xref: /freebsd/sys/amd64/linux/linux_sysvec.c (revision 2f68cdb94454035989fa370bcb68db2f4e96b1ed)
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/syscallsubr.h>
57 #include <sys/sysctl.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 #include <machine/trap.h>
76 
77 #include <amd64/linux/linux.h>
78 #include <amd64/linux/linux_proto.h>
79 #include <compat/linux/linux_emul.h>
80 #include <compat/linux/linux_futex.h>
81 #include <compat/linux/linux_ioctl.h>
82 #include <compat/linux/linux_mib.h>
83 #include <compat/linux/linux_misc.h>
84 #include <compat/linux/linux_signal.h>
85 #include <compat/linux/linux_sysproto.h>
86 #include <compat/linux/linux_util.h>
87 #include <compat/linux/linux_vdso.h>
88 
89 MODULE_VERSION(linux64, 1);
90 
91 #if BYTE_ORDER == LITTLE_ENDIAN
92 #define SHELLMAGIC      0x2123 /* #! */
93 #else
94 #define SHELLMAGIC      0x2321
95 #endif
96 
97 #if defined(DEBUG)
98 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
99 	    CTLTYPE_STRING | CTLFLAG_RW,
100 	    0, 0, linux_sysctl_debug, "A",
101 	    "Linux 64 debugging control");
102 #endif
103 
104 /*
105  * Allow the sendsig functions to use the ldebug() facility even though they
106  * are not syscalls themselves.  Map them to syscall 0.  This is slightly less
107  * bogus than using ldebug(sigreturn).
108  */
109 #define	LINUX_SYS_linux_rt_sendsig	0
110 
111 const char *linux_kplatform;
112 static int linux_szsigcode;
113 static vm_object_t linux_shared_page_obj;
114 static char *linux_shared_page_mapping;
115 extern char _binary_linux_locore_o_start;
116 extern char _binary_linux_locore_o_end;
117 
118 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
119 
120 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
121 
122 static register_t * linux_copyout_strings(struct image_params *imgp);
123 static int	linux_fixup_elf(register_t **stack_base,
124 		    struct image_params *iparams);
125 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
126 static void	linux_vdso_install(void *param);
127 static void	linux_vdso_deinstall(void *param);
128 static void	linux_set_syscall_retval(struct thread *td, int error);
129 static int	linux_fetch_syscall_args(struct thread *td);
130 static int	linux_exec_imgact_try(struct image_params *iparams);
131 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
132 		    u_long stack);
133 static int	linux_vsyscall(struct thread *td);
134 
135 #define LINUX_T_UNKNOWN  255
136 static int _bsd_to_linux_trapcode[] = {
137 	LINUX_T_UNKNOWN,	/* 0 */
138 	6,			/* 1  T_PRIVINFLT */
139 	LINUX_T_UNKNOWN,	/* 2 */
140 	3,			/* 3  T_BPTFLT */
141 	LINUX_T_UNKNOWN,	/* 4 */
142 	LINUX_T_UNKNOWN,	/* 5 */
143 	16,			/* 6  T_ARITHTRAP */
144 	254,			/* 7  T_ASTFLT */
145 	LINUX_T_UNKNOWN,	/* 8 */
146 	13,			/* 9  T_PROTFLT */
147 	1,			/* 10 T_TRCTRAP */
148 	LINUX_T_UNKNOWN,	/* 11 */
149 	14,			/* 12 T_PAGEFLT */
150 	LINUX_T_UNKNOWN,	/* 13 */
151 	17,			/* 14 T_ALIGNFLT */
152 	LINUX_T_UNKNOWN,	/* 15 */
153 	LINUX_T_UNKNOWN,	/* 16 */
154 	LINUX_T_UNKNOWN,	/* 17 */
155 	0,			/* 18 T_DIVIDE */
156 	2,			/* 19 T_NMI */
157 	4,			/* 20 T_OFLOW */
158 	5,			/* 21 T_BOUND */
159 	7,			/* 22 T_DNA */
160 	8,			/* 23 T_DOUBLEFLT */
161 	9,			/* 24 T_FPOPFLT */
162 	10,			/* 25 T_TSSFLT */
163 	11,			/* 26 T_SEGNPFLT */
164 	12,			/* 27 T_STKFLT */
165 	18,			/* 28 T_MCHK */
166 	19,			/* 29 T_XMMFLT */
167 	15			/* 30 T_RESERVED */
168 };
169 #define bsd_to_linux_trapcode(code) \
170     ((code)<nitems(_bsd_to_linux_trapcode)? \
171      _bsd_to_linux_trapcode[(code)]: \
172      LINUX_T_UNKNOWN)
173 
174 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
175 LINUX_VDSO_SYM_CHAR(linux_platform);
176 
177 /*
178  * If FreeBSD & Linux have a difference of opinion about what a trap
179  * means, deal with it here.
180  *
181  * MPSAFE
182  */
183 static int
184 linux_translate_traps(int signal, int trap_code)
185 {
186 
187 	if (signal != SIGBUS)
188 		return (signal);
189 	switch (trap_code) {
190 	case T_PROTFLT:
191 	case T_TSSFLT:
192 	case T_DOUBLEFLT:
193 	case T_PAGEFLT:
194 		return (SIGSEGV);
195 	default:
196 		return (signal);
197 	}
198 }
199 
200 static int
201 linux_fetch_syscall_args(struct thread *td)
202 {
203 	struct proc *p;
204 	struct trapframe *frame;
205 	struct syscall_args *sa;
206 
207 	p = td->td_proc;
208 	frame = td->td_frame;
209 	sa = &td->td_sa;
210 
211 	sa->args[0] = frame->tf_rdi;
212 	sa->args[1] = frame->tf_rsi;
213 	sa->args[2] = frame->tf_rdx;
214 	sa->args[3] = frame->tf_rcx;
215 	sa->args[4] = frame->tf_r8;
216 	sa->args[5] = frame->tf_r9;
217 	sa->code = frame->tf_rax;
218 
219 	if (sa->code >= p->p_sysent->sv_size)
220 		/* nosys */
221 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
222 	else
223 		sa->callp = &p->p_sysent->sv_table[sa->code];
224 	sa->narg = sa->callp->sy_narg;
225 
226 	td->td_retval[0] = 0;
227 	return (0);
228 }
229 
230 static void
231 linux_set_syscall_retval(struct thread *td, int error)
232 {
233 	struct trapframe *frame = td->td_frame;
234 
235 	/*
236 	 * On Linux only %rcx and %r11 values are not preserved across
237 	 * the syscall.  So, do not clobber %rdx and %r10.
238 	 */
239 	td->td_retval[1] = frame->tf_rdx;
240 	frame->tf_r10 = frame->tf_rcx;
241 
242 	cpu_set_syscall_retval(td, error);
243 
244 	 /* Restore all registers. */
245 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
246 }
247 
248 static int
249 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
250 {
251 	Elf_Auxargs *args;
252 	Elf_Addr *base;
253 	Elf_Addr *pos;
254 	struct ps_strings *arginfo;
255 	struct proc *p;
256 	int issetugid;
257 
258 	p = imgp->proc;
259 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
260 
261 	KASSERT(curthread->td_proc == imgp->proc,
262 	    ("unsafe linux_fixup_elf(), should be curproc"));
263 	base = (Elf64_Addr *)*stack_base;
264 	args = (Elf64_Auxargs *)imgp->auxargs;
265 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
266 
267 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
268 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
269 	    imgp->proc->p_sysent->sv_shared_page_base);
270 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
271 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
272 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
273 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
274 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
275 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
276 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
277 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
278 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
279 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
280 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
281 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
282 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
283 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
284 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
285 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
286 	if (imgp->execpathp != 0)
287 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
288 	if (args->execfd != -1)
289 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
290 	AUXARGS_ENTRY(pos, AT_NULL, 0);
291 	free(imgp->auxargs, M_TEMP);
292 	imgp->auxargs = NULL;
293 
294 	base--;
295 	suword(base, (uint64_t)imgp->args->argc);
296 
297 	*stack_base = (register_t *)base;
298 	return (0);
299 }
300 
301 /*
302  * Copy strings out to the new process address space, constructing new arg
303  * and env vector tables. Return a pointer to the base so that it can be used
304  * as the initial stack pointer.
305  */
306 static register_t *
307 linux_copyout_strings(struct image_params *imgp)
308 {
309 	int argc, envc;
310 	char **vectp;
311 	char *stringp, *destp;
312 	register_t *stack_base;
313 	struct ps_strings *arginfo;
314 	char canary[LINUX_AT_RANDOM_LEN];
315 	size_t execpath_len;
316 	struct proc *p;
317 
318 	/* Calculate string base and vector table pointers. */
319 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
320 		execpath_len = strlen(imgp->execpath) + 1;
321 	else
322 		execpath_len = 0;
323 
324 	p = imgp->proc;
325 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
326 	destp = (caddr_t)arginfo - SPARE_USRSPACE -
327 	    roundup(sizeof(canary), sizeof(char *)) -
328 	    roundup(execpath_len, sizeof(char *)) -
329 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
330 
331 	if (execpath_len != 0) {
332 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
333 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
334 	}
335 
336 	/* Prepare the canary for SSP. */
337 	arc4rand(canary, sizeof(canary), 0);
338 	imgp->canary = (uintptr_t)arginfo -
339 	    roundup(execpath_len, sizeof(char *)) -
340 	    roundup(sizeof(canary), sizeof(char *));
341 	copyout(canary, (void *)imgp->canary, sizeof(canary));
342 
343 	/* If we have a valid auxargs ptr, prepare some room on the stack. */
344 	if (imgp->auxargs) {
345 		/*
346 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
347 		 * lower compatibility.
348 		 */
349 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
350 		    (LINUX_AT_COUNT * 2);
351 
352 		/*
353 		 * The '+ 2' is for the null pointers at the end of each of
354 		 * the arg and env vector sets,and imgp->auxarg_size is room
355 		 * for argument of Runtime loader.
356 		 */
357 		vectp = (char **)(destp - (imgp->args->argc +
358 		    imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
359 
360 	} else {
361 		/*
362 		 * The '+ 2' is for the null pointers at the end of each of
363 		 * the arg and env vector sets
364 		 */
365 		vectp = (char **)(destp - (imgp->args->argc +
366 		    imgp->args->envc + 2) * sizeof(char *));
367 	}
368 
369 	/* vectp also becomes our initial stack base. */
370 	stack_base = (register_t *)vectp;
371 
372 	stringp = imgp->args->begin_argv;
373 	argc = imgp->args->argc;
374 	envc = imgp->args->envc;
375 
376 	/* Copy out strings - arguments and environment. */
377 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
378 
379 	/* Fill in "ps_strings" struct for ps, w, etc. */
380 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
381 	suword(&arginfo->ps_nargvstr, argc);
382 
383 	/* Fill in argument portion of vector table. */
384 	for (; argc > 0; --argc) {
385 		suword(vectp++, (long)(intptr_t)destp);
386 		while (*stringp++ != 0)
387 			destp++;
388 		destp++;
389 	}
390 
391 	/* A null vector table pointer separates the argp's from the envp's. */
392 	suword(vectp++, 0);
393 
394 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
395 	suword(&arginfo->ps_nenvstr, envc);
396 
397 	/* Fill in environment portion of vector table. */
398 	for (; envc > 0; --envc) {
399 		suword(vectp++, (long)(intptr_t)destp);
400 		while (*stringp++ != 0)
401 			destp++;
402 		destp++;
403 	}
404 
405 	/* The end of the vector table is a null pointer. */
406 	suword(vectp, 0);
407 	return (stack_base);
408 }
409 
410 /*
411  * Reset registers to default values on exec.
412  */
413 static void
414 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
415 {
416 	struct trapframe *regs = td->td_frame;
417 	struct pcb *pcb = td->td_pcb;
418 
419 	if (td->td_proc->p_md.md_ldt != NULL)
420 		user_ldt_free(td);
421 
422 	pcb->pcb_fsbase = 0;
423 	pcb->pcb_gsbase = 0;
424 	clear_pcb_flags(pcb, PCB_32BIT);
425 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
426 	set_pcb_flags(pcb, PCB_FULL_IRET);
427 
428 	bzero((char *)regs, sizeof(struct trapframe));
429 	regs->tf_rip = imgp->entry_addr;
430 	regs->tf_rsp = stack;
431 	regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
432 	regs->tf_ss = _udatasel;
433 	regs->tf_cs = _ucodesel;
434 	regs->tf_ds = _udatasel;
435 	regs->tf_es = _udatasel;
436 	regs->tf_fs = _ufssel;
437 	regs->tf_gs = _ugssel;
438 	regs->tf_flags = TF_HASSEGS;
439 
440 	/*
441 	 * Reset the hardware debug registers if they were in use.
442 	 * They won't have any meaning for the newly exec'd process.
443 	 */
444 	if (pcb->pcb_flags & PCB_DBREGS) {
445 		pcb->pcb_dr0 = 0;
446 		pcb->pcb_dr1 = 0;
447 		pcb->pcb_dr2 = 0;
448 		pcb->pcb_dr3 = 0;
449 		pcb->pcb_dr6 = 0;
450 		pcb->pcb_dr7 = 0;
451 		if (pcb == curpcb) {
452 			/*
453 			 * Clear the debug registers on the running
454 			 * CPU, otherwise they will end up affecting
455 			 * the next process we switch to.
456 			 */
457 			reset_dbregs();
458 		}
459 		clear_pcb_flags(pcb, PCB_DBREGS);
460 	}
461 
462 	/*
463 	 * Drop the FP state if we hold it, so that the process gets a
464 	 * clean FP state if it uses the FPU again.
465 	 */
466 	fpstate_drop(td);
467 }
468 
469 /*
470  * Copied from amd64/amd64/machdep.c
471  *
472  * XXX fpu state need? don't think so
473  */
474 int
475 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
476 {
477 	struct proc *p;
478 	struct l_ucontext uc;
479 	struct l_sigcontext *context;
480 	struct trapframe *regs;
481 	unsigned long rflags;
482 	int error;
483 	ksiginfo_t ksi;
484 
485 	regs = td->td_frame;
486 	error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
487 	if (error != 0)
488 		return (error);
489 
490 	p = td->td_proc;
491 	context = &uc.uc_mcontext;
492 	rflags = context->sc_rflags;
493 
494 	/*
495 	 * Don't allow users to change privileged or reserved flags.
496 	 */
497 	/*
498 	 * XXX do allow users to change the privileged flag PSL_RF.
499 	 * The cpu sets PSL_RF in tf_rflags for faults.  Debuggers
500 	 * should sometimes set it there too.  tf_rflags is kept in
501 	 * the signal context during signal handling and there is no
502 	 * other place to remember it, so the PSL_RF bit may be
503 	 * corrupted by the signal handler without us knowing.
504 	 * Corruption of the PSL_RF bit at worst causes one more or
505 	 * one less debugger trap, so allowing it is fairly harmless.
506 	 */
507 
508 #define RFLAG_SECURE(ef, oef)     ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
509 	if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
510 		printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
511 		return (EINVAL);
512 	}
513 
514 	/*
515 	 * Don't allow users to load a valid privileged %cs.  Let the
516 	 * hardware check for invalid selectors, excess privilege in
517 	 * other selectors, invalid %eip's and invalid %esp's.
518 	 */
519 #define CS_SECURE(cs)           (ISPL(cs) == SEL_UPL)
520 	if (!CS_SECURE(context->sc_cs)) {
521 		printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
522 		ksiginfo_init_trap(&ksi);
523 		ksi.ksi_signo = SIGBUS;
524 		ksi.ksi_code = BUS_OBJERR;
525 		ksi.ksi_trapno = T_PROTFLT;
526 		ksi.ksi_addr = (void *)regs->tf_rip;
527 		trapsignal(td, &ksi);
528 		return (EINVAL);
529 	}
530 
531 	PROC_LOCK(p);
532 	linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
533 	SIG_CANTMASK(td->td_sigmask);
534 	signotify(td);
535 	PROC_UNLOCK(p);
536 
537 	regs->tf_rdi    = context->sc_rdi;
538 	regs->tf_rsi    = context->sc_rsi;
539 	regs->tf_rdx    = context->sc_rdx;
540 	regs->tf_rbp    = context->sc_rbp;
541 	regs->tf_rbx    = context->sc_rbx;
542 	regs->tf_rcx    = context->sc_rcx;
543 	regs->tf_rax    = context->sc_rax;
544 	regs->tf_rip    = context->sc_rip;
545 	regs->tf_rsp    = context->sc_rsp;
546 	regs->tf_r8     = context->sc_r8;
547 	regs->tf_r9     = context->sc_r9;
548 	regs->tf_r10    = context->sc_r10;
549 	regs->tf_r11    = context->sc_r11;
550 	regs->tf_r12    = context->sc_r12;
551 	regs->tf_r13    = context->sc_r13;
552 	regs->tf_r14    = context->sc_r14;
553 	regs->tf_r15    = context->sc_r15;
554 	regs->tf_cs     = context->sc_cs;
555 	regs->tf_err    = context->sc_err;
556 	regs->tf_rflags = rflags;
557 
558 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
559 	return (EJUSTRETURN);
560 }
561 
562 /*
563  * copied from amd64/amd64/machdep.c
564  *
565  * Send an interrupt to process.
566  */
567 static void
568 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
569 {
570 	struct l_rt_sigframe sf, *sfp;
571 	struct proc *p;
572 	struct thread *td;
573 	struct sigacts *psp;
574 	caddr_t sp;
575 	struct trapframe *regs;
576 	int sig, code;
577 	int oonstack;
578 
579 	td = curthread;
580 	p = td->td_proc;
581 	PROC_LOCK_ASSERT(p, MA_OWNED);
582 	sig = ksi->ksi_signo;
583 	psp = p->p_sigacts;
584 	code = ksi->ksi_code;
585 	mtx_assert(&psp->ps_mtx, MA_OWNED);
586 	regs = td->td_frame;
587 	oonstack = sigonstack(regs->tf_rsp);
588 
589 	LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
590 	    catcher, sig, mask, code);
591 
592 	/* Allocate space for the signal handler context. */
593 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
594 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
595 		sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
596 		    sizeof(struct l_rt_sigframe);
597 	} else
598 		sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
599 	/* Align to 16 bytes. */
600 	sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
601 	mtx_unlock(&psp->ps_mtx);
602 
603 	/* Translate the signal. */
604 	sig = bsd_to_linux_signal(sig);
605 
606 	/* Save user context. */
607 	bzero(&sf, sizeof(sf));
608 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
609 	bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
610 
611 	sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
612 	sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
613 	sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
614 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
615 	PROC_UNLOCK(p);
616 
617 	sf.sf_sc.uc_mcontext.sc_rdi    = regs->tf_rdi;
618 	sf.sf_sc.uc_mcontext.sc_rsi    = regs->tf_rsi;
619 	sf.sf_sc.uc_mcontext.sc_rdx    = regs->tf_rdx;
620 	sf.sf_sc.uc_mcontext.sc_rbp    = regs->tf_rbp;
621 	sf.sf_sc.uc_mcontext.sc_rbx    = regs->tf_rbx;
622 	sf.sf_sc.uc_mcontext.sc_rcx    = regs->tf_rcx;
623 	sf.sf_sc.uc_mcontext.sc_rax    = regs->tf_rax;
624 	sf.sf_sc.uc_mcontext.sc_rip    = regs->tf_rip;
625 	sf.sf_sc.uc_mcontext.sc_rsp    = regs->tf_rsp;
626 	sf.sf_sc.uc_mcontext.sc_r8     = regs->tf_r8;
627 	sf.sf_sc.uc_mcontext.sc_r9     = regs->tf_r9;
628 	sf.sf_sc.uc_mcontext.sc_r10    = regs->tf_r10;
629 	sf.sf_sc.uc_mcontext.sc_r11    = regs->tf_r11;
630 	sf.sf_sc.uc_mcontext.sc_r12    = regs->tf_r12;
631 	sf.sf_sc.uc_mcontext.sc_r13    = regs->tf_r13;
632 	sf.sf_sc.uc_mcontext.sc_r14    = regs->tf_r14;
633 	sf.sf_sc.uc_mcontext.sc_r15    = regs->tf_r15;
634 	sf.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
635 	sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
636 	sf.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
637 	sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
638 	sf.sf_sc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
639 
640 	/* Build the argument list for the signal handler. */
641 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
642 	regs->tf_rax = 0;
643 	regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
644 	regs->tf_rdx = (register_t)&sfp->sf_sc;	/* arg 3 in %rdx */
645 
646 	sf.sf_handler = catcher;
647 	/* Fill in POSIX parts. */
648 	ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
649 
650 	/* Copy the sigframe out to the user's stack. */
651 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
652 #ifdef DEBUG
653 		printf("process %ld has trashed its stack\n", (long)p->p_pid);
654 #endif
655 		PROC_LOCK(p);
656 		sigexit(td, SIGILL);
657 	}
658 
659 	regs->tf_rsp = (long)sfp;
660 	regs->tf_rip = linux_rt_sigcode;
661 	regs->tf_rflags &= ~(PSL_T | PSL_D);
662 	regs->tf_cs = _ucodesel;
663 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
664 	PROC_LOCK(p);
665 	mtx_lock(&psp->ps_mtx);
666 }
667 
668 /*
669  * If a Linux binary is exec'ing something, try this image activator
670  * first.  We override standard shell script execution in order to
671  * be able to modify the interpreter path.  We only do this if a Linux
672  * binary is doing the exec, so we do not create an EXEC module for it.
673  */
674 static int
675 linux_exec_imgact_try(struct image_params *imgp)
676 {
677 	const char *head = (const char *)imgp->image_header;
678 	char *rpath;
679 	int error = -1;
680 
681 	/*
682 	 * The interpreter for shell scripts run from a Linux binary needs
683 	 * to be located in /compat/linux if possible in order to recursively
684 	 * maintain Linux path emulation.
685 	 */
686 	if (((const short *)head)[0] == SHELLMAGIC) {
687 		/*
688 		 * Run our normal shell image activator.  If it succeeds then
689 		 * attempt to use the alternate path for the interpreter.  If
690 		 * an alternate path is found, use our stringspace to store it.
691 		 */
692 		if ((error = exec_shell_imgact(imgp)) == 0) {
693 			linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
694 			    imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
695 			    AT_FDCWD);
696 			if (rpath != NULL)
697 				imgp->args->fname_buf =
698 				    imgp->interpreter_name = rpath;
699 		}
700 	}
701 	return (error);
702 }
703 
704 #define	LINUX_VSYSCALL_START		(-10UL << 20)
705 #define	LINUX_VSYSCALL_SZ		1024
706 
707 const unsigned long linux_vsyscall_vector[] = {
708 	LINUX_SYS_gettimeofday,
709 	LINUX_SYS_linux_time,
710 				/* getcpu not implemented */
711 };
712 
713 static int
714 linux_vsyscall(struct thread *td)
715 {
716 	struct trapframe *frame;
717 	uint64_t retqaddr;
718 	int code, traced;
719 	int error;
720 
721 	frame = td->td_frame;
722 
723 	/* Check %rip for vsyscall area. */
724 	if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
725 		return (EINVAL);
726 	if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
727 		return (EINVAL);
728 	code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
729 	if (code >= nitems(linux_vsyscall_vector))
730 		return (EINVAL);
731 
732 	/*
733 	 * vsyscall called as callq *(%rax), so we must
734 	 * use return address from %rsp and also fixup %rsp.
735 	 */
736 	error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
737 	if (error)
738 		return (error);
739 
740 	frame->tf_rip = retqaddr;
741 	frame->tf_rax = linux_vsyscall_vector[code];
742 	frame->tf_rsp += 8;
743 
744 	traced = (frame->tf_flags & PSL_T);
745 
746 	amd64_syscall(td, traced);
747 
748 	return (0);
749 }
750 
751 struct sysentvec elf_linux_sysvec = {
752 	.sv_size	= LINUX_SYS_MAXSYSCALL,
753 	.sv_table	= linux_sysent,
754 	.sv_mask	= 0,
755 	.sv_errsize	= ELAST + 1,
756 	.sv_errtbl	= linux_errtbl,
757 	.sv_transtrap	= linux_translate_traps,
758 	.sv_fixup	= linux_fixup_elf,
759 	.sv_sendsig	= linux_rt_sendsig,
760 	.sv_sigcode	= &_binary_linux_locore_o_start,
761 	.sv_szsigcode	= &linux_szsigcode,
762 	.sv_name	= "Linux ELF64",
763 	.sv_coredump	= elf64_coredump,
764 	.sv_imgact_try	= linux_exec_imgact_try,
765 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
766 	.sv_pagesize	= PAGE_SIZE,
767 	.sv_minuser	= VM_MIN_ADDRESS,
768 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
769 	.sv_usrstack	= USRSTACK,
770 	.sv_psstrings	= PS_STRINGS,
771 	.sv_stackprot	= VM_PROT_ALL,
772 	.sv_copyout_strings = linux_copyout_strings,
773 	.sv_setregs	= linux_exec_setregs,
774 	.sv_fixlimit	= NULL,
775 	.sv_maxssiz	= NULL,
776 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
777 	.sv_set_syscall_retval = linux_set_syscall_retval,
778 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
779 	.sv_syscallnames = NULL,
780 	.sv_shared_page_base = SHAREDPAGE,
781 	.sv_shared_page_len = PAGE_SIZE,
782 	.sv_schedtail	= linux_schedtail,
783 	.sv_thread_detach = linux_thread_detach,
784 	.sv_trap	= linux_vsyscall,
785 };
786 
787 static void
788 linux_vdso_install(void *param)
789 {
790 
791 	amd64_lower_shared_page(&elf_linux_sysvec);
792 
793 	linux_szsigcode = (&_binary_linux_locore_o_end -
794 	    &_binary_linux_locore_o_start);
795 
796 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
797 		panic("Linux invalid vdso size\n");
798 
799 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
800 
801 	linux_shared_page_obj = __elfN(linux_shared_page_init)
802 	    (&linux_shared_page_mapping);
803 
804 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
805 
806 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
807 	    linux_szsigcode);
808 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
809 
810 	linux_kplatform = linux_shared_page_mapping +
811 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
812 }
813 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
814     linux_vdso_install, NULL);
815 
816 static void
817 linux_vdso_deinstall(void *param)
818 {
819 
820 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
821 }
822 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
823     linux_vdso_deinstall, NULL);
824 
825 static char GNULINUX_ABI_VENDOR[] = "GNU";
826 static int GNULINUX_ABI_DESC = 0;
827 
828 static bool
829 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
830 {
831 	const Elf32_Word *desc;
832 	uintptr_t p;
833 
834 	p = (uintptr_t)(note + 1);
835 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
836 
837 	desc = (const Elf32_Word *)p;
838 	if (desc[0] != GNULINUX_ABI_DESC)
839 		return (false);
840 
841 	/*
842 	 * For Linux we encode osrel as follows (see linux_mib.c):
843 	 * VVVMMMIII (version, major, minor), see linux_mib.c.
844 	 */
845 	*osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
846 
847 	return (true);
848 }
849 
850 static Elf_Brandnote linux64_brandnote = {
851 	.hdr.n_namesz	= sizeof(GNULINUX_ABI_VENDOR),
852 	.hdr.n_descsz	= 16,
853 	.hdr.n_type	= 1,
854 	.vendor		= GNULINUX_ABI_VENDOR,
855 	.flags		= BN_TRANSLATE_OSREL,
856 	.trans_osrel	= linux_trans_osrel
857 };
858 
859 static Elf64_Brandinfo linux_glibc2brand = {
860 	.brand		= ELFOSABI_LINUX,
861 	.machine	= EM_X86_64,
862 	.compat_3_brand	= "Linux",
863 	.emul_path	= "/compat/linux",
864 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
865 	.sysvec		= &elf_linux_sysvec,
866 	.interp_newpath	= NULL,
867 	.brand_note	= &linux64_brandnote,
868 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
869 };
870 
871 static Elf64_Brandinfo linux_glibc2brandshort = {
872 	.brand		= ELFOSABI_LINUX,
873 	.machine	= EM_X86_64,
874 	.compat_3_brand	= "Linux",
875 	.emul_path	= "/compat/linux",
876 	.interp_path	= "/lib64/ld-linux.so.2",
877 	.sysvec		= &elf_linux_sysvec,
878 	.interp_newpath	= NULL,
879 	.brand_note	= &linux64_brandnote,
880 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
881 };
882 
883 static Elf64_Brandinfo linux_muslbrand = {
884 	.brand		= ELFOSABI_LINUX,
885 	.machine	= EM_X86_64,
886 	.compat_3_brand	= "Linux",
887 	.emul_path	= "/compat/linux",
888 	.interp_path	= "/lib/ld-musl-x86_64.so.1",
889 	.sysvec		= &elf_linux_sysvec,
890 	.interp_newpath	= NULL,
891 	.brand_note	= &linux64_brandnote,
892 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
893 };
894 
895 Elf64_Brandinfo *linux_brandlist[] = {
896 	&linux_glibc2brand,
897 	&linux_glibc2brandshort,
898 	&linux_muslbrand,
899 	NULL
900 };
901 
902 static int
903 linux64_elf_modevent(module_t mod, int type, void *data)
904 {
905 	Elf64_Brandinfo **brandinfo;
906 	int error;
907 	struct linux_ioctl_handler **lihp;
908 
909 	error = 0;
910 
911 	switch(type) {
912 	case MOD_LOAD:
913 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
914 		     ++brandinfo)
915 			if (elf64_insert_brand_entry(*brandinfo) < 0)
916 				error = EINVAL;
917 		if (error == 0) {
918 			SET_FOREACH(lihp, linux_ioctl_handler_set)
919 				linux_ioctl_register_handler(*lihp);
920 			LIST_INIT(&futex_list);
921 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
922 			stclohz = (stathz ? stathz : hz);
923 			if (bootverbose)
924 				printf("Linux x86-64 ELF exec handler installed\n");
925 		} else
926 			printf("cannot insert Linux x86-64 ELF brand handler\n");
927 		break;
928 	case MOD_UNLOAD:
929 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
930 		     ++brandinfo)
931 			if (elf64_brand_inuse(*brandinfo))
932 				error = EBUSY;
933 		if (error == 0) {
934 			for (brandinfo = &linux_brandlist[0];
935 			     *brandinfo != NULL; ++brandinfo)
936 				if (elf64_remove_brand_entry(*brandinfo) < 0)
937 					error = EINVAL;
938 		}
939 		if (error == 0) {
940 			SET_FOREACH(lihp, linux_ioctl_handler_set)
941 				linux_ioctl_unregister_handler(*lihp);
942 			mtx_destroy(&futex_mtx);
943 			if (bootverbose)
944 				printf("Linux ELF exec handler removed\n");
945 		} else
946 			printf("Could not deinstall ELF interpreter entry\n");
947 		break;
948 	default:
949 		return (EOPNOTSUPP);
950 	}
951 	return (error);
952 }
953 
954 static moduledata_t linux64_elf_mod = {
955 	"linux64elf",
956 	linux64_elf_modevent,
957 	0
958 };
959 
960 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
961 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
962 FEATURE(linux64, "Linux 64bit support");
963