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 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/imgact_aout.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/signalvar.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysctl.h>
48 #include <sys/sysent.h>
49 #include <sys/sysproto.h>
50 #include <sys/vnode.h>
51
52 #include <vm/vm.h>
53 #include <vm/pmap.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_param.h>
59
60 #include <machine/cpu.h>
61 #include <machine/cputypes.h>
62 #include <machine/md_var.h>
63 #include <machine/pcb.h>
64 #include <machine/trap.h>
65
66 #include <i386/linux/linux.h>
67 #include <i386/linux/linux_proto.h>
68 #include <compat/linux/linux_emul.h>
69 #include <compat/linux/linux_ioctl.h>
70 #include <compat/linux/linux_mib.h>
71 #include <compat/linux/linux_misc.h>
72 #include <compat/linux/linux_signal.h>
73 #include <compat/linux/linux_util.h>
74 #include <compat/linux/linux_vdso.h>
75
76 MODULE_VERSION(linux, 1);
77
78 #define LINUX_PS_STRINGS (LINUX_USRSTACK - sizeof(struct ps_strings))
79
80 static int linux_szsigcode;
81 static vm_object_t linux_shared_page_obj;
82 static char *linux_shared_page_mapping;
83 extern char _binary_linux_locore_o_start;
84 extern char _binary_linux_locore_o_end;
85
86 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
87
88 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
89
90 static int linux_fixup(uintptr_t *stack_base,
91 struct image_params *iparams);
92 static int linux_fixup_elf(uintptr_t *stack_base,
93 struct image_params *iparams);
94 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
95 static void linux_exec_setregs(struct thread *td,
96 struct image_params *imgp, uintptr_t stack);
97 static int linux_copyout_strings(struct image_params *imgp,
98 uintptr_t *stack_base);
99 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static void linux_vdso_install(void *param);
101 static void linux_vdso_deinstall(void *param);
102
103 static int linux_szplatform;
104 const char *linux_kplatform;
105
106 #define LINUX_T_UNKNOWN 255
107 static int _bsd_to_linux_trapcode[] = {
108 LINUX_T_UNKNOWN, /* 0 */
109 6, /* 1 T_PRIVINFLT */
110 LINUX_T_UNKNOWN, /* 2 */
111 3, /* 3 T_BPTFLT */
112 LINUX_T_UNKNOWN, /* 4 */
113 LINUX_T_UNKNOWN, /* 5 */
114 16, /* 6 T_ARITHTRAP */
115 254, /* 7 T_ASTFLT */
116 LINUX_T_UNKNOWN, /* 8 */
117 13, /* 9 T_PROTFLT */
118 1, /* 10 T_TRCTRAP */
119 LINUX_T_UNKNOWN, /* 11 */
120 14, /* 12 T_PAGEFLT */
121 LINUX_T_UNKNOWN, /* 13 */
122 17, /* 14 T_ALIGNFLT */
123 LINUX_T_UNKNOWN, /* 15 */
124 LINUX_T_UNKNOWN, /* 16 */
125 LINUX_T_UNKNOWN, /* 17 */
126 0, /* 18 T_DIVIDE */
127 2, /* 19 T_NMI */
128 4, /* 20 T_OFLOW */
129 5, /* 21 T_BOUND */
130 7, /* 22 T_DNA */
131 8, /* 23 T_DOUBLEFLT */
132 9, /* 24 T_FPOPFLT */
133 10, /* 25 T_TSSFLT */
134 11, /* 26 T_SEGNPFLT */
135 12, /* 27 T_STKFLT */
136 18, /* 28 T_MCHK */
137 19, /* 29 T_XMMFLT */
138 15 /* 30 T_RESERVED */
139 };
140 #define bsd_to_linux_trapcode(code) \
141 ((code)<nitems(_bsd_to_linux_trapcode)? \
142 _bsd_to_linux_trapcode[(code)]: \
143 LINUX_T_UNKNOWN)
144
145 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
146 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
147 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
148
149 /*
150 * If FreeBSD & Linux have a difference of opinion about what a trap
151 * means, deal with it here.
152 *
153 * MPSAFE
154 */
155 static int
linux_translate_traps(int signal,int trap_code)156 linux_translate_traps(int signal, int trap_code)
157 {
158 if (signal != SIGBUS)
159 return (signal);
160 switch (trap_code) {
161 case T_PROTFLT:
162 case T_TSSFLT:
163 case T_DOUBLEFLT:
164 case T_PAGEFLT:
165 return (SIGSEGV);
166 default:
167 return (signal);
168 }
169 }
170
171 static int
linux_fixup(uintptr_t * stack_base,struct image_params * imgp)172 linux_fixup(uintptr_t *stack_base, struct image_params *imgp)
173 {
174 register_t *base, *argv, *envp;
175
176 base = (register_t *)*stack_base;
177 argv = base;
178 envp = base + (imgp->args->argc + 1);
179 base--;
180 suword(base, (intptr_t)envp);
181 base--;
182 suword(base, (intptr_t)argv);
183 base--;
184 suword(base, imgp->args->argc);
185 *stack_base = (uintptr_t)base;
186 return (0);
187 }
188
189 static int
linux_copyout_auxargs(struct image_params * imgp,uintptr_t base)190 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
191 {
192 struct proc *p;
193 Elf32_Auxargs *args;
194 Elf32_Auxinfo *argarray, *pos;
195 Elf32_Addr *uplatform;
196 struct ps_strings *arginfo;
197 int error, issetugid;
198
199 p = imgp->proc;
200 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
201 arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
202 uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
203 args = (Elf32_Auxargs *)imgp->auxargs;
204 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
205 M_WAITOK | M_ZERO);
206
207 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
208 imgp->proc->p_sysent->sv_shared_page_base);
209 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
210 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
211
212 /*
213 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
214 * as it has appeared in the 2.4.0-rc7 first time.
215 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
216 * glibc falls back to the hard-coded CLK_TCK value when aux entry
217 * is not present.
218 * Also see linux_times() implementation.
219 */
220 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
221 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
222 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
223 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
224 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
225 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
226 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
227 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
228 AUXARGS_ENTRY(pos, AT_BASE, args->base);
229 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
230 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
231 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
232 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
233 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
234 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
235 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
236 if (imgp->execpathp != 0)
237 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
238 if (args->execfd != -1)
239 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
240 AUXARGS_ENTRY(pos, AT_NULL, 0);
241
242 free(imgp->auxargs, M_TEMP);
243 imgp->auxargs = NULL;
244 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
245
246 error = copyout(argarray, (void *)base,
247 sizeof(*argarray) * LINUX_AT_COUNT);
248 free(argarray, M_TEMP);
249 return (error);
250 }
251
252 static int
linux_fixup_elf(uintptr_t * stack_base,struct image_params * imgp)253 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
254 {
255 register_t *base;
256
257 base = (register_t *)*stack_base;
258 base--;
259 if (suword(base, (register_t)imgp->args->argc) == -1)
260 return (EFAULT);
261 *stack_base = (uintptr_t)base;
262 return (0);
263 }
264
265 /*
266 * Copied from kern/kern_exec.c
267 */
268 static int
linux_copyout_strings(struct image_params * imgp,uintptr_t * stack_base)269 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
270 {
271 int argc, envc, error;
272 char **vectp;
273 char *stringp;
274 uintptr_t destp, ustringp;
275 struct ps_strings *arginfo;
276 char canary[LINUX_AT_RANDOM_LEN];
277 size_t execpath_len;
278 struct proc *p;
279
280 p = imgp->proc;
281 arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
282 destp = (uintptr_t)arginfo;
283
284 /* Install LINUX_PLATFORM. */
285 destp -= linux_szplatform;
286 destp = rounddown2(destp, sizeof(void *));
287 error = copyout(linux_kplatform, (void *)destp, linux_szplatform);
288 if (error != 0)
289 return (error);
290
291 if (imgp->execpath != NULL && imgp->auxargs != NULL) {
292 execpath_len = strlen(imgp->execpath) + 1;
293 destp -= execpath_len;
294 destp = rounddown2(destp, sizeof(void *));
295 imgp->execpathp = (void *)destp;
296 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
297 if (error != 0)
298 return (error);
299 }
300
301 /* Prepare the canary for SSP. */
302 arc4rand(canary, sizeof(canary), 0);
303 destp -= roundup(sizeof(canary), sizeof(void *));
304 imgp->canary = (void *)destp;
305 error = copyout(canary, imgp->canary, sizeof(canary));
306 if (error != 0)
307 return (error);
308
309 /* Allocate room for the argument and environment strings. */
310 destp -= ARG_MAX - imgp->args->stringspace;
311 destp = rounddown2(destp, sizeof(void *));
312 ustringp = destp;
313
314 if (imgp->auxargs) {
315 /*
316 * Allocate room on the stack for the ELF auxargs
317 * array. It has LINUX_AT_COUNT entries.
318 */
319 destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
320 destp = rounddown2(destp, sizeof(void *));
321 }
322
323 vectp = (char **)destp;
324
325 /*
326 * Allocate room for the argv[] and env vectors including the
327 * terminating NULL pointers.
328 */
329 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
330
331 /* vectp also becomes our initial stack base. */
332 *stack_base = (uintptr_t)vectp;
333
334 stringp = imgp->args->begin_argv;
335 argc = imgp->args->argc;
336 envc = imgp->args->envc;
337
338 /* Copy out strings - arguments and environment. */
339 error = copyout(stringp, (void *)ustringp,
340 ARG_MAX - imgp->args->stringspace);
341 if (error != 0)
342 return (error);
343
344 /* Fill in "ps_strings" struct for ps, w, etc. */
345 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
346 suword(&arginfo->ps_nargvstr, argc) != 0)
347 return (EFAULT);
348
349 /* Fill in argument portion of vector table. */
350 for (; argc > 0; --argc) {
351 if (suword(vectp++, ustringp) != 0)
352 return (EFAULT);
353 while (*stringp++ != 0)
354 ustringp++;
355 ustringp++;
356 }
357
358 /* A null vector table pointer separates the argp's from the envp's. */
359 if (suword(vectp++, 0) != 0)
360 return (EFAULT);
361
362 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
363 suword(&arginfo->ps_nenvstr, envc) != 0)
364 return (EFAULT);
365
366 /* Fill in environment portion of vector table. */
367 for (; envc > 0; --envc) {
368 if (suword(vectp++, ustringp) != 0)
369 return (EFAULT);
370 while (*stringp++ != 0)
371 ustringp++;
372 ustringp++;
373 }
374
375 /* The end of the vector table is a null pointer. */
376 if (suword(vectp, 0) != 0)
377 return (EFAULT);
378
379 if (imgp->auxargs) {
380 vectp++;
381 error = imgp->sysent->sv_copyout_auxargs(imgp,
382 (uintptr_t)vectp);
383 if (error != 0)
384 return (error);
385 }
386
387 return (0);
388 }
389
390 static void
linux_rt_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)391 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
392 {
393 struct thread *td = curthread;
394 struct proc *p = td->td_proc;
395 struct sigacts *psp;
396 struct trapframe *regs;
397 struct l_rt_sigframe *fp, frame;
398 int sig, code;
399 int oonstack;
400
401 sig = ksi->ksi_signo;
402 code = ksi->ksi_code;
403 PROC_LOCK_ASSERT(p, MA_OWNED);
404 psp = p->p_sigacts;
405 mtx_assert(&psp->ps_mtx, MA_OWNED);
406 regs = td->td_frame;
407 oonstack = sigonstack(regs->tf_esp);
408
409 /* Allocate space for the signal handler context. */
410 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
411 SIGISMEMBER(psp->ps_sigonstack, sig)) {
412 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
413 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
414 } else
415 fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
416 mtx_unlock(&psp->ps_mtx);
417
418 /* Build the argument list for the signal handler. */
419 sig = bsd_to_linux_signal(sig);
420
421 bzero(&frame, sizeof(frame));
422
423 frame.sf_handler = catcher;
424 frame.sf_sig = sig;
425 frame.sf_siginfo = &fp->sf_si;
426 frame.sf_ucontext = &fp->sf_sc;
427
428 /* Fill in POSIX parts. */
429 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
430
431 /* Build the signal context to be used by sigreturn. */
432 frame.sf_sc.uc_flags = 0; /* XXX ??? */
433 frame.sf_sc.uc_link = NULL; /* XXX ??? */
434
435 frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
436 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
437 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
438 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
439 PROC_UNLOCK(p);
440
441 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
442
443 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask;
444 frame.sf_sc.uc_mcontext.sc_gs = rgs();
445 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
446 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
447 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
448 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_edi;
449 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_esi;
450 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_ebp;
451 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_ebx;
452 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_esp;
453 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_edx;
454 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_ecx;
455 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_eax;
456 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_eip;
457 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
458 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
459 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
460 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
461 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
462 frame.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
463 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
464
465 if (copyout(&frame, fp, sizeof(frame)) != 0) {
466 /*
467 * Process has trashed its stack; give it an illegal
468 * instruction to halt it in its tracks.
469 */
470 PROC_LOCK(p);
471 sigexit(td, SIGILL);
472 }
473
474 /* Build context to run handler in. */
475 regs->tf_esp = (int)fp;
476 regs->tf_eip = linux_rt_sigcode;
477 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
478 regs->tf_cs = _ucodesel;
479 regs->tf_ds = _udatasel;
480 regs->tf_es = _udatasel;
481 regs->tf_fs = _udatasel;
482 regs->tf_ss = _udatasel;
483 PROC_LOCK(p);
484 mtx_lock(&psp->ps_mtx);
485 }
486
487 /*
488 * Send an interrupt to process.
489 *
490 * Stack is set up to allow sigcode stored
491 * in u. to call routine, followed by kcall
492 * to sigreturn routine below. After sigreturn
493 * resets the signal mask, the stack, and the
494 * frame pointer, it returns to the user
495 * specified pc, psl.
496 */
497 static void
linux_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)498 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
499 {
500 struct thread *td = curthread;
501 struct proc *p = td->td_proc;
502 struct sigacts *psp;
503 struct trapframe *regs;
504 struct l_sigframe *fp, frame;
505 l_sigset_t lmask;
506 int sig, code;
507 int oonstack;
508
509 PROC_LOCK_ASSERT(p, MA_OWNED);
510 psp = p->p_sigacts;
511 sig = ksi->ksi_signo;
512 code = ksi->ksi_code;
513 mtx_assert(&psp->ps_mtx, MA_OWNED);
514 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
515 /* Signal handler installed with SA_SIGINFO. */
516 linux_rt_sendsig(catcher, ksi, mask);
517 return;
518 }
519 regs = td->td_frame;
520 oonstack = sigonstack(regs->tf_esp);
521
522 /* Allocate space for the signal handler context. */
523 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
524 SIGISMEMBER(psp->ps_sigonstack, sig)) {
525 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
526 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
527 } else
528 fp = (struct l_sigframe *)regs->tf_esp - 1;
529 mtx_unlock(&psp->ps_mtx);
530 PROC_UNLOCK(p);
531
532 /* Build the argument list for the signal handler. */
533 sig = bsd_to_linux_signal(sig);
534
535 bzero(&frame, sizeof(frame));
536
537 frame.sf_handler = catcher;
538 frame.sf_sig = sig;
539
540 bsd_to_linux_sigset(mask, &lmask);
541
542 /* Build the signal context to be used by sigreturn. */
543 frame.sf_sc.sc_mask = lmask.__mask;
544 frame.sf_sc.sc_gs = rgs();
545 frame.sf_sc.sc_fs = regs->tf_fs;
546 frame.sf_sc.sc_es = regs->tf_es;
547 frame.sf_sc.sc_ds = regs->tf_ds;
548 frame.sf_sc.sc_edi = regs->tf_edi;
549 frame.sf_sc.sc_esi = regs->tf_esi;
550 frame.sf_sc.sc_ebp = regs->tf_ebp;
551 frame.sf_sc.sc_ebx = regs->tf_ebx;
552 frame.sf_sc.sc_esp = regs->tf_esp;
553 frame.sf_sc.sc_edx = regs->tf_edx;
554 frame.sf_sc.sc_ecx = regs->tf_ecx;
555 frame.sf_sc.sc_eax = regs->tf_eax;
556 frame.sf_sc.sc_eip = regs->tf_eip;
557 frame.sf_sc.sc_cs = regs->tf_cs;
558 frame.sf_sc.sc_eflags = regs->tf_eflags;
559 frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
560 frame.sf_sc.sc_ss = regs->tf_ss;
561 frame.sf_sc.sc_err = regs->tf_err;
562 frame.sf_sc.sc_cr2 = (register_t)ksi->ksi_addr;
563 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
564
565 frame.sf_extramask[0] = lmask.__mask;
566
567 if (copyout(&frame, fp, sizeof(frame)) != 0) {
568 /*
569 * Process has trashed its stack; give it an illegal
570 * instruction to halt it in its tracks.
571 */
572 PROC_LOCK(p);
573 sigexit(td, SIGILL);
574 }
575
576 /* Build context to run handler in. */
577 regs->tf_esp = (int)fp;
578 regs->tf_eip = linux_sigcode;
579 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
580 regs->tf_cs = _ucodesel;
581 regs->tf_ds = _udatasel;
582 regs->tf_es = _udatasel;
583 regs->tf_fs = _udatasel;
584 regs->tf_ss = _udatasel;
585 PROC_LOCK(p);
586 mtx_lock(&psp->ps_mtx);
587 }
588
589 /*
590 * System call to cleanup state after a signal
591 * has been taken. Reset signal mask and
592 * stack state from context left by sendsig (above).
593 * Return to previous pc and psl as specified by
594 * context left by sendsig. Check carefully to
595 * make sure that the user has not modified the
596 * psl to gain improper privileges or to cause
597 * a machine fault.
598 */
599 int
linux_sigreturn(struct thread * td,struct linux_sigreturn_args * args)600 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
601 {
602 struct l_sigframe frame;
603 struct trapframe *regs;
604 l_sigset_t lmask;
605 sigset_t bmask;
606 int eflags;
607 ksiginfo_t ksi;
608
609 regs = td->td_frame;
610
611 /*
612 * The trampoline code hands us the sigframe.
613 * It is unsafe to keep track of it ourselves, in the event that a
614 * program jumps out of a signal handler.
615 */
616 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
617 return (EFAULT);
618
619 /* Check for security violations. */
620 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
621 eflags = frame.sf_sc.sc_eflags;
622 if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
623 return (EINVAL);
624
625 /*
626 * Don't allow users to load a valid privileged %cs. Let the
627 * hardware check for invalid selectors, excess privilege in
628 * other selectors, invalid %eip's and invalid %esp's.
629 */
630 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
631 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
632 ksiginfo_init_trap(&ksi);
633 ksi.ksi_signo = SIGBUS;
634 ksi.ksi_code = BUS_OBJERR;
635 ksi.ksi_trapno = T_PROTFLT;
636 ksi.ksi_addr = (void *)regs->tf_eip;
637 trapsignal(td, &ksi);
638 return (EINVAL);
639 }
640
641 lmask.__mask = frame.sf_sc.sc_mask;
642 linux_to_bsd_sigset(&lmask, &bmask);
643 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
644
645 /* Restore signal context. */
646 /* %gs was restored by the trampoline. */
647 regs->tf_fs = frame.sf_sc.sc_fs;
648 regs->tf_es = frame.sf_sc.sc_es;
649 regs->tf_ds = frame.sf_sc.sc_ds;
650 regs->tf_edi = frame.sf_sc.sc_edi;
651 regs->tf_esi = frame.sf_sc.sc_esi;
652 regs->tf_ebp = frame.sf_sc.sc_ebp;
653 regs->tf_ebx = frame.sf_sc.sc_ebx;
654 regs->tf_edx = frame.sf_sc.sc_edx;
655 regs->tf_ecx = frame.sf_sc.sc_ecx;
656 regs->tf_eax = frame.sf_sc.sc_eax;
657 regs->tf_eip = frame.sf_sc.sc_eip;
658 regs->tf_cs = frame.sf_sc.sc_cs;
659 regs->tf_eflags = eflags;
660 regs->tf_esp = frame.sf_sc.sc_esp_at_signal;
661 regs->tf_ss = frame.sf_sc.sc_ss;
662
663 return (EJUSTRETURN);
664 }
665
666 /*
667 * System call to cleanup state after a signal
668 * has been taken. Reset signal mask and
669 * stack state from context left by rt_sendsig (above).
670 * Return to previous pc and psl as specified by
671 * context left by sendsig. Check carefully to
672 * make sure that the user has not modified the
673 * psl to gain improper privileges or to cause
674 * a machine fault.
675 */
676 int
linux_rt_sigreturn(struct thread * td,struct linux_rt_sigreturn_args * args)677 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
678 {
679 struct l_ucontext uc;
680 struct l_sigcontext *context;
681 sigset_t bmask;
682 l_stack_t *lss;
683 stack_t ss;
684 struct trapframe *regs;
685 int eflags;
686 ksiginfo_t ksi;
687
688 regs = td->td_frame;
689
690 /*
691 * The trampoline code hands us the ucontext.
692 * It is unsafe to keep track of it ourselves, in the event that a
693 * program jumps out of a signal handler.
694 */
695 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
696 return (EFAULT);
697
698 context = &uc.uc_mcontext;
699
700 /* Check for security violations. */
701 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
702 eflags = context->sc_eflags;
703 if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
704 return (EINVAL);
705
706 /*
707 * Don't allow users to load a valid privileged %cs. Let the
708 * hardware check for invalid selectors, excess privilege in
709 * other selectors, invalid %eip's and invalid %esp's.
710 */
711 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
712 if (!CS_SECURE(context->sc_cs)) {
713 ksiginfo_init_trap(&ksi);
714 ksi.ksi_signo = SIGBUS;
715 ksi.ksi_code = BUS_OBJERR;
716 ksi.ksi_trapno = T_PROTFLT;
717 ksi.ksi_addr = (void *)regs->tf_eip;
718 trapsignal(td, &ksi);
719 return (EINVAL);
720 }
721
722 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
723 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
724
725 /* Restore signal context. */
726 /* %gs was restored by the trampoline. */
727 regs->tf_fs = context->sc_fs;
728 regs->tf_es = context->sc_es;
729 regs->tf_ds = context->sc_ds;
730 regs->tf_edi = context->sc_edi;
731 regs->tf_esi = context->sc_esi;
732 regs->tf_ebp = context->sc_ebp;
733 regs->tf_ebx = context->sc_ebx;
734 regs->tf_edx = context->sc_edx;
735 regs->tf_ecx = context->sc_ecx;
736 regs->tf_eax = context->sc_eax;
737 regs->tf_eip = context->sc_eip;
738 regs->tf_cs = context->sc_cs;
739 regs->tf_eflags = eflags;
740 regs->tf_esp = context->sc_esp_at_signal;
741 regs->tf_ss = context->sc_ss;
742
743 /* Call sigaltstack & ignore results. */
744 lss = &uc.uc_stack;
745 ss.ss_sp = lss->ss_sp;
746 ss.ss_size = lss->ss_size;
747 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
748
749 (void)kern_sigaltstack(td, &ss, NULL);
750
751 return (EJUSTRETURN);
752 }
753
754 static int
linux_fetch_syscall_args(struct thread * td)755 linux_fetch_syscall_args(struct thread *td)
756 {
757 struct proc *p;
758 struct trapframe *frame;
759 struct syscall_args *sa;
760
761 p = td->td_proc;
762 frame = td->td_frame;
763 sa = &td->td_sa;
764
765 sa->code = frame->tf_eax;
766 sa->args[0] = frame->tf_ebx;
767 sa->args[1] = frame->tf_ecx;
768 sa->args[2] = frame->tf_edx;
769 sa->args[3] = frame->tf_esi;
770 sa->args[4] = frame->tf_edi;
771 sa->args[5] = frame->tf_ebp; /* Unconfirmed */
772
773 if (sa->code >= p->p_sysent->sv_size)
774 /* nosys */
775 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
776 else
777 sa->callp = &p->p_sysent->sv_table[sa->code];
778
779 td->td_retval[0] = 0;
780 td->td_retval[1] = frame->tf_edx;
781
782 return (0);
783 }
784
785 static void
linux_set_syscall_retval(struct thread * td,int error)786 linux_set_syscall_retval(struct thread *td, int error)
787 {
788 struct trapframe *frame = td->td_frame;
789
790 cpu_set_syscall_retval(td, error);
791
792 if (__predict_false(error != 0)) {
793 if (error != ERESTART && error != EJUSTRETURN)
794 frame->tf_eax = bsd_to_linux_errno(error);
795 }
796 }
797
798 /*
799 * exec_setregs may initialize some registers differently than Linux
800 * does, thus potentially confusing Linux binaries. If necessary, we
801 * override the exec_setregs default(s) here.
802 */
803 static void
linux_exec_setregs(struct thread * td,struct image_params * imgp,uintptr_t stack)804 linux_exec_setregs(struct thread *td, struct image_params *imgp,
805 uintptr_t stack)
806 {
807 struct pcb *pcb = td->td_pcb;
808
809 exec_setregs(td, imgp, stack);
810
811 /* Linux sets %gs to 0, we default to _udatasel. */
812 pcb->pcb_gs = 0;
813 load_gs(0);
814
815 pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
816 }
817
818 static void
linux_get_machine(const char ** dst)819 linux_get_machine(const char **dst)
820 {
821
822 switch (cpu_class) {
823 case CPUCLASS_686:
824 *dst = "i686";
825 break;
826 case CPUCLASS_586:
827 *dst = "i586";
828 break;
829 case CPUCLASS_486:
830 *dst = "i486";
831 break;
832 default:
833 *dst = "i386";
834 }
835 }
836
837 struct sysentvec linux_sysvec = {
838 .sv_size = LINUX_SYS_MAXSYSCALL,
839 .sv_table = linux_sysent,
840 .sv_transtrap = linux_translate_traps,
841 .sv_fixup = linux_fixup,
842 .sv_sendsig = linux_sendsig,
843 .sv_sigcode = &_binary_linux_locore_o_start,
844 .sv_szsigcode = &linux_szsigcode,
845 .sv_name = "Linux a.out",
846 .sv_coredump = NULL,
847 .sv_imgact_try = linux_exec_imgact_try,
848 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
849 .sv_minuser = VM_MIN_ADDRESS,
850 .sv_maxuser = VM_MAXUSER_ADDRESS,
851 .sv_usrstack = LINUX_USRSTACK,
852 .sv_psstrings = PS_STRINGS,
853 .sv_psstringssz = sizeof(struct ps_strings),
854 .sv_stackprot = VM_PROT_ALL,
855 .sv_copyout_strings = exec_copyout_strings,
856 .sv_setregs = linux_exec_setregs,
857 .sv_fixlimit = NULL,
858 .sv_maxssiz = NULL,
859 .sv_flags = SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32 |
860 SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
861 .sv_set_syscall_retval = linux_set_syscall_retval,
862 .sv_fetch_syscall_args = linux_fetch_syscall_args,
863 .sv_syscallnames = NULL,
864 .sv_shared_page_base = LINUX_SHAREDPAGE,
865 .sv_shared_page_len = PAGE_SIZE,
866 .sv_schedtail = linux_schedtail,
867 .sv_thread_detach = linux_thread_detach,
868 .sv_trap = NULL,
869 .sv_onexec = linux_on_exec,
870 .sv_onexit = linux_on_exit,
871 .sv_ontdexit = linux_thread_dtor,
872 .sv_setid_allowed = &linux_setid_allowed_query,
873 };
874 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
875
876 struct sysentvec elf_linux_sysvec = {
877 .sv_size = LINUX_SYS_MAXSYSCALL,
878 .sv_table = linux_sysent,
879 .sv_transtrap = linux_translate_traps,
880 .sv_fixup = linux_fixup_elf,
881 .sv_sendsig = linux_sendsig,
882 .sv_sigcode = &_binary_linux_locore_o_start,
883 .sv_szsigcode = &linux_szsigcode,
884 .sv_name = "Linux ELF32",
885 .sv_coredump = elf32_coredump,
886 .sv_imgact_try = linux_exec_imgact_try,
887 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
888 .sv_minuser = VM_MIN_ADDRESS,
889 .sv_maxuser = VM_MAXUSER_ADDRESS,
890 .sv_usrstack = LINUX_USRSTACK,
891 .sv_psstrings = LINUX_PS_STRINGS,
892 .sv_psstringssz = sizeof(struct ps_strings),
893 .sv_stackprot = VM_PROT_ALL,
894 .sv_copyout_auxargs = linux_copyout_auxargs,
895 .sv_copyout_strings = linux_copyout_strings,
896 .sv_setregs = linux_exec_setregs,
897 .sv_fixlimit = NULL,
898 .sv_maxssiz = NULL,
899 .sv_flags = SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP |
900 SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
901 .sv_set_syscall_retval = linux_set_syscall_retval,
902 .sv_fetch_syscall_args = linux_fetch_syscall_args,
903 .sv_syscallnames = NULL,
904 .sv_shared_page_base = LINUX_SHAREDPAGE,
905 .sv_shared_page_len = PAGE_SIZE,
906 .sv_schedtail = linux_schedtail,
907 .sv_thread_detach = linux_thread_detach,
908 .sv_trap = NULL,
909 .sv_onexec = linux_on_exec,
910 .sv_onexit = linux_on_exit,
911 .sv_ontdexit = linux_thread_dtor,
912 .sv_setid_allowed = &linux_setid_allowed_query,
913 };
914
915 static void
linux_vdso_install(void * param)916 linux_vdso_install(void *param)
917 {
918
919 linux_szsigcode = (&_binary_linux_locore_o_end -
920 &_binary_linux_locore_o_start);
921
922 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
923 panic("Linux invalid vdso size\n");
924
925 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
926
927 linux_shared_page_obj = __elfN(linux_shared_page_init)
928 (&linux_shared_page_mapping);
929
930 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
931
932 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
933 linux_szsigcode);
934 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
935 }
936 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
937 linux_vdso_install, NULL);
938
939 static void
linux_vdso_deinstall(void * param)940 linux_vdso_deinstall(void *param)
941 {
942
943 __elfN(linux_shared_page_fini)(linux_shared_page_obj,
944 linux_shared_page_mapping);
945 }
946 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
947 linux_vdso_deinstall, NULL);
948
949 static char GNU_ABI_VENDOR[] = "GNU";
950 static int GNULINUX_ABI_DESC = 0;
951
952 static bool
linux_trans_osrel(const Elf_Note * note,int32_t * osrel)953 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
954 {
955 const Elf32_Word *desc;
956 uintptr_t p;
957
958 p = (uintptr_t)(note + 1);
959 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
960
961 desc = (const Elf32_Word *)p;
962 if (desc[0] != GNULINUX_ABI_DESC)
963 return (false);
964
965 /*
966 * For Linux we encode osrel using the Linux convention of
967 * (version << 16) | (major << 8) | (minor)
968 * See macro in linux_mib.h
969 */
970 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
971
972 return (true);
973 }
974
975 static Elf_Brandnote linux_brandnote = {
976 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
977 .hdr.n_descsz = 16, /* XXX at least 16 */
978 .hdr.n_type = 1,
979 .vendor = GNU_ABI_VENDOR,
980 .flags = BN_TRANSLATE_OSREL,
981 .trans_osrel = linux_trans_osrel
982 };
983
984 static Elf32_Brandinfo linux_brand = {
985 .brand = ELFOSABI_LINUX,
986 .machine = EM_386,
987 .compat_3_brand = "Linux",
988 .emul_path = linux_emul_path,
989 .interp_path = "/lib/ld-linux.so.1",
990 .sysvec = &elf_linux_sysvec,
991 .interp_newpath = NULL,
992 .brand_note = &linux_brandnote,
993 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
994 };
995
996 static Elf32_Brandinfo linux_glibc2brand = {
997 .brand = ELFOSABI_LINUX,
998 .machine = EM_386,
999 .compat_3_brand = "Linux",
1000 .emul_path = linux_emul_path,
1001 .interp_path = "/lib/ld-linux.so.2",
1002 .sysvec = &elf_linux_sysvec,
1003 .interp_newpath = NULL,
1004 .brand_note = &linux_brandnote,
1005 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1006 };
1007
1008 static Elf32_Brandinfo linux_muslbrand = {
1009 .brand = ELFOSABI_LINUX,
1010 .machine = EM_386,
1011 .compat_3_brand = "Linux",
1012 .emul_path = linux_emul_path,
1013 .interp_path = "/lib/ld-musl-i386.so.1",
1014 .sysvec = &elf_linux_sysvec,
1015 .interp_newpath = NULL,
1016 .brand_note = &linux_brandnote,
1017 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1018 };
1019
1020 Elf32_Brandinfo *linux_brandlist[] = {
1021 &linux_brand,
1022 &linux_glibc2brand,
1023 &linux_muslbrand,
1024 NULL
1025 };
1026
1027 static int
linux_elf_modevent(module_t mod,int type,void * data)1028 linux_elf_modevent(module_t mod, int type, void *data)
1029 {
1030 Elf32_Brandinfo **brandinfo;
1031 int error;
1032 struct linux_ioctl_handler **lihp;
1033
1034 error = 0;
1035
1036 switch(type) {
1037 case MOD_LOAD:
1038 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1039 ++brandinfo)
1040 if (elf32_insert_brand_entry(*brandinfo) < 0)
1041 error = EINVAL;
1042 if (error == 0) {
1043 SET_FOREACH(lihp, linux_ioctl_handler_set)
1044 linux_ioctl_register_handler(*lihp);
1045 LIST_INIT(&futex_list);
1046 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1047 linux_get_machine(&linux_kplatform);
1048 linux_szplatform = roundup(strlen(linux_kplatform) + 1,
1049 sizeof(char *));
1050 linux_dev_shm_create();
1051 linux_osd_jail_register();
1052 stclohz = (stathz ? stathz : hz);
1053 if (bootverbose)
1054 printf("Linux ELF exec handler installed\n");
1055 } else
1056 printf("cannot insert Linux ELF brand handler\n");
1057 break;
1058 case MOD_UNLOAD:
1059 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1060 ++brandinfo)
1061 if (elf32_brand_inuse(*brandinfo))
1062 error = EBUSY;
1063 if (error == 0) {
1064 for (brandinfo = &linux_brandlist[0];
1065 *brandinfo != NULL; ++brandinfo)
1066 if (elf32_remove_brand_entry(*brandinfo) < 0)
1067 error = EINVAL;
1068 }
1069 if (error == 0) {
1070 SET_FOREACH(lihp, linux_ioctl_handler_set)
1071 linux_ioctl_unregister_handler(*lihp);
1072 mtx_destroy(&futex_mtx);
1073 linux_dev_shm_destroy();
1074 linux_osd_jail_deregister();
1075 if (bootverbose)
1076 printf("Linux ELF exec handler removed\n");
1077 } else
1078 printf("Could not deinstall ELF interpreter entry\n");
1079 break;
1080 default:
1081 return (EOPNOTSUPP);
1082 }
1083 return (error);
1084 }
1085
1086 static moduledata_t linux_elf_mod = {
1087 "linuxelf",
1088 linux_elf_modevent,
1089 0
1090 };
1091
1092 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1093 FEATURE(linux, "Linux 32bit support");
1094