1 /*-
2 * Copyright (c) 2013 Dmitry Chagin
3 * Copyright (c) 2004 Tim J. Robbins
4 * Copyright (c) 2003 Peter Wemm
5 * Copyright (c) 2002 Doug Rabson
6 * Copyright (c) 1998-1999 Andrew Gallatin
7 * Copyright (c) 1994-1996 Søren Schmidt
8 * All rights reserved.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #define __ELF_WORD_SIZE 64
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/exec.h>
42 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/kernel.h>
46 #include <sys/ktr.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/module.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/resourcevar.h>
53 #include <sys/signalvar.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysent.h>
57 #include <sys/sysproto.h>
58 #include <sys/vnode.h>
59 #include <sys/eventhandler.h>
60
61 #include <vm/vm.h>
62 #include <vm/pmap.h>
63 #include <vm/vm_extern.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_param.h>
68
69 #include <machine/cpu.h>
70 #include <machine/md_var.h>
71 #include <machine/pcb.h>
72 #include <machine/specialreg.h>
73 #include <machine/trap.h>
74
75 #include <amd64/linux/linux.h>
76 #include <amd64/linux/linux_proto.h>
77 #include <compat/linux/linux_emul.h>
78 #include <compat/linux/linux_ioctl.h>
79 #include <compat/linux/linux_mib.h>
80 #include <compat/linux/linux_misc.h>
81 #include <compat/linux/linux_signal.h>
82 #include <compat/linux/linux_sysproto.h>
83 #include <compat/linux/linux_util.h>
84 #include <compat/linux/linux_vdso.h>
85
86 MODULE_VERSION(linux64, 1);
87
88 const char *linux_kplatform;
89 static int linux_szsigcode;
90 static vm_object_t linux_shared_page_obj;
91 static char *linux_shared_page_mapping;
92 extern char _binary_linux_locore_o_start;
93 extern char _binary_linux_locore_o_end;
94
95 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
96
97 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
98
99 static int linux_copyout_strings(struct image_params *imgp,
100 uintptr_t *stack_base);
101 static int linux_fixup_elf(uintptr_t *stack_base,
102 struct image_params *iparams);
103 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
104 static void linux_vdso_install(void *param);
105 static void linux_vdso_deinstall(void *param);
106 static void linux_set_syscall_retval(struct thread *td, int error);
107 static int linux_fetch_syscall_args(struct thread *td);
108 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
109 uintptr_t stack);
110 static int linux_vsyscall(struct thread *td);
111
112 #define LINUX_T_UNKNOWN 255
113 static int _bsd_to_linux_trapcode[] = {
114 LINUX_T_UNKNOWN, /* 0 */
115 6, /* 1 T_PRIVINFLT */
116 LINUX_T_UNKNOWN, /* 2 */
117 3, /* 3 T_BPTFLT */
118 LINUX_T_UNKNOWN, /* 4 */
119 LINUX_T_UNKNOWN, /* 5 */
120 16, /* 6 T_ARITHTRAP */
121 254, /* 7 T_ASTFLT */
122 LINUX_T_UNKNOWN, /* 8 */
123 13, /* 9 T_PROTFLT */
124 1, /* 10 T_TRCTRAP */
125 LINUX_T_UNKNOWN, /* 11 */
126 14, /* 12 T_PAGEFLT */
127 LINUX_T_UNKNOWN, /* 13 */
128 17, /* 14 T_ALIGNFLT */
129 LINUX_T_UNKNOWN, /* 15 */
130 LINUX_T_UNKNOWN, /* 16 */
131 LINUX_T_UNKNOWN, /* 17 */
132 0, /* 18 T_DIVIDE */
133 2, /* 19 T_NMI */
134 4, /* 20 T_OFLOW */
135 5, /* 21 T_BOUND */
136 7, /* 22 T_DNA */
137 8, /* 23 T_DOUBLEFLT */
138 9, /* 24 T_FPOPFLT */
139 10, /* 25 T_TSSFLT */
140 11, /* 26 T_SEGNPFLT */
141 12, /* 27 T_STKFLT */
142 18, /* 28 T_MCHK */
143 19, /* 29 T_XMMFLT */
144 15 /* 30 T_RESERVED */
145 };
146 #define bsd_to_linux_trapcode(code) \
147 ((code)<nitems(_bsd_to_linux_trapcode)? \
148 _bsd_to_linux_trapcode[(code)]: \
149 LINUX_T_UNKNOWN)
150
151 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
152 LINUX_VDSO_SYM_CHAR(linux_platform);
153
154 /*
155 * If FreeBSD & Linux have a difference of opinion about what a trap
156 * means, deal with it here.
157 *
158 * MPSAFE
159 */
160 static int
linux_translate_traps(int signal,int trap_code)161 linux_translate_traps(int signal, int trap_code)
162 {
163
164 if (signal != SIGBUS)
165 return (signal);
166 switch (trap_code) {
167 case T_PROTFLT:
168 case T_TSSFLT:
169 case T_DOUBLEFLT:
170 case T_PAGEFLT:
171 return (SIGSEGV);
172 default:
173 return (signal);
174 }
175 }
176
177 static int
linux_fetch_syscall_args(struct thread * td)178 linux_fetch_syscall_args(struct thread *td)
179 {
180 struct proc *p;
181 struct trapframe *frame;
182 struct syscall_args *sa;
183
184 p = td->td_proc;
185 frame = td->td_frame;
186 sa = &td->td_sa;
187
188 sa->args[0] = frame->tf_rdi;
189 sa->args[1] = frame->tf_rsi;
190 sa->args[2] = frame->tf_rdx;
191 sa->args[3] = frame->tf_rcx;
192 sa->args[4] = frame->tf_r8;
193 sa->args[5] = frame->tf_r9;
194 sa->code = frame->tf_rax;
195
196 if (sa->code >= p->p_sysent->sv_size)
197 /* nosys */
198 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
199 else
200 sa->callp = &p->p_sysent->sv_table[sa->code];
201
202 td->td_retval[0] = 0;
203 return (0);
204 }
205
206 static void
linux_set_syscall_retval(struct thread * td,int error)207 linux_set_syscall_retval(struct thread *td, int error)
208 {
209 struct trapframe *frame;
210
211 frame = td->td_frame;
212
213 switch (error) {
214 case 0:
215 frame->tf_rax = td->td_retval[0];
216 frame->tf_r10 = frame->tf_rcx;
217 break;
218
219 case ERESTART:
220 /*
221 * Reconstruct pc, we know that 'syscall' is 2 bytes,
222 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
223 * We saved this in tf_err.
224 *
225 */
226 frame->tf_rip -= frame->tf_err;
227 frame->tf_r10 = frame->tf_rcx;
228 break;
229
230 case EJUSTRETURN:
231 break;
232
233 default:
234 frame->tf_rax = bsd_to_linux_errno(error);
235 frame->tf_r10 = frame->tf_rcx;
236 break;
237 }
238
239 /*
240 * Differently from FreeBSD native ABI, on Linux only %rcx
241 * and %r11 values are not preserved across the syscall.
242 * Require full context restore to get all registers except
243 * those two restored at return to usermode.
244 *
245 * XXX: Would be great to be able to avoid PCB_FULL_IRET
246 * for the error == 0 case.
247 */
248 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
249 }
250
251 static int
linux_copyout_auxargs(struct image_params * imgp,uintptr_t base)252 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
253 {
254 Elf_Auxargs *args;
255 Elf_Auxinfo *argarray, *pos;
256 struct proc *p;
257 int error, issetugid;
258
259 p = imgp->proc;
260 args = (Elf64_Auxargs *)imgp->auxargs;
261 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
262 M_WAITOK | M_ZERO);
263
264 issetugid = p->p_flag & P_SUGID ? 1 : 0;
265 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
266 imgp->proc->p_sysent->sv_shared_page_base);
267 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
268 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
269 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
270 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
271 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
272 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
273 AUXARGS_ENTRY(pos, AT_BASE, args->base);
274 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
275 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
276 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
277 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
278 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
279 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
280 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
281 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
282 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
283 if (imgp->execpathp != 0)
284 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
285 if (args->execfd != -1)
286 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
287 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
288 AUXARGS_ENTRY(pos, AT_NULL, 0);
289
290 free(imgp->auxargs, M_TEMP);
291 imgp->auxargs = NULL;
292 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
293
294 error = copyout(argarray, (void *)base,
295 sizeof(*argarray) * LINUX_AT_COUNT);
296 free(argarray, M_TEMP);
297 return (error);
298 }
299
300 static int
linux_fixup_elf(uintptr_t * stack_base,struct image_params * imgp)301 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
302 {
303 Elf_Addr *base;
304
305 base = (Elf64_Addr *)*stack_base;
306 base--;
307 if (suword(base, (uint64_t)imgp->args->argc) == -1)
308 return (EFAULT);
309
310 *stack_base = (uintptr_t)base;
311 return (0);
312 }
313
314 /*
315 * Copy strings out to the new process address space, constructing new arg
316 * and env vector tables. Return a pointer to the base so that it can be used
317 * as the initial stack pointer.
318 */
319 static int
linux_copyout_strings(struct image_params * imgp,uintptr_t * stack_base)320 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
321 {
322 int argc, envc, error;
323 char **vectp;
324 char *stringp;
325 uintptr_t destp, ustringp;
326 struct ps_strings *arginfo;
327 char canary[LINUX_AT_RANDOM_LEN];
328 size_t execpath_len;
329 struct proc *p;
330
331 p = imgp->proc;
332 arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
333 destp = (uintptr_t)arginfo;
334
335 if (imgp->execpath != NULL && imgp->auxargs != NULL) {
336 execpath_len = strlen(imgp->execpath) + 1;
337 destp -= execpath_len;
338 destp = rounddown2(destp, sizeof(void *));
339 imgp->execpathp = (void *)destp;
340 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
341 if (error != 0)
342 return (error);
343 }
344
345 /* Prepare the canary for SSP. */
346 arc4rand(canary, sizeof(canary), 0);
347 destp -= roundup(sizeof(canary), sizeof(void *));
348 imgp->canary = (void *)destp;
349 error = copyout(canary, imgp->canary, sizeof(canary));
350 if (error != 0)
351 return (error);
352
353 /* Allocate room for the argument and environment strings. */
354 destp -= ARG_MAX - imgp->args->stringspace;
355 destp = rounddown2(destp, sizeof(void *));
356 ustringp = destp;
357
358 if (imgp->auxargs) {
359 /*
360 * Allocate room on the stack for the ELF auxargs
361 * array. It has LINUX_AT_COUNT entries.
362 */
363 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
364 destp = rounddown2(destp, sizeof(void *));
365 }
366
367 vectp = (char **)destp;
368
369 /*
370 * Allocate room for the argv[] and env vectors including the
371 * terminating NULL pointers.
372 */
373 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
374
375 /*
376 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
377 * One "long argc" will be prepended later.
378 */
379 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
380
381 /* vectp also becomes our initial stack base. */
382 *stack_base = (uintptr_t)vectp;
383
384 stringp = imgp->args->begin_argv;
385 argc = imgp->args->argc;
386 envc = imgp->args->envc;
387
388 /* Copy out strings - arguments and environment. */
389 error = copyout(stringp, (void *)ustringp,
390 ARG_MAX - imgp->args->stringspace);
391 if (error != 0)
392 return (error);
393
394 /* Fill in "ps_strings" struct for ps, w, etc. */
395 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
396 suword(&arginfo->ps_nargvstr, argc) != 0)
397 return (EFAULT);
398
399 /* Fill in argument portion of vector table. */
400 for (; argc > 0; --argc) {
401 if (suword(vectp++, ustringp) != 0)
402 return (EFAULT);
403 while (*stringp++ != 0)
404 ustringp++;
405 ustringp++;
406 }
407
408 /* A null vector table pointer separates the argp's from the envp's. */
409 if (suword(vectp++, 0) != 0)
410 return (EFAULT);
411
412 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
413 suword(&arginfo->ps_nenvstr, envc) != 0)
414 return (EFAULT);
415
416 /* Fill in environment portion of vector table. */
417 for (; envc > 0; --envc) {
418 if (suword(vectp++, ustringp) != 0)
419 return (EFAULT);
420 while (*stringp++ != 0)
421 ustringp++;
422 ustringp++;
423 }
424
425 /* The end of the vector table is a null pointer. */
426 if (suword(vectp, 0) != 0)
427 return (EFAULT);
428
429 if (imgp->auxargs) {
430 vectp++;
431 error = imgp->sysent->sv_copyout_auxargs(imgp,
432 (uintptr_t)vectp);
433 if (error != 0)
434 return (error);
435 }
436
437 return (0);
438 }
439
440 /*
441 * Reset registers to default values on exec.
442 */
443 static void
linux_exec_setregs(struct thread * td,struct image_params * imgp,uintptr_t stack)444 linux_exec_setregs(struct thread *td, struct image_params *imgp,
445 uintptr_t stack)
446 {
447 struct trapframe *regs;
448 struct pcb *pcb;
449 register_t saved_rflags;
450
451 regs = td->td_frame;
452 pcb = td->td_pcb;
453
454 if (td->td_proc->p_md.md_ldt != NULL)
455 user_ldt_free(td);
456
457 pcb->pcb_fsbase = 0;
458 pcb->pcb_gsbase = 0;
459 clear_pcb_flags(pcb, PCB_32BIT);
460 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
461 set_pcb_flags(pcb, PCB_FULL_IRET);
462
463 saved_rflags = regs->tf_rflags & PSL_T;
464 bzero((char *)regs, sizeof(struct trapframe));
465 regs->tf_rip = imgp->entry_addr;
466 regs->tf_rsp = stack;
467 regs->tf_rflags = PSL_USER | saved_rflags;
468 regs->tf_ss = _udatasel;
469 regs->tf_cs = _ucodesel;
470 regs->tf_ds = _udatasel;
471 regs->tf_es = _udatasel;
472 regs->tf_fs = _ufssel;
473 regs->tf_gs = _ugssel;
474 regs->tf_flags = TF_HASSEGS;
475
476 x86_clear_dbregs(pcb);
477
478 /*
479 * Drop the FP state if we hold it, so that the process gets a
480 * clean FP state if it uses the FPU again.
481 */
482 fpstate_drop(td);
483 }
484
485 /*
486 * Copied from amd64/amd64/machdep.c
487 *
488 * XXX fpu state need? don't think so
489 */
490 int
linux_rt_sigreturn(struct thread * td,struct linux_rt_sigreturn_args * args)491 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
492 {
493 struct proc *p;
494 struct l_ucontext uc;
495 struct l_sigcontext *context;
496 struct trapframe *regs;
497 unsigned long rflags;
498 int error;
499 ksiginfo_t ksi;
500
501 regs = td->td_frame;
502 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
503 if (error != 0)
504 return (error);
505
506 p = td->td_proc;
507 context = &uc.uc_mcontext;
508 rflags = context->sc_rflags;
509
510 /*
511 * Don't allow users to change privileged or reserved flags.
512 */
513 /*
514 * XXX do allow users to change the privileged flag PSL_RF.
515 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
516 * should sometimes set it there too. tf_rflags is kept in
517 * the signal context during signal handling and there is no
518 * other place to remember it, so the PSL_RF bit may be
519 * corrupted by the signal handler without us knowing.
520 * Corruption of the PSL_RF bit at worst causes one more or
521 * one less debugger trap, so allowing it is fairly harmless.
522 */
523 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
524 uprintf("pid %d comm %s linux mangled rflags %#lx\n",
525 p->p_pid, p->p_comm, rflags);
526 return (EINVAL);
527 }
528
529 /*
530 * Don't allow users to load a valid privileged %cs. Let the
531 * hardware check for invalid selectors, excess privilege in
532 * other selectors, invalid %eip's and invalid %esp's.
533 */
534 if (!CS_SECURE(context->sc_cs)) {
535 uprintf("pid %d comm %s linux mangled cs %#x\n",
536 p->p_pid, p->p_comm, context->sc_cs);
537 ksiginfo_init_trap(&ksi);
538 ksi.ksi_signo = SIGBUS;
539 ksi.ksi_code = BUS_OBJERR;
540 ksi.ksi_trapno = T_PROTFLT;
541 ksi.ksi_addr = (void *)regs->tf_rip;
542 trapsignal(td, &ksi);
543 return (EINVAL);
544 }
545
546 PROC_LOCK(p);
547 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
548 SIG_CANTMASK(td->td_sigmask);
549 signotify(td);
550 PROC_UNLOCK(p);
551
552 regs->tf_rdi = context->sc_rdi;
553 regs->tf_rsi = context->sc_rsi;
554 regs->tf_rdx = context->sc_rdx;
555 regs->tf_rbp = context->sc_rbp;
556 regs->tf_rbx = context->sc_rbx;
557 regs->tf_rcx = context->sc_rcx;
558 regs->tf_rax = context->sc_rax;
559 regs->tf_rip = context->sc_rip;
560 regs->tf_rsp = context->sc_rsp;
561 regs->tf_r8 = context->sc_r8;
562 regs->tf_r9 = context->sc_r9;
563 regs->tf_r10 = context->sc_r10;
564 regs->tf_r11 = context->sc_r11;
565 regs->tf_r12 = context->sc_r12;
566 regs->tf_r13 = context->sc_r13;
567 regs->tf_r14 = context->sc_r14;
568 regs->tf_r15 = context->sc_r15;
569 regs->tf_cs = context->sc_cs;
570 regs->tf_err = context->sc_err;
571 regs->tf_rflags = rflags;
572
573 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
574 return (EJUSTRETURN);
575 }
576
577 /*
578 * copied from amd64/amd64/machdep.c
579 *
580 * Send an interrupt to process.
581 */
582 static void
linux_rt_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)583 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
584 {
585 struct l_rt_sigframe sf, *sfp;
586 struct proc *p;
587 struct thread *td;
588 struct sigacts *psp;
589 caddr_t sp;
590 struct trapframe *regs;
591 int sig, code;
592 int oonstack;
593
594 td = curthread;
595 p = td->td_proc;
596 PROC_LOCK_ASSERT(p, MA_OWNED);
597 sig = ksi->ksi_signo;
598 psp = p->p_sigacts;
599 code = ksi->ksi_code;
600 mtx_assert(&psp->ps_mtx, MA_OWNED);
601 regs = td->td_frame;
602 oonstack = sigonstack(regs->tf_rsp);
603
604 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
605 catcher, sig, mask, code);
606
607 /* Allocate space for the signal handler context. */
608 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
609 SIGISMEMBER(psp->ps_sigonstack, sig)) {
610 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
611 sizeof(struct l_rt_sigframe);
612 } else
613 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
614 /* Align to 16 bytes. */
615 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
616 mtx_unlock(&psp->ps_mtx);
617
618 /* Translate the signal. */
619 sig = bsd_to_linux_signal(sig);
620
621 /* Save user context. */
622 bzero(&sf, sizeof(sf));
623 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
624 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
625
626 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
627 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
628 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
629 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
630 PROC_UNLOCK(p);
631
632 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi;
633 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi;
634 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx;
635 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp;
636 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx;
637 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx;
638 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax;
639 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip;
640 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp;
641 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8;
642 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9;
643 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10;
644 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11;
645 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12;
646 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13;
647 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14;
648 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15;
649 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
650 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
651 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err;
652 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
653 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
654
655 /* Build the argument list for the signal handler. */
656 regs->tf_rdi = sig; /* arg 1 in %rdi */
657 regs->tf_rax = 0;
658 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
659 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */
660
661 sf.sf_handler = catcher;
662 /* Fill in POSIX parts. */
663 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
664
665 /* Copy the sigframe out to the user's stack. */
666 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
667 PROC_LOCK(p);
668 sigexit(td, SIGILL);
669 }
670
671 regs->tf_rsp = (long)sfp;
672 regs->tf_rip = linux_rt_sigcode;
673 regs->tf_rflags &= ~(PSL_T | PSL_D);
674 regs->tf_cs = _ucodesel;
675 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
676 PROC_LOCK(p);
677 mtx_lock(&psp->ps_mtx);
678 }
679
680 #define LINUX_VSYSCALL_START (-10UL << 20)
681 #define LINUX_VSYSCALL_SZ 1024
682
683 const unsigned long linux_vsyscall_vector[] = {
684 LINUX_SYS_gettimeofday,
685 LINUX_SYS_linux_time,
686 LINUX_SYS_linux_getcpu,
687 };
688
689 static int
linux_vsyscall(struct thread * td)690 linux_vsyscall(struct thread *td)
691 {
692 struct trapframe *frame;
693 uint64_t retqaddr;
694 int code, traced;
695 int error;
696
697 frame = td->td_frame;
698
699 /* Check %rip for vsyscall area. */
700 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
701 return (EINVAL);
702 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
703 return (EINVAL);
704 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
705 if (code >= nitems(linux_vsyscall_vector))
706 return (EINVAL);
707
708 /*
709 * vsyscall called as callq *(%rax), so we must
710 * use return address from %rsp and also fixup %rsp.
711 */
712 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
713 if (error)
714 return (error);
715
716 frame->tf_rip = retqaddr;
717 frame->tf_rax = linux_vsyscall_vector[code];
718 frame->tf_rsp += 8;
719
720 traced = (frame->tf_flags & PSL_T);
721
722 amd64_syscall(td, traced);
723
724 return (0);
725 }
726
727 struct sysentvec elf_linux_sysvec = {
728 .sv_size = LINUX_SYS_MAXSYSCALL,
729 .sv_table = linux_sysent,
730 .sv_transtrap = linux_translate_traps,
731 .sv_fixup = linux_fixup_elf,
732 .sv_sendsig = linux_rt_sendsig,
733 .sv_sigcode = &_binary_linux_locore_o_start,
734 .sv_szsigcode = &linux_szsigcode,
735 .sv_name = "Linux ELF64",
736 .sv_coredump = elf64_coredump,
737 .sv_imgact_try = linux_exec_imgact_try,
738 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
739 .sv_minuser = VM_MIN_ADDRESS,
740 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48,
741 .sv_usrstack = USRSTACK_LA48,
742 .sv_psstrings = PS_STRINGS_LA48,
743 .sv_psstringssz = sizeof(struct ps_strings),
744 .sv_stackprot = VM_PROT_ALL,
745 .sv_copyout_auxargs = linux_copyout_auxargs,
746 .sv_copyout_strings = linux_copyout_strings,
747 .sv_setregs = linux_exec_setregs,
748 .sv_fixlimit = NULL,
749 .sv_maxssiz = NULL,
750 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
751 SV_SIG_WAITNDQ,
752 .sv_set_syscall_retval = linux_set_syscall_retval,
753 .sv_fetch_syscall_args = linux_fetch_syscall_args,
754 .sv_syscallnames = NULL,
755 .sv_shared_page_base = SHAREDPAGE_LA48,
756 .sv_shared_page_len = PAGE_SIZE,
757 .sv_schedtail = linux_schedtail,
758 .sv_thread_detach = linux_thread_detach,
759 .sv_trap = linux_vsyscall,
760 .sv_onexec = linux_on_exec,
761 .sv_onexit = linux_on_exit,
762 .sv_ontdexit = linux_thread_dtor,
763 .sv_setid_allowed = &linux_setid_allowed_query,
764 };
765
766 static void
linux_vdso_install(void * param)767 linux_vdso_install(void *param)
768 {
769
770 amd64_lower_shared_page(&elf_linux_sysvec);
771
772 linux_szsigcode = (&_binary_linux_locore_o_end -
773 &_binary_linux_locore_o_start);
774
775 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
776 panic("Linux invalid vdso size\n");
777
778 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
779
780 linux_shared_page_obj = __elfN(linux_shared_page_init)
781 (&linux_shared_page_mapping);
782
783 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
784
785 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
786 linux_szsigcode);
787 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
788
789 linux_kplatform = linux_shared_page_mapping +
790 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
791 }
792 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
793 linux_vdso_install, NULL);
794
795 static void
linux_vdso_deinstall(void * param)796 linux_vdso_deinstall(void *param)
797 {
798
799 __elfN(linux_shared_page_fini)(linux_shared_page_obj,
800 linux_shared_page_mapping);
801 }
802 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
803 linux_vdso_deinstall, NULL);
804
805 static char GNULINUX_ABI_VENDOR[] = "GNU";
806 static int GNULINUX_ABI_DESC = 0;
807
808 static bool
linux_trans_osrel(const Elf_Note * note,int32_t * osrel)809 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
810 {
811 const Elf32_Word *desc;
812 uintptr_t p;
813
814 p = (uintptr_t)(note + 1);
815 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
816
817 desc = (const Elf32_Word *)p;
818 if (desc[0] != GNULINUX_ABI_DESC)
819 return (false);
820
821 /*
822 * For Linux we encode osrel using the Linux convention of
823 * (version << 16) | (major << 8) | (minor)
824 * See macro in linux_mib.h
825 */
826 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
827
828 return (true);
829 }
830
831 static Elf_Brandnote linux64_brandnote = {
832 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR),
833 .hdr.n_descsz = 16,
834 .hdr.n_type = 1,
835 .vendor = GNULINUX_ABI_VENDOR,
836 .flags = BN_TRANSLATE_OSREL,
837 .trans_osrel = linux_trans_osrel
838 };
839
840 static Elf64_Brandinfo linux_glibc2brand = {
841 .brand = ELFOSABI_LINUX,
842 .machine = EM_X86_64,
843 .compat_3_brand = "Linux",
844 .emul_path = linux_emul_path,
845 .interp_path = "/lib64/ld-linux-x86-64.so.2",
846 .sysvec = &elf_linux_sysvec,
847 .interp_newpath = NULL,
848 .brand_note = &linux64_brandnote,
849 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
850 };
851
852 static Elf64_Brandinfo linux_glibc2brandshort = {
853 .brand = ELFOSABI_LINUX,
854 .machine = EM_X86_64,
855 .compat_3_brand = "Linux",
856 .emul_path = linux_emul_path,
857 .interp_path = "/lib64/ld-linux.so.2",
858 .sysvec = &elf_linux_sysvec,
859 .interp_newpath = NULL,
860 .brand_note = &linux64_brandnote,
861 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
862 };
863
864 static Elf64_Brandinfo linux_muslbrand = {
865 .brand = ELFOSABI_LINUX,
866 .machine = EM_X86_64,
867 .compat_3_brand = "Linux",
868 .emul_path = linux_emul_path,
869 .interp_path = "/lib/ld-musl-x86_64.so.1",
870 .sysvec = &elf_linux_sysvec,
871 .interp_newpath = NULL,
872 .brand_note = &linux64_brandnote,
873 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
874 };
875
876 Elf64_Brandinfo *linux_brandlist[] = {
877 &linux_glibc2brand,
878 &linux_glibc2brandshort,
879 &linux_muslbrand,
880 NULL
881 };
882
883 static int
linux64_elf_modevent(module_t mod,int type,void * data)884 linux64_elf_modevent(module_t mod, int type, void *data)
885 {
886 Elf64_Brandinfo **brandinfo;
887 int error;
888 struct linux_ioctl_handler **lihp;
889
890 error = 0;
891
892 switch(type) {
893 case MOD_LOAD:
894 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
895 ++brandinfo)
896 if (elf64_insert_brand_entry(*brandinfo) < 0)
897 error = EINVAL;
898 if (error == 0) {
899 SET_FOREACH(lihp, linux_ioctl_handler_set)
900 linux_ioctl_register_handler(*lihp);
901 stclohz = (stathz ? stathz : hz);
902 if (bootverbose)
903 printf("Linux x86-64 ELF exec handler installed\n");
904 } else
905 printf("cannot insert Linux x86-64 ELF brand handler\n");
906 break;
907 case MOD_UNLOAD:
908 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
909 ++brandinfo)
910 if (elf64_brand_inuse(*brandinfo))
911 error = EBUSY;
912 if (error == 0) {
913 for (brandinfo = &linux_brandlist[0];
914 *brandinfo != NULL; ++brandinfo)
915 if (elf64_remove_brand_entry(*brandinfo) < 0)
916 error = EINVAL;
917 }
918 if (error == 0) {
919 SET_FOREACH(lihp, linux_ioctl_handler_set)
920 linux_ioctl_unregister_handler(*lihp);
921 if (bootverbose)
922 printf("Linux ELF exec handler removed\n");
923 } else
924 printf("Could not deinstall ELF interpreter entry\n");
925 break;
926 default:
927 return (EOPNOTSUPP);
928 }
929 return (error);
930 }
931
932 static moduledata_t linux64_elf_mod = {
933 "linux64elf",
934 linux64_elf_modevent,
935 0
936 };
937
938 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
939 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
940 FEATURE(linux64, "Linux 64bit support");
941