xref: /freebsd/sys/i386/linux/linux_sysvec.c (revision a5c0d5517771d84514cdeff2c6a655d120bfd71e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * All rights reserved.
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 #define __ELF_WORD_SIZE	32
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/exec.h>
37 #include <sys/fcntl.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_aout.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kernel.h>
42 #include <sys/malloc.h>
43 #include <sys/module.h>
44 #include <sys/proc.h>
45 #include <sys/stddef.h>
46 #include <sys/signalvar.h>
47 #include <sys/syscallsubr.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
50 #include <sys/sysproto.h>
51 #include <sys/vnode.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_extern.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_param.h>
60 
61 #include <machine/cpu.h>
62 #include <machine/cputypes.h>
63 #include <machine/md_var.h>
64 #include <machine/pcb.h>
65 #include <machine/trap.h>
66 
67 #include <x86/linux/linux_x86.h>
68 #include <i386/linux/linux.h>
69 #include <i386/linux/linux_proto.h>
70 #include <compat/linux/linux_elf.h>
71 #include <compat/linux/linux_emul.h>
72 #include <compat/linux/linux_fork.h>
73 #include <compat/linux/linux_ioctl.h>
74 #include <compat/linux/linux_mib.h>
75 #include <compat/linux/linux_misc.h>
76 #include <compat/linux/linux_signal.h>
77 #include <compat/linux/linux_util.h>
78 #include <compat/linux/linux_vdso.h>
79 
80 #include <x86/linux/linux_x86_sigframe.h>
81 
82 MODULE_VERSION(linux, 1);
83 
84 #define	LINUX_VDSOPAGE_SIZE	PAGE_SIZE * 2
85 #define	LINUX_VDSOPAGE		(VM_MAXUSER_ADDRESS - LINUX_VDSOPAGE_SIZE)
86 #define	LINUX_SHAREDPAGE	(LINUX_VDSOPAGE - PAGE_SIZE)
87 				/*
88 				 * PAGE_SIZE - the size
89 				 * of the native SHAREDPAGE
90 				 */
91 #define	LINUX_USRSTACK		LINUX_SHAREDPAGE
92 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - sizeof(struct ps_strings))
93 
94 static int linux_szsigcode;
95 static vm_object_t linux_vdso_obj;
96 static char *linux_vdso_mapping;
97 extern char _binary_linux_vdso_so_o_start;
98 extern char _binary_linux_vdso_so_o_end;
99 static vm_offset_t linux_vdso_base;
100 
101 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
102 extern const char *linux_syscallnames[];
103 
104 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
105 
106 static int	linux_fixup(uintptr_t *stack_base,
107 		    struct image_params *iparams);
108 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
109 static void	linux_exec_setregs(struct thread *td,
110 		    struct image_params *imgp, uintptr_t stack);
111 static void	linux_exec_sysvec_init(void *param);
112 static int	linux_on_exec_vmspace(struct proc *p,
113 		    struct image_params *imgp);
114 static void	linux_set_fork_retval(struct thread *td);
115 static void	linux_vdso_install(const void *param);
116 static void	linux_vdso_deinstall(const void *param);
117 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
118 
119 LINUX_VDSO_SYM_CHAR(linux_platform);
120 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
121 LINUX_VDSO_SYM_INTPTR(linux_vdso_sigcode);
122 LINUX_VDSO_SYM_INTPTR(linux_vdso_rt_sigcode);
123 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
124 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
125 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
126 
127 static int
128 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
129 {
130 	register_t *base, *argv, *envp;
131 
132 	base = (register_t *)*stack_base;
133 	argv = base;
134 	envp = base + (imgp->args->argc + 1);
135 	base--;
136 	suword(base, (intptr_t)envp);
137 	base--;
138 	suword(base, (intptr_t)argv);
139 	base--;
140 	suword(base, imgp->args->argc);
141 	*stack_base = (uintptr_t)base;
142 	return (0);
143 }
144 
145 static int
146 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
147 {
148 	Elf32_Auxargs *args;
149 	Elf32_Auxinfo *argarray, *pos;
150 	int error, issetugid;
151 
152 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
153 	args = (Elf32_Auxargs *)imgp->auxargs;
154 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
155 	    M_WAITOK | M_ZERO);
156 
157 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
158 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, __kernel_vsyscall);
159 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
160 
161 	/*
162 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
163 	 * as it has appeared in the 2.4.0-rc7 first time.
164 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
165 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
166 	 * is not present.
167 	 * Also see linux_times() implementation.
168 	 */
169 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
170 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
171 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
172 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
173 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
174 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
175 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
176 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
177 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
178 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
179 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
180 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
181 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
182 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
183 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
184 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
185 	if (imgp->execpathp != 0)
186 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
187 	if (args->execfd != -1)
188 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
189 	AUXARGS_ENTRY(pos, AT_NULL, 0);
190 
191 	free(imgp->auxargs, M_TEMP);
192 	imgp->auxargs = NULL;
193 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
194 
195 	error = copyout(argarray, (void *)base,
196 	    sizeof(*argarray) * LINUX_AT_COUNT);
197 	free(argarray, M_TEMP);
198 	return (error);
199 }
200 
201 static void
202 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
203 {
204 	struct thread *td = curthread;
205 	struct proc *p = td->td_proc;
206 	struct sigacts *psp;
207 	struct trapframe *regs;
208 	struct l_rt_sigframe *fp, frame;
209 	int sig, code;
210 	int oonstack;
211 
212 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
213 	code = ksi->ksi_code;
214 	PROC_LOCK_ASSERT(p, MA_OWNED);
215 	psp = p->p_sigacts;
216 	mtx_assert(&psp->ps_mtx, MA_OWNED);
217 	regs = td->td_frame;
218 	oonstack = sigonstack(regs->tf_esp);
219 
220 	/* Allocate space for the signal handler context. */
221 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
222 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
223 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
224 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
225 	} else
226 		fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
227 	mtx_unlock(&psp->ps_mtx);
228 
229 	/* Build the argument list for the signal handler. */
230 	sig = bsd_to_linux_signal(sig);
231 
232 	bzero(&frame, sizeof(frame));
233 
234 	frame.sf_sig = sig;
235 	frame.sf_siginfo = PTROUT(&fp->sf_si);
236 	frame.sf_ucontext = PTROUT(&fp->sf_uc);
237 
238 	/* Fill in POSIX parts. */
239 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
240 
241 	/* Build the signal context to be used by sigreturn. */
242 	frame.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
243 	frame.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
244 	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
245 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
246 	PROC_UNLOCK(p);
247 
248 	bsd_to_linux_sigset(mask, &frame.sf_uc.uc_sigmask);
249 
250 	frame.sf_uc.uc_mcontext.sc_mask   = frame.sf_uc.uc_sigmask.__mask;
251 	frame.sf_uc.uc_mcontext.sc_gs     = rgs();
252 	frame.sf_uc.uc_mcontext.sc_fs     = regs->tf_fs;
253 	frame.sf_uc.uc_mcontext.sc_es     = regs->tf_es;
254 	frame.sf_uc.uc_mcontext.sc_ds     = regs->tf_ds;
255 	frame.sf_uc.uc_mcontext.sc_edi    = regs->tf_edi;
256 	frame.sf_uc.uc_mcontext.sc_esi    = regs->tf_esi;
257 	frame.sf_uc.uc_mcontext.sc_ebp    = regs->tf_ebp;
258 	frame.sf_uc.uc_mcontext.sc_ebx    = regs->tf_ebx;
259 	frame.sf_uc.uc_mcontext.sc_esp    = regs->tf_esp;
260 	frame.sf_uc.uc_mcontext.sc_edx    = regs->tf_edx;
261 	frame.sf_uc.uc_mcontext.sc_ecx    = regs->tf_ecx;
262 	frame.sf_uc.uc_mcontext.sc_eax    = regs->tf_eax;
263 	frame.sf_uc.uc_mcontext.sc_eip    = regs->tf_eip;
264 	frame.sf_uc.uc_mcontext.sc_cs     = regs->tf_cs;
265 	frame.sf_uc.uc_mcontext.sc_eflags = regs->tf_eflags;
266 	frame.sf_uc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
267 	frame.sf_uc.uc_mcontext.sc_ss     = regs->tf_ss;
268 	frame.sf_uc.uc_mcontext.sc_err    = regs->tf_err;
269 	frame.sf_uc.uc_mcontext.sc_cr2    = (register_t)ksi->ksi_addr;
270 	frame.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
271 
272 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
273 		/*
274 		 * Process has trashed its stack; give it an illegal
275 		 * instruction to halt it in its tracks.
276 		 */
277 		PROC_LOCK(p);
278 		sigexit(td, SIGILL);
279 	}
280 
281 	/* Build context to run handler in. */
282 	regs->tf_esp = PTROUT(fp);
283 	regs->tf_eip = linux_vdso_rt_sigcode;
284 	regs->tf_edi = PTROUT(catcher);
285 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
286 	regs->tf_cs = _ucodesel;
287 	regs->tf_ds = _udatasel;
288 	regs->tf_es = _udatasel;
289 	regs->tf_fs = _udatasel;
290 	regs->tf_ss = _udatasel;
291 	PROC_LOCK(p);
292 	mtx_lock(&psp->ps_mtx);
293 }
294 
295 /*
296  * Send an interrupt to process.
297  *
298  * Stack is set up to allow sigcode stored
299  * in u. to call routine, followed by kcall
300  * to sigreturn routine below.  After sigreturn
301  * resets the signal mask, the stack, and the
302  * frame pointer, it returns to the user
303  * specified pc, psl.
304  */
305 static void
306 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
307 {
308 	struct thread *td = curthread;
309 	struct proc *p = td->td_proc;
310 	struct sigacts *psp;
311 	struct trapframe *regs;
312 	struct l_sigframe *fp, frame;
313 	l_sigset_t lmask;
314 	int sig;
315 	int oonstack;
316 
317 	PROC_LOCK_ASSERT(p, MA_OWNED);
318 	psp = p->p_sigacts;
319 	sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
320 	mtx_assert(&psp->ps_mtx, MA_OWNED);
321 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
322 		/* Signal handler installed with SA_SIGINFO. */
323 		linux_rt_sendsig(catcher, ksi, mask);
324 		return;
325 	}
326 	regs = td->td_frame;
327 	oonstack = sigonstack(regs->tf_esp);
328 
329 	/* Allocate space for the signal handler context. */
330 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
331 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
332 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
333 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
334 	} else
335 		fp = (struct l_sigframe *)regs->tf_esp - 1;
336 	mtx_unlock(&psp->ps_mtx);
337 	PROC_UNLOCK(p);
338 
339 	/* Build the argument list for the signal handler. */
340 	sig = bsd_to_linux_signal(sig);
341 
342 	bzero(&frame, sizeof(frame));
343 
344 	frame.sf_sig = sig;
345 	frame.sf_sigmask = *mask;
346 	bsd_to_linux_sigset(mask, &lmask);
347 
348 	/* Build the signal context to be used by sigreturn. */
349 	frame.sf_sc.sc_mask   = lmask.__mask;
350 	frame.sf_sc.sc_gs     = rgs();
351 	frame.sf_sc.sc_fs     = regs->tf_fs;
352 	frame.sf_sc.sc_es     = regs->tf_es;
353 	frame.sf_sc.sc_ds     = regs->tf_ds;
354 	frame.sf_sc.sc_edi    = regs->tf_edi;
355 	frame.sf_sc.sc_esi    = regs->tf_esi;
356 	frame.sf_sc.sc_ebp    = regs->tf_ebp;
357 	frame.sf_sc.sc_ebx    = regs->tf_ebx;
358 	frame.sf_sc.sc_esp    = regs->tf_esp;
359 	frame.sf_sc.sc_edx    = regs->tf_edx;
360 	frame.sf_sc.sc_ecx    = regs->tf_ecx;
361 	frame.sf_sc.sc_eax    = regs->tf_eax;
362 	frame.sf_sc.sc_eip    = regs->tf_eip;
363 	frame.sf_sc.sc_cs     = regs->tf_cs;
364 	frame.sf_sc.sc_eflags = regs->tf_eflags;
365 	frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
366 	frame.sf_sc.sc_ss     = regs->tf_ss;
367 	frame.sf_sc.sc_err    = regs->tf_err;
368 	frame.sf_sc.sc_cr2    = (register_t)ksi->ksi_addr;
369 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
370 
371 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
372 		/*
373 		 * Process has trashed its stack; give it an illegal
374 		 * instruction to halt it in its tracks.
375 		 */
376 		PROC_LOCK(p);
377 		sigexit(td, SIGILL);
378 	}
379 
380 	/* Build context to run handler in. */
381 	regs->tf_esp = PTROUT(fp);
382 	regs->tf_eip = linux_vdso_sigcode;
383 	regs->tf_edi = PTROUT(catcher);
384 	regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
385 	regs->tf_cs = _ucodesel;
386 	regs->tf_ds = _udatasel;
387 	regs->tf_es = _udatasel;
388 	regs->tf_fs = _udatasel;
389 	regs->tf_ss = _udatasel;
390 	PROC_LOCK(p);
391 	mtx_lock(&psp->ps_mtx);
392 }
393 
394 /*
395  * System call to cleanup state after a signal
396  * has been taken.  Reset signal mask and
397  * stack state from context left by sendsig (above).
398  * Return to previous pc and psl as specified by
399  * context left by sendsig. Check carefully to
400  * make sure that the user has not modified the
401  * psl to gain improper privileges or to cause
402  * a machine fault.
403  */
404 int
405 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
406 {
407 	struct l_sigframe frame;
408 	struct trapframe *regs;
409 	int eflags;
410 	ksiginfo_t ksi;
411 
412 	regs = td->td_frame;
413 
414 	/*
415 	 * The trampoline code hands us the sigframe.
416 	 * It is unsafe to keep track of it ourselves, in the event that a
417 	 * program jumps out of a signal handler.
418 	 */
419 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
420 		return (EFAULT);
421 
422 	/* Check for security violations. */
423 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
424 	eflags = frame.sf_sc.sc_eflags;
425 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
426 		return (EINVAL);
427 
428 	/*
429 	 * Don't allow users to load a valid privileged %cs.  Let the
430 	 * hardware check for invalid selectors, excess privilege in
431 	 * other selectors, invalid %eip's and invalid %esp's.
432 	 */
433 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
434 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
435 		ksiginfo_init_trap(&ksi);
436 		ksi.ksi_signo = SIGBUS;
437 		ksi.ksi_code = BUS_OBJERR;
438 		ksi.ksi_trapno = T_PROTFLT;
439 		ksi.ksi_addr = (void *)regs->tf_eip;
440 		trapsignal(td, &ksi);
441 		return (EINVAL);
442 	}
443 
444 	kern_sigprocmask(td, SIG_SETMASK, &frame.sf_sigmask, NULL, 0);
445 
446 	/* Restore signal context. */
447 	/* %gs was restored by the trampoline. */
448 	regs->tf_fs     = frame.sf_sc.sc_fs;
449 	regs->tf_es     = frame.sf_sc.sc_es;
450 	regs->tf_ds     = frame.sf_sc.sc_ds;
451 	regs->tf_edi    = frame.sf_sc.sc_edi;
452 	regs->tf_esi    = frame.sf_sc.sc_esi;
453 	regs->tf_ebp    = frame.sf_sc.sc_ebp;
454 	regs->tf_ebx    = frame.sf_sc.sc_ebx;
455 	regs->tf_edx    = frame.sf_sc.sc_edx;
456 	regs->tf_ecx    = frame.sf_sc.sc_ecx;
457 	regs->tf_eax    = frame.sf_sc.sc_eax;
458 	regs->tf_eip    = frame.sf_sc.sc_eip;
459 	regs->tf_cs     = frame.sf_sc.sc_cs;
460 	regs->tf_eflags = eflags;
461 	regs->tf_esp    = frame.sf_sc.sc_esp_at_signal;
462 	regs->tf_ss     = frame.sf_sc.sc_ss;
463 
464 	return (EJUSTRETURN);
465 }
466 
467 /*
468  * System call to cleanup state after a signal
469  * has been taken.  Reset signal mask and
470  * stack state from context left by rt_sendsig (above).
471  * Return to previous pc and psl as specified by
472  * context left by sendsig. Check carefully to
473  * make sure that the user has not modified the
474  * psl to gain improper privileges or to cause
475  * a machine fault.
476  */
477 int
478 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
479 {
480 	struct l_ucontext uc;
481 	struct l_sigcontext *context;
482 	sigset_t bmask;
483 	l_stack_t *lss;
484 	stack_t ss;
485 	struct trapframe *regs;
486 	int eflags;
487 	ksiginfo_t ksi;
488 
489 	regs = td->td_frame;
490 
491 	/*
492 	 * The trampoline code hands us the ucontext.
493 	 * It is unsafe to keep track of it ourselves, in the event that a
494 	 * program jumps out of a signal handler.
495 	 */
496 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
497 		return (EFAULT);
498 
499 	context = &uc.uc_mcontext;
500 
501 	/* Check for security violations. */
502 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
503 	eflags = context->sc_eflags;
504 	if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
505 		return (EINVAL);
506 
507 	/*
508 	 * Don't allow users to load a valid privileged %cs.  Let the
509 	 * hardware check for invalid selectors, excess privilege in
510 	 * other selectors, invalid %eip's and invalid %esp's.
511 	 */
512 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
513 	if (!CS_SECURE(context->sc_cs)) {
514 		ksiginfo_init_trap(&ksi);
515 		ksi.ksi_signo = SIGBUS;
516 		ksi.ksi_code = BUS_OBJERR;
517 		ksi.ksi_trapno = T_PROTFLT;
518 		ksi.ksi_addr = (void *)regs->tf_eip;
519 		trapsignal(td, &ksi);
520 		return (EINVAL);
521 	}
522 
523 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
524 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
525 
526 	/* Restore signal context. */
527 	/* %gs was restored by the trampoline. */
528 	regs->tf_fs     = context->sc_fs;
529 	regs->tf_es     = context->sc_es;
530 	regs->tf_ds     = context->sc_ds;
531 	regs->tf_edi    = context->sc_edi;
532 	regs->tf_esi    = context->sc_esi;
533 	regs->tf_ebp    = context->sc_ebp;
534 	regs->tf_ebx    = context->sc_ebx;
535 	regs->tf_edx    = context->sc_edx;
536 	regs->tf_ecx    = context->sc_ecx;
537 	regs->tf_eax    = context->sc_eax;
538 	regs->tf_eip    = context->sc_eip;
539 	regs->tf_cs     = context->sc_cs;
540 	regs->tf_eflags = eflags;
541 	regs->tf_esp    = context->sc_esp_at_signal;
542 	regs->tf_ss     = context->sc_ss;
543 
544 	/* Call sigaltstack & ignore results. */
545 	lss = &uc.uc_stack;
546 	ss.ss_sp = PTRIN(lss->ss_sp);
547 	ss.ss_size = lss->ss_size;
548 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
549 
550 	(void)kern_sigaltstack(td, &ss, NULL);
551 
552 	return (EJUSTRETURN);
553 }
554 
555 static int
556 linux_fetch_syscall_args(struct thread *td)
557 {
558 	struct proc *p;
559 	struct trapframe *frame;
560 	struct syscall_args *sa;
561 
562 	p = td->td_proc;
563 	frame = td->td_frame;
564 	sa = &td->td_sa;
565 
566 	sa->code = frame->tf_eax;
567 	sa->original_code = sa->code;
568 	sa->args[0] = frame->tf_ebx;
569 	sa->args[1] = frame->tf_ecx;
570 	sa->args[2] = frame->tf_edx;
571 	sa->args[3] = frame->tf_esi;
572 	sa->args[4] = frame->tf_edi;
573 	sa->args[5] = frame->tf_ebp;	/* Unconfirmed */
574 
575 	if (sa->code >= p->p_sysent->sv_size)
576 		/* nosys */
577 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
578 	else
579 		sa->callp = &p->p_sysent->sv_table[sa->code];
580 
581 	td->td_retval[0] = 0;
582 	td->td_retval[1] = frame->tf_edx;
583 
584 	return (0);
585 }
586 
587 static void
588 linux_set_syscall_retval(struct thread *td, int error)
589 {
590 	struct trapframe *frame = td->td_frame;
591 
592 	cpu_set_syscall_retval(td, error);
593 
594 	if (__predict_false(error != 0)) {
595 		if (error != ERESTART && error != EJUSTRETURN)
596 			frame->tf_eax = bsd_to_linux_errno(error);
597 	}
598 }
599 
600 static void
601 linux_set_fork_retval(struct thread *td)
602 {
603 	struct trapframe *frame = td->td_frame;
604 
605 	frame->tf_eax = 0;
606 }
607 
608 /*
609  * exec_setregs may initialize some registers differently than Linux
610  * does, thus potentially confusing Linux binaries. If necessary, we
611  * override the exec_setregs default(s) here.
612  */
613 static void
614 linux_exec_setregs(struct thread *td, struct image_params *imgp,
615     uintptr_t stack)
616 {
617 	struct pcb *pcb = td->td_pcb;
618 
619 	exec_setregs(td, imgp, stack);
620 
621 	/* Linux sets %gs to 0, we default to _udatasel. */
622 	pcb->pcb_gs = 0;
623 	load_gs(0);
624 
625 	pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
626 }
627 
628 struct sysentvec linux_sysvec = {
629 	.sv_size	= LINUX_SYS_MAXSYSCALL,
630 	.sv_table	= linux_sysent,
631 	.sv_fixup	= linux_fixup,
632 	.sv_sendsig	= linux_sendsig,
633 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
634 	.sv_szsigcode	= &linux_szsigcode,
635 	.sv_name	= "Linux a.out",
636 	.sv_coredump	= NULL,
637 	.sv_imgact_try	= linux_exec_imgact_try,
638 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
639 	.sv_minuser	= VM_MIN_ADDRESS,
640 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
641 	.sv_usrstack	= LINUX_USRSTACK,
642 	.sv_psstrings	= PS_STRINGS,
643 	.sv_psstringssz	= sizeof(struct ps_strings),
644 	.sv_stackprot	= VM_PROT_ALL,
645 	.sv_copyout_strings = exec_copyout_strings,
646 	.sv_setregs	= linux_exec_setregs,
647 	.sv_fixlimit	= NULL,
648 	.sv_maxssiz	= NULL,
649 	.sv_flags	= SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32 |
650 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
651 	.sv_set_syscall_retval = linux_set_syscall_retval,
652 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
653 	.sv_syscallnames = linux_syscallnames,
654 	.sv_schedtail	= linux_schedtail,
655 	.sv_thread_detach = linux_thread_detach,
656 	.sv_trap	= NULL,
657 	.sv_onexec	= linux_on_exec_vmspace,
658 	.sv_onexit	= linux_on_exit,
659 	.sv_ontdexit	= linux_thread_dtor,
660 	.sv_setid_allowed = &linux_setid_allowed_query,
661 	.sv_set_fork_retval = linux_set_fork_retval,
662 };
663 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
664 
665 struct sysentvec elf_linux_sysvec = {
666 	.sv_size	= LINUX_SYS_MAXSYSCALL,
667 	.sv_table	= linux_sysent,
668 	.sv_fixup	= __elfN(freebsd_fixup),
669 	.sv_sendsig	= linux_sendsig,
670 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
671 	.sv_szsigcode	= &linux_szsigcode,
672 	.sv_name	= "Linux ELF32",
673 	.sv_coredump	= elf32_coredump,
674 	.sv_elf_core_osabi = ELFOSABI_NONE,
675 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
676 	.sv_elf_core_prepare_notes = __linuxN(prepare_notes),
677 	.sv_imgact_try	= linux_exec_imgact_try,
678 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
679 	.sv_minuser	= VM_MIN_ADDRESS,
680 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
681 	.sv_usrstack	= LINUX_USRSTACK,
682 	.sv_psstrings	= LINUX_PS_STRINGS,
683 	.sv_psstringssz	= sizeof(struct ps_strings),
684 	.sv_stackprot	= VM_PROT_ALL,
685 	.sv_copyout_auxargs = linux_copyout_auxargs,
686 	.sv_copyout_strings = __linuxN(copyout_strings),
687 	.sv_setregs	= linux_exec_setregs,
688 	.sv_fixlimit	= NULL,
689 	.sv_maxssiz	= NULL,
690 	.sv_flags	= SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP |
691 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
692 	.sv_set_syscall_retval = linux_set_syscall_retval,
693 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
694 	.sv_syscallnames = NULL,
695 	.sv_shared_page_base = LINUX_SHAREDPAGE,
696 	.sv_shared_page_len = PAGE_SIZE,
697 	.sv_schedtail	= linux_schedtail,
698 	.sv_thread_detach = linux_thread_detach,
699 	.sv_trap	= NULL,
700 	.sv_onexec	= linux_on_exec_vmspace,
701 	.sv_onexit	= linux_on_exit,
702 	.sv_ontdexit	= linux_thread_dtor,
703 	.sv_setid_allowed = &linux_setid_allowed_query,
704 	.sv_set_fork_retval = linux_set_fork_retval,
705 };
706 
707 static int
708 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
709 {
710 	int error = 0;
711 
712 	if (SV_PROC_FLAG(p, SV_SHP) != 0)
713 		error = linux_map_vdso(p, linux_vdso_obj,
714 		    linux_vdso_base, LINUX_VDSOPAGE_SIZE, imgp);
715 	if (error == 0)
716 		linux_on_exec(p, imgp);
717 	return (error);
718 }
719 
720 /*
721  * linux_vdso_install() and linux_exec_sysvec_init() must be called
722  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
723  */
724 static void
725 linux_exec_sysvec_init(void *param)
726 {
727 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
728 	struct sysentvec *sv;
729 	ptrdiff_t tkoff;
730 
731 	sv = param;
732 	/* Fill timekeep_base */
733 	exec_sysvec_init(sv);
734 
735 	tkoff = kern_timekeep_base - linux_vdso_base;
736 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
737 	*ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
738 
739 	tkoff = kern_tsc_selector - linux_vdso_base;
740 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
741 	*ktsc_selector = linux_vdso_tsc_selector_idx();
742 	if (bootverbose)
743 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
744 
745 	tkoff = kern_cpu_selector - linux_vdso_base;
746 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
747 	*ktsc_selector = linux_vdso_cpu_selector_idx();
748 	if (bootverbose)
749 		printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector);
750 }
751 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
752     linux_exec_sysvec_init, &elf_linux_sysvec);
753 
754 static void
755 linux_vdso_install(const void *param)
756 {
757 	char *vdso_start = &_binary_linux_vdso_so_o_start;
758 	char *vdso_end = &_binary_linux_vdso_so_o_end;
759 
760 	linux_szsigcode = vdso_end - vdso_start;
761 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
762 
763 	linux_vdso_base = LINUX_VDSOPAGE;
764 
765 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
766 
767 	linux_vdso_obj = __elfN(linux_shared_page_init)
768 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
769 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
770 
771 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
772 }
773 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
774     linux_vdso_install, NULL);
775 
776 static void
777 linux_vdso_deinstall(const void *param)
778 {
779 
780 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
781 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
782 }
783 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
784     linux_vdso_deinstall, NULL);
785 
786 static void
787 linux_vdso_reloc(char *mapping, Elf_Addr offset)
788 {
789 	const Elf_Shdr *shdr;
790 	const Elf_Rel *rel;
791 	const Elf_Ehdr *ehdr;
792 	Elf_Addr *where;
793 	Elf_Size rtype, symidx;
794 	Elf_Addr addr, addend;
795 	int i, relcnt;
796 
797 	MPASS(offset != 0);
798 
799 	relcnt = 0;
800 	ehdr = (const Elf_Ehdr *)mapping;
801 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
802 	for (i = 0; i < ehdr->e_shnum; i++)
803 	{
804 		switch (shdr[i].sh_type) {
805 		case SHT_REL:
806 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
807 			relcnt = shdr[i].sh_size / sizeof(*rel);
808 			break;
809 		case SHT_RELA:
810 			printf("Linux i386 vDSO: unexpected Rela section\n");
811 			break;
812 		}
813 	}
814 
815 	for (i = 0; i < relcnt; i++, rel++) {
816 		where = (Elf_Addr *)(mapping + rel->r_offset);
817 		addend = *where;
818 		rtype = ELF_R_TYPE(rel->r_info);
819 		symidx = ELF_R_SYM(rel->r_info);
820 
821 		switch (rtype) {
822 		case R_386_NONE:	/* none */
823 			break;
824 
825 		case R_386_RELATIVE:	/* B + A */
826 			addr = (Elf_Addr)PTROUT(offset + addend);
827 			if (*where != addr)
828 				*where = addr;
829 			break;
830 
831 		case R_386_IRELATIVE:
832 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
833 			    "symbol index %d\n", symidx);
834 			break;
835 		default:
836 			printf("Linux i386 vDSO: unexpected relocation type %d, "
837 			    "symbol index %d\n", rtype, symidx);
838 		}
839 	}
840 }
841 
842 static Elf_Brandnote linux_brandnote = {
843 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
844 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
845 	.hdr.n_type	= 1,
846 	.vendor		= GNU_ABI_VENDOR,
847 	.flags		= BN_TRANSLATE_OSREL,
848 	.trans_osrel	= linux_trans_osrel
849 };
850 
851 static Elf32_Brandinfo linux_brand = {
852 	.brand		= ELFOSABI_LINUX,
853 	.machine	= EM_386,
854 	.compat_3_brand	= "Linux",
855 	.emul_path	= linux_emul_path,
856 	.interp_path	= "/lib/ld-linux.so.1",
857 	.sysvec		= &elf_linux_sysvec,
858 	.interp_newpath	= NULL,
859 	.brand_note	= &linux_brandnote,
860 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
861 };
862 
863 static Elf32_Brandinfo linux_glibc2brand = {
864 	.brand		= ELFOSABI_LINUX,
865 	.machine	= EM_386,
866 	.compat_3_brand	= "Linux",
867 	.emul_path	= linux_emul_path,
868 	.interp_path	= "/lib/ld-linux.so.2",
869 	.sysvec		= &elf_linux_sysvec,
870 	.interp_newpath	= NULL,
871 	.brand_note	= &linux_brandnote,
872 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
873 };
874 
875 static Elf32_Brandinfo linux_muslbrand = {
876 	.brand		= ELFOSABI_LINUX,
877 	.machine	= EM_386,
878 	.compat_3_brand	= "Linux",
879 	.emul_path	= linux_emul_path,
880 	.interp_path	= "/lib/ld-musl-i386.so.1",
881 	.sysvec		= &elf_linux_sysvec,
882 	.interp_newpath	= NULL,
883 	.brand_note	= &linux_brandnote,
884 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
885 			    LINUX_BI_FUTEX_REQUEUE
886 };
887 
888 Elf32_Brandinfo *linux_brandlist[] = {
889 	&linux_brand,
890 	&linux_glibc2brand,
891 	&linux_muslbrand,
892 	NULL
893 };
894 
895 static int
896 linux_elf_modevent(module_t mod, int type, void *data)
897 {
898 	Elf32_Brandinfo **brandinfo;
899 	int error;
900 	struct linux_ioctl_handler **lihp;
901 
902 	error = 0;
903 
904 	switch(type) {
905 	case MOD_LOAD:
906 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
907 		     ++brandinfo)
908 			if (elf32_insert_brand_entry(*brandinfo) < 0)
909 				error = EINVAL;
910 		if (error == 0) {
911 			SET_FOREACH(lihp, linux_ioctl_handler_set)
912 				linux_ioctl_register_handler(*lihp);
913 			linux_dev_shm_create();
914 			linux_osd_jail_register();
915 			stclohz = (stathz ? stathz : hz);
916 			if (bootverbose)
917 				printf("Linux ELF exec handler installed\n");
918 		} else
919 			printf("cannot insert Linux ELF brand handler\n");
920 		break;
921 	case MOD_UNLOAD:
922 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
923 		     ++brandinfo)
924 			if (elf32_brand_inuse(*brandinfo))
925 				error = EBUSY;
926 		if (error == 0) {
927 			for (brandinfo = &linux_brandlist[0];
928 			     *brandinfo != NULL; ++brandinfo)
929 				if (elf32_remove_brand_entry(*brandinfo) < 0)
930 					error = EINVAL;
931 		}
932 		if (error == 0) {
933 			SET_FOREACH(lihp, linux_ioctl_handler_set)
934 				linux_ioctl_unregister_handler(*lihp);
935 			linux_dev_shm_destroy();
936 			linux_osd_jail_deregister();
937 			if (bootverbose)
938 				printf("Linux ELF exec handler removed\n");
939 		} else
940 			printf("Could not deinstall ELF interpreter entry\n");
941 		break;
942 	default:
943 		return (EOPNOTSUPP);
944 	}
945 	return (error);
946 }
947 
948 static moduledata_t linux_elf_mod = {
949 	"linuxelf",
950 	linux_elf_modevent,
951 	0
952 };
953 
954 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
955 FEATURE(linux, "Linux 32bit support");
956