xref: /freebsd/sys/amd64/linux32/linux32_sysvec.c (revision 5caa67fa840e33df3a8a8edb16392116aa028ba1)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2004 Tim J. Robbins
5  * Copyright (c) 2003 Peter Wemm
6  * Copyright (c) 2002 Doug Rabson
7  * Copyright (c) 1998-1999 Andrew Gallatin
8  * Copyright (c) 1994-1996 Søren Schmidt
9  * All rights reserved.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer
16  *    in this position and unchanged.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. The name of the author may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "opt_compat.h"
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #ifndef COMPAT_FREEBSD32
41 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!"
42 #endif
43 
44 #define	__ELF_WORD_SIZE	32
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/exec.h>
49 #include <sys/fcntl.h>
50 #include <sys/imgact.h>
51 #include <sys/imgact_elf.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/module.h>
56 #include <sys/mutex.h>
57 #include <sys/proc.h>
58 #include <sys/resourcevar.h>
59 #include <sys/signalvar.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysent.h>
63 #include <sys/sysproto.h>
64 #include <sys/vnode.h>
65 #include <sys/eventhandler.h>
66 
67 #include <vm/vm.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_param.h>
74 
75 #include <machine/cpu.h>
76 #include <machine/md_var.h>
77 #include <machine/pcb.h>
78 #include <machine/specialreg.h>
79 #include <machine/trap.h>
80 
81 #include <amd64/linux32/linux.h>
82 #include <amd64/linux32/linux32_proto.h>
83 #include <compat/linux/linux_emul.h>
84 #include <compat/linux/linux_ioctl.h>
85 #include <compat/linux/linux_mib.h>
86 #include <compat/linux/linux_misc.h>
87 #include <compat/linux/linux_signal.h>
88 #include <compat/linux/linux_util.h>
89 #include <compat/linux/linux_vdso.h>
90 
91 MODULE_VERSION(linux, 1);
92 
93 const char *linux_kplatform;
94 static int linux_szsigcode;
95 static vm_object_t linux_shared_page_obj;
96 static char *linux_shared_page_mapping;
97 extern char _binary_linux32_locore_o_start;
98 extern char _binary_linux32_locore_o_end;
99 
100 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
101 
102 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
103 
104 static int	linux_fixup_elf(register_t **stack_base,
105 		    struct image_params *iparams);
106 static register_t *linux_copyout_strings(struct image_params *imgp);
107 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
108 static void	linux_exec_setregs(struct thread *td,
109 				   struct image_params *imgp, u_long stack);
110 static void	linux32_fixlimit(struct rlimit *rl, int which);
111 static bool	linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
112 static void	linux_vdso_install(void *param);
113 static void	linux_vdso_deinstall(void *param);
114 
115 #define LINUX_T_UNKNOWN  255
116 static int _bsd_to_linux_trapcode[] = {
117 	LINUX_T_UNKNOWN,	/* 0 */
118 	6,			/* 1  T_PRIVINFLT */
119 	LINUX_T_UNKNOWN,	/* 2 */
120 	3,			/* 3  T_BPTFLT */
121 	LINUX_T_UNKNOWN,	/* 4 */
122 	LINUX_T_UNKNOWN,	/* 5 */
123 	16,			/* 6  T_ARITHTRAP */
124 	254,			/* 7  T_ASTFLT */
125 	LINUX_T_UNKNOWN,	/* 8 */
126 	13,			/* 9  T_PROTFLT */
127 	1,			/* 10 T_TRCTRAP */
128 	LINUX_T_UNKNOWN,	/* 11 */
129 	14,			/* 12 T_PAGEFLT */
130 	LINUX_T_UNKNOWN,	/* 13 */
131 	17,			/* 14 T_ALIGNFLT */
132 	LINUX_T_UNKNOWN,	/* 15 */
133 	LINUX_T_UNKNOWN,	/* 16 */
134 	LINUX_T_UNKNOWN,	/* 17 */
135 	0,			/* 18 T_DIVIDE */
136 	2,			/* 19 T_NMI */
137 	4,			/* 20 T_OFLOW */
138 	5,			/* 21 T_BOUND */
139 	7,			/* 22 T_DNA */
140 	8,			/* 23 T_DOUBLEFLT */
141 	9,			/* 24 T_FPOPFLT */
142 	10,			/* 25 T_TSSFLT */
143 	11,			/* 26 T_SEGNPFLT */
144 	12,			/* 27 T_STKFLT */
145 	18,			/* 28 T_MCHK */
146 	19,			/* 29 T_XMMFLT */
147 	15			/* 30 T_RESERVED */
148 };
149 #define bsd_to_linux_trapcode(code) \
150     ((code)<nitems(_bsd_to_linux_trapcode)? \
151      _bsd_to_linux_trapcode[(code)]: \
152      LINUX_T_UNKNOWN)
153 
154 struct linux32_ps_strings {
155 	u_int32_t ps_argvstr;	/* first of 0 or more argument strings */
156 	u_int ps_nargvstr;	/* the number of argument strings */
157 	u_int32_t ps_envstr;	/* first of 0 or more environment strings */
158 	u_int ps_nenvstr;	/* the number of environment strings */
159 };
160 
161 LINUX_VDSO_SYM_INTPTR(linux32_sigcode);
162 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode);
163 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall);
164 LINUX_VDSO_SYM_CHAR(linux_platform);
165 
166 /*
167  * If FreeBSD & Linux have a difference of opinion about what a trap
168  * means, deal with it here.
169  *
170  * MPSAFE
171  */
172 static int
173 linux_translate_traps(int signal, int trap_code)
174 {
175 	if (signal != SIGBUS)
176 		return (signal);
177 	switch (trap_code) {
178 	case T_PROTFLT:
179 	case T_TSSFLT:
180 	case T_DOUBLEFLT:
181 	case T_PAGEFLT:
182 		return (SIGSEGV);
183 	default:
184 		return (signal);
185 	}
186 }
187 
188 static void
189 linux_copyout_auxargs(struct image_params *imgp, u_long *base)
190 {
191 	Elf32_Auxargs *args;
192 	Elf32_Auxinfo *argarray, *pos;
193 	u_long auxlen;
194 	int issetugid;
195 
196 	args = (Elf32_Auxargs *)imgp->auxargs;
197 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
198 	    M_WAITOK | M_ZERO);
199 
200 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
201 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
202 	    imgp->proc->p_sysent->sv_shared_page_base);
203 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux32_vsyscall);
204 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
205 
206 	/*
207 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
208 	 * as it has appeared in the 2.4.0-rc7 first time.
209 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
210 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
211 	 * is not present.
212 	 * Also see linux_times() implementation.
213 	 */
214 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
215 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
216 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
217 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
218 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
219 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
220 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
221 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
222 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
223 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
224 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
225 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
226 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
227 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
228 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
229 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
230 	if (imgp->execpathp != 0)
231 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
232 	if (args->execfd != -1)
233 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
234 	AUXARGS_ENTRY(pos, AT_NULL, 0);
235 
236 	free(imgp->auxargs, M_TEMP);
237 	imgp->auxargs = NULL;
238 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
239 
240 	auxlen = sizeof(*argarray) * (pos - argarray);
241 	*base -= auxlen;
242 	copyout(argarray, (void *)*base, auxlen);
243 	free(argarray, M_TEMP);
244 }
245 
246 static int
247 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
248 {
249 	Elf32_Addr *base;
250 
251 	base = (Elf32_Addr *)*stack_base;
252 	base--;
253 	if (suword32(base, (uint32_t)imgp->args->argc) == -1)
254 		return (EFAULT);
255 	*stack_base = (register_t *)base;
256 	return (0);
257 }
258 
259 static void
260 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
261 {
262 	struct thread *td = curthread;
263 	struct proc *p = td->td_proc;
264 	struct sigacts *psp;
265 	struct trapframe *regs;
266 	struct l_rt_sigframe *fp, frame;
267 	int oonstack;
268 	int sig;
269 	int code;
270 
271 	sig = ksi->ksi_signo;
272 	code = ksi->ksi_code;
273 	PROC_LOCK_ASSERT(p, MA_OWNED);
274 	psp = p->p_sigacts;
275 	mtx_assert(&psp->ps_mtx, MA_OWNED);
276 	regs = td->td_frame;
277 	oonstack = sigonstack(regs->tf_rsp);
278 
279 	/* Allocate space for the signal handler context. */
280 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
281 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
282 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
283 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
284 	} else
285 		fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
286 	mtx_unlock(&psp->ps_mtx);
287 
288 	/* Build the argument list for the signal handler. */
289 	sig = bsd_to_linux_signal(sig);
290 
291 	bzero(&frame, sizeof(frame));
292 
293 	frame.sf_handler = PTROUT(catcher);
294 	frame.sf_sig = sig;
295 	frame.sf_siginfo = PTROUT(&fp->sf_si);
296 	frame.sf_ucontext = PTROUT(&fp->sf_sc);
297 
298 	/* Fill in POSIX parts. */
299 	ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
300 
301 	/*
302 	 * Build the signal context to be used by sigreturn and libgcc unwind.
303 	 */
304 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
305 	frame.sf_sc.uc_link = 0;		/* XXX ??? */
306 
307 	frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
308 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
309 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
310 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
311 	PROC_UNLOCK(p);
312 
313 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
314 
315 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
316 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_rdi;
317 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_rsi;
318 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_rbp;
319 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_rbx;
320 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_rsp;
321 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_rdx;
322 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_rcx;
323 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_rax;
324 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_rip;
325 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
326 	frame.sf_sc.uc_mcontext.sc_gs     = regs->tf_gs;
327 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
328 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
329 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
330 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
331 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
332 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
333 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
334 	frame.sf_sc.uc_mcontext.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
335 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
336 
337 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
338 		/*
339 		 * Process has trashed its stack; give it an illegal
340 		 * instruction to halt it in its tracks.
341 		 */
342 		PROC_LOCK(p);
343 		sigexit(td, SIGILL);
344 	}
345 
346 	/* Build context to run handler in. */
347 	regs->tf_rsp = PTROUT(fp);
348 	regs->tf_rip = linux32_rt_sigcode;
349 	regs->tf_rflags &= ~(PSL_T | PSL_D);
350 	regs->tf_cs = _ucode32sel;
351 	regs->tf_ss = _udatasel;
352 	regs->tf_ds = _udatasel;
353 	regs->tf_es = _udatasel;
354 	regs->tf_fs = _ufssel;
355 	regs->tf_gs = _ugssel;
356 	regs->tf_flags = TF_HASSEGS;
357 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
358 	PROC_LOCK(p);
359 	mtx_lock(&psp->ps_mtx);
360 }
361 
362 
363 /*
364  * Send an interrupt to process.
365  *
366  * Stack is set up to allow sigcode stored
367  * in u. to call routine, followed by kcall
368  * to sigreturn routine below.  After sigreturn
369  * resets the signal mask, the stack, and the
370  * frame pointer, it returns to the user
371  * specified pc, psl.
372  */
373 static void
374 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
375 {
376 	struct thread *td = curthread;
377 	struct proc *p = td->td_proc;
378 	struct sigacts *psp;
379 	struct trapframe *regs;
380 	struct l_sigframe *fp, frame;
381 	l_sigset_t lmask;
382 	int oonstack;
383 	int sig, code;
384 
385 	sig = ksi->ksi_signo;
386 	code = ksi->ksi_code;
387 	PROC_LOCK_ASSERT(p, MA_OWNED);
388 	psp = p->p_sigacts;
389 	mtx_assert(&psp->ps_mtx, MA_OWNED);
390 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
391 		/* Signal handler installed with SA_SIGINFO. */
392 		linux_rt_sendsig(catcher, ksi, mask);
393 		return;
394 	}
395 
396 	regs = td->td_frame;
397 	oonstack = sigonstack(regs->tf_rsp);
398 
399 	/* Allocate space for the signal handler context. */
400 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
401 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
402 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
403 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
404 	} else
405 		fp = (struct l_sigframe *)regs->tf_rsp - 1;
406 	mtx_unlock(&psp->ps_mtx);
407 	PROC_UNLOCK(p);
408 
409 	/* Build the argument list for the signal handler. */
410 	sig = bsd_to_linux_signal(sig);
411 
412 	bzero(&frame, sizeof(frame));
413 
414 	frame.sf_handler = PTROUT(catcher);
415 	frame.sf_sig = sig;
416 
417 	bsd_to_linux_sigset(mask, &lmask);
418 
419 	/* Build the signal context to be used by sigreturn. */
420 	frame.sf_sc.sc_mask   = lmask.__mask;
421 	frame.sf_sc.sc_gs     = regs->tf_gs;
422 	frame.sf_sc.sc_fs     = regs->tf_fs;
423 	frame.sf_sc.sc_es     = regs->tf_es;
424 	frame.sf_sc.sc_ds     = regs->tf_ds;
425 	frame.sf_sc.sc_edi    = regs->tf_rdi;
426 	frame.sf_sc.sc_esi    = regs->tf_rsi;
427 	frame.sf_sc.sc_ebp    = regs->tf_rbp;
428 	frame.sf_sc.sc_ebx    = regs->tf_rbx;
429 	frame.sf_sc.sc_esp    = regs->tf_rsp;
430 	frame.sf_sc.sc_edx    = regs->tf_rdx;
431 	frame.sf_sc.sc_ecx    = regs->tf_rcx;
432 	frame.sf_sc.sc_eax    = regs->tf_rax;
433 	frame.sf_sc.sc_eip    = regs->tf_rip;
434 	frame.sf_sc.sc_cs     = regs->tf_cs;
435 	frame.sf_sc.sc_eflags = regs->tf_rflags;
436 	frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
437 	frame.sf_sc.sc_ss     = regs->tf_ss;
438 	frame.sf_sc.sc_err    = regs->tf_err;
439 	frame.sf_sc.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
440 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
441 
442 	frame.sf_extramask[0] = lmask.__mask;
443 
444 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
445 		/*
446 		 * Process has trashed its stack; give it an illegal
447 		 * instruction to halt it in its tracks.
448 		 */
449 		PROC_LOCK(p);
450 		sigexit(td, SIGILL);
451 	}
452 
453 	/* Build context to run handler in. */
454 	regs->tf_rsp = PTROUT(fp);
455 	regs->tf_rip = linux32_sigcode;
456 	regs->tf_rflags &= ~(PSL_T | PSL_D);
457 	regs->tf_cs = _ucode32sel;
458 	regs->tf_ss = _udatasel;
459 	regs->tf_ds = _udatasel;
460 	regs->tf_es = _udatasel;
461 	regs->tf_fs = _ufssel;
462 	regs->tf_gs = _ugssel;
463 	regs->tf_flags = TF_HASSEGS;
464 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
465 	PROC_LOCK(p);
466 	mtx_lock(&psp->ps_mtx);
467 }
468 
469 /*
470  * System call to cleanup state after a signal
471  * has been taken.  Reset signal mask and
472  * stack state from context left by sendsig (above).
473  * Return to previous pc and psl as specified by
474  * context left by sendsig. Check carefully to
475  * make sure that the user has not modified the
476  * psl to gain improper privileges or to cause
477  * a machine fault.
478  */
479 int
480 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
481 {
482 	struct l_sigframe frame;
483 	struct trapframe *regs;
484 	sigset_t bmask;
485 	l_sigset_t lmask;
486 	int eflags;
487 	ksiginfo_t ksi;
488 
489 	regs = td->td_frame;
490 
491 	/*
492 	 * The trampoline code hands us the sigframe.
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->sfp, &frame, sizeof(frame)) != 0)
497 		return (EFAULT);
498 
499 	/* Check for security violations. */
500 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
501 	eflags = frame.sf_sc.sc_eflags;
502 	if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
503 		return(EINVAL);
504 
505 	/*
506 	 * Don't allow users to load a valid privileged %cs.  Let the
507 	 * hardware check for invalid selectors, excess privilege in
508 	 * other selectors, invalid %eip's and invalid %esp's.
509 	 */
510 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
511 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
512 		ksiginfo_init_trap(&ksi);
513 		ksi.ksi_signo = SIGBUS;
514 		ksi.ksi_code = BUS_OBJERR;
515 		ksi.ksi_trapno = T_PROTFLT;
516 		ksi.ksi_addr = (void *)regs->tf_rip;
517 		trapsignal(td, &ksi);
518 		return(EINVAL);
519 	}
520 
521 	lmask.__mask = frame.sf_sc.sc_mask;
522 	lmask.__mask = frame.sf_extramask[0];
523 	linux_to_bsd_sigset(&lmask, &bmask);
524 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
525 
526 	/* Restore signal context. */
527 	regs->tf_rdi    = frame.sf_sc.sc_edi;
528 	regs->tf_rsi    = frame.sf_sc.sc_esi;
529 	regs->tf_rbp    = frame.sf_sc.sc_ebp;
530 	regs->tf_rbx    = frame.sf_sc.sc_ebx;
531 	regs->tf_rdx    = frame.sf_sc.sc_edx;
532 	regs->tf_rcx    = frame.sf_sc.sc_ecx;
533 	regs->tf_rax    = frame.sf_sc.sc_eax;
534 	regs->tf_rip    = frame.sf_sc.sc_eip;
535 	regs->tf_cs     = frame.sf_sc.sc_cs;
536 	regs->tf_ds     = frame.sf_sc.sc_ds;
537 	regs->tf_es     = frame.sf_sc.sc_es;
538 	regs->tf_fs     = frame.sf_sc.sc_fs;
539 	regs->tf_gs     = frame.sf_sc.sc_gs;
540 	regs->tf_rflags = eflags;
541 	regs->tf_rsp    = frame.sf_sc.sc_esp_at_signal;
542 	regs->tf_ss     = frame.sf_sc.sc_ss;
543 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
544 
545 	return (EJUSTRETURN);
546 }
547 
548 /*
549  * System call to cleanup state after a signal
550  * has been taken.  Reset signal mask and
551  * stack state from context left by rt_sendsig (above).
552  * Return to previous pc and psl as specified by
553  * context left by sendsig. Check carefully to
554  * make sure that the user has not modified the
555  * psl to gain improper privileges or to cause
556  * a machine fault.
557  */
558 int
559 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
560 {
561 	struct l_ucontext uc;
562 	struct l_sigcontext *context;
563 	sigset_t bmask;
564 	l_stack_t *lss;
565 	stack_t ss;
566 	struct trapframe *regs;
567 	int eflags;
568 	ksiginfo_t ksi;
569 
570 	regs = td->td_frame;
571 
572 	/*
573 	 * The trampoline code hands us the ucontext.
574 	 * It is unsafe to keep track of it ourselves, in the event that a
575 	 * program jumps out of a signal handler.
576 	 */
577 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
578 		return (EFAULT);
579 
580 	context = &uc.uc_mcontext;
581 
582 	/* Check for security violations. */
583 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
584 	eflags = context->sc_eflags;
585 	if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
586 		return(EINVAL);
587 
588 	/*
589 	 * Don't allow users to load a valid privileged %cs.  Let the
590 	 * hardware check for invalid selectors, excess privilege in
591 	 * other selectors, invalid %eip's and invalid %esp's.
592 	 */
593 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
594 	if (!CS_SECURE(context->sc_cs)) {
595 		ksiginfo_init_trap(&ksi);
596 		ksi.ksi_signo = SIGBUS;
597 		ksi.ksi_code = BUS_OBJERR;
598 		ksi.ksi_trapno = T_PROTFLT;
599 		ksi.ksi_addr = (void *)regs->tf_rip;
600 		trapsignal(td, &ksi);
601 		return(EINVAL);
602 	}
603 
604 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
605 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
606 
607 	/*
608 	 * Restore signal context
609 	 */
610 	regs->tf_gs	= context->sc_gs;
611 	regs->tf_fs	= context->sc_fs;
612 	regs->tf_es	= context->sc_es;
613 	regs->tf_ds	= context->sc_ds;
614 	regs->tf_rdi    = context->sc_edi;
615 	regs->tf_rsi    = context->sc_esi;
616 	regs->tf_rbp    = context->sc_ebp;
617 	regs->tf_rbx    = context->sc_ebx;
618 	regs->tf_rdx    = context->sc_edx;
619 	regs->tf_rcx    = context->sc_ecx;
620 	regs->tf_rax    = context->sc_eax;
621 	regs->tf_rip    = context->sc_eip;
622 	regs->tf_cs     = context->sc_cs;
623 	regs->tf_rflags = eflags;
624 	regs->tf_rsp    = context->sc_esp_at_signal;
625 	regs->tf_ss     = context->sc_ss;
626 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
627 
628 	/*
629 	 * call sigaltstack & ignore results..
630 	 */
631 	lss = &uc.uc_stack;
632 	ss.ss_sp = PTRIN(lss->ss_sp);
633 	ss.ss_size = lss->ss_size;
634 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
635 
636 	(void)kern_sigaltstack(td, &ss, NULL);
637 
638 	return (EJUSTRETURN);
639 }
640 
641 static int
642 linux32_fetch_syscall_args(struct thread *td)
643 {
644 	struct proc *p;
645 	struct trapframe *frame;
646 	struct syscall_args *sa;
647 
648 	p = td->td_proc;
649 	frame = td->td_frame;
650 	sa = &td->td_sa;
651 
652 	sa->args[0] = frame->tf_rbx;
653 	sa->args[1] = frame->tf_rcx;
654 	sa->args[2] = frame->tf_rdx;
655 	sa->args[3] = frame->tf_rsi;
656 	sa->args[4] = frame->tf_rdi;
657 	sa->args[5] = frame->tf_rbp;	/* Unconfirmed */
658 	sa->code = frame->tf_rax;
659 
660 	if (sa->code >= p->p_sysent->sv_size)
661 		/* nosys */
662 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
663 	else
664 		sa->callp = &p->p_sysent->sv_table[sa->code];
665 	sa->narg = sa->callp->sy_narg;
666 
667 	td->td_retval[0] = 0;
668 	td->td_retval[1] = frame->tf_rdx;
669 
670 	return (0);
671 }
672 
673 /*
674  * Clear registers on exec
675  * XXX copied from ia32_signal.c.
676  */
677 static void
678 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
679 {
680 	struct trapframe *regs = td->td_frame;
681 	struct pcb *pcb = td->td_pcb;
682 	register_t saved_rflags;
683 
684 	regs = td->td_frame;
685 	pcb = td->td_pcb;
686 
687 	if (td->td_proc->p_md.md_ldt != NULL)
688 		user_ldt_free(td);
689 
690 	critical_enter();
691 	wrmsr(MSR_FSBASE, 0);
692 	wrmsr(MSR_KGSBASE, 0);	/* User value while we're in the kernel */
693 	pcb->pcb_fsbase = 0;
694 	pcb->pcb_gsbase = 0;
695 	critical_exit();
696 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
697 
698 	saved_rflags = regs->tf_rflags & PSL_T;
699 	bzero((char *)regs, sizeof(struct trapframe));
700 	regs->tf_rip = imgp->entry_addr;
701 	regs->tf_rsp = stack;
702 	regs->tf_rflags = PSL_USER | saved_rflags;
703 	regs->tf_gs = _ugssel;
704 	regs->tf_fs = _ufssel;
705 	regs->tf_es = _udatasel;
706 	regs->tf_ds = _udatasel;
707 	regs->tf_ss = _udatasel;
708 	regs->tf_flags = TF_HASSEGS;
709 	regs->tf_cs = _ucode32sel;
710 	regs->tf_rbx = imgp->ps_strings;
711 
712 	fpstate_drop(td);
713 
714 	/* Do full restore on return so that we can change to a different %cs */
715 	set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
716 }
717 
718 /*
719  * XXX copied from ia32_sysvec.c.
720  */
721 static register_t *
722 linux_copyout_strings(struct image_params *imgp)
723 {
724 	int argc, envc;
725 	u_int32_t *vectp;
726 	char *stringp, *destp;
727 	u_int32_t *stack_base;
728 	struct linux32_ps_strings *arginfo;
729 	char canary[LINUX_AT_RANDOM_LEN];
730 	size_t execpath_len;
731 
732 	/* Calculate string base and vector table pointers. */
733 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
734 		execpath_len = strlen(imgp->execpath) + 1;
735 	else
736 		execpath_len = 0;
737 
738 	arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
739 	destp =	(caddr_t)arginfo - SPARE_USRSPACE -
740 	    roundup(sizeof(canary), sizeof(char *)) -
741 	    roundup(execpath_len, sizeof(char *)) -
742 	    roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
743 
744 	if (execpath_len != 0) {
745 		imgp->execpathp = (uintptr_t)arginfo - execpath_len;
746 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
747 	}
748 
749 	/* Prepare the canary for SSP. */
750 	arc4rand(canary, sizeof(canary), 0);
751 	imgp->canary = (uintptr_t)arginfo -
752 	    roundup(execpath_len, sizeof(char *)) -
753 	    roundup(sizeof(canary), sizeof(char *));
754 	copyout(canary, (void *)imgp->canary, sizeof(canary));
755 
756 	vectp = (uint32_t *)destp;
757 	if (imgp->auxargs)
758 		imgp->sysent->sv_copyout_auxargs(imgp, (u_long *)&vectp);
759 
760 	/*
761 	 * Allocate room for the argv[] and env vectors including the
762 	 * terminating NULL pointers.
763 	 */
764 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
765 
766 	/* vectp also becomes our initial stack base. */
767 	stack_base = vectp;
768 
769 	stringp = imgp->args->begin_argv;
770 	argc = imgp->args->argc;
771 	envc = imgp->args->envc;
772 	/* Copy out strings - arguments and environment. */
773 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
774 
775 	/* Fill in "ps_strings" struct for ps, w, etc. */
776 	suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp);
777 	suword32(&arginfo->ps_nargvstr, argc);
778 
779 	/* Fill in argument portion of vector table. */
780 	for (; argc > 0; --argc) {
781 		suword32(vectp++, (uint32_t)(intptr_t)destp);
782 		while (*stringp++ != 0)
783 			destp++;
784 		destp++;
785 	}
786 
787 	/* A null vector table pointer separates the argp's from the envp's. */
788 	suword32(vectp++, 0);
789 
790 	suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp);
791 	suword32(&arginfo->ps_nenvstr, envc);
792 
793 	/* Fill in environment portion of vector table. */
794 	for (; envc > 0; --envc) {
795 		suword32(vectp++, (uint32_t)(intptr_t)destp);
796 		while (*stringp++ != 0)
797 			destp++;
798 		destp++;
799 	}
800 
801 	/* The end of the vector table is a null pointer. */
802 	suword32(vectp, 0);
803 
804 	return ((register_t *)stack_base);
805 }
806 
807 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0,
808     "32-bit Linux emulation");
809 
810 static u_long	linux32_maxdsiz = LINUX32_MAXDSIZ;
811 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
812     &linux32_maxdsiz, 0, "");
813 static u_long	linux32_maxssiz = LINUX32_MAXSSIZ;
814 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
815     &linux32_maxssiz, 0, "");
816 static u_long	linux32_maxvmem = LINUX32_MAXVMEM;
817 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
818     &linux32_maxvmem, 0, "");
819 
820 static void
821 linux32_fixlimit(struct rlimit *rl, int which)
822 {
823 
824 	switch (which) {
825 	case RLIMIT_DATA:
826 		if (linux32_maxdsiz != 0) {
827 			if (rl->rlim_cur > linux32_maxdsiz)
828 				rl->rlim_cur = linux32_maxdsiz;
829 			if (rl->rlim_max > linux32_maxdsiz)
830 				rl->rlim_max = linux32_maxdsiz;
831 		}
832 		break;
833 	case RLIMIT_STACK:
834 		if (linux32_maxssiz != 0) {
835 			if (rl->rlim_cur > linux32_maxssiz)
836 				rl->rlim_cur = linux32_maxssiz;
837 			if (rl->rlim_max > linux32_maxssiz)
838 				rl->rlim_max = linux32_maxssiz;
839 		}
840 		break;
841 	case RLIMIT_VMEM:
842 		if (linux32_maxvmem != 0) {
843 			if (rl->rlim_cur > linux32_maxvmem)
844 				rl->rlim_cur = linux32_maxvmem;
845 			if (rl->rlim_max > linux32_maxvmem)
846 				rl->rlim_max = linux32_maxvmem;
847 		}
848 		break;
849 	}
850 }
851 
852 struct sysentvec elf_linux_sysvec = {
853 	.sv_size	= LINUX32_SYS_MAXSYSCALL,
854 	.sv_table	= linux32_sysent,
855 	.sv_errsize	= ELAST + 1,
856 	.sv_errtbl	= linux_errtbl,
857 	.sv_transtrap	= linux_translate_traps,
858 	.sv_fixup	= linux_fixup_elf,
859 	.sv_sendsig	= linux_sendsig,
860 	.sv_sigcode	= &_binary_linux32_locore_o_start,
861 	.sv_szsigcode	= &linux_szsigcode,
862 	.sv_name	= "Linux ELF32",
863 	.sv_coredump	= elf32_coredump,
864 	.sv_imgact_try	= linux_exec_imgact_try,
865 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
866 	.sv_minuser	= VM_MIN_ADDRESS,
867 	.sv_maxuser	= LINUX32_MAXUSER,
868 	.sv_usrstack	= LINUX32_USRSTACK,
869 	.sv_psstrings	= LINUX32_PS_STRINGS,
870 	.sv_stackprot	= VM_PROT_ALL,
871 	.sv_copyout_auxargs = linux_copyout_auxargs,
872 	.sv_copyout_strings = linux_copyout_strings,
873 	.sv_setregs	= linux_exec_setregs,
874 	.sv_fixlimit	= linux32_fixlimit,
875 	.sv_maxssiz	= &linux32_maxssiz,
876 	.sv_flags	= SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP,
877 	.sv_set_syscall_retval = cpu_set_syscall_retval,
878 	.sv_fetch_syscall_args = linux32_fetch_syscall_args,
879 	.sv_syscallnames = NULL,
880 	.sv_shared_page_base = LINUX32_SHAREDPAGE,
881 	.sv_shared_page_len = PAGE_SIZE,
882 	.sv_schedtail	= linux_schedtail,
883 	.sv_thread_detach = linux_thread_detach,
884 	.sv_trap	= NULL,
885 };
886 
887 static void
888 linux_vdso_install(void *param)
889 {
890 
891 	linux_szsigcode = (&_binary_linux32_locore_o_end -
892 	    &_binary_linux32_locore_o_start);
893 
894 	if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
895 		panic("Linux invalid vdso size\n");
896 
897 	__elfN(linux_vdso_fixup)(&elf_linux_sysvec);
898 
899 	linux_shared_page_obj = __elfN(linux_shared_page_init)
900 	    (&linux_shared_page_mapping);
901 
902 	__elfN(linux_vdso_reloc)(&elf_linux_sysvec);
903 
904 	bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
905 	    linux_szsigcode);
906 	elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
907 
908 	linux_kplatform = linux_shared_page_mapping +
909 	    (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
910 }
911 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
912     linux_vdso_install, NULL);
913 
914 static void
915 linux_vdso_deinstall(void *param)
916 {
917 
918 	__elfN(linux_shared_page_fini)(linux_shared_page_obj);
919 }
920 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
921     linux_vdso_deinstall, NULL);
922 
923 static char GNU_ABI_VENDOR[] = "GNU";
924 static int GNULINUX_ABI_DESC = 0;
925 
926 static bool
927 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
928 {
929 	const Elf32_Word *desc;
930 	uintptr_t p;
931 
932 	p = (uintptr_t)(note + 1);
933 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
934 
935 	desc = (const Elf32_Word *)p;
936 	if (desc[0] != GNULINUX_ABI_DESC)
937 		return (false);
938 
939 	/*
940 	 * For Linux we encode osrel using the Linux convention of
941 	 * 	(version << 16) | (major << 8) | (minor)
942 	 * See macro in linux_mib.h
943 	 */
944 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
945 
946 	return (true);
947 }
948 
949 static Elf_Brandnote linux32_brandnote = {
950 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
951 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
952 	.hdr.n_type	= 1,
953 	.vendor		= GNU_ABI_VENDOR,
954 	.flags		= BN_TRANSLATE_OSREL,
955 	.trans_osrel	= linux32_trans_osrel
956 };
957 
958 static Elf32_Brandinfo linux_brand = {
959 	.brand		= ELFOSABI_LINUX,
960 	.machine	= EM_386,
961 	.compat_3_brand	= "Linux",
962 	.emul_path	= "/compat/linux",
963 	.interp_path	= "/lib/ld-linux.so.1",
964 	.sysvec		= &elf_linux_sysvec,
965 	.interp_newpath	= NULL,
966 	.brand_note	= &linux32_brandnote,
967 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
968 };
969 
970 static Elf32_Brandinfo linux_glibc2brand = {
971 	.brand		= ELFOSABI_LINUX,
972 	.machine	= EM_386,
973 	.compat_3_brand	= "Linux",
974 	.emul_path	= "/compat/linux",
975 	.interp_path	= "/lib/ld-linux.so.2",
976 	.sysvec		= &elf_linux_sysvec,
977 	.interp_newpath	= NULL,
978 	.brand_note	= &linux32_brandnote,
979 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
980 };
981 
982 static Elf32_Brandinfo linux_muslbrand = {
983 	.brand		= ELFOSABI_LINUX,
984 	.machine	= EM_386,
985 	.compat_3_brand	= "Linux",
986 	.emul_path	= "/compat/linux",
987 	.interp_path	= "/lib/ld-musl-i386.so.1",
988 	.sysvec		= &elf_linux_sysvec,
989 	.interp_newpath	= NULL,
990 	.brand_note	= &linux32_brandnote,
991 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
992 };
993 
994 Elf32_Brandinfo *linux_brandlist[] = {
995 	&linux_brand,
996 	&linux_glibc2brand,
997 	&linux_muslbrand,
998 	NULL
999 };
1000 
1001 static int
1002 linux_elf_modevent(module_t mod, int type, void *data)
1003 {
1004 	Elf32_Brandinfo **brandinfo;
1005 	int error;
1006 	struct linux_ioctl_handler **lihp;
1007 
1008 	error = 0;
1009 
1010 	switch(type) {
1011 	case MOD_LOAD:
1012 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1013 		     ++brandinfo)
1014 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1015 				error = EINVAL;
1016 		if (error == 0) {
1017 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1018 				linux32_ioctl_register_handler(*lihp);
1019 			stclohz = (stathz ? stathz : hz);
1020 			if (bootverbose)
1021 				printf("Linux ELF exec handler installed\n");
1022 		} else
1023 			printf("cannot insert Linux ELF brand handler\n");
1024 		break;
1025 	case MOD_UNLOAD:
1026 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1027 		     ++brandinfo)
1028 			if (elf32_brand_inuse(*brandinfo))
1029 				error = EBUSY;
1030 		if (error == 0) {
1031 			for (brandinfo = &linux_brandlist[0];
1032 			     *brandinfo != NULL; ++brandinfo)
1033 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1034 					error = EINVAL;
1035 		}
1036 		if (error == 0) {
1037 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1038 				linux32_ioctl_unregister_handler(*lihp);
1039 			if (bootverbose)
1040 				printf("Linux ELF exec handler removed\n");
1041 		} else
1042 			printf("Could not deinstall ELF interpreter entry\n");
1043 		break;
1044 	default:
1045 		return (EOPNOTSUPP);
1046 	}
1047 	return (error);
1048 }
1049 
1050 static moduledata_t linux_elf_mod = {
1051 	"linuxelf",
1052 	linux_elf_modevent,
1053 	0
1054 };
1055 
1056 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1057 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1058 FEATURE(linux, "Linux 32bit support");
1059