xref: /freebsd/sys/amd64/linux32/linux32_sysvec.c (revision 77013d11e6483b970af25e13c9b892075742f7e5)
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/stddef.h>
60 #include <sys/signalvar.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysent.h>
64 #include <sys/sysproto.h>
65 #include <sys/vnode.h>
66 #include <sys/eventhandler.h>
67 
68 #include <vm/vm.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_param.h>
75 
76 #include <machine/cpu.h>
77 #include <machine/md_var.h>
78 #include <machine/pcb.h>
79 #include <machine/specialreg.h>
80 #include <machine/trap.h>
81 
82 #include <x86/linux/linux_x86.h>
83 #include <amd64/linux32/linux.h>
84 #include <amd64/linux32/linux32_proto.h>
85 #include <compat/linux/linux_emul.h>
86 #include <compat/linux/linux_ioctl.h>
87 #include <compat/linux/linux_mib.h>
88 #include <compat/linux/linux_misc.h>
89 #include <compat/linux/linux_signal.h>
90 #include <compat/linux/linux_util.h>
91 #include <compat/linux/linux_vdso.h>
92 
93 MODULE_VERSION(linux, 1);
94 
95 #define	LINUX32_MAXUSER		((1ul << 32) - PAGE_SIZE)
96 #define	LINUX32_VDSOPAGE_SIZE	PAGE_SIZE * 2
97 #define	LINUX32_VDSOPAGE	(LINUX32_MAXUSER - LINUX32_VDSOPAGE_SIZE)
98 #define	LINUX32_SHAREDPAGE	(LINUX32_VDSOPAGE - PAGE_SIZE)
99 				/*
100 				 * PAGE_SIZE - the size
101 				 * of the native SHAREDPAGE
102 				 */
103 #define	LINUX32_USRSTACK	LINUX32_SHAREDPAGE
104 
105 static int linux_szsigcode;
106 static vm_object_t linux_vdso_obj;
107 static char *linux_vdso_mapping;
108 extern char _binary_linux32_vdso_so_o_start;
109 extern char _binary_linux32_vdso_so_o_end;
110 static vm_offset_t linux_vdso_base;
111 
112 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
113 
114 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
115 
116 static int	linux_fixup_elf(uintptr_t *stack_base,
117 		    struct image_params *iparams);
118 static int	linux_copyout_strings(struct image_params *imgp,
119 		    uintptr_t *stack_base);
120 static void     linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
121 static void	linux_exec_setregs(struct thread *td,
122 				   struct image_params *imgp, uintptr_t stack);
123 static void	linux_exec_sysvec_init(void *param);
124 static int	linux_on_exec_vmspace(struct proc *p,
125 		    struct image_params *imgp);
126 static void	linux32_fixlimit(struct rlimit *rl, int which);
127 static bool	linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
128 static void	linux_vdso_install(const void *param);
129 static void	linux_vdso_deinstall(const void *param);
130 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
131 static void	linux32_set_syscall_retval(struct thread *td, int error);
132 
133 #define LINUX_T_UNKNOWN  255
134 static int _bsd_to_linux_trapcode[] = {
135 	LINUX_T_UNKNOWN,	/* 0 */
136 	6,			/* 1  T_PRIVINFLT */
137 	LINUX_T_UNKNOWN,	/* 2 */
138 	3,			/* 3  T_BPTFLT */
139 	LINUX_T_UNKNOWN,	/* 4 */
140 	LINUX_T_UNKNOWN,	/* 5 */
141 	16,			/* 6  T_ARITHTRAP */
142 	254,			/* 7  T_ASTFLT */
143 	LINUX_T_UNKNOWN,	/* 8 */
144 	13,			/* 9  T_PROTFLT */
145 	1,			/* 10 T_TRCTRAP */
146 	LINUX_T_UNKNOWN,	/* 11 */
147 	14,			/* 12 T_PAGEFLT */
148 	LINUX_T_UNKNOWN,	/* 13 */
149 	17,			/* 14 T_ALIGNFLT */
150 	LINUX_T_UNKNOWN,	/* 15 */
151 	LINUX_T_UNKNOWN,	/* 16 */
152 	LINUX_T_UNKNOWN,	/* 17 */
153 	0,			/* 18 T_DIVIDE */
154 	2,			/* 19 T_NMI */
155 	4,			/* 20 T_OFLOW */
156 	5,			/* 21 T_BOUND */
157 	7,			/* 22 T_DNA */
158 	8,			/* 23 T_DOUBLEFLT */
159 	9,			/* 24 T_FPOPFLT */
160 	10,			/* 25 T_TSSFLT */
161 	11,			/* 26 T_SEGNPFLT */
162 	12,			/* 27 T_STKFLT */
163 	18,			/* 28 T_MCHK */
164 	19,			/* 29 T_XMMFLT */
165 	15			/* 30 T_RESERVED */
166 };
167 #define bsd_to_linux_trapcode(code) \
168     ((code)<nitems(_bsd_to_linux_trapcode)? \
169      _bsd_to_linux_trapcode[(code)]: \
170      LINUX_T_UNKNOWN)
171 
172 struct linux32_ps_strings {
173 	u_int32_t ps_argvstr;	/* first of 0 or more argument strings */
174 	u_int ps_nargvstr;	/* the number of argument strings */
175 	u_int32_t ps_envstr;	/* first of 0 or more environment strings */
176 	u_int ps_nenvstr;	/* the number of environment strings */
177 };
178 #define	LINUX32_PS_STRINGS	(LINUX32_USRSTACK - \
179 				    sizeof(struct linux32_ps_strings))
180 
181 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
182 LINUX_VDSO_SYM_INTPTR(__kernel_sigreturn);
183 LINUX_VDSO_SYM_INTPTR(__kernel_rt_sigreturn);
184 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
185 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
186 LINUX_VDSO_SYM_CHAR(linux_platform);
187 
188 /*
189  * If FreeBSD & Linux have a difference of opinion about what a trap
190  * means, deal with it here.
191  *
192  * MPSAFE
193  */
194 static int
195 linux_translate_traps(int signal, int trap_code)
196 {
197 	if (signal != SIGBUS)
198 		return (signal);
199 	switch (trap_code) {
200 	case T_PROTFLT:
201 	case T_TSSFLT:
202 	case T_DOUBLEFLT:
203 	case T_PAGEFLT:
204 		return (SIGSEGV);
205 	default:
206 		return (signal);
207 	}
208 }
209 
210 static int
211 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
212 {
213 	Elf32_Auxargs *args;
214 	Elf32_Auxinfo *argarray, *pos;
215 	int error, issetugid;
216 
217 	args = (Elf32_Auxargs *)imgp->auxargs;
218 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
219 	    M_WAITOK | M_ZERO);
220 
221 	issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
222 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, __kernel_vsyscall);
223 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
224 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
225 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
226 
227 	/*
228 	 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
229 	 * as it has appeared in the 2.4.0-rc7 first time.
230 	 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
231 	 * glibc falls back to the hard-coded CLK_TCK value when aux entry
232 	 * is not present.
233 	 * Also see linux_times() implementation.
234 	 */
235 	if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
236 		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
237 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
238 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
239 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
240 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
241 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
242 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
243 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
244 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
245 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
246 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
247 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
248 	AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
249 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
250 	if (imgp->execpathp != 0)
251 		AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
252 	if (args->execfd != -1)
253 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
254 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
255 	AUXARGS_ENTRY(pos, AT_NULL, 0);
256 
257 	free(imgp->auxargs, M_TEMP);
258 	imgp->auxargs = NULL;
259 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
260 
261 	error = copyout(argarray, (void *)base,
262 	    sizeof(*argarray) * LINUX_AT_COUNT);
263 	free(argarray, M_TEMP);
264 	return (error);
265 }
266 
267 static int
268 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
269 {
270 	Elf32_Addr *base;
271 
272 	base = (Elf32_Addr *)*stack_base;
273 	base--;
274 	if (suword32(base, (uint32_t)imgp->args->argc) == -1)
275 		return (EFAULT);
276 	*stack_base = (uintptr_t)base;
277 	return (0);
278 }
279 
280 static void
281 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
282 {
283 	struct thread *td = curthread;
284 	struct proc *p = td->td_proc;
285 	struct sigacts *psp;
286 	struct trapframe *regs;
287 	struct l_rt_sigframe *fp, frame;
288 	int oonstack;
289 	int sig;
290 	int code;
291 
292 	sig = ksi->ksi_signo;
293 	code = ksi->ksi_code;
294 	PROC_LOCK_ASSERT(p, MA_OWNED);
295 	psp = p->p_sigacts;
296 	mtx_assert(&psp->ps_mtx, MA_OWNED);
297 	regs = td->td_frame;
298 	oonstack = sigonstack(regs->tf_rsp);
299 
300 	/* Allocate space for the signal handler context. */
301 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
302 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
303 		fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
304 		    td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
305 	} else
306 		fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
307 	mtx_unlock(&psp->ps_mtx);
308 
309 	/* Build the argument list for the signal handler. */
310 	sig = bsd_to_linux_signal(sig);
311 
312 	bzero(&frame, sizeof(frame));
313 
314 	frame.sf_handler = PTROUT(catcher);
315 	frame.sf_sig = sig;
316 	frame.sf_siginfo = PTROUT(&fp->sf_si);
317 	frame.sf_ucontext = PTROUT(&fp->sf_sc);
318 
319 	/* Fill in POSIX parts. */
320 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
321 
322 	/*
323 	 * Build the signal context to be used by sigreturn and libgcc unwind.
324 	 */
325 	frame.sf_sc.uc_flags = 0;		/* XXX ??? */
326 	frame.sf_sc.uc_link = 0;		/* XXX ??? */
327 
328 	frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
329 	frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
330 	frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
331 	    ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
332 	PROC_UNLOCK(p);
333 
334 	bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
335 
336 	frame.sf_sc.uc_mcontext.sc_mask   = frame.sf_sc.uc_sigmask.__mask;
337 	frame.sf_sc.uc_mcontext.sc_edi    = regs->tf_rdi;
338 	frame.sf_sc.uc_mcontext.sc_esi    = regs->tf_rsi;
339 	frame.sf_sc.uc_mcontext.sc_ebp    = regs->tf_rbp;
340 	frame.sf_sc.uc_mcontext.sc_ebx    = regs->tf_rbx;
341 	frame.sf_sc.uc_mcontext.sc_esp    = regs->tf_rsp;
342 	frame.sf_sc.uc_mcontext.sc_edx    = regs->tf_rdx;
343 	frame.sf_sc.uc_mcontext.sc_ecx    = regs->tf_rcx;
344 	frame.sf_sc.uc_mcontext.sc_eax    = regs->tf_rax;
345 	frame.sf_sc.uc_mcontext.sc_eip    = regs->tf_rip;
346 	frame.sf_sc.uc_mcontext.sc_cs     = regs->tf_cs;
347 	frame.sf_sc.uc_mcontext.sc_gs     = regs->tf_gs;
348 	frame.sf_sc.uc_mcontext.sc_fs     = regs->tf_fs;
349 	frame.sf_sc.uc_mcontext.sc_es     = regs->tf_es;
350 	frame.sf_sc.uc_mcontext.sc_ds     = regs->tf_ds;
351 	frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
352 	frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
353 	frame.sf_sc.uc_mcontext.sc_ss     = regs->tf_ss;
354 	frame.sf_sc.uc_mcontext.sc_err    = regs->tf_err;
355 	frame.sf_sc.uc_mcontext.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
356 	frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
357 
358 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
359 		/*
360 		 * Process has trashed its stack; give it an illegal
361 		 * instruction to halt it in its tracks.
362 		 */
363 		PROC_LOCK(p);
364 		sigexit(td, SIGILL);
365 	}
366 
367 	/* Build context to run handler in. */
368 	regs->tf_rsp = PTROUT(fp);
369 	regs->tf_rip = __kernel_rt_sigreturn;
370 	regs->tf_rflags &= ~(PSL_T | PSL_D);
371 	regs->tf_cs = _ucode32sel;
372 	regs->tf_ss = _udatasel;
373 	regs->tf_ds = _udatasel;
374 	regs->tf_es = _udatasel;
375 	regs->tf_fs = _ufssel;
376 	regs->tf_gs = _ugssel;
377 	regs->tf_flags = TF_HASSEGS;
378 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
379 	PROC_LOCK(p);
380 	mtx_lock(&psp->ps_mtx);
381 }
382 
383 /*
384  * Send an interrupt to process.
385  *
386  * Stack is set up to allow sigcode stored
387  * in u. to call routine, followed by kcall
388  * to sigreturn routine below.  After sigreturn
389  * resets the signal mask, the stack, and the
390  * frame pointer, it returns to the user
391  * specified pc, psl.
392  */
393 static void
394 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
395 {
396 	struct thread *td = curthread;
397 	struct proc *p = td->td_proc;
398 	struct sigacts *psp;
399 	struct trapframe *regs;
400 	struct l_sigframe *fp, frame;
401 	l_sigset_t lmask;
402 	int oonstack;
403 	int sig, code;
404 
405 	sig = ksi->ksi_signo;
406 	code = ksi->ksi_code;
407 	PROC_LOCK_ASSERT(p, MA_OWNED);
408 	psp = p->p_sigacts;
409 	mtx_assert(&psp->ps_mtx, MA_OWNED);
410 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
411 		/* Signal handler installed with SA_SIGINFO. */
412 		linux_rt_sendsig(catcher, ksi, mask);
413 		return;
414 	}
415 
416 	regs = td->td_frame;
417 	oonstack = sigonstack(regs->tf_rsp);
418 
419 	/* Allocate space for the signal handler context. */
420 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
421 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
422 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
423 		    td->td_sigstk.ss_size - sizeof(struct l_sigframe));
424 	} else
425 		fp = (struct l_sigframe *)regs->tf_rsp - 1;
426 	mtx_unlock(&psp->ps_mtx);
427 	PROC_UNLOCK(p);
428 
429 	/* Build the argument list for the signal handler. */
430 	sig = bsd_to_linux_signal(sig);
431 
432 	bzero(&frame, sizeof(frame));
433 
434 	frame.sf_handler = PTROUT(catcher);
435 	frame.sf_sig = sig;
436 
437 	bsd_to_linux_sigset(mask, &lmask);
438 
439 	/* Build the signal context to be used by sigreturn. */
440 	frame.sf_sc.sc_mask   = lmask.__mask;
441 	frame.sf_sc.sc_gs     = regs->tf_gs;
442 	frame.sf_sc.sc_fs     = regs->tf_fs;
443 	frame.sf_sc.sc_es     = regs->tf_es;
444 	frame.sf_sc.sc_ds     = regs->tf_ds;
445 	frame.sf_sc.sc_edi    = regs->tf_rdi;
446 	frame.sf_sc.sc_esi    = regs->tf_rsi;
447 	frame.sf_sc.sc_ebp    = regs->tf_rbp;
448 	frame.sf_sc.sc_ebx    = regs->tf_rbx;
449 	frame.sf_sc.sc_esp    = regs->tf_rsp;
450 	frame.sf_sc.sc_edx    = regs->tf_rdx;
451 	frame.sf_sc.sc_ecx    = regs->tf_rcx;
452 	frame.sf_sc.sc_eax    = regs->tf_rax;
453 	frame.sf_sc.sc_eip    = regs->tf_rip;
454 	frame.sf_sc.sc_cs     = regs->tf_cs;
455 	frame.sf_sc.sc_eflags = regs->tf_rflags;
456 	frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
457 	frame.sf_sc.sc_ss     = regs->tf_ss;
458 	frame.sf_sc.sc_err    = regs->tf_err;
459 	frame.sf_sc.sc_cr2    = (u_int32_t)(uintptr_t)ksi->ksi_addr;
460 	frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
461 
462 	frame.sf_extramask[0] = lmask.__mask;
463 
464 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
465 		/*
466 		 * Process has trashed its stack; give it an illegal
467 		 * instruction to halt it in its tracks.
468 		 */
469 		PROC_LOCK(p);
470 		sigexit(td, SIGILL);
471 	}
472 
473 	/* Build context to run handler in. */
474 	regs->tf_rsp = PTROUT(fp);
475 	regs->tf_rip = __kernel_sigreturn;
476 	regs->tf_rflags &= ~(PSL_T | PSL_D);
477 	regs->tf_cs = _ucode32sel;
478 	regs->tf_ss = _udatasel;
479 	regs->tf_ds = _udatasel;
480 	regs->tf_es = _udatasel;
481 	regs->tf_fs = _ufssel;
482 	regs->tf_gs = _ugssel;
483 	regs->tf_flags = TF_HASSEGS;
484 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
485 	PROC_LOCK(p);
486 	mtx_lock(&psp->ps_mtx);
487 }
488 
489 /*
490  * System call to cleanup state after a signal
491  * has been taken.  Reset signal mask and
492  * stack state from context left by sendsig (above).
493  * Return to previous pc and psl as specified by
494  * context left by sendsig. Check carefully to
495  * make sure that the user has not modified the
496  * psl to gain improper privileges or to cause
497  * a machine fault.
498  */
499 int
500 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
501 {
502 	struct l_sigframe frame;
503 	struct trapframe *regs;
504 	sigset_t bmask;
505 	l_sigset_t lmask;
506 	int eflags;
507 	ksiginfo_t ksi;
508 
509 	regs = td->td_frame;
510 
511 	/*
512 	 * The trampoline code hands us the sigframe.
513 	 * It is unsafe to keep track of it ourselves, in the event that a
514 	 * program jumps out of a signal handler.
515 	 */
516 	if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
517 		return (EFAULT);
518 
519 	/* Check for security violations. */
520 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
521 	eflags = frame.sf_sc.sc_eflags;
522 	if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
523 		return(EINVAL);
524 
525 	/*
526 	 * Don't allow users to load a valid privileged %cs.  Let the
527 	 * hardware check for invalid selectors, excess privilege in
528 	 * other selectors, invalid %eip's and invalid %esp's.
529 	 */
530 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
531 	if (!CS_SECURE(frame.sf_sc.sc_cs)) {
532 		ksiginfo_init_trap(&ksi);
533 		ksi.ksi_signo = SIGBUS;
534 		ksi.ksi_code = BUS_OBJERR;
535 		ksi.ksi_trapno = T_PROTFLT;
536 		ksi.ksi_addr = (void *)regs->tf_rip;
537 		trapsignal(td, &ksi);
538 		return(EINVAL);
539 	}
540 
541 	lmask.__mask = frame.sf_sc.sc_mask;
542 	lmask.__mask = frame.sf_extramask[0];
543 	linux_to_bsd_sigset(&lmask, &bmask);
544 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
545 
546 	/* Restore signal context. */
547 	regs->tf_rdi    = frame.sf_sc.sc_edi;
548 	regs->tf_rsi    = frame.sf_sc.sc_esi;
549 	regs->tf_rbp    = frame.sf_sc.sc_ebp;
550 	regs->tf_rbx    = frame.sf_sc.sc_ebx;
551 	regs->tf_rdx    = frame.sf_sc.sc_edx;
552 	regs->tf_rcx    = frame.sf_sc.sc_ecx;
553 	regs->tf_rax    = frame.sf_sc.sc_eax;
554 	regs->tf_rip    = frame.sf_sc.sc_eip;
555 	regs->tf_cs     = frame.sf_sc.sc_cs;
556 	regs->tf_ds     = frame.sf_sc.sc_ds;
557 	regs->tf_es     = frame.sf_sc.sc_es;
558 	regs->tf_fs     = frame.sf_sc.sc_fs;
559 	regs->tf_gs     = frame.sf_sc.sc_gs;
560 	regs->tf_rflags = eflags;
561 	regs->tf_rsp    = frame.sf_sc.sc_esp_at_signal;
562 	regs->tf_ss     = frame.sf_sc.sc_ss;
563 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
564 
565 	return (EJUSTRETURN);
566 }
567 
568 /*
569  * System call to cleanup state after a signal
570  * has been taken.  Reset signal mask and
571  * stack state from context left by rt_sendsig (above).
572  * Return to previous pc and psl as specified by
573  * context left by sendsig. Check carefully to
574  * make sure that the user has not modified the
575  * psl to gain improper privileges or to cause
576  * a machine fault.
577  */
578 int
579 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
580 {
581 	struct l_ucontext uc;
582 	struct l_sigcontext *context;
583 	sigset_t bmask;
584 	l_stack_t *lss;
585 	stack_t ss;
586 	struct trapframe *regs;
587 	int eflags;
588 	ksiginfo_t ksi;
589 
590 	regs = td->td_frame;
591 
592 	/*
593 	 * The trampoline code hands us the ucontext.
594 	 * It is unsafe to keep track of it ourselves, in the event that a
595 	 * program jumps out of a signal handler.
596 	 */
597 	if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
598 		return (EFAULT);
599 
600 	context = &uc.uc_mcontext;
601 
602 	/* Check for security violations. */
603 #define	EFLAGS_SECURE(ef, oef)	((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
604 	eflags = context->sc_eflags;
605 	if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
606 		return(EINVAL);
607 
608 	/*
609 	 * Don't allow users to load a valid privileged %cs.  Let the
610 	 * hardware check for invalid selectors, excess privilege in
611 	 * other selectors, invalid %eip's and invalid %esp's.
612 	 */
613 #define	CS_SECURE(cs)	(ISPL(cs) == SEL_UPL)
614 	if (!CS_SECURE(context->sc_cs)) {
615 		ksiginfo_init_trap(&ksi);
616 		ksi.ksi_signo = SIGBUS;
617 		ksi.ksi_code = BUS_OBJERR;
618 		ksi.ksi_trapno = T_PROTFLT;
619 		ksi.ksi_addr = (void *)regs->tf_rip;
620 		trapsignal(td, &ksi);
621 		return(EINVAL);
622 	}
623 
624 	linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
625 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
626 
627 	/*
628 	 * Restore signal context
629 	 */
630 	regs->tf_gs	= context->sc_gs;
631 	regs->tf_fs	= context->sc_fs;
632 	regs->tf_es	= context->sc_es;
633 	regs->tf_ds	= context->sc_ds;
634 	regs->tf_rdi    = context->sc_edi;
635 	regs->tf_rsi    = context->sc_esi;
636 	regs->tf_rbp    = context->sc_ebp;
637 	regs->tf_rbx    = context->sc_ebx;
638 	regs->tf_rdx    = context->sc_edx;
639 	regs->tf_rcx    = context->sc_ecx;
640 	regs->tf_rax    = context->sc_eax;
641 	regs->tf_rip    = context->sc_eip;
642 	regs->tf_cs     = context->sc_cs;
643 	regs->tf_rflags = eflags;
644 	regs->tf_rsp    = context->sc_esp_at_signal;
645 	regs->tf_ss     = context->sc_ss;
646 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
647 
648 	/*
649 	 * call sigaltstack & ignore results..
650 	 */
651 	lss = &uc.uc_stack;
652 	ss.ss_sp = PTRIN(lss->ss_sp);
653 	ss.ss_size = lss->ss_size;
654 	ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
655 
656 	(void)kern_sigaltstack(td, &ss, NULL);
657 
658 	return (EJUSTRETURN);
659 }
660 
661 static int
662 linux32_fetch_syscall_args(struct thread *td)
663 {
664 	struct proc *p;
665 	struct trapframe *frame;
666 	struct syscall_args *sa;
667 
668 	p = td->td_proc;
669 	frame = td->td_frame;
670 	sa = &td->td_sa;
671 
672 	sa->args[0] = frame->tf_rbx;
673 	sa->args[1] = frame->tf_rcx;
674 	sa->args[2] = frame->tf_rdx;
675 	sa->args[3] = frame->tf_rsi;
676 	sa->args[4] = frame->tf_rdi;
677 	sa->args[5] = frame->tf_rbp;	/* Unconfirmed */
678 	sa->code = frame->tf_rax;
679 	sa->original_code = sa->code;
680 
681 	if (sa->code >= p->p_sysent->sv_size)
682 		/* nosys */
683 		sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
684 	else
685 		sa->callp = &p->p_sysent->sv_table[sa->code];
686 
687 	td->td_retval[0] = 0;
688 	td->td_retval[1] = frame->tf_rdx;
689 
690 	return (0);
691 }
692 
693 static void
694 linux32_set_syscall_retval(struct thread *td, int error)
695 {
696 	struct trapframe *frame = td->td_frame;
697 
698 	cpu_set_syscall_retval(td, error);
699 
700 	if (__predict_false(error != 0)) {
701 		if (error != ERESTART && error != EJUSTRETURN)
702 			frame->tf_rax = bsd_to_linux_errno(error);
703 	}
704 }
705 
706 /*
707  * Clear registers on exec
708  * XXX copied from ia32_signal.c.
709  */
710 static void
711 linux_exec_setregs(struct thread *td, struct image_params *imgp,
712     uintptr_t stack)
713 {
714 	struct trapframe *regs = td->td_frame;
715 	struct pcb *pcb = td->td_pcb;
716 	register_t saved_rflags;
717 
718 	regs = td->td_frame;
719 	pcb = td->td_pcb;
720 
721 	if (td->td_proc->p_md.md_ldt != NULL)
722 		user_ldt_free(td);
723 
724 	critical_enter();
725 	wrmsr(MSR_FSBASE, 0);
726 	wrmsr(MSR_KGSBASE, 0);	/* User value while we're in the kernel */
727 	pcb->pcb_fsbase = 0;
728 	pcb->pcb_gsbase = 0;
729 	critical_exit();
730 	pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
731 
732 	saved_rflags = regs->tf_rflags & PSL_T;
733 	bzero((char *)regs, sizeof(struct trapframe));
734 	regs->tf_rip = imgp->entry_addr;
735 	regs->tf_rsp = stack;
736 	regs->tf_rflags = PSL_USER | saved_rflags;
737 	regs->tf_gs = _ugssel;
738 	regs->tf_fs = _ufssel;
739 	regs->tf_es = _udatasel;
740 	regs->tf_ds = _udatasel;
741 	regs->tf_ss = _udatasel;
742 	regs->tf_flags = TF_HASSEGS;
743 	regs->tf_cs = _ucode32sel;
744 	regs->tf_rbx = (register_t)imgp->ps_strings;
745 
746 	x86_clear_dbregs(pcb);
747 
748 	fpstate_drop(td);
749 
750 	/* Do full restore on return so that we can change to a different %cs */
751 	set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
752 }
753 
754 /*
755  * XXX copied from ia32_sysvec.c.
756  */
757 static int
758 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
759 {
760 	int argc, envc, error;
761 	u_int32_t *vectp;
762 	char *stringp;
763 	uintptr_t destp, ustringp;
764 	struct linux32_ps_strings *arginfo;
765 	char canary[LINUX_AT_RANDOM_LEN];
766 	size_t execpath_len;
767 
768 	/* Calculate string base and vector table pointers. */
769 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
770 		execpath_len = strlen(imgp->execpath) + 1;
771 	else
772 		execpath_len = 0;
773 
774 	arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
775 	destp = (uintptr_t)arginfo;
776 
777 	if (execpath_len != 0) {
778 		destp -= execpath_len;
779 		destp = rounddown2(destp, sizeof(uint32_t));
780 		imgp->execpathp = (void *)destp;
781 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
782 		if (error != 0)
783 			return (error);
784 	}
785 
786 	/* Prepare the canary for SSP. */
787 	arc4rand(canary, sizeof(canary), 0);
788 	destp -= roundup(sizeof(canary), sizeof(uint32_t));
789 	imgp->canary = (void *)destp;
790 	error = copyout(canary, imgp->canary, sizeof(canary));
791 	if (error != 0)
792 		return (error);
793 
794 	/* Allocate room for the argument and environment strings. */
795 	destp -= ARG_MAX - imgp->args->stringspace;
796 	destp = rounddown2(destp, sizeof(uint32_t));
797 	ustringp = destp;
798 
799 	if (imgp->auxargs) {
800 		/*
801 		 * Allocate room on the stack for the ELF auxargs
802 		 * array.  It has LINUX_AT_COUNT entries.
803 		 */
804 		destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
805 		destp = rounddown2(destp, sizeof(uint32_t));
806 	}
807 
808 	vectp = (uint32_t *)destp;
809 
810 	/*
811 	 * Allocate room for the argv[] and env vectors including the
812 	 * terminating NULL pointers.
813 	 */
814 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
815 
816 	/* vectp also becomes our initial stack base. */
817 	*stack_base = (uintptr_t)vectp;
818 
819 	stringp = imgp->args->begin_argv;
820 	argc = imgp->args->argc;
821 	envc = imgp->args->envc;
822 
823 	/* Copy out strings - arguments and environment. */
824 	error = copyout(stringp, (void *)ustringp,
825 	    ARG_MAX - imgp->args->stringspace);
826 	if (error != 0)
827 		return (error);
828 
829 	/* Fill in "ps_strings" struct for ps, w, etc. */
830 	if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 ||
831 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
832 		return (EFAULT);
833 
834 	/* Fill in argument portion of vector table. */
835 	for (; argc > 0; --argc) {
836 		if (suword32(vectp++, ustringp) != 0)
837 			return (EFAULT);
838 		while (*stringp++ != 0)
839 			ustringp++;
840 		ustringp++;
841 	}
842 
843 	/* A null vector table pointer separates the argp's from the envp's. */
844 	if (suword32(vectp++, 0) != 0)
845 		return (EFAULT);
846 
847 	if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 ||
848 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
849 		return (EFAULT);
850 
851 	/* Fill in environment portion of vector table. */
852 	for (; envc > 0; --envc) {
853 		if (suword32(vectp++, ustringp) != 0)
854 			return (EFAULT);
855 		while (*stringp++ != 0)
856 			ustringp++;
857 		ustringp++;
858 	}
859 
860 	/* The end of the vector table is a null pointer. */
861 	if (suword32(vectp, 0) != 0)
862 		return (EFAULT);
863 
864 	if (imgp->auxargs) {
865 		vectp++;
866 		error = imgp->sysent->sv_copyout_auxargs(imgp,
867 		    (uintptr_t)vectp);
868 		if (error != 0)
869 			return (error);
870 	}
871 
872 	return (0);
873 }
874 
875 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
876     "32-bit Linux emulation");
877 
878 static u_long	linux32_maxdsiz = LINUX32_MAXDSIZ;
879 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
880     &linux32_maxdsiz, 0, "");
881 static u_long	linux32_maxssiz = LINUX32_MAXSSIZ;
882 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
883     &linux32_maxssiz, 0, "");
884 static u_long	linux32_maxvmem = LINUX32_MAXVMEM;
885 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
886     &linux32_maxvmem, 0, "");
887 
888 static void
889 linux32_fixlimit(struct rlimit *rl, int which)
890 {
891 
892 	switch (which) {
893 	case RLIMIT_DATA:
894 		if (linux32_maxdsiz != 0) {
895 			if (rl->rlim_cur > linux32_maxdsiz)
896 				rl->rlim_cur = linux32_maxdsiz;
897 			if (rl->rlim_max > linux32_maxdsiz)
898 				rl->rlim_max = linux32_maxdsiz;
899 		}
900 		break;
901 	case RLIMIT_STACK:
902 		if (linux32_maxssiz != 0) {
903 			if (rl->rlim_cur > linux32_maxssiz)
904 				rl->rlim_cur = linux32_maxssiz;
905 			if (rl->rlim_max > linux32_maxssiz)
906 				rl->rlim_max = linux32_maxssiz;
907 		}
908 		break;
909 	case RLIMIT_VMEM:
910 		if (linux32_maxvmem != 0) {
911 			if (rl->rlim_cur > linux32_maxvmem)
912 				rl->rlim_cur = linux32_maxvmem;
913 			if (rl->rlim_max > linux32_maxvmem)
914 				rl->rlim_max = linux32_maxvmem;
915 		}
916 		break;
917 	}
918 }
919 
920 struct sysentvec elf_linux_sysvec = {
921 	.sv_size	= LINUX32_SYS_MAXSYSCALL,
922 	.sv_table	= linux32_sysent,
923 	.sv_transtrap	= linux_translate_traps,
924 	.sv_fixup	= linux_fixup_elf,
925 	.sv_sendsig	= linux_sendsig,
926 	.sv_sigcode	= &_binary_linux32_vdso_so_o_start,
927 	.sv_szsigcode	= &linux_szsigcode,
928 	.sv_name	= "Linux ELF32",
929 	.sv_coredump	= elf32_coredump,
930 	.sv_elf_core_osabi = ELFOSABI_NONE,
931 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
932 	.sv_elf_core_prepare_notes = linux32_prepare_notes,
933 	.sv_imgact_try	= linux_exec_imgact_try,
934 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
935 	.sv_minuser	= VM_MIN_ADDRESS,
936 	.sv_maxuser	= LINUX32_MAXUSER,
937 	.sv_usrstack	= LINUX32_USRSTACK,
938 	.sv_psstrings	= LINUX32_PS_STRINGS,
939 	.sv_stackprot	= VM_PROT_ALL,
940 	.sv_copyout_auxargs = linux_copyout_auxargs,
941 	.sv_copyout_strings = linux_copyout_strings,
942 	.sv_setregs	= linux_exec_setregs,
943 	.sv_fixlimit	= linux32_fixlimit,
944 	.sv_maxssiz	= &linux32_maxssiz,
945 	.sv_flags	= SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP |
946 	    SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
947 	.sv_set_syscall_retval = linux32_set_syscall_retval,
948 	.sv_fetch_syscall_args = linux32_fetch_syscall_args,
949 	.sv_syscallnames = NULL,
950 	.sv_shared_page_base = LINUX32_SHAREDPAGE,
951 	.sv_shared_page_len = PAGE_SIZE,
952 	.sv_schedtail	= linux_schedtail,
953 	.sv_thread_detach = linux_thread_detach,
954 	.sv_trap	= NULL,
955 	.sv_onexec	= linux_on_exec_vmspace,
956 	.sv_onexit	= linux_on_exit,
957 	.sv_ontdexit	= linux_thread_dtor,
958 	.sv_setid_allowed = &linux_setid_allowed_query,
959 };
960 
961 static int
962 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
963 {
964 	int error;
965 
966 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
967 	    LINUX32_VDSOPAGE_SIZE, imgp);
968 	if (error == 0)
969 		linux_on_exec(p, imgp);
970 	return (error);
971 }
972 
973 /*
974  * linux_vdso_install() and linux_exec_sysvec_init() must be called
975  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
976  */
977 static void
978 linux_exec_sysvec_init(void *param)
979 {
980 	l_uintptr_t *ktimekeep_base, *ktsc_selector;
981 	struct sysentvec *sv;
982 	ptrdiff_t tkoff;
983 
984 	sv = param;
985 	/* Fill timekeep_base */
986 	exec_sysvec_init(sv);
987 
988 	tkoff = kern_timekeep_base - linux_vdso_base;
989 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
990 	*ktimekeep_base = sv->sv_timekeep_base;
991 
992 	tkoff = kern_tsc_selector - linux_vdso_base;
993 	ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
994 	*ktsc_selector = linux_vdso_tsc_selector_idx();
995 	if (bootverbose)
996 		printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
997 }
998 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
999     linux_exec_sysvec_init, &elf_linux_sysvec);
1000 
1001 static void
1002 linux_vdso_install(const void *param)
1003 {
1004 	char *vdso_start = &_binary_linux32_vdso_so_o_start;
1005 	char *vdso_end = &_binary_linux32_vdso_so_o_end;
1006 
1007 	linux_szsigcode = vdso_end - vdso_start;
1008 	MPASS(linux_szsigcode <= LINUX32_VDSOPAGE_SIZE);
1009 
1010 	linux_vdso_base = LINUX32_VDSOPAGE;
1011 
1012 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
1013 
1014 	linux_vdso_obj = __elfN(linux_shared_page_init)
1015 	    (&linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
1016 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
1017 
1018 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
1019 }
1020 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
1021     linux_vdso_install, NULL);
1022 
1023 static void
1024 linux_vdso_deinstall(const void *param)
1025 {
1026 
1027 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
1028 	    linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
1029 }
1030 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1031     linux_vdso_deinstall, NULL);
1032 
1033 static void
1034 linux_vdso_reloc(char *mapping, Elf_Addr offset)
1035 {
1036 	const Elf_Shdr *shdr;
1037 	const Elf_Rel *rel;
1038 	const Elf_Ehdr *ehdr;
1039 	Elf32_Addr *where;
1040 	Elf_Size rtype, symidx;
1041 	Elf32_Addr addr, addend;
1042 	int i, relcnt;
1043 
1044 	MPASS(offset != 0);
1045 
1046 	relcnt = 0;
1047 	ehdr = (const Elf_Ehdr *)mapping;
1048 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
1049 	for (i = 0; i < ehdr->e_shnum; i++)
1050 	{
1051 		switch (shdr[i].sh_type) {
1052 		case SHT_REL:
1053 			rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
1054 			relcnt = shdr[i].sh_size / sizeof(*rel);
1055 			break;
1056 		case SHT_RELA:
1057 			printf("Linux i386 vDSO: unexpected Rela section\n");
1058 			break;
1059 		}
1060 	}
1061 
1062 	for (i = 0; i < relcnt; i++, rel++) {
1063 		where = (Elf32_Addr *)(mapping + rel->r_offset);
1064 		addend = *where;
1065 		rtype = ELF_R_TYPE(rel->r_info);
1066 		symidx = ELF_R_SYM(rel->r_info);
1067 
1068 		switch (rtype) {
1069 		case R_386_NONE:	/* none */
1070 			break;
1071 
1072 		case R_386_RELATIVE:	/* B + A */
1073 			addr = (Elf32_Addr)PTROUT(offset + addend);
1074 			if (*where != addr)
1075 				*where = addr;
1076 			break;
1077 
1078 		case R_386_IRELATIVE:
1079 			printf("Linux i386 vDSO: unexpected ifunc relocation, "
1080 			    "symbol index %ld\n", (intmax_t)symidx);
1081 			break;
1082 		default:
1083 			printf("Linux i386 vDSO: unexpected relocation type %ld, "
1084 			    "symbol index %ld\n", (intmax_t)rtype, (intmax_t)symidx);
1085 		}
1086 	}
1087 }
1088 
1089 static char GNU_ABI_VENDOR[] = "GNU";
1090 static int GNULINUX_ABI_DESC = 0;
1091 
1092 static bool
1093 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
1094 {
1095 	const Elf32_Word *desc;
1096 	uintptr_t p;
1097 
1098 	p = (uintptr_t)(note + 1);
1099 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1100 
1101 	desc = (const Elf32_Word *)p;
1102 	if (desc[0] != GNULINUX_ABI_DESC)
1103 		return (false);
1104 
1105 	/*
1106 	 * For Linux we encode osrel using the Linux convention of
1107 	 * 	(version << 16) | (major << 8) | (minor)
1108 	 * See macro in linux_mib.h
1109 	 */
1110 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
1111 
1112 	return (true);
1113 }
1114 
1115 static Elf_Brandnote linux32_brandnote = {
1116 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
1117 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
1118 	.hdr.n_type	= 1,
1119 	.vendor		= GNU_ABI_VENDOR,
1120 	.flags		= BN_TRANSLATE_OSREL,
1121 	.trans_osrel	= linux32_trans_osrel
1122 };
1123 
1124 static Elf32_Brandinfo linux_brand = {
1125 	.brand		= ELFOSABI_LINUX,
1126 	.machine	= EM_386,
1127 	.compat_3_brand	= "Linux",
1128 	.emul_path	= linux_emul_path,
1129 	.interp_path	= "/lib/ld-linux.so.1",
1130 	.sysvec		= &elf_linux_sysvec,
1131 	.interp_newpath	= NULL,
1132 	.brand_note	= &linux32_brandnote,
1133 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1134 };
1135 
1136 static Elf32_Brandinfo linux_glibc2brand = {
1137 	.brand		= ELFOSABI_LINUX,
1138 	.machine	= EM_386,
1139 	.compat_3_brand	= "Linux",
1140 	.emul_path	= linux_emul_path,
1141 	.interp_path	= "/lib/ld-linux.so.2",
1142 	.sysvec		= &elf_linux_sysvec,
1143 	.interp_newpath	= NULL,
1144 	.brand_note	= &linux32_brandnote,
1145 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1146 };
1147 
1148 static Elf32_Brandinfo linux_muslbrand = {
1149 	.brand		= ELFOSABI_LINUX,
1150 	.machine	= EM_386,
1151 	.compat_3_brand	= "Linux",
1152 	.emul_path	= linux_emul_path,
1153 	.interp_path	= "/lib/ld-musl-i386.so.1",
1154 	.sysvec		= &elf_linux_sysvec,
1155 	.interp_newpath	= NULL,
1156 	.brand_note	= &linux32_brandnote,
1157 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
1158 			    LINUX_BI_FUTEX_REQUEUE
1159 };
1160 
1161 Elf32_Brandinfo *linux_brandlist[] = {
1162 	&linux_brand,
1163 	&linux_glibc2brand,
1164 	&linux_muslbrand,
1165 	NULL
1166 };
1167 
1168 static int
1169 linux_elf_modevent(module_t mod, int type, void *data)
1170 {
1171 	Elf32_Brandinfo **brandinfo;
1172 	int error;
1173 	struct linux_ioctl_handler **lihp;
1174 
1175 	error = 0;
1176 
1177 	switch(type) {
1178 	case MOD_LOAD:
1179 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1180 		     ++brandinfo)
1181 			if (elf32_insert_brand_entry(*brandinfo) < 0)
1182 				error = EINVAL;
1183 		if (error == 0) {
1184 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1185 				linux32_ioctl_register_handler(*lihp);
1186 			stclohz = (stathz ? stathz : hz);
1187 			if (bootverbose)
1188 				printf("Linux i386 ELF exec handler installed\n");
1189 		} else
1190 			printf("cannot insert Linux i386 ELF brand handler\n");
1191 		break;
1192 	case MOD_UNLOAD:
1193 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1194 		     ++brandinfo)
1195 			if (elf32_brand_inuse(*brandinfo))
1196 				error = EBUSY;
1197 		if (error == 0) {
1198 			for (brandinfo = &linux_brandlist[0];
1199 			     *brandinfo != NULL; ++brandinfo)
1200 				if (elf32_remove_brand_entry(*brandinfo) < 0)
1201 					error = EINVAL;
1202 		}
1203 		if (error == 0) {
1204 			SET_FOREACH(lihp, linux_ioctl_handler_set)
1205 				linux32_ioctl_unregister_handler(*lihp);
1206 			if (bootverbose)
1207 				printf("Linux i386 ELF exec handler removed\n");
1208 		} else
1209 			printf("Could not deinstall Linux i386 ELF interpreter entry\n");
1210 		break;
1211 	default:
1212 		return (EOPNOTSUPP);
1213 	}
1214 	return (error);
1215 }
1216 
1217 static moduledata_t linux_elf_mod = {
1218 	"linuxelf",
1219 	linux_elf_modevent,
1220 	0
1221 };
1222 
1223 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1224 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1225 FEATURE(linux, "Linux 32bit support");
1226