xref: /freebsd/sys/arm64/linux/linux_sysvec.c (revision 86aa9539fef591a363b06a0ebd3aa7a07f4c1579) 
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * Copyright (c) 2018 Turing Robotic Industries Inc.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/cdefs.h>
35 #include <sys/elf.h>
36 #include <sys/exec.h>
37 #include <sys/imgact.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/module.h>
42 #include <sys/mutex.h>
43 #include <sys/proc.h>
44 #include <sys/signalvar.h>
45 #include <sys/sysctl.h>
46 #include <sys/sysent.h>
47 
48 #include <vm/vm_param.h>
49 
50 #include <arm64/linux/linux.h>
51 #include <arm64/linux/linux_proto.h>
52 #include <compat/linux/linux_dtrace.h>
53 #include <compat/linux/linux_emul.h>
54 #include <compat/linux/linux_futex.h>
55 #include <compat/linux/linux_ioctl.h>
56 #include <compat/linux/linux_mib.h>
57 #include <compat/linux/linux_misc.h>
58 #include <compat/linux/linux_vdso.h>
59 
60 MODULE_VERSION(linux64elf, 1);
61 
62 const char *linux_kplatform;
63 static int linux_szsigcode;
64 static vm_object_t linux_shared_page_obj;
65 static char *linux_shared_page_mapping;
66 extern char _binary_linux_locore_o_start;
67 extern char _binary_linux_locore_o_end;
68 
69 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
70 
71 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
72 
73 static register_t *linux_copyout_strings(struct image_params *imgp);
74 static int	linux_elf_fixup(register_t **stack_base,
75 		    struct image_params *iparams);
76 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
77 static void	linux_vdso_install(const void *param);
78 static void	linux_vdso_deinstall(const void *param);
79 static void	linux_set_syscall_retval(struct thread *td, int error);
80 static int	linux_fetch_syscall_args(struct thread *td);
81 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
82 		    u_long stack);
83 static int	linux_vsyscall(struct thread *td);
84 
85 /* DTrace init */
86 LIN_SDT_PROVIDER_DECLARE(LINUX_DTRACE);
87 
88 /* DTrace probes */
89 LIN_SDT_PROBE_DEFINE2(sysvec, linux_translate_traps, todo, "int", "int");
90 LIN_SDT_PROBE_DEFINE0(sysvec, linux_exec_setregs, todo);
91 LIN_SDT_PROBE_DEFINE0(sysvec, linux_elf_fixup, todo);
92 LIN_SDT_PROBE_DEFINE0(sysvec, linux_rt_sigreturn, todo);
93 LIN_SDT_PROBE_DEFINE0(sysvec, linux_rt_sendsig, todo);
94 LIN_SDT_PROBE_DEFINE0(sysvec, linux_vsyscall, todo);
95 LIN_SDT_PROBE_DEFINE0(sysvec, linux_vdso_install, todo);
96 LIN_SDT_PROBE_DEFINE0(sysvec, linux_vdso_deinstall, todo);
97 
98 /* LINUXTODO: do we have traps to translate? */
99 static int
100 linux_translate_traps(int signal, int trap_code)
101 {
102 
103 	LIN_SDT_PROBE2(sysvec, linux_translate_traps, todo, signal, trap_code);
104 	return (signal);
105 }
106 
107 LINUX_VDSO_SYM_CHAR(linux_platform);
108 
109 static int
110 linux_fetch_syscall_args(struct thread *td)
111 {
112 	struct proc *p;
113 	struct syscall_args *sa;
114 	register_t *ap;
115 
116 	p = td->td_proc;
117 	ap = td->td_frame->tf_x;
118 	sa = &td->td_sa;
119 
120 	sa->code = td->td_frame->tf_x[8];
121 	/* LINUXTODO: generic syscall? */
122 	if (sa->code >= p->p_sysent->sv_size)
123 		sa->callp = &p->p_sysent->sv_table[0];
124 	else
125 		sa->callp = &p->p_sysent->sv_table[sa->code];
126 
127 	sa->narg = sa->callp->sy_narg;
128 	if (sa->narg > 8)
129 		panic("ARM64TODO: Could we have more than 8 args?");
130 	memcpy(sa->args, ap, 8 * sizeof(register_t));
131 
132 	td->td_retval[0] = 0;
133 	return (0);
134 }
135 
136 static void
137 linux_set_syscall_retval(struct thread *td, int error)
138 {
139 
140 	td->td_retval[1] = td->td_frame->tf_x[1];
141 	cpu_set_syscall_retval(td, error);
142 }
143 
144 static int
145 linux_elf_fixup(register_t **stack_base, struct image_params *imgp)
146 {
147 	Elf_Auxargs *args;
148 	Elf_Auxinfo *argarray, *pos;
149 	Elf_Addr *auxbase, *base;
150 	struct ps_strings *arginfo;
151 	struct proc *p;
152 	int error, issetugid;
153 
154 	LIN_SDT_PROBE0(sysvec, linux_elf_fixup, todo);
155 	p = imgp->proc;
156 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
157 
158 	KASSERT(curthread->td_proc == imgp->proc,
159 	    ("unsafe linux_elf_fixup(), should be curproc"));
160 	base = (Elf64_Addr *)*stack_base;
161 	args = (Elf64_Auxargs *)imgp->auxargs;
162 	/* Auxargs after argc, and NULL-terminated argv and envv lists. */
163 	auxbase = base + 1 + imgp->args->argc + 1 + imgp->args->envc + 1;
164 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
165 	    M_WAITOK | M_ZERO);
166 
167 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
168 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
169 	    imgp->proc->p_sysent->sv_shared_page_base);
170 #if 0	/* LINUXTODO: implement arm64 LINUX_AT_HWCAP */
171 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
172 #endif
173 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
174 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
175 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
176 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
177 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
178 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
179 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
180 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
181 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
182 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
183 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
184 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
185 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
186 #if 0	/* LINUXTODO: implement arm64 LINUX_AT_PLATFORM */
187 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
188 #endif
189 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
190 	if (imgp->execpathp != 0)
191 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
192 	if (args->execfd != -1)
193 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
194 	AUXARGS_ENTRY(pos, AT_NULL, 0);
195 	free(imgp->auxargs, M_TEMP);
196 	imgp->auxargs = NULL;
197 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
198 
199 	error = copyout(argarray, auxbase, sizeof(*argarray) * LINUX_AT_COUNT);
200 	free(argarray, M_TEMP);
201 	if (error != 0)
202 		return (error);
203 
204 	return (0);
205 }
206 
207 /*
208  * Copy strings out to the new process address space, constructing new arg
209  * and env vector tables. Return a pointer to the base so that it can be used
210  * as the initial stack pointer.
211  * LINUXTODO: deduplicate against other linuxulator archs
212  */
213 static register_t *
214 linux_copyout_strings(struct image_params *imgp)
215 {
216 	char **vectp;
217 	char *stringp, *destp;
218 	register_t *stack_base;
219 	struct ps_strings *arginfo;
220 	char canary[LINUX_AT_RANDOM_LEN];
221 	size_t execpath_len;
222 	struct proc *p;
223 	int argc, envc;
224 
225 	/* Calculate string base and vector table pointers. */
226 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
227 		execpath_len = strlen(imgp->execpath) + 1;
228 	else
229 		execpath_len = 0;
230 
231 	p = imgp->proc;
232 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
233 	destp = (caddr_t)arginfo - SPARE_USRSPACE -
234 	    roundup(sizeof(canary), sizeof(char *)) -
235 	    roundup(execpath_len, sizeof(char *)) -
236 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
237 
238 	if (execpath_len != 0) {
239 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
240 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
241 	}
242 
243 	/* Prepare the canary for SSP. */
244 	arc4rand(canary, sizeof(canary), 0);
245 	imgp->canary = (uintptr_t)arginfo -
246 	    roundup(execpath_len, sizeof(char *)) -
247 	    roundup(sizeof(canary), sizeof(char *));
248 	copyout(canary, (void *)imgp->canary, sizeof(canary));
249 
250 	vectp = (char **)destp;
251 	if (imgp->auxargs) {
252 		/*
253 		 * Allocate room on the stack for the ELF auxargs
254 		 * array.  It has up to LINUX_AT_COUNT entries.
255 		 */
256 		vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf64_Auxinfo),
257 		    sizeof(*vectp));
258 	}
259 
260 	/*
261 	 * Allocate room for argc and the argv[] and env vectors including the
262 	 * terminating NULL pointers.
263 	 */
264 	vectp -= 1 + imgp->args->argc + 1 + imgp->args->envc + 1;
265 	vectp = (char **)STACKALIGN(vectp);
266 
267 	/* vectp also becomes our initial stack base. */
268 	stack_base = (register_t *)vectp;
269 
270 	stringp = imgp->args->begin_argv;
271 	argc = imgp->args->argc;
272 	envc = imgp->args->envc;
273 
274 	/* Copy out strings - arguments and environment. */
275 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
276 
277 	/* Fill in "ps_strings" struct for ps, w, etc. */
278 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
279 	suword(&arginfo->ps_nargvstr, argc);
280 
281 	suword(vectp++, argc);
282 	/* Fill in argument portion of vector table. */
283 	for (; argc > 0; --argc) {
284 		suword(vectp++, (long)(intptr_t)destp);
285 		while (*stringp++ != 0)
286 			destp++;
287 		destp++;
288 	}
289 
290 	/* A null vector table pointer separates the argp's from the envp's. */
291 	suword(vectp++, 0);
292 
293 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
294 	suword(&arginfo->ps_nenvstr, envc);
295 
296 	/* Fill in environment portion of vector table. */
297 	for (; envc > 0; --envc) {
298 		suword(vectp++, (long)(intptr_t)destp);
299 		while (*stringp++ != 0)
300 			destp++;
301 		destp++;
302 	}
303 
304 	/* The end of the vector table is a null pointer. */
305 	suword(vectp, 0);
306 	return (stack_base);
307 }
308 
309 /*
310  * Reset registers to default values on exec.
311  */
312 static void
313 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
314 {
315 	struct trapframe *regs = td->td_frame;
316 
317 	/* LINUXTODO: validate */
318 	LIN_SDT_PROBE0(sysvec, linux_exec_setregs, todo);
319 
320 	memset(regs, 0, sizeof(*regs));
321 	/* glibc start.S registers function pointer in x0 with atexit. */
322         regs->tf_sp = stack;
323 #if 0	/* LINUXTODO: See if this is used. */
324 	regs->tf_lr = imgp->entry_addr;
325 #else
326         regs->tf_lr = 0xffffffffffffffff;
327 #endif
328         regs->tf_elr = imgp->entry_addr;
329 }
330 
331 int
332 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
333 {
334 
335 	/* LINUXTODO: implement */
336 	LIN_SDT_PROBE0(sysvec, linux_rt_sigreturn, todo);
337 	return (EDOOFUS);
338 }
339 
340 static void
341 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
342 {
343 
344 	/* LINUXTODO: implement */
345 	LIN_SDT_PROBE0(sysvec, linux_rt_sendsig, todo);
346 }
347 
348 static int
349 linux_vsyscall(struct thread *td)
350 {
351 
352 	/* LINUXTODO: implement */
353 	LIN_SDT_PROBE0(sysvec, linux_vsyscall, todo);
354 	return (EDOOFUS);
355 }
356 
357 struct sysentvec elf_linux_sysvec = {
358 	.sv_size	= LINUX_SYS_MAXSYSCALL,
359 	.sv_table	= linux_sysent,
360 	.sv_errsize	= ELAST + 1,
361 	.sv_errtbl	= linux_errtbl,
362 	.sv_transtrap	= linux_translate_traps,
363 	.sv_fixup	= linux_elf_fixup,
364 	.sv_sendsig	= linux_rt_sendsig,
365 	.sv_sigcode	= &_binary_linux_locore_o_start,
366 	.sv_szsigcode	= &linux_szsigcode,
367 	.sv_name	= "Linux ELF64",
368 	.sv_coredump	= elf64_coredump,
369 	.sv_imgact_try	= linux_exec_imgact_try,
370 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
371 	.sv_minuser	= VM_MIN_ADDRESS,
372 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
373 	.sv_usrstack	= USRSTACK,
374 	.sv_psstrings	= PS_STRINGS, /* XXX */
375 	.sv_stackprot	= VM_PROT_ALL, /* XXX */
376 	.sv_copyout_strings = linux_copyout_strings,
377 	.sv_setregs	= linux_exec_setregs,
378 	.sv_fixlimit	= NULL,
379 	.sv_maxssiz	= NULL,
380 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP,
381 	.sv_set_syscall_retval = linux_set_syscall_retval,
382 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
383 	.sv_syscallnames = NULL,
384 	.sv_shared_page_base = SHAREDPAGE,
385 	.sv_shared_page_len = PAGE_SIZE,
386 	.sv_schedtail	= linux_schedtail,
387 	.sv_thread_detach = linux_thread_detach,
388 	.sv_trap	= linux_vsyscall,
389 };
390 
391 static void
392 linux_vdso_install(const void *param)
393 {
394 
395 	linux_szsigcode = (&_binary_linux_locore_o_end -
396 	    &_binary_linux_locore_o_start);
397 
398 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
399 		panic("invalid Linux VDSO size\n");
400 
401 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
402 
403 	linux_shared_page_obj = __elfN(linux_shared_page_init)
404 	    (&linux_shared_page_mapping);
405 
406 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
407 
408 	memcpy(linux_shared_page_mapping, elf_linux_sysvec.sv_sigcode,
409 	    linux_szsigcode);
410 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
411 
412 	printf("LINUXTODO: %s: fix linux_kplatform\n", __func__);
413 #if 0
414 	linux_kplatform = linux_shared_page_mapping +
415 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
416 #else
417 	linux_kplatform = "arm64";
418 #endif
419 }
420 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
421     linux_vdso_install, NULL);
422 
423 static void
424 linux_vdso_deinstall(const void *param)
425 {
426 
427 	LIN_SDT_PROBE0(sysvec, linux_vdso_deinstall, todo);
428 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
429 }
430 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
431     linux_vdso_deinstall, NULL);
432 
433 static char GNU_ABI_VENDOR[] = "GNU";
434 static int GNU_ABI_LINUX = 0;
435 
436 /* LINUXTODO: deduplicate */
437 static bool
438 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
439 {
440 	const Elf32_Word *desc;
441 	uintptr_t p;
442 
443 	p = (uintptr_t)(note + 1);
444 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
445 
446 	desc = (const Elf32_Word *)p;
447 	if (desc[0] != GNU_ABI_LINUX)
448 		return (false);
449 
450 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
451 	return (true);
452 }
453 
454 static Elf_Brandnote linux64_brandnote = {
455 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
456 	.hdr.n_descsz	= 16,
457 	.hdr.n_type	= 1,
458 	.vendor		= GNU_ABI_VENDOR,
459 	.flags		= BN_TRANSLATE_OSREL,
460 	.trans_osrel	= linux_trans_osrel
461 };
462 
463 static Elf64_Brandinfo linux_glibc2brand = {
464 	.brand		= ELFOSABI_LINUX,
465 	.machine	= EM_AARCH64,
466 	.compat_3_brand	= "Linux",
467 	.emul_path	= "/compat/linux",
468 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
469 	.sysvec		= &elf_linux_sysvec,
470 	.interp_newpath	= NULL,
471 	.brand_note	= &linux64_brandnote,
472 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
473 };
474 
475 Elf64_Brandinfo *linux_brandlist[] = {
476 	&linux_glibc2brand,
477 	NULL
478 };
479 
480 static int
481 linux64_elf_modevent(module_t mod, int type, void *data)
482 {
483 	Elf64_Brandinfo **brandinfo;
484 	struct linux_ioctl_handler**lihp;
485 	int error;
486 
487 	error = 0;
488 	switch(type) {
489 	case MOD_LOAD:
490 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
491 		    ++brandinfo)
492 			if (elf64_insert_brand_entry(*brandinfo) < 0)
493 				error = EINVAL;
494 		if (error == 0) {
495 			SET_FOREACH(lihp, linux_ioctl_handler_set)
496 				linux_ioctl_register_handler(*lihp);
497 			LIST_INIT(&futex_list);
498 			mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
499 			stclohz = (stathz ? stathz : hz);
500 			if (bootverbose)
501 				printf("Linux arm64 ELF exec handler installed\n");
502 		}
503 		break;
504 	case MOD_UNLOAD:
505 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
506 		    ++brandinfo)
507 			if (elf64_brand_inuse(*brandinfo))
508 				error = EBUSY;
509 		if (error == 0) {
510 			for (brandinfo = &linux_brandlist[0];
511 			    *brandinfo != NULL; ++brandinfo)
512 				if (elf64_remove_brand_entry(*brandinfo) < 0)
513 					error = EINVAL;
514 		}
515 		if (error == 0) {
516 			SET_FOREACH(lihp, linux_ioctl_handler_set)
517 				linux_ioctl_unregister_handler(*lihp);
518 			mtx_destroy(&futex_mtx);
519 			if (bootverbose)
520 				printf("Linux ELF exec handler removed\n");
521 		} else
522 			printf("Could not deinstall ELF interpreter entry\n");
523 		break;
524 	default:
525 		return (EOPNOTSUPP);
526 	}
527 	return (error);
528 }
529 
530 static moduledata_t linux64_elf_mod = {
531 	"linux64elf",
532 	linux64_elf_modevent,
533 	0
534 };
535 
536 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
537 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
538 FEATURE(linux64, "AArch64 Linux 64bit support");
539