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: stable/12/sys/amd64/linux32/linux32_sysvec.c 369693 2021-05-01 00:41:05Z kib $");
39
40 #ifndef COMPAT_FREEBSD32
41 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!"
42 #endif
43
44 #define __ELF_WORD_SIZE 32
45
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/exec.h>
49 #include <sys/fcntl.h>
50 #include <sys/imgact.h>
51 #include <sys/imgact_elf.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/module.h>
56 #include <sys/mutex.h>
57 #include <sys/proc.h>
58 #include <sys/resourcevar.h>
59 #include <sys/signalvar.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysent.h>
63 #include <sys/sysproto.h>
64 #include <sys/vnode.h>
65 #include <sys/eventhandler.h>
66
67 #include <vm/vm.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_param.h>
74
75 #include <machine/cpu.h>
76 #include <machine/md_var.h>
77 #include <machine/pcb.h>
78 #include <machine/specialreg.h>
79 #include <machine/trap.h>
80
81 #include <amd64/linux32/linux.h>
82 #include <amd64/linux32/linux32_proto.h>
83 #include <compat/linux/linux_emul.h>
84 #include <compat/linux/linux_ioctl.h>
85 #include <compat/linux/linux_mib.h>
86 #include <compat/linux/linux_misc.h>
87 #include <compat/linux/linux_signal.h>
88 #include <compat/linux/linux_util.h>
89 #include <compat/linux/linux_vdso.h>
90
91 MODULE_VERSION(linux, 1);
92
93 const char *linux_kplatform;
94 static int linux_szsigcode;
95 static vm_object_t linux_shared_page_obj;
96 static char *linux_shared_page_mapping;
97 extern char _binary_linux32_locore_o_start;
98 extern char _binary_linux32_locore_o_end;
99
100 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
101
102 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
103
104 static int linux_fixup_elf(register_t **stack_base,
105 struct image_params *iparams);
106 static register_t *linux_copyout_strings(struct image_params *imgp);
107 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
108 static void linux_exec_setregs(struct thread *td,
109 struct image_params *imgp, u_long stack);
110 static void linux32_fixlimit(struct rlimit *rl, int which);
111 static bool linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
112 static void linux_vdso_install(void *param);
113 static void linux_vdso_deinstall(void *param);
114
115 #define LINUX_T_UNKNOWN 255
116 static int _bsd_to_linux_trapcode[] = {
117 LINUX_T_UNKNOWN, /* 0 */
118 6, /* 1 T_PRIVINFLT */
119 LINUX_T_UNKNOWN, /* 2 */
120 3, /* 3 T_BPTFLT */
121 LINUX_T_UNKNOWN, /* 4 */
122 LINUX_T_UNKNOWN, /* 5 */
123 16, /* 6 T_ARITHTRAP */
124 254, /* 7 T_ASTFLT */
125 LINUX_T_UNKNOWN, /* 8 */
126 13, /* 9 T_PROTFLT */
127 1, /* 10 T_TRCTRAP */
128 LINUX_T_UNKNOWN, /* 11 */
129 14, /* 12 T_PAGEFLT */
130 LINUX_T_UNKNOWN, /* 13 */
131 17, /* 14 T_ALIGNFLT */
132 LINUX_T_UNKNOWN, /* 15 */
133 LINUX_T_UNKNOWN, /* 16 */
134 LINUX_T_UNKNOWN, /* 17 */
135 0, /* 18 T_DIVIDE */
136 2, /* 19 T_NMI */
137 4, /* 20 T_OFLOW */
138 5, /* 21 T_BOUND */
139 7, /* 22 T_DNA */
140 8, /* 23 T_DOUBLEFLT */
141 9, /* 24 T_FPOPFLT */
142 10, /* 25 T_TSSFLT */
143 11, /* 26 T_SEGNPFLT */
144 12, /* 27 T_STKFLT */
145 18, /* 28 T_MCHK */
146 19, /* 29 T_XMMFLT */
147 15 /* 30 T_RESERVED */
148 };
149 #define bsd_to_linux_trapcode(code) \
150 ((code)<nitems(_bsd_to_linux_trapcode)? \
151 _bsd_to_linux_trapcode[(code)]: \
152 LINUX_T_UNKNOWN)
153
154 struct linux32_ps_strings {
155 u_int32_t ps_argvstr; /* first of 0 or more argument strings */
156 u_int ps_nargvstr; /* the number of argument strings */
157 u_int32_t ps_envstr; /* first of 0 or more environment strings */
158 u_int ps_nenvstr; /* the number of environment strings */
159 };
160
161 LINUX_VDSO_SYM_INTPTR(linux32_sigcode);
162 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode);
163 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall);
164 LINUX_VDSO_SYM_CHAR(linux_platform);
165
166 /*
167 * If FreeBSD & Linux have a difference of opinion about what a trap
168 * means, deal with it here.
169 *
170 * MPSAFE
171 */
172 static int
linux_translate_traps(int signal,int trap_code)173 linux_translate_traps(int signal, int trap_code)
174 {
175 if (signal != SIGBUS)
176 return (signal);
177 switch (trap_code) {
178 case T_PROTFLT:
179 case T_TSSFLT:
180 case T_DOUBLEFLT:
181 case T_PAGEFLT:
182 return (SIGSEGV);
183 default:
184 return (signal);
185 }
186 }
187
188 static int
linux_fixup_elf(register_t ** stack_base,struct image_params * imgp)189 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
190 {
191 Elf32_Auxargs *args;
192 Elf32_Auxinfo *argarray, *pos;
193 Elf32_Addr *auxbase, *base;
194 struct linux32_ps_strings *arginfo;
195 int error, issetugid;
196
197 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
198
199 KASSERT(curthread->td_proc == imgp->proc,
200 ("unsafe linux_fixup_elf(), should be curproc"));
201 base = (Elf32_Addr *)*stack_base;
202 args = (Elf32_Auxargs *)imgp->auxargs;
203 auxbase = base + (imgp->args->argc + 1 + imgp->args->envc + 1);
204 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
205 M_WAITOK | M_ZERO);
206
207 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
208 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
209 imgp->proc->p_sysent->sv_shared_page_base);
210 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux32_vsyscall);
211 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
212
213 /*
214 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
215 * as it has appeared in the 2.4.0-rc7 first time.
216 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
217 * glibc falls back to the hard-coded CLK_TCK value when aux entry
218 * is not present.
219 * Also see linux_times() implementation.
220 */
221 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
222 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
223 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
224 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
225 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
226 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
227 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
228 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
229 AUXARGS_ENTRY(pos, AT_BASE, args->base);
230 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
231 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
232 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
233 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
234 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
235 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
236 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
237 if (imgp->execpathp != 0)
238 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
239 if (args->execfd != -1)
240 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
241 AUXARGS_ENTRY(pos, AT_NULL, 0);
242
243 free(imgp->auxargs, M_TEMP);
244 imgp->auxargs = NULL;
245 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
246
247 error = copyout(&argarray[0], auxbase,
248 sizeof(*argarray) * LINUX_AT_COUNT);
249 free(argarray, M_TEMP);
250 if (error != 0)
251 return (error);
252
253 base--;
254 if (suword32(base, (uint32_t)imgp->args->argc) == -1)
255 return (EFAULT);
256 *stack_base = (register_t *)base;
257 return (0);
258 }
259
260 static void
linux_rt_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)261 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
262 {
263 struct thread *td = curthread;
264 struct proc *p = td->td_proc;
265 struct sigacts *psp;
266 struct trapframe *regs;
267 struct l_rt_sigframe *fp, frame;
268 int oonstack;
269 int sig;
270 int code;
271
272 sig = ksi->ksi_signo;
273 code = ksi->ksi_code;
274 PROC_LOCK_ASSERT(p, MA_OWNED);
275 psp = p->p_sigacts;
276 mtx_assert(&psp->ps_mtx, MA_OWNED);
277 regs = td->td_frame;
278 oonstack = sigonstack(regs->tf_rsp);
279
280 /* Allocate space for the signal handler context. */
281 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
282 SIGISMEMBER(psp->ps_sigonstack, sig)) {
283 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
284 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
285 } else
286 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
287 mtx_unlock(&psp->ps_mtx);
288
289 /* Build the argument list for the signal handler. */
290 sig = bsd_to_linux_signal(sig);
291
292 bzero(&frame, sizeof(frame));
293
294 frame.sf_handler = PTROUT(catcher);
295 frame.sf_sig = sig;
296 frame.sf_siginfo = PTROUT(&fp->sf_si);
297 frame.sf_ucontext = PTROUT(&fp->sf_sc);
298
299 /* Fill in POSIX parts. */
300 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
301
302 /*
303 * Build the signal context to be used by sigreturn and libgcc unwind.
304 */
305 frame.sf_sc.uc_flags = 0; /* XXX ??? */
306 frame.sf_sc.uc_link = 0; /* XXX ??? */
307
308 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
309 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
310 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
311 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
312 PROC_UNLOCK(p);
313
314 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
315
316 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask;
317 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi;
318 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi;
319 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp;
320 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx;
321 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp;
322 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx;
323 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx;
324 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax;
325 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip;
326 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
327 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs;
328 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
329 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
330 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
331 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
332 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
333 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
334 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
335 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
336 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
337
338 if (copyout(&frame, fp, sizeof(frame)) != 0) {
339 /*
340 * Process has trashed its stack; give it an illegal
341 * instruction to halt it in its tracks.
342 */
343 PROC_LOCK(p);
344 sigexit(td, SIGILL);
345 }
346
347 /* Build context to run handler in. */
348 regs->tf_rsp = PTROUT(fp);
349 regs->tf_rip = linux32_rt_sigcode;
350 regs->tf_rflags &= ~(PSL_T | PSL_D);
351 regs->tf_cs = _ucode32sel;
352 regs->tf_ss = _udatasel;
353 regs->tf_ds = _udatasel;
354 regs->tf_es = _udatasel;
355 regs->tf_fs = _ufssel;
356 regs->tf_gs = _ugssel;
357 regs->tf_flags = TF_HASSEGS;
358 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
359 PROC_LOCK(p);
360 mtx_lock(&psp->ps_mtx);
361 }
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 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
502 eflags = frame.sf_sc.sc_eflags;
503 if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
504 return(EINVAL);
505
506 /*
507 * Don't allow users to load a valid privileged %cs. Let the
508 * hardware check for invalid selectors, excess privilege in
509 * other selectors, invalid %eip's and invalid %esp's.
510 */
511 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
512 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
513 ksiginfo_init_trap(&ksi);
514 ksi.ksi_signo = SIGBUS;
515 ksi.ksi_code = BUS_OBJERR;
516 ksi.ksi_trapno = T_PROTFLT;
517 ksi.ksi_addr = (void *)regs->tf_rip;
518 trapsignal(td, &ksi);
519 return(EINVAL);
520 }
521
522 lmask.__mask = frame.sf_sc.sc_mask;
523 lmask.__mask = frame.sf_extramask[0];
524 linux_to_bsd_sigset(&lmask, &bmask);
525 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
526
527 /* Restore signal context. */
528 regs->tf_rdi = frame.sf_sc.sc_edi;
529 regs->tf_rsi = frame.sf_sc.sc_esi;
530 regs->tf_rbp = frame.sf_sc.sc_ebp;
531 regs->tf_rbx = frame.sf_sc.sc_ebx;
532 regs->tf_rdx = frame.sf_sc.sc_edx;
533 regs->tf_rcx = frame.sf_sc.sc_ecx;
534 regs->tf_rax = frame.sf_sc.sc_eax;
535 regs->tf_rip = frame.sf_sc.sc_eip;
536 regs->tf_cs = frame.sf_sc.sc_cs;
537 regs->tf_ds = frame.sf_sc.sc_ds;
538 regs->tf_es = frame.sf_sc.sc_es;
539 regs->tf_fs = frame.sf_sc.sc_fs;
540 regs->tf_gs = frame.sf_sc.sc_gs;
541 regs->tf_rflags = eflags;
542 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal;
543 regs->tf_ss = frame.sf_sc.sc_ss;
544 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
545
546 return (EJUSTRETURN);
547 }
548
549 /*
550 * System call to cleanup state after a signal
551 * has been taken. Reset signal mask and
552 * stack state from context left by rt_sendsig (above).
553 * Return to previous pc and psl as specified by
554 * context left by sendsig. Check carefully to
555 * make sure that the user has not modified the
556 * psl to gain improper privileges or to cause
557 * a machine fault.
558 */
559 int
linux_rt_sigreturn(struct thread * td,struct linux_rt_sigreturn_args * args)560 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
561 {
562 struct l_ucontext uc;
563 struct l_sigcontext *context;
564 sigset_t bmask;
565 l_stack_t *lss;
566 stack_t ss;
567 struct trapframe *regs;
568 int eflags;
569 ksiginfo_t ksi;
570
571 regs = td->td_frame;
572
573 /*
574 * The trampoline code hands us the ucontext.
575 * It is unsafe to keep track of it ourselves, in the event that a
576 * program jumps out of a signal handler.
577 */
578 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
579 return (EFAULT);
580
581 context = &uc.uc_mcontext;
582
583 /* Check for security violations. */
584 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
585 eflags = context->sc_eflags;
586 if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
587 return(EINVAL);
588
589 /*
590 * Don't allow users to load a valid privileged %cs. Let the
591 * hardware check for invalid selectors, excess privilege in
592 * other selectors, invalid %eip's and invalid %esp's.
593 */
594 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
595 if (!CS_SECURE(context->sc_cs)) {
596 ksiginfo_init_trap(&ksi);
597 ksi.ksi_signo = SIGBUS;
598 ksi.ksi_code = BUS_OBJERR;
599 ksi.ksi_trapno = T_PROTFLT;
600 ksi.ksi_addr = (void *)regs->tf_rip;
601 trapsignal(td, &ksi);
602 return(EINVAL);
603 }
604
605 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
606 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
607
608 /*
609 * Restore signal context
610 */
611 regs->tf_gs = context->sc_gs;
612 regs->tf_fs = context->sc_fs;
613 regs->tf_es = context->sc_es;
614 regs->tf_ds = context->sc_ds;
615 regs->tf_rdi = context->sc_edi;
616 regs->tf_rsi = context->sc_esi;
617 regs->tf_rbp = context->sc_ebp;
618 regs->tf_rbx = context->sc_ebx;
619 regs->tf_rdx = context->sc_edx;
620 regs->tf_rcx = context->sc_ecx;
621 regs->tf_rax = context->sc_eax;
622 regs->tf_rip = context->sc_eip;
623 regs->tf_cs = context->sc_cs;
624 regs->tf_rflags = eflags;
625 regs->tf_rsp = context->sc_esp_at_signal;
626 regs->tf_ss = context->sc_ss;
627 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
628
629 /*
630 * call sigaltstack & ignore results..
631 */
632 lss = &uc.uc_stack;
633 ss.ss_sp = PTRIN(lss->ss_sp);
634 ss.ss_size = lss->ss_size;
635 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
636
637 (void)kern_sigaltstack(td, &ss, NULL);
638
639 return (EJUSTRETURN);
640 }
641
642 static int
linux32_fetch_syscall_args(struct thread * td)643 linux32_fetch_syscall_args(struct thread *td)
644 {
645 struct proc *p;
646 struct trapframe *frame;
647 struct syscall_args *sa;
648
649 p = td->td_proc;
650 frame = td->td_frame;
651 sa = &td->td_sa;
652
653 sa->args[0] = frame->tf_rbx;
654 sa->args[1] = frame->tf_rcx;
655 sa->args[2] = frame->tf_rdx;
656 sa->args[3] = frame->tf_rsi;
657 sa->args[4] = frame->tf_rdi;
658 sa->args[5] = frame->tf_rbp; /* Unconfirmed */
659 sa->code = frame->tf_rax;
660
661 if (sa->code >= p->p_sysent->sv_size)
662 /* nosys */
663 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
664 else
665 sa->callp = &p->p_sysent->sv_table[sa->code];
666 sa->narg = sa->callp->sy_narg;
667
668 td->td_retval[0] = 0;
669 td->td_retval[1] = frame->tf_rdx;
670
671 return (0);
672 }
673
674 /*
675 * Clear registers on exec
676 * XXX copied from ia32_signal.c.
677 */
678 static void
linux_exec_setregs(struct thread * td,struct image_params * imgp,u_long stack)679 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
680 {
681 struct trapframe *regs = td->td_frame;
682 struct pcb *pcb = td->td_pcb;
683 register_t saved_rflags;
684
685 regs = td->td_frame;
686 pcb = td->td_pcb;
687
688 if (td->td_proc->p_md.md_ldt != NULL)
689 user_ldt_free(td);
690
691 critical_enter();
692 wrmsr(MSR_FSBASE, 0);
693 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */
694 pcb->pcb_fsbase = 0;
695 pcb->pcb_gsbase = 0;
696 critical_exit();
697 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
698
699 saved_rflags = regs->tf_rflags & PSL_T;
700 bzero((char *)regs, sizeof(struct trapframe));
701 regs->tf_rip = imgp->entry_addr;
702 regs->tf_rsp = stack;
703 regs->tf_rflags = PSL_USER | saved_rflags;
704 regs->tf_gs = _ugssel;
705 regs->tf_fs = _ufssel;
706 regs->tf_es = _udatasel;
707 regs->tf_ds = _udatasel;
708 regs->tf_ss = _udatasel;
709 regs->tf_flags = TF_HASSEGS;
710 regs->tf_cs = _ucode32sel;
711 regs->tf_rbx = imgp->ps_strings;
712
713 x86_clear_dbregs(pcb);
714
715 fpstate_drop(td);
716
717 /* Do full restore on return so that we can change to a different %cs */
718 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
719 }
720
721 /*
722 * XXX copied from ia32_sysvec.c.
723 */
724 static register_t *
linux_copyout_strings(struct image_params * imgp)725 linux_copyout_strings(struct image_params *imgp)
726 {
727 int argc, envc;
728 u_int32_t *vectp;
729 char *stringp, *destp;
730 u_int32_t *stack_base;
731 struct linux32_ps_strings *arginfo;
732 char canary[LINUX_AT_RANDOM_LEN];
733 size_t execpath_len;
734
735 /* Calculate string base and vector table pointers. */
736 if (imgp->execpath != NULL && imgp->auxargs != NULL)
737 execpath_len = strlen(imgp->execpath) + 1;
738 else
739 execpath_len = 0;
740
741 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
742 destp = (caddr_t)arginfo - SPARE_USRSPACE -
743 roundup(sizeof(canary), sizeof(char *)) -
744 roundup(execpath_len, sizeof(char *)) -
745 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
746
747 if (execpath_len != 0) {
748 imgp->execpathp = (uintptr_t)arginfo - execpath_len;
749 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
750 }
751
752 /* Prepare the canary for SSP. */
753 arc4rand(canary, sizeof(canary), 0);
754 imgp->canary = (uintptr_t)arginfo -
755 roundup(execpath_len, sizeof(char *)) -
756 roundup(sizeof(canary), sizeof(char *));
757 copyout(canary, (void *)imgp->canary, sizeof(canary));
758
759 vectp = (uint32_t *)destp;
760 if (imgp->auxargs) {
761 /*
762 * Allocate room on the stack for the ELF auxargs
763 * array. It has LINUX_AT_COUNT entries.
764 */
765 vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf32_Auxinfo),
766 sizeof(*vectp));
767 }
768
769 /*
770 * Allocate room for the argv[] and env vectors including the
771 * terminating NULL pointers.
772 */
773 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
774
775 /* vectp also becomes our initial stack base. */
776 stack_base = vectp;
777
778 stringp = imgp->args->begin_argv;
779 argc = imgp->args->argc;
780 envc = imgp->args->envc;
781 /* Copy out strings - arguments and environment. */
782 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
783
784 /* Fill in "ps_strings" struct for ps, w, etc. */
785 suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp);
786 suword32(&arginfo->ps_nargvstr, argc);
787
788 /* Fill in argument portion of vector table. */
789 for (; argc > 0; --argc) {
790 suword32(vectp++, (uint32_t)(intptr_t)destp);
791 while (*stringp++ != 0)
792 destp++;
793 destp++;
794 }
795
796 /* A null vector table pointer separates the argp's from the envp's. */
797 suword32(vectp++, 0);
798
799 suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp);
800 suword32(&arginfo->ps_nenvstr, envc);
801
802 /* Fill in environment portion of vector table. */
803 for (; envc > 0; --envc) {
804 suword32(vectp++, (uint32_t)(intptr_t)destp);
805 while (*stringp++ != 0)
806 destp++;
807 destp++;
808 }
809
810 /* The end of the vector table is a null pointer. */
811 suword32(vectp, 0);
812
813 return ((register_t *)stack_base);
814 }
815
816 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0,
817 "32-bit Linux emulation");
818
819 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ;
820 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
821 &linux32_maxdsiz, 0, "");
822 static u_long linux32_maxssiz = LINUX32_MAXSSIZ;
823 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
824 &linux32_maxssiz, 0, "");
825 static u_long linux32_maxvmem = LINUX32_MAXVMEM;
826 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
827 &linux32_maxvmem, 0, "");
828
829 static void
linux32_fixlimit(struct rlimit * rl,int which)830 linux32_fixlimit(struct rlimit *rl, int which)
831 {
832
833 switch (which) {
834 case RLIMIT_DATA:
835 if (linux32_maxdsiz != 0) {
836 if (rl->rlim_cur > linux32_maxdsiz)
837 rl->rlim_cur = linux32_maxdsiz;
838 if (rl->rlim_max > linux32_maxdsiz)
839 rl->rlim_max = linux32_maxdsiz;
840 }
841 break;
842 case RLIMIT_STACK:
843 if (linux32_maxssiz != 0) {
844 if (rl->rlim_cur > linux32_maxssiz)
845 rl->rlim_cur = linux32_maxssiz;
846 if (rl->rlim_max > linux32_maxssiz)
847 rl->rlim_max = linux32_maxssiz;
848 }
849 break;
850 case RLIMIT_VMEM:
851 if (linux32_maxvmem != 0) {
852 if (rl->rlim_cur > linux32_maxvmem)
853 rl->rlim_cur = linux32_maxvmem;
854 if (rl->rlim_max > linux32_maxvmem)
855 rl->rlim_max = linux32_maxvmem;
856 }
857 break;
858 }
859 }
860
861 struct sysentvec elf_linux_sysvec = {
862 .sv_size = LINUX32_SYS_MAXSYSCALL,
863 .sv_table = linux32_sysent,
864 .sv_mask = 0,
865 .sv_errsize = ELAST + 1,
866 .sv_errtbl = linux_errtbl,
867 .sv_transtrap = linux_translate_traps,
868 .sv_fixup = linux_fixup_elf,
869 .sv_sendsig = linux_sendsig,
870 .sv_sigcode = &_binary_linux32_locore_o_start,
871 .sv_szsigcode = &linux_szsigcode,
872 .sv_name = "Linux ELF32",
873 .sv_coredump = elf32_coredump,
874 .sv_imgact_try = linux_exec_imgact_try,
875 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
876 .sv_minuser = VM_MIN_ADDRESS,
877 .sv_maxuser = LINUX32_MAXUSER,
878 .sv_usrstack = LINUX32_USRSTACK,
879 .sv_psstrings = LINUX32_PS_STRINGS,
880 .sv_stackprot = VM_PROT_ALL,
881 .sv_copyout_strings = linux_copyout_strings,
882 .sv_setregs = linux_exec_setregs,
883 .sv_fixlimit = linux32_fixlimit,
884 .sv_maxssiz = &linux32_maxssiz,
885 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP,
886 .sv_set_syscall_retval = cpu_set_syscall_retval,
887 .sv_fetch_syscall_args = linux32_fetch_syscall_args,
888 .sv_syscallnames = NULL,
889 .sv_shared_page_base = LINUX32_SHAREDPAGE,
890 .sv_shared_page_len = PAGE_SIZE,
891 .sv_schedtail = linux_schedtail,
892 .sv_thread_detach = linux_thread_detach,
893 .sv_trap = NULL,
894 };
895
896 static void
linux_vdso_install(void * param)897 linux_vdso_install(void *param)
898 {
899
900 linux_szsigcode = (&_binary_linux32_locore_o_end -
901 &_binary_linux32_locore_o_start);
902
903 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
904 panic("Linux invalid vdso size\n");
905
906 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
907
908 linux_shared_page_obj = __elfN(linux_shared_page_init)
909 (&linux_shared_page_mapping);
910
911 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
912
913 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
914 linux_szsigcode);
915 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
916
917 linux_kplatform = linux_shared_page_mapping +
918 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
919 }
920 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
921 linux_vdso_install, NULL);
922
923 static void
linux_vdso_deinstall(void * param)924 linux_vdso_deinstall(void *param)
925 {
926
927 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
928 }
929 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
930 linux_vdso_deinstall, NULL);
931
932 static char GNU_ABI_VENDOR[] = "GNU";
933 static int GNULINUX_ABI_DESC = 0;
934
935 static bool
linux32_trans_osrel(const Elf_Note * note,int32_t * osrel)936 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
937 {
938 const Elf32_Word *desc;
939 uintptr_t p;
940
941 p = (uintptr_t)(note + 1);
942 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
943
944 desc = (const Elf32_Word *)p;
945 if (desc[0] != GNULINUX_ABI_DESC)
946 return (false);
947
948 /*
949 * For Linux we encode osrel using the Linux convention of
950 * (version << 16) | (major << 8) | (minor)
951 * See macro in linux_mib.h
952 */
953 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
954
955 return (true);
956 }
957
958 static Elf_Brandnote linux32_brandnote = {
959 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
960 .hdr.n_descsz = 16, /* XXX at least 16 */
961 .hdr.n_type = 1,
962 .vendor = GNU_ABI_VENDOR,
963 .flags = BN_TRANSLATE_OSREL,
964 .trans_osrel = linux32_trans_osrel
965 };
966
967 static Elf32_Brandinfo linux_brand = {
968 .brand = ELFOSABI_LINUX,
969 .machine = EM_386,
970 .compat_3_brand = "Linux",
971 .emul_path = linux_emul_path,
972 .interp_path = "/lib/ld-linux.so.1",
973 .sysvec = &elf_linux_sysvec,
974 .interp_newpath = NULL,
975 .brand_note = &linux32_brandnote,
976 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
977 };
978
979 static Elf32_Brandinfo linux_glibc2brand = {
980 .brand = ELFOSABI_LINUX,
981 .machine = EM_386,
982 .compat_3_brand = "Linux",
983 .emul_path = linux_emul_path,
984 .interp_path = "/lib/ld-linux.so.2",
985 .sysvec = &elf_linux_sysvec,
986 .interp_newpath = NULL,
987 .brand_note = &linux32_brandnote,
988 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
989 };
990
991 static Elf32_Brandinfo linux_muslbrand = {
992 .brand = ELFOSABI_LINUX,
993 .machine = EM_386,
994 .compat_3_brand = "Linux",
995 .emul_path = linux_emul_path,
996 .interp_path = "/lib/ld-musl-i386.so.1",
997 .sysvec = &elf_linux_sysvec,
998 .interp_newpath = NULL,
999 .brand_note = &linux32_brandnote,
1000 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1001 };
1002
1003 Elf32_Brandinfo *linux_brandlist[] = {
1004 &linux_brand,
1005 &linux_glibc2brand,
1006 &linux_muslbrand,
1007 NULL
1008 };
1009
1010 static int
linux_elf_modevent(module_t mod,int type,void * data)1011 linux_elf_modevent(module_t mod, int type, void *data)
1012 {
1013 Elf32_Brandinfo **brandinfo;
1014 int error;
1015 struct linux_ioctl_handler **lihp;
1016
1017 error = 0;
1018
1019 switch(type) {
1020 case MOD_LOAD:
1021 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1022 ++brandinfo)
1023 if (elf32_insert_brand_entry(*brandinfo) < 0)
1024 error = EINVAL;
1025 if (error == 0) {
1026 SET_FOREACH(lihp, linux_ioctl_handler_set)
1027 linux32_ioctl_register_handler(*lihp);
1028 stclohz = (stathz ? stathz : hz);
1029 if (bootverbose)
1030 printf("Linux ELF exec handler installed\n");
1031 } else
1032 printf("cannot insert Linux ELF brand handler\n");
1033 break;
1034 case MOD_UNLOAD:
1035 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1036 ++brandinfo)
1037 if (elf32_brand_inuse(*brandinfo))
1038 error = EBUSY;
1039 if (error == 0) {
1040 for (brandinfo = &linux_brandlist[0];
1041 *brandinfo != NULL; ++brandinfo)
1042 if (elf32_remove_brand_entry(*brandinfo) < 0)
1043 error = EINVAL;
1044 }
1045 if (error == 0) {
1046 SET_FOREACH(lihp, linux_ioctl_handler_set)
1047 linux32_ioctl_unregister_handler(*lihp);
1048 if (bootverbose)
1049 printf("Linux ELF exec handler removed\n");
1050 } else
1051 printf("Could not deinstall ELF interpreter entry\n");
1052 break;
1053 default:
1054 return (EOPNOTSUPP);
1055 }
1056 return (error);
1057 }
1058
1059 static moduledata_t linux_elf_mod = {
1060 "linuxelf",
1061 linux_elf_modevent,
1062 0
1063 };
1064
1065 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1066 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1067 FEATURE(linux, "Linux 32bit support");
1068