 
506 	if (xfpusave != NULL)
507 		return (linux_xsave(&mc, xfpusave, ufp));
508 	else
509 		return (linux_fxsave(&mc, ufp));
510 }
511 
512 /*
513  * copied from amd64/amd64/machdep.c
514  *
515  * Send an interrupt to process.
516  */
517 static void
518 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
519 {
520 	struct l_rt_sigframe sf, *sfp;
521 	struct proc *p;
522 	struct thread *td;
523 	struct sigacts *psp;
524 	char *sp;
525 	struct trapframe *regs;
526 	int sig, code;
527 	int oonstack, issiginfo;
528 
529 	td = curthread;
530 	p = td->td_proc;
531 	PROC_LOCK_ASSERT(p, MA_OWNED);
532 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
533 	psp = p->p_sigacts;
534 	issiginfo = SIGISMEMBER(psp->ps_siginfo, sig);
535 	code = ksi->ksi_code;
536 	mtx_assert(&psp->ps_mtx, MA_OWNED);
537 	regs = td->td_frame;
538 	oonstack = sigonstack(regs->tf_rsp);
539 
540 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
541 	    catcher, sig, mask, code);
542 
543 	bzero(&sf, sizeof(sf));
544 	sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
545 	sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
546 	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
547 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
548 
549 	/* Allocate space for the signal handler context. */
550 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
551 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
552 		sp = (char *)td->td_sigstk.ss_sp + td->td_sigstk.ss_size;
553 	} else
554 		sp = (char *)regs->tf_rsp - 128;
555 
556 	mtx_unlock(&psp->ps_mtx);
557 	PROC_UNLOCK(p);
558 
559 	if (linux_copyout_fpstate(td, &sf.sf_uc, &sp) != 0) {
560 		uprintf("pid %d comm %s linux can't save fpu state, killing\n",
561 		    p->p_pid, p->p_comm);
562 		PROC_LOCK(p);
563 		sigexit(td, SIGILL);
564 	}
565 	sf.sf_uc.uc_mcontext.sc_fpstate = (register_t)sp;
566 
567 	/* Make room, keeping the stack aligned. */
568 	sp -= sizeof(struct l_rt_sigframe);
569 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
570 
571 	/* Save user context. */
572 	bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask);
573 	sf.sf_uc.uc_mcontext.sc_mask   = sf.sf_uc.uc_sigmask;
574 	sf.sf_uc.uc_mcontext.sc_rdi    = regs->tf_rdi;
575 	sf.sf_uc.uc_mcontext.sc_rsi    = regs->tf_rsi;
576 	sf.sf_uc.uc_mcontext.sc_rdx    = regs->tf_rdx;
577 	sf.sf_uc.uc_mcontext.sc_rbp    = regs->tf_rbp;
578 	sf.sf_uc.uc_mcontext.sc_rbx    = regs->tf_rbx;
579 	sf.sf_uc.uc_mcontext.sc_rcx    = regs->tf_rcx;
580 	sf.sf_uc.uc_mcontext.sc_rax    = regs->tf_rax;
581 	sf.sf_uc.uc_mcontext.sc_rip    = regs->tf_rip;
582 	sf.sf_uc.uc_mcontext.sc_rsp    = regs->tf_rsp;
583 	sf.sf_uc.uc_mcontext.sc_r8     = regs->tf_r8;
584 	sf.sf_uc.uc_mcontext.sc_r9     = regs->tf_r9;
585 	sf.sf_uc.uc_mcontext.sc_r10    = regs->tf_r10;
586 	sf.sf_uc.uc_mcontext.sc_r11    = regs->tf_r11;
587 	sf.sf_uc.uc_mcontext.sc_r12    = regs->tf_r12;
588 	sf.sf_uc.uc_mcontext.sc_r13    = regs->tf_r13;
589 	sf.sf_uc.uc_mcontext.sc_r14    = regs->tf_r14;
590 	sf.sf_uc.uc_mcontext.sc_r15    = regs->tf_r15;
591 	sf.sf_uc.uc_mcontext.sc_cs     = regs->tf_cs;
592 	sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags;
593 	sf.sf_uc.uc_mcontext.sc_err    = regs->tf_err;
594 	sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
595 	sf.sf_uc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
596 
597 	/* Translate the signal. */
598 	sig = bsd_to_linux_signal(sig);
599 	/* Fill in POSIX parts. */
600 	siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig);
601 
602 	/* Copy the sigframe out to the user's stack. */
603 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
604 		uprintf("pid %d comm %s has trashed its stack, killing\n",
605 		    p->p_pid, p->p_comm);
606 		PROC_LOCK(p);
607 		sigexit(td, SIGILL);
608 	}
609 
610 	fpstate_drop(td);
611 	/* Build the argument list for the signal handler. */
612 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
613 	regs->tf_rax = 0;
614 	if (issiginfo) {
615 		regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
616 		regs->tf_rdx = (register_t)&sfp->sf_uc;	/* arg 3 in %rdx */
617 	} else {
618 		regs->tf_rsi = 0;
619 		regs->tf_rdx = 0;
620 	}
621 	regs->tf_rcx = (register_t)catcher;
622 	regs->tf_rsp = (long)sfp;
623 	regs->tf_rip = linux_rt_sigcode;
624 	regs->tf_rflags &= ~(PSL_T | PSL_D);
625 	regs->tf_cs = _ucodesel;
626 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
627 	PROC_LOCK(p);
628 	mtx_lock(&psp->ps_mtx);
629 }
630 
631 #define	LINUX_VSYSCALL_START		(-10UL << 20)
632 #define	LINUX_VSYSCALL_SZ		1024
633 
634 const unsigned long linux_vsyscall_vector[] = {
635 	LINUX_SYS_gettimeofday,
636 	LINUX_SYS_linux_time,
637 	LINUX_SYS_linux_getcpu,
638 };
639 
640 static int
641 linux_vsyscall(struct thread *td)
642 {
643 	struct trapframe *frame;
644 	uint64_t retqaddr;
645 	int code, traced;
646 	int error;
647 
648 	frame = td->td_frame;
649 
650 	/* Check %rip for vsyscall area. */
651 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
652 		return (EINVAL);
653 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
654 		return (EINVAL);
655 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
656 	if (code >= nitems(linux_vsyscall_vector))
657 		return (EINVAL);
658 
659 	/*
660 	 * vsyscall called as callq *(%rax), so we must
661 	 * use return address from %rsp and also fixup %rsp.
662 	 */
663 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
664 	if (error)
665 		return (error);
666 
667 	frame->tf_rip = retqaddr;
668 	frame->tf_rax = linux_vsyscall_vector[code];
669 	frame->tf_rsp += 8;
670 
671 	traced = (frame->tf_flags & PSL_T);
672 
673 	amd64_syscall(td, traced);
674 
675 	return (0);
676 }
677 
678 struct sysentvec elf_linux_sysvec = {
679 	.sv_size	= LINUX_SYS_MAXSYSCALL,
680 	.sv_table	= linux_sysent,
681 	.sv_fixup	= __elfN(freebsd_fixup),
682 	.sv_sendsig	= linux_rt_sendsig,
683 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
684 	.sv_szsigcode	= &linux_szsigcode,
685 	.sv_name	= "Linux ELF64",
686 	.sv_coredump	= elf64_coredump,
687 	.sv_elf_core_osabi = ELFOSABI_NONE,
688 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
689 	.sv_elf_core_prepare_notes = linux64_prepare_notes,
690 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
691 	.sv_minuser	= VM_MIN_ADDRESS,
692 	.sv_maxuser	= VM_MAXUSER_ADDRESS_LA48,
693 	.sv_usrstack	= LINUX_USRSTACK_LA48,
694 	.sv_psstrings	= LINUX_PS_STRINGS_LA48,
695 	.sv_psstringssz	= sizeof(struct ps_strings),
696 	.sv_stackprot	= VM_PROT_ALL,
697 	.sv_copyout_auxargs = __linuxN(copyout_auxargs),
698 	.sv_copyout_strings = __linuxN(copyout_strings),
699 	.sv_setregs	= linux_exec_setregs,
700 	.sv_fixlimit	= NULL,
701 	.sv_maxssiz	= NULL,
702 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
703 	    SV_SIG_WAITNDQ | SV_TIMEKEEP,
704 	.sv_set_syscall_retval = linux_set_syscall_retval,
705 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
706 	.sv_syscallnames = linux_syscallnames,
707 	.sv_shared_page_base = LINUX_SHAREDPAGE_LA48,
708 	.sv_shared_page_len = PAGE_SIZE,
709 	.sv_schedtail	= linux_schedtail,
710 	.sv_thread_detach = linux_thread_detach,
711 	.sv_trap	= linux_vsyscall,
712 	.sv_hwcap	= NULL,
713 	.sv_hwcap2	= NULL,
714 	.sv_onexec	= linux_on_exec_vmspace,
715 	.sv_onexit	= linux_on_exit,
716 	.sv_ontdexit	= linux_thread_dtor,
717 	.sv_setid_allowed = &linux_setid_allowed_query,
718 	.sv_set_fork_retval = linux_set_fork_retval,
719 };
720 
721 static int
722 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
723 {
724 	int error;
725 
726 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
727 	    LINUX_VDSOPAGE_SIZE, imgp);
728 	if (error == 0)
729 		error = linux_on_exec(p, imgp);
730 	return (error);
731 }
732 
733 /*
734  * linux_vdso_install() and linux_exec_sysvec_init() must be called
735  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
736  */
737 static void
738 linux_exec_sysvec_init(void *param)
739 {
740 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
741 	struct sysentvec *sv;
742 	ptrdiff_t tkoff;
743 
744 	sv = param;
745 	amd64_lower_shared_page(sv);
746 	/* Fill timekeep_base */
747 	exec_sysvec_init(sv);
748 
749 	tkoff = kern_timekeep_base - linux_vdso_base;
750 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
751 	*ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
752 
753 	tkoff = kern_tsc_selector - linux_vdso_base;
754 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
755 	*ktsc_selector = linux_vdso_tsc_selector_idx();
756 	if (bootverbose)
757 		printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector);
758 
759 	tkoff = kern_cpu_selector - linux_vdso_base;
760 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
761 	*ktsc_selector = linux_vdso_cpu_selector_idx();
762 	if (bootverbose)
763 		printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector);
764 }
765 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
766     linux_exec_sysvec_init, &elf_linux_sysvec);
767 
768 static void
769 linux_vdso_install(const void *param)
770 {
771 	char *vdso_start = &_binary_linux_vdso_so_o_start;
772 	char *vdso_end = &_binary_linux_vdso_so_o_end;
773 
774 	linux_szsigcode = vdso_end - vdso_start;
775 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
776 
777 	linux_vdso_base = LINUX_VDSOPAGE_LA48;
778 	if (hw_lower_amd64_sharedpage != 0)
779 		linux_vdso_base -= PAGE_SIZE;
780 
781 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
782 
783 	linux_vdso_obj = __elfN(linux_shared_page_init)
784 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
785 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
786 
787 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
788 }
789 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
790     linux_vdso_install, NULL);
791 
792 static void
793 linux_vdso_deinstall(const void *param)
794 {
795 
796 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
797 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
798 }
799 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
800     linux_vdso_deinstall, NULL);
801 
802 static void
803 linux_vdso_reloc(char *mapping, Elf_Addr offset)
804 {
805 	const Elf_Ehdr *ehdr;
806 	const Elf_Shdr *shdr;
807 	Elf64_Addr *where, val;
808 	Elf_Size rtype, symidx;
809 	const Elf_Rela *rela;
810 	Elf_Addr addr, addend;
811 	int relacnt;
812 	int i, j;
813 
814 	MPASS(offset != 0);
815 
816 	relacnt = 0;
817 	ehdr = (const Elf_Ehdr *)mapping;
818 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
819 	for (i = 0; i < ehdr->e_shnum; i++)
820 	{
821 		switch (shdr[i].sh_type) {
822 		case SHT_REL:
823 			printf("Linux x86_64 vDSO: unexpected Rel section\n");
824 			break;
825 		case SHT_RELA:
826 			rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
827 			relacnt = shdr[i].sh_size / sizeof(*rela);
828 		}
829 	}
830 
831 	for (j = 0; j < relacnt; j++, rela++) {
832 		where = (Elf_Addr *)(mapping + rela->r_offset);
833 		addend = rela->r_addend;
834 		rtype = ELF_R_TYPE(rela->r_info);
835 		symidx = ELF_R_SYM(rela->r_info);
836 
837 		switch (rtype) {
838 		case R_X86_64_NONE:	/* none */
839 			break;
840 
841 		case R_X86_64_RELATIVE:	/* B + A */
842 			addr = (Elf_Addr)(offset + addend);
843 			val = addr;
844 			if (*where != val)
845 				*where = val;
846 			break;
847 		case R_X86_64_IRELATIVE:
848 			printf("Linux x86_64 vDSO: unexpected ifunc relocation, "
849 			    "symbol index %ld\n", symidx);
850 			break;
851 		default:
852 			printf("Linux x86_64 vDSO: unexpected relocation type %ld, "
853 			    "symbol index %ld\n", rtype, symidx);
854 		}
855 	}
856 }
857 
858 static Elf_Brandnote linux64_brandnote = {
859 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
860 	.hdr.n_descsz	= 16,
861 	.hdr.n_type	= 1,
862 	.vendor		= GNU_ABI_VENDOR,
863 	.flags		= BN_TRANSLATE_OSREL,
864 	.trans_osrel	= linux_trans_osrel
865 };
866 
867 static Elf64_Brandinfo linux_glibc2brand = {
868 	.brand		= ELFOSABI_LINUX,
869 	.machine	= EM_X86_64,
870 	.compat_3_brand	= "Linux",
871 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
872 	.sysvec		= &elf_linux_sysvec,
873 	.interp_newpath	= NULL,
874 	.brand_note	= &linux64_brandnote,
875 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
876 };
877 
878 static Elf64_Brandinfo linux_glibc2brandshort = {
879 	.brand		= ELFOSABI_LINUX,
880 	.machine	= EM_X86_64,
881 	.compat_3_brand	= "Linux",
882 	.interp_path	= "/lib64/ld-linux.so.2",
883 	.sysvec		= &elf_linux_sysvec,
884 	.interp_newpath	= NULL,
885 	.brand_note	= &linux64_brandnote,
886 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
887 };
888 
889 static Elf64_Brandinfo linux_muslbrand = {
890 	.brand		= ELFOSABI_LINUX,
891 	.machine	= EM_X86_64,
892 	.compat_3_brand	= "Linux",
893 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
894 	.sysvec		= &elf_linux_sysvec,
895 	.interp_newpath	= NULL,
896 	.brand_note	= &linux64_brandnote,
897 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
898 			    LINUX_BI_FUTEX_REQUEUE
899 };
900 
901 static Elf64_Brandinfo *linux_brandlist[] = {
902 	&linux_glibc2brand,
903 	&linux_glibc2brandshort,
904 	&linux_muslbrand,
905 	NULL
906 };
907 
908 static int
909 linux64_elf_modevent(module_t mod, int type, void *data)
910 {
911 	Elf64_Brandinfo **brandinfo;
912 	int error;
913 	struct linux_ioctl_handler **lihp;
914 
915 	error = 0;
916 
917 	switch(type) {
918 	case MOD_LOAD:
919 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
920 		     ++brandinfo)
921 			if (elf64_insert_brand_entry(*brandinfo) < 0)
922 				error = EINVAL;
923 		if (error == 0) {
924 			SET_FOREACH(lihp, linux_ioctl_handler_set)
925 				linux_ioctl_register_handler(*lihp);
926 			stclohz = (stathz ? stathz : hz);
927 			if (bootverbose)
928 				printf("Linux x86-64 ELF exec handler installed\n");
929 		} else
930 			printf("cannot insert Linux x86-64 ELF brand handler\n");
931 		break;
932 	case MOD_UNLOAD:
933 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
934 		     ++brandinfo)
935 			if (elf64_brand_inuse(*brandinfo))
936 				error = EBUSY;
937 		if (error == 0) {
938 			for (brandinfo = &linux_brandlist[0];
939 			     *brandinfo != NULL; ++brandinfo)
940 				if (elf64_remove_brand_entry(*brandinfo) < 0)
941 					error = EINVAL;
942 		}
943 		if (error == 0) {
944 			SET_FOREACH(lihp, linux_ioctl_handler_set)
945 				linux_ioctl_unregister_handler(*lihp);
946 			if (bootverbose)
947 				printf("Linux x86_64 ELF exec handler removed\n");
948 		} else
949 			printf("Could not deinstall Linux x86_64 ELF interpreter entry\n");
950 		break;
951 	default:
952 		return (EOPNOTSUPP);
953 	}
954 	return (error);
955 }
956 
957 static moduledata_t linux64_elf_mod = {
958 	"linux64elf",
959 	linux64_elf_modevent,
960 	0
961 };
962 
963 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
964 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
965 FEATURE(linux64, "Linux 64bit support");
966