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(uintptr_t *stack_base,
105 struct image_params *iparams);
106 static int linux_copyout_strings(struct image_params *imgp,
107 uintptr_t *stack_base);
108 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
109 static void linux_exec_setregs(struct thread *td,
110 struct image_params *imgp, uintptr_t stack);
111 static void linux32_fixlimit(struct rlimit *rl, int which);
112 static bool linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
113 static void linux_vdso_install(void *param);
114 static void linux_vdso_deinstall(void *param);
115 static void linux32_set_syscall_retval(struct thread *td, int error);
116
117 #define LINUX_T_UNKNOWN 255
118 static int _bsd_to_linux_trapcode[] = {
119 LINUX_T_UNKNOWN, /* 0 */
120 6, /* 1 T_PRIVINFLT */
121 LINUX_T_UNKNOWN, /* 2 */
122 3, /* 3 T_BPTFLT */
123 LINUX_T_UNKNOWN, /* 4 */
124 LINUX_T_UNKNOWN, /* 5 */
125 16, /* 6 T_ARITHTRAP */
126 254, /* 7 T_ASTFLT */
127 LINUX_T_UNKNOWN, /* 8 */
128 13, /* 9 T_PROTFLT */
129 1, /* 10 T_TRCTRAP */
130 LINUX_T_UNKNOWN, /* 11 */
131 14, /* 12 T_PAGEFLT */
132 LINUX_T_UNKNOWN, /* 13 */
133 17, /* 14 T_ALIGNFLT */
134 LINUX_T_UNKNOWN, /* 15 */
135 LINUX_T_UNKNOWN, /* 16 */
136 LINUX_T_UNKNOWN, /* 17 */
137 0, /* 18 T_DIVIDE */
138 2, /* 19 T_NMI */
139 4, /* 20 T_OFLOW */
140 5, /* 21 T_BOUND */
141 7, /* 22 T_DNA */
142 8, /* 23 T_DOUBLEFLT */
143 9, /* 24 T_FPOPFLT */
144 10, /* 25 T_TSSFLT */
145 11, /* 26 T_SEGNPFLT */
146 12, /* 27 T_STKFLT */
147 18, /* 28 T_MCHK */
148 19, /* 29 T_XMMFLT */
149 15 /* 30 T_RESERVED */
150 };
151 #define bsd_to_linux_trapcode(code) \
152 ((code)<nitems(_bsd_to_linux_trapcode)? \
153 _bsd_to_linux_trapcode[(code)]: \
154 LINUX_T_UNKNOWN)
155
156 struct linux32_ps_strings {
157 u_int32_t ps_argvstr; /* first of 0 or more argument strings */
158 u_int ps_nargvstr; /* the number of argument strings */
159 u_int32_t ps_envstr; /* first of 0 or more environment strings */
160 u_int ps_nenvstr; /* the number of environment strings */
161 };
162
163 LINUX_VDSO_SYM_INTPTR(linux32_sigcode);
164 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode);
165 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall);
166 LINUX_VDSO_SYM_CHAR(linux_platform);
167
168 /*
169 * If FreeBSD & Linux have a difference of opinion about what a trap
170 * means, deal with it here.
171 *
172 * MPSAFE
173 */
174 static int
linux_translate_traps(int signal,int trap_code)175 linux_translate_traps(int signal, int trap_code)
176 {
177 if (signal != SIGBUS)
178 return (signal);
179 switch (trap_code) {
180 case T_PROTFLT:
181 case T_TSSFLT:
182 case T_DOUBLEFLT:
183 case T_PAGEFLT:
184 return (SIGSEGV);
185 default:
186 return (signal);
187 }
188 }
189
190 static int
linux_copyout_auxargs(struct image_params * imgp,uintptr_t base)191 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
192 {
193 Elf32_Auxargs *args;
194 Elf32_Auxinfo *argarray, *pos;
195 int error, issetugid;
196
197 args = (Elf32_Auxargs *)imgp->auxargs;
198 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
199 M_WAITOK | M_ZERO);
200
201 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
202 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux32_vsyscall);
203 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
204 imgp->proc->p_sysent->sv_shared_page_base);
205 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
206 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
207
208 /*
209 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
210 * as it has appeared in the 2.4.0-rc7 first time.
211 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
212 * glibc falls back to the hard-coded CLK_TCK value when aux entry
213 * is not present.
214 * Also see linux_times() implementation.
215 */
216 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
217 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
218 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
219 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
220 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
221 AUXARGS_ENTRY(pos, AT_BASE, args->base);
222 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
223 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
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_SECURE, issetugid);
229 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
230 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
231 if (imgp->execpathp != 0)
232 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
233 if (args->execfd != -1)
234 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
235 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
236 AUXARGS_ENTRY(pos, AT_NULL, 0);
237
238 free(imgp->auxargs, M_TEMP);
239 imgp->auxargs = NULL;
240 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
241
242 error = copyout(argarray, (void *)base,
243 sizeof(*argarray) * LINUX_AT_COUNT);
244 free(argarray, M_TEMP);
245 return (error);
246 }
247
248 static int
linux_fixup_elf(uintptr_t * stack_base,struct image_params * imgp)249 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
250 {
251 Elf32_Addr *base;
252
253 base = (Elf32_Addr *)*stack_base;
254 base--;
255 if (suword32(base, (uint32_t)imgp->args->argc) == -1)
256 return (EFAULT);
257 *stack_base = (uintptr_t)base;
258 return (0);
259 }
260
261 static void
linux_rt_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)262 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
263 {
264 struct thread *td = curthread;
265 struct proc *p = td->td_proc;
266 struct sigacts *psp;
267 struct trapframe *regs;
268 struct l_rt_sigframe *fp, frame;
269 int oonstack;
270 int sig;
271 int code;
272
273 sig = ksi->ksi_signo;
274 code = ksi->ksi_code;
275 PROC_LOCK_ASSERT(p, MA_OWNED);
276 psp = p->p_sigacts;
277 mtx_assert(&psp->ps_mtx, MA_OWNED);
278 regs = td->td_frame;
279 oonstack = sigonstack(regs->tf_rsp);
280
281 /* Allocate space for the signal handler context. */
282 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
283 SIGISMEMBER(psp->ps_sigonstack, sig)) {
284 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
285 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
286 } else
287 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
288 mtx_unlock(&psp->ps_mtx);
289
290 /* Build the argument list for the signal handler. */
291 sig = bsd_to_linux_signal(sig);
292
293 bzero(&frame, sizeof(frame));
294
295 frame.sf_handler = PTROUT(catcher);
296 frame.sf_sig = sig;
297 frame.sf_siginfo = PTROUT(&fp->sf_si);
298 frame.sf_ucontext = PTROUT(&fp->sf_sc);
299
300 /* Fill in POSIX parts. */
301 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
302
303 /*
304 * Build the signal context to be used by sigreturn and libgcc unwind.
305 */
306 frame.sf_sc.uc_flags = 0; /* XXX ??? */
307 frame.sf_sc.uc_link = 0; /* XXX ??? */
308
309 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
310 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
311 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
312 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
313 PROC_UNLOCK(p);
314
315 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
316
317 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask;
318 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi;
319 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi;
320 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp;
321 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx;
322 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp;
323 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx;
324 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx;
325 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax;
326 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip;
327 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
328 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs;
329 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
330 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
331 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
332 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
333 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
334 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
335 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
336 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
337 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
338
339 if (copyout(&frame, fp, sizeof(frame)) != 0) {
340 /*
341 * Process has trashed its stack; give it an illegal
342 * instruction to halt it in its tracks.
343 */
344 PROC_LOCK(p);
345 sigexit(td, SIGILL);
346 }
347
348 /* Build context to run handler in. */
349 regs->tf_rsp = PTROUT(fp);
350 regs->tf_rip = linux32_rt_sigcode;
351 regs->tf_rflags &= ~(PSL_T | PSL_D);
352 regs->tf_cs = _ucode32sel;
353 regs->tf_ss = _udatasel;
354 regs->tf_ds = _udatasel;
355 regs->tf_es = _udatasel;
356 regs->tf_fs = _ufssel;
357 regs->tf_gs = _ugssel;
358 regs->tf_flags = TF_HASSEGS;
359 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
360 PROC_LOCK(p);
361 mtx_lock(&psp->ps_mtx);
362 }
363
364 /*
365 * Send an interrupt to process.
366 *
367 * Stack is set up to allow sigcode stored
368 * in u. to call routine, followed by kcall
369 * to sigreturn routine below. After sigreturn
370 * resets the signal mask, the stack, and the
371 * frame pointer, it returns to the user
372 * specified pc, psl.
373 */
374 static void
linux_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)375 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
376 {
377 struct thread *td = curthread;
378 struct proc *p = td->td_proc;
379 struct sigacts *psp;
380 struct trapframe *regs;
381 struct l_sigframe *fp, frame;
382 l_sigset_t lmask;
383 int oonstack;
384 int sig, code;
385
386 sig = ksi->ksi_signo;
387 code = ksi->ksi_code;
388 PROC_LOCK_ASSERT(p, MA_OWNED);
389 psp = p->p_sigacts;
390 mtx_assert(&psp->ps_mtx, MA_OWNED);
391 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
392 /* Signal handler installed with SA_SIGINFO. */
393 linux_rt_sendsig(catcher, ksi, mask);
394 return;
395 }
396
397 regs = td->td_frame;
398 oonstack = sigonstack(regs->tf_rsp);
399
400 /* Allocate space for the signal handler context. */
401 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
402 SIGISMEMBER(psp->ps_sigonstack, sig)) {
403 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
404 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
405 } else
406 fp = (struct l_sigframe *)regs->tf_rsp - 1;
407 mtx_unlock(&psp->ps_mtx);
408 PROC_UNLOCK(p);
409
410 /* Build the argument list for the signal handler. */
411 sig = bsd_to_linux_signal(sig);
412
413 bzero(&frame, sizeof(frame));
414
415 frame.sf_handler = PTROUT(catcher);
416 frame.sf_sig = sig;
417
418 bsd_to_linux_sigset(mask, &lmask);
419
420 /* Build the signal context to be used by sigreturn. */
421 frame.sf_sc.sc_mask = lmask.__mask;
422 frame.sf_sc.sc_gs = regs->tf_gs;
423 frame.sf_sc.sc_fs = regs->tf_fs;
424 frame.sf_sc.sc_es = regs->tf_es;
425 frame.sf_sc.sc_ds = regs->tf_ds;
426 frame.sf_sc.sc_edi = regs->tf_rdi;
427 frame.sf_sc.sc_esi = regs->tf_rsi;
428 frame.sf_sc.sc_ebp = regs->tf_rbp;
429 frame.sf_sc.sc_ebx = regs->tf_rbx;
430 frame.sf_sc.sc_esp = regs->tf_rsp;
431 frame.sf_sc.sc_edx = regs->tf_rdx;
432 frame.sf_sc.sc_ecx = regs->tf_rcx;
433 frame.sf_sc.sc_eax = regs->tf_rax;
434 frame.sf_sc.sc_eip = regs->tf_rip;
435 frame.sf_sc.sc_cs = regs->tf_cs;
436 frame.sf_sc.sc_eflags = regs->tf_rflags;
437 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
438 frame.sf_sc.sc_ss = regs->tf_ss;
439 frame.sf_sc.sc_err = regs->tf_err;
440 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
441 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
442
443 frame.sf_extramask[0] = lmask.__mask;
444
445 if (copyout(&frame, fp, sizeof(frame)) != 0) {
446 /*
447 * Process has trashed its stack; give it an illegal
448 * instruction to halt it in its tracks.
449 */
450 PROC_LOCK(p);
451 sigexit(td, SIGILL);
452 }
453
454 /* Build context to run handler in. */
455 regs->tf_rsp = PTROUT(fp);
456 regs->tf_rip = linux32_sigcode;
457 regs->tf_rflags &= ~(PSL_T | PSL_D);
458 regs->tf_cs = _ucode32sel;
459 regs->tf_ss = _udatasel;
460 regs->tf_ds = _udatasel;
461 regs->tf_es = _udatasel;
462 regs->tf_fs = _ufssel;
463 regs->tf_gs = _ugssel;
464 regs->tf_flags = TF_HASSEGS;
465 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
466 PROC_LOCK(p);
467 mtx_lock(&psp->ps_mtx);
468 }
469
470 /*
471 * System call to cleanup state after a signal
472 * has been taken. Reset signal mask and
473 * stack state from context left by sendsig (above).
474 * Return to previous pc and psl as specified by
475 * context left by sendsig. Check carefully to
476 * make sure that the user has not modified the
477 * psl to gain improper privileges or to cause
478 * a machine fault.
479 */
480 int
linux_sigreturn(struct thread * td,struct linux_sigreturn_args * args)481 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
482 {
483 struct l_sigframe frame;
484 struct trapframe *regs;
485 sigset_t bmask;
486 l_sigset_t lmask;
487 int eflags;
488 ksiginfo_t ksi;
489
490 regs = td->td_frame;
491
492 /*
493 * The trampoline code hands us the sigframe.
494 * It is unsafe to keep track of it ourselves, in the event that a
495 * program jumps out of a signal handler.
496 */
497 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
498 return (EFAULT);
499
500 /* Check for security violations. */
501 eflags = frame.sf_sc.sc_eflags;
502 if (!EFL_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 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
511 ksiginfo_init_trap(&ksi);
512 ksi.ksi_signo = SIGBUS;
513 ksi.ksi_code = BUS_OBJERR;
514 ksi.ksi_trapno = T_PROTFLT;
515 ksi.ksi_addr = (void *)regs->tf_rip;
516 trapsignal(td, &ksi);
517 return(EINVAL);
518 }
519
520 lmask.__mask = frame.sf_sc.sc_mask;
521 lmask.__mask = frame.sf_extramask[0];
522 linux_to_bsd_sigset(&lmask, &bmask);
523 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
524
525 /* Restore signal context. */
526 regs->tf_rdi = frame.sf_sc.sc_edi;
527 regs->tf_rsi = frame.sf_sc.sc_esi;
528 regs->tf_rbp = frame.sf_sc.sc_ebp;
529 regs->tf_rbx = frame.sf_sc.sc_ebx;
530 regs->tf_rdx = frame.sf_sc.sc_edx;
531 regs->tf_rcx = frame.sf_sc.sc_ecx;
532 regs->tf_rax = frame.sf_sc.sc_eax;
533 regs->tf_rip = frame.sf_sc.sc_eip;
534 regs->tf_cs = frame.sf_sc.sc_cs;
535 regs->tf_ds = frame.sf_sc.sc_ds;
536 regs->tf_es = frame.sf_sc.sc_es;
537 regs->tf_fs = frame.sf_sc.sc_fs;
538 regs->tf_gs = frame.sf_sc.sc_gs;
539 regs->tf_rflags = eflags;
540 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal;
541 regs->tf_ss = frame.sf_sc.sc_ss;
542 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
543
544 return (EJUSTRETURN);
545 }
546
547 /*
548 * System call to cleanup state after a signal
549 * has been taken. Reset signal mask and
550 * stack state from context left by rt_sendsig (above).
551 * Return to previous pc and psl as specified by
552 * context left by sendsig. Check carefully to
553 * make sure that the user has not modified the
554 * psl to gain improper privileges or to cause
555 * a machine fault.
556 */
557 int
linux_rt_sigreturn(struct thread * td,struct linux_rt_sigreturn_args * args)558 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
559 {
560 struct l_ucontext uc;
561 struct l_sigcontext *context;
562 sigset_t bmask;
563 l_stack_t *lss;
564 stack_t ss;
565 struct trapframe *regs;
566 int eflags;
567 ksiginfo_t ksi;
568
569 regs = td->td_frame;
570
571 /*
572 * The trampoline code hands us the ucontext.
573 * It is unsafe to keep track of it ourselves, in the event that a
574 * program jumps out of a signal handler.
575 */
576 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
577 return (EFAULT);
578
579 context = &uc.uc_mcontext;
580
581 /* Check for security violations. */
582 eflags = context->sc_eflags;
583 if (!EFL_SECURE(eflags, regs->tf_rflags))
584 return(EINVAL);
585
586 /*
587 * Don't allow users to load a valid privileged %cs. Let the
588 * hardware check for invalid selectors, excess privilege in
589 * other selectors, invalid %eip's and invalid %esp's.
590 */
591 if (!CS_SECURE(context->sc_cs)) {
592 ksiginfo_init_trap(&ksi);
593 ksi.ksi_signo = SIGBUS;
594 ksi.ksi_code = BUS_OBJERR;
595 ksi.ksi_trapno = T_PROTFLT;
596 ksi.ksi_addr = (void *)regs->tf_rip;
597 trapsignal(td, &ksi);
598 return(EINVAL);
599 }
600
601 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
602 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
603
604 /*
605 * Restore signal context
606 */
607 regs->tf_gs = context->sc_gs;
608 regs->tf_fs = context->sc_fs;
609 regs->tf_es = context->sc_es;
610 regs->tf_ds = context->sc_ds;
611 regs->tf_rdi = context->sc_edi;
612 regs->tf_rsi = context->sc_esi;
613 regs->tf_rbp = context->sc_ebp;
614 regs->tf_rbx = context->sc_ebx;
615 regs->tf_rdx = context->sc_edx;
616 regs->tf_rcx = context->sc_ecx;
617 regs->tf_rax = context->sc_eax;
618 regs->tf_rip = context->sc_eip;
619 regs->tf_cs = context->sc_cs;
620 regs->tf_rflags = eflags;
621 regs->tf_rsp = context->sc_esp_at_signal;
622 regs->tf_ss = context->sc_ss;
623 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
624
625 /*
626 * call sigaltstack & ignore results..
627 */
628 lss = &uc.uc_stack;
629 ss.ss_sp = PTRIN(lss->ss_sp);
630 ss.ss_size = lss->ss_size;
631 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
632
633 (void)kern_sigaltstack(td, &ss, NULL);
634
635 return (EJUSTRETURN);
636 }
637
638 static int
linux32_fetch_syscall_args(struct thread * td)639 linux32_fetch_syscall_args(struct thread *td)
640 {
641 struct proc *p;
642 struct trapframe *frame;
643 struct syscall_args *sa;
644
645 p = td->td_proc;
646 frame = td->td_frame;
647 sa = &td->td_sa;
648
649 sa->args[0] = frame->tf_rbx;
650 sa->args[1] = frame->tf_rcx;
651 sa->args[2] = frame->tf_rdx;
652 sa->args[3] = frame->tf_rsi;
653 sa->args[4] = frame->tf_rdi;
654 sa->args[5] = frame->tf_rbp; /* Unconfirmed */
655 sa->code = frame->tf_rax;
656
657 if (sa->code >= p->p_sysent->sv_size)
658 /* nosys */
659 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
660 else
661 sa->callp = &p->p_sysent->sv_table[sa->code];
662
663 td->td_retval[0] = 0;
664 td->td_retval[1] = frame->tf_rdx;
665
666 return (0);
667 }
668
669 static void
linux32_set_syscall_retval(struct thread * td,int error)670 linux32_set_syscall_retval(struct thread *td, int error)
671 {
672 struct trapframe *frame = td->td_frame;
673
674 cpu_set_syscall_retval(td, error);
675
676 if (__predict_false(error != 0)) {
677 if (error != ERESTART && error != EJUSTRETURN)
678 frame->tf_rax = bsd_to_linux_errno(error);
679 }
680 }
681
682 /*
683 * Clear registers on exec
684 * XXX copied from ia32_signal.c.
685 */
686 static void
linux_exec_setregs(struct thread * td,struct image_params * imgp,uintptr_t stack)687 linux_exec_setregs(struct thread *td, struct image_params *imgp,
688 uintptr_t stack)
689 {
690 struct trapframe *regs = td->td_frame;
691 struct pcb *pcb = td->td_pcb;
692 register_t saved_rflags;
693
694 regs = td->td_frame;
695 pcb = td->td_pcb;
696
697 if (td->td_proc->p_md.md_ldt != NULL)
698 user_ldt_free(td);
699
700 critical_enter();
701 wrmsr(MSR_FSBASE, 0);
702 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */
703 pcb->pcb_fsbase = 0;
704 pcb->pcb_gsbase = 0;
705 critical_exit();
706 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
707
708 saved_rflags = regs->tf_rflags & PSL_T;
709 bzero((char *)regs, sizeof(struct trapframe));
710 regs->tf_rip = imgp->entry_addr;
711 regs->tf_rsp = stack;
712 regs->tf_rflags = PSL_USER | saved_rflags;
713 regs->tf_gs = _ugssel;
714 regs->tf_fs = _ufssel;
715 regs->tf_es = _udatasel;
716 regs->tf_ds = _udatasel;
717 regs->tf_ss = _udatasel;
718 regs->tf_flags = TF_HASSEGS;
719 regs->tf_cs = _ucode32sel;
720 regs->tf_rbx = (register_t)imgp->ps_strings;
721
722 x86_clear_dbregs(pcb);
723
724 fpstate_drop(td);
725
726 /* Do full restore on return so that we can change to a different %cs */
727 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
728 }
729
730 /*
731 * XXX copied from ia32_sysvec.c.
732 */
733 static int
linux_copyout_strings(struct image_params * imgp,uintptr_t * stack_base)734 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
735 {
736 int argc, envc, error;
737 u_int32_t *vectp;
738 char *stringp;
739 uintptr_t destp, ustringp;
740 struct linux32_ps_strings *arginfo;
741 char canary[LINUX_AT_RANDOM_LEN];
742 size_t execpath_len;
743
744 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
745 destp = (uintptr_t)arginfo;
746
747 if (imgp->execpath != NULL && imgp->auxargs != NULL) {
748 execpath_len = strlen(imgp->execpath) + 1;
749 destp -= execpath_len;
750 destp = rounddown2(destp, sizeof(uint32_t));
751 imgp->execpathp = (void *)destp;
752 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
753 if (error != 0)
754 return (error);
755 }
756
757 /* Prepare the canary for SSP. */
758 arc4rand(canary, sizeof(canary), 0);
759 destp -= roundup(sizeof(canary), sizeof(uint32_t));
760 imgp->canary = (void *)destp;
761 error = copyout(canary, imgp->canary, sizeof(canary));
762 if (error != 0)
763 return (error);
764
765 /* Allocate room for the argument and environment strings. */
766 destp -= ARG_MAX - imgp->args->stringspace;
767 destp = rounddown2(destp, sizeof(uint32_t));
768 ustringp = destp;
769
770 if (imgp->auxargs) {
771 /*
772 * Allocate room on the stack for the ELF auxargs
773 * array. It has LINUX_AT_COUNT entries.
774 */
775 destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
776 destp = rounddown2(destp, sizeof(uint32_t));
777 }
778
779 vectp = (uint32_t *)destp;
780
781 /*
782 * Allocate room for the argv[] and env vectors including the
783 * terminating NULL pointers.
784 */
785 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
786
787 /* vectp also becomes our initial stack base. */
788 *stack_base = (uintptr_t)vectp;
789
790 stringp = imgp->args->begin_argv;
791 argc = imgp->args->argc;
792 envc = imgp->args->envc;
793
794 /* Copy out strings - arguments and environment. */
795 error = copyout(stringp, (void *)ustringp,
796 ARG_MAX - imgp->args->stringspace);
797 if (error != 0)
798 return (error);
799
800 /* Fill in "ps_strings" struct for ps, w, etc. */
801 if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 ||
802 suword32(&arginfo->ps_nargvstr, argc) != 0)
803 return (EFAULT);
804
805 /* Fill in argument portion of vector table. */
806 for (; argc > 0; --argc) {
807 if (suword32(vectp++, ustringp) != 0)
808 return (EFAULT);
809 while (*stringp++ != 0)
810 ustringp++;
811 ustringp++;
812 }
813
814 /* A null vector table pointer separates the argp's from the envp's. */
815 if (suword32(vectp++, 0) != 0)
816 return (EFAULT);
817
818 if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 ||
819 suword32(&arginfo->ps_nenvstr, envc) != 0)
820 return (EFAULT);
821
822 /* Fill in environment portion of vector table. */
823 for (; envc > 0; --envc) {
824 if (suword32(vectp++, ustringp) != 0)
825 return (EFAULT);
826 while (*stringp++ != 0)
827 ustringp++;
828 ustringp++;
829 }
830
831 /* The end of the vector table is a null pointer. */
832 if (suword32(vectp, 0) != 0)
833 return (EFAULT);
834
835 if (imgp->auxargs) {
836 vectp++;
837 error = imgp->sysent->sv_copyout_auxargs(imgp,
838 (uintptr_t)vectp);
839 if (error != 0)
840 return (error);
841 }
842
843 return (0);
844 }
845
846 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
847 "32-bit Linux emulation");
848
849 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ;
850 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
851 &linux32_maxdsiz, 0, "");
852 static u_long linux32_maxssiz = LINUX32_MAXSSIZ;
853 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
854 &linux32_maxssiz, 0, "");
855 static u_long linux32_maxvmem = LINUX32_MAXVMEM;
856 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
857 &linux32_maxvmem, 0, "");
858
859 static void
linux32_fixlimit(struct rlimit * rl,int which)860 linux32_fixlimit(struct rlimit *rl, int which)
861 {
862
863 switch (which) {
864 case RLIMIT_DATA:
865 if (linux32_maxdsiz != 0) {
866 if (rl->rlim_cur > linux32_maxdsiz)
867 rl->rlim_cur = linux32_maxdsiz;
868 if (rl->rlim_max > linux32_maxdsiz)
869 rl->rlim_max = linux32_maxdsiz;
870 }
871 break;
872 case RLIMIT_STACK:
873 if (linux32_maxssiz != 0) {
874 if (rl->rlim_cur > linux32_maxssiz)
875 rl->rlim_cur = linux32_maxssiz;
876 if (rl->rlim_max > linux32_maxssiz)
877 rl->rlim_max = linux32_maxssiz;
878 }
879 break;
880 case RLIMIT_VMEM:
881 if (linux32_maxvmem != 0) {
882 if (rl->rlim_cur > linux32_maxvmem)
883 rl->rlim_cur = linux32_maxvmem;
884 if (rl->rlim_max > linux32_maxvmem)
885 rl->rlim_max = linux32_maxvmem;
886 }
887 break;
888 }
889 }
890
891 struct sysentvec elf_linux_sysvec = {
892 .sv_size = LINUX32_SYS_MAXSYSCALL,
893 .sv_table = linux32_sysent,
894 .sv_transtrap = linux_translate_traps,
895 .sv_fixup = linux_fixup_elf,
896 .sv_sendsig = linux_sendsig,
897 .sv_sigcode = &_binary_linux32_locore_o_start,
898 .sv_szsigcode = &linux_szsigcode,
899 .sv_name = "Linux ELF32",
900 .sv_coredump = elf32_coredump,
901 .sv_imgact_try = linux_exec_imgact_try,
902 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
903 .sv_minuser = VM_MIN_ADDRESS,
904 .sv_maxuser = LINUX32_MAXUSER,
905 .sv_usrstack = LINUX32_USRSTACK,
906 .sv_psstrings = LINUX32_PS_STRINGS,
907 .sv_psstringssz = sizeof(struct linux32_ps_strings),
908 .sv_stackprot = VM_PROT_ALL,
909 .sv_copyout_auxargs = linux_copyout_auxargs,
910 .sv_copyout_strings = linux_copyout_strings,
911 .sv_setregs = linux_exec_setregs,
912 .sv_fixlimit = linux32_fixlimit,
913 .sv_maxssiz = &linux32_maxssiz,
914 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP |
915 SV_SIG_DISCIGN | SV_SIG_WAITNDQ,
916 .sv_set_syscall_retval = linux32_set_syscall_retval,
917 .sv_fetch_syscall_args = linux32_fetch_syscall_args,
918 .sv_syscallnames = NULL,
919 .sv_shared_page_base = LINUX32_SHAREDPAGE,
920 .sv_shared_page_len = PAGE_SIZE,
921 .sv_schedtail = linux_schedtail,
922 .sv_thread_detach = linux_thread_detach,
923 .sv_trap = NULL,
924 .sv_onexec = linux_on_exec,
925 .sv_onexit = linux_on_exit,
926 .sv_ontdexit = linux_thread_dtor,
927 .sv_setid_allowed = &linux_setid_allowed_query,
928 };
929
930 static void
linux_vdso_install(void * param)931 linux_vdso_install(void *param)
932 {
933
934 linux_szsigcode = (&_binary_linux32_locore_o_end -
935 &_binary_linux32_locore_o_start);
936
937 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
938 panic("Linux invalid vdso size\n");
939
940 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
941
942 linux_shared_page_obj = __elfN(linux_shared_page_init)
943 (&linux_shared_page_mapping);
944
945 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
946
947 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
948 linux_szsigcode);
949 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
950
951 linux_kplatform = linux_shared_page_mapping +
952 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
953 }
954 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
955 linux_vdso_install, NULL);
956
957 static void
linux_vdso_deinstall(void * param)958 linux_vdso_deinstall(void *param)
959 {
960
961 __elfN(linux_shared_page_fini)(linux_shared_page_obj,
962 linux_shared_page_mapping);
963 }
964 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
965 linux_vdso_deinstall, NULL);
966
967 static char GNU_ABI_VENDOR[] = "GNU";
968 static int GNULINUX_ABI_DESC = 0;
969
970 static bool
linux32_trans_osrel(const Elf_Note * note,int32_t * osrel)971 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
972 {
973 const Elf32_Word *desc;
974 uintptr_t p;
975
976 p = (uintptr_t)(note + 1);
977 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
978
979 desc = (const Elf32_Word *)p;
980 if (desc[0] != GNULINUX_ABI_DESC)
981 return (false);
982
983 /*
984 * For Linux we encode osrel using the Linux convention of
985 * (version << 16) | (major << 8) | (minor)
986 * See macro in linux_mib.h
987 */
988 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
989
990 return (true);
991 }
992
993 static Elf_Brandnote linux32_brandnote = {
994 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
995 .hdr.n_descsz = 16, /* XXX at least 16 */
996 .hdr.n_type = 1,
997 .vendor = GNU_ABI_VENDOR,
998 .flags = BN_TRANSLATE_OSREL,
999 .trans_osrel = linux32_trans_osrel
1000 };
1001
1002 static Elf32_Brandinfo linux_brand = {
1003 .brand = ELFOSABI_LINUX,
1004 .machine = EM_386,
1005 .compat_3_brand = "Linux",
1006 .emul_path = linux_emul_path,
1007 .interp_path = "/lib/ld-linux.so.1",
1008 .sysvec = &elf_linux_sysvec,
1009 .interp_newpath = NULL,
1010 .brand_note = &linux32_brandnote,
1011 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1012 };
1013
1014 static Elf32_Brandinfo linux_glibc2brand = {
1015 .brand = ELFOSABI_LINUX,
1016 .machine = EM_386,
1017 .compat_3_brand = "Linux",
1018 .emul_path = linux_emul_path,
1019 .interp_path = "/lib/ld-linux.so.2",
1020 .sysvec = &elf_linux_sysvec,
1021 .interp_newpath = NULL,
1022 .brand_note = &linux32_brandnote,
1023 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1024 };
1025
1026 static Elf32_Brandinfo linux_muslbrand = {
1027 .brand = ELFOSABI_LINUX,
1028 .machine = EM_386,
1029 .compat_3_brand = "Linux",
1030 .emul_path = linux_emul_path,
1031 .interp_path = "/lib/ld-musl-i386.so.1",
1032 .sysvec = &elf_linux_sysvec,
1033 .interp_newpath = NULL,
1034 .brand_note = &linux32_brandnote,
1035 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1036 };
1037
1038 Elf32_Brandinfo *linux_brandlist[] = {
1039 &linux_brand,
1040 &linux_glibc2brand,
1041 &linux_muslbrand,
1042 NULL
1043 };
1044
1045 static int
linux_elf_modevent(module_t mod,int type,void * data)1046 linux_elf_modevent(module_t mod, int type, void *data)
1047 {
1048 Elf32_Brandinfo **brandinfo;
1049 int error;
1050 struct linux_ioctl_handler **lihp;
1051
1052 error = 0;
1053
1054 switch(type) {
1055 case MOD_LOAD:
1056 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1057 ++brandinfo)
1058 if (elf32_insert_brand_entry(*brandinfo) < 0)
1059 error = EINVAL;
1060 if (error == 0) {
1061 SET_FOREACH(lihp, linux_ioctl_handler_set)
1062 linux32_ioctl_register_handler(*lihp);
1063 stclohz = (stathz ? stathz : hz);
1064 if (bootverbose)
1065 printf("Linux ELF exec handler installed\n");
1066 } else
1067 printf("cannot insert Linux ELF brand handler\n");
1068 break;
1069 case MOD_UNLOAD:
1070 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1071 ++brandinfo)
1072 if (elf32_brand_inuse(*brandinfo))
1073 error = EBUSY;
1074 if (error == 0) {
1075 for (brandinfo = &linux_brandlist[0];
1076 *brandinfo != NULL; ++brandinfo)
1077 if (elf32_remove_brand_entry(*brandinfo) < 0)
1078 error = EINVAL;
1079 }
1080 if (error == 0) {
1081 SET_FOREACH(lihp, linux_ioctl_handler_set)
1082 linux32_ioctl_unregister_handler(*lihp);
1083 if (bootverbose)
1084 printf("Linux ELF exec handler removed\n");
1085 } else
1086 printf("Could not deinstall ELF interpreter entry\n");
1087 break;
1088 default:
1089 return (EOPNOTSUPP);
1090 }
1091 return (error);
1092 }
1093
1094 static moduledata_t linux_elf_mod = {
1095 "linuxelf",
1096 linux_elf_modevent,
1097 0
1098 };
1099
1100 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1101 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1102 FEATURE(linux, "Linux 32bit support");
1103