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, LINUX_AT_CLKTCK, stclohz);
269 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
270 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
271 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
272 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
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(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
282 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
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, AT_NULL, 0);
288
289 free(imgp->auxargs, M_TEMP);
290 imgp->auxargs = NULL;
291 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
292
293 error = copyout(argarray, (void *)base,
294 sizeof(*argarray) * LINUX_AT_COUNT);
295 free(argarray, M_TEMP);
296 return (error);
297 }
298
299 static int
linux_fixup_elf(uintptr_t * stack_base,struct image_params * imgp)300 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
301 {
302 Elf_Addr *base;
303
304 base = (Elf64_Addr *)*stack_base;
305 base--;
306 if (suword(base, (uint64_t)imgp->args->argc) == -1)
307 return (EFAULT);
308
309 *stack_base = (uintptr_t)base;
310 return (0);
311 }
312
313 /*
314 * Copy strings out to the new process address space, constructing new arg
315 * and env vector tables. Return a pointer to the base so that it can be used
316 * as the initial stack pointer.
317 */
318 static int
linux_copyout_strings(struct image_params * imgp,uintptr_t * stack_base)319 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
320 {
321 int argc, envc, error;
322 char **vectp;
323 char *stringp;
324 uintptr_t destp, ustringp;
325 struct ps_strings *arginfo;
326 char canary[LINUX_AT_RANDOM_LEN];
327 size_t execpath_len;
328 struct proc *p;
329
330 /* Calculate string base and vector table pointers. */
331 if (imgp->execpath != NULL && imgp->auxargs != NULL)
332 execpath_len = strlen(imgp->execpath) + 1;
333 else
334 execpath_len = 0;
335
336 p = imgp->proc;
337 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
338 destp = (uintptr_t)arginfo;
339
340 if (execpath_len != 0) {
341 destp -= execpath_len;
342 destp = rounddown2(destp, sizeof(void *));
343 imgp->execpathp = (void *)destp;
344 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
345 if (error != 0)
346 return (error);
347 }
348
349 /* Prepare the canary for SSP. */
350 arc4rand(canary, sizeof(canary), 0);
351 destp -= roundup(sizeof(canary), sizeof(void *));
352 imgp->canary = (void *)destp;
353 error = copyout(canary, imgp->canary, sizeof(canary));
354 if (error != 0)
355 return (error);
356
357 /* Allocate room for the argument and environment strings. */
358 destp -= ARG_MAX - imgp->args->stringspace;
359 destp = rounddown2(destp, sizeof(void *));
360 ustringp = destp;
361
362 if (imgp->auxargs) {
363 /*
364 * Allocate room on the stack for the ELF auxargs
365 * array. It has LINUX_AT_COUNT entries.
366 */
367 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
368 destp = rounddown2(destp, sizeof(void *));
369 }
370
371 vectp = (char **)destp;
372
373 /*
374 * Allocate room for the argv[] and env vectors including the
375 * terminating NULL pointers.
376 */
377 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
378
379 /*
380 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
381 * One "long argc" will be prepended later.
382 */
383 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
384
385 /* vectp also becomes our initial stack base. */
386 *stack_base = (uintptr_t)vectp;
387
388 stringp = imgp->args->begin_argv;
389 argc = imgp->args->argc;
390 envc = imgp->args->envc;
391
392 /* Copy out strings - arguments and environment. */
393 error = copyout(stringp, (void *)ustringp,
394 ARG_MAX - imgp->args->stringspace);
395 if (error != 0)
396 return (error);
397
398 /* Fill in "ps_strings" struct for ps, w, etc. */
399 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
400 suword(&arginfo->ps_nargvstr, argc) != 0)
401 return (EFAULT);
402
403 /* Fill in argument portion of vector table. */
404 for (; argc > 0; --argc) {
405 if (suword(vectp++, ustringp) != 0)
406 return (EFAULT);
407 while (*stringp++ != 0)
408 ustringp++;
409 ustringp++;
410 }
411
412 /* A null vector table pointer separates the argp's from the envp's. */
413 if (suword(vectp++, 0) != 0)
414 return (EFAULT);
415
416 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
417 suword(&arginfo->ps_nenvstr, envc) != 0)
418 return (EFAULT);
419
420 /* Fill in environment portion of vector table. */
421 for (; envc > 0; --envc) {
422 if (suword(vectp++, ustringp) != 0)
423 return (EFAULT);
424 while (*stringp++ != 0)
425 ustringp++;
426 ustringp++;
427 }
428
429 /* The end of the vector table is a null pointer. */
430 if (suword(vectp, 0) != 0)
431 return (EFAULT);
432
433 if (imgp->auxargs) {
434 vectp++;
435 error = imgp->sysent->sv_copyout_auxargs(imgp,
436 (uintptr_t)vectp);
437 if (error != 0)
438 return (error);
439 }
440
441 return (0);
442 }
443
444 /*
445 * Reset registers to default values on exec.
446 */
447 static void
linux_exec_setregs(struct thread * td,struct image_params * imgp,uintptr_t stack)448 linux_exec_setregs(struct thread *td, struct image_params *imgp,
449 uintptr_t stack)
450 {
451 struct trapframe *regs;
452 struct pcb *pcb;
453 register_t saved_rflags;
454
455 regs = td->td_frame;
456 pcb = td->td_pcb;
457
458 if (td->td_proc->p_md.md_ldt != NULL)
459 user_ldt_free(td);
460
461 pcb->pcb_fsbase = 0;
462 pcb->pcb_gsbase = 0;
463 clear_pcb_flags(pcb, PCB_32BIT);
464 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
465 set_pcb_flags(pcb, PCB_FULL_IRET);
466
467 saved_rflags = regs->tf_rflags & PSL_T;
468 bzero((char *)regs, sizeof(struct trapframe));
469 regs->tf_rip = imgp->entry_addr;
470 regs->tf_rsp = stack;
471 regs->tf_rflags = PSL_USER | saved_rflags;
472 regs->tf_ss = _udatasel;
473 regs->tf_cs = _ucodesel;
474 regs->tf_ds = _udatasel;
475 regs->tf_es = _udatasel;
476 regs->tf_fs = _ufssel;
477 regs->tf_gs = _ugssel;
478 regs->tf_flags = TF_HASSEGS;
479
480 /*
481 * Reset the hardware debug registers if they were in use.
482 * They won't have any meaning for the newly exec'd process.
483 */
484 if (pcb->pcb_flags & PCB_DBREGS) {
485 pcb->pcb_dr0 = 0;
486 pcb->pcb_dr1 = 0;
487 pcb->pcb_dr2 = 0;
488 pcb->pcb_dr3 = 0;
489 pcb->pcb_dr6 = 0;
490 pcb->pcb_dr7 = 0;
491 if (pcb == curpcb) {
492 /*
493 * Clear the debug registers on the running
494 * CPU, otherwise they will end up affecting
495 * the next process we switch to.
496 */
497 reset_dbregs();
498 }
499 clear_pcb_flags(pcb, PCB_DBREGS);
500 }
501
502 /*
503 * Drop the FP state if we hold it, so that the process gets a
504 * clean FP state if it uses the FPU again.
505 */
506 fpstate_drop(td);
507 }
508
509 /*
510 * Copied from amd64/amd64/machdep.c
511 *
512 * XXX fpu state need? don't think so
513 */
514 int
linux_rt_sigreturn(struct thread * td,struct linux_rt_sigreturn_args * args)515 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
516 {
517 struct proc *p;
518 struct l_ucontext uc;
519 struct l_sigcontext *context;
520 struct trapframe *regs;
521 unsigned long rflags;
522 int error;
523 ksiginfo_t ksi;
524
525 regs = td->td_frame;
526 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
527 if (error != 0)
528 return (error);
529
530 p = td->td_proc;
531 context = &uc.uc_mcontext;
532 rflags = context->sc_rflags;
533
534 /*
535 * Don't allow users to change privileged or reserved flags.
536 */
537 /*
538 * XXX do allow users to change the privileged flag PSL_RF.
539 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
540 * should sometimes set it there too. tf_rflags is kept in
541 * the signal context during signal handling and there is no
542 * other place to remember it, so the PSL_RF bit may be
543 * corrupted by the signal handler without us knowing.
544 * Corruption of the PSL_RF bit at worst causes one more or
545 * one less debugger trap, so allowing it is fairly harmless.
546 */
547
548 #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
549 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
550 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
551 return (EINVAL);
552 }
553
554 /*
555 * Don't allow users to load a valid privileged %cs. Let the
556 * hardware check for invalid selectors, excess privilege in
557 * other selectors, invalid %eip's and invalid %esp's.
558 */
559 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
560 if (!CS_SECURE(context->sc_cs)) {
561 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
562 ksiginfo_init_trap(&ksi);
563 ksi.ksi_signo = SIGBUS;
564 ksi.ksi_code = BUS_OBJERR;
565 ksi.ksi_trapno = T_PROTFLT;
566 ksi.ksi_addr = (void *)regs->tf_rip;
567 trapsignal(td, &ksi);
568 return (EINVAL);
569 }
570
571 PROC_LOCK(p);
572 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
573 SIG_CANTMASK(td->td_sigmask);
574 signotify(td);
575 PROC_UNLOCK(p);
576
577 regs->tf_rdi = context->sc_rdi;
578 regs->tf_rsi = context->sc_rsi;
579 regs->tf_rdx = context->sc_rdx;
580 regs->tf_rbp = context->sc_rbp;
581 regs->tf_rbx = context->sc_rbx;
582 regs->tf_rcx = context->sc_rcx;
583 regs->tf_rax = context->sc_rax;
584 regs->tf_rip = context->sc_rip;
585 regs->tf_rsp = context->sc_rsp;
586 regs->tf_r8 = context->sc_r8;
587 regs->tf_r9 = context->sc_r9;
588 regs->tf_r10 = context->sc_r10;
589 regs->tf_r11 = context->sc_r11;
590 regs->tf_r12 = context->sc_r12;
591 regs->tf_r13 = context->sc_r13;
592 regs->tf_r14 = context->sc_r14;
593 regs->tf_r15 = context->sc_r15;
594 regs->tf_cs = context->sc_cs;
595 regs->tf_err = context->sc_err;
596 regs->tf_rflags = rflags;
597
598 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
599 return (EJUSTRETURN);
600 }
601
602 /*
603 * copied from amd64/amd64/machdep.c
604 *
605 * Send an interrupt to process.
606 */
607 static void
linux_rt_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)608 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
609 {
610 struct l_rt_sigframe sf, *sfp;
611 struct proc *p;
612 struct thread *td;
613 struct sigacts *psp;
614 caddr_t sp;
615 struct trapframe *regs;
616 int sig, code;
617 int oonstack;
618
619 td = curthread;
620 p = td->td_proc;
621 PROC_LOCK_ASSERT(p, MA_OWNED);
622 sig = ksi->ksi_signo;
623 psp = p->p_sigacts;
624 code = ksi->ksi_code;
625 mtx_assert(&psp->ps_mtx, MA_OWNED);
626 regs = td->td_frame;
627 oonstack = sigonstack(regs->tf_rsp);
628
629 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
630 catcher, sig, mask, code);
631
632 /* Allocate space for the signal handler context. */
633 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
634 SIGISMEMBER(psp->ps_sigonstack, sig)) {
635 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
636 sizeof(struct l_rt_sigframe);
637 } else
638 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
639 /* Align to 16 bytes. */
640 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
641 mtx_unlock(&psp->ps_mtx);
642
643 /* Translate the signal. */
644 sig = bsd_to_linux_signal(sig);
645
646 /* Save user context. */
647 bzero(&sf, sizeof(sf));
648 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
649 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
650
651 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
652 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
653 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
654 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
655 PROC_UNLOCK(p);
656
657 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi;
658 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi;
659 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx;
660 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp;
661 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx;
662 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx;
663 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax;
664 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip;
665 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp;
666 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8;
667 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9;
668 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10;
669 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11;
670 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12;
671 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13;
672 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14;
673 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15;
674 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
675 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
676 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err;
677 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
678 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
679
680 /* Build the argument list for the signal handler. */
681 regs->tf_rdi = sig; /* arg 1 in %rdi */
682 regs->tf_rax = 0;
683 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
684 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */
685
686 sf.sf_handler = catcher;
687 /* Fill in POSIX parts. */
688 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
689
690 /* Copy the sigframe out to the user's stack. */
691 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
692 PROC_LOCK(p);
693 sigexit(td, SIGILL);
694 }
695
696 regs->tf_rsp = (long)sfp;
697 regs->tf_rip = linux_rt_sigcode;
698 regs->tf_rflags &= ~(PSL_T | PSL_D);
699 regs->tf_cs = _ucodesel;
700 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
701 PROC_LOCK(p);
702 mtx_lock(&psp->ps_mtx);
703 }
704
705 #define LINUX_VSYSCALL_START (-10UL << 20)
706 #define LINUX_VSYSCALL_SZ 1024
707
708 const unsigned long linux_vsyscall_vector[] = {
709 LINUX_SYS_gettimeofday,
710 LINUX_SYS_linux_time,
711 LINUX_SYS_linux_getcpu,
712 };
713
714 static int
linux_vsyscall(struct thread * td)715 linux_vsyscall(struct thread *td)
716 {
717 struct trapframe *frame;
718 uint64_t retqaddr;
719 int code, traced;
720 int error;
721
722 frame = td->td_frame;
723
724 /* Check %rip for vsyscall area. */
725 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
726 return (EINVAL);
727 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
728 return (EINVAL);
729 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
730 if (code >= nitems(linux_vsyscall_vector))
731 return (EINVAL);
732
733 /*
734 * vsyscall called as callq *(%rax), so we must
735 * use return address from %rsp and also fixup %rsp.
736 */
737 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
738 if (error)
739 return (error);
740
741 frame->tf_rip = retqaddr;
742 frame->tf_rax = linux_vsyscall_vector[code];
743 frame->tf_rsp += 8;
744
745 traced = (frame->tf_flags & PSL_T);
746
747 amd64_syscall(td, traced);
748
749 return (0);
750 }
751
752 struct sysentvec elf_linux_sysvec = {
753 .sv_size = LINUX_SYS_MAXSYSCALL,
754 .sv_table = linux_sysent,
755 .sv_transtrap = linux_translate_traps,
756 .sv_fixup = linux_fixup_elf,
757 .sv_sendsig = linux_rt_sendsig,
758 .sv_sigcode = &_binary_linux_locore_o_start,
759 .sv_szsigcode = &linux_szsigcode,
760 .sv_name = "Linux ELF64",
761 .sv_coredump = elf64_coredump,
762 .sv_imgact_try = linux_exec_imgact_try,
763 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
764 .sv_minuser = VM_MIN_ADDRESS,
765 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48,
766 .sv_usrstack = USRSTACK_LA48,
767 .sv_psstrings = PS_STRINGS_LA48,
768 .sv_stackprot = VM_PROT_ALL,
769 .sv_copyout_auxargs = linux_copyout_auxargs,
770 .sv_copyout_strings = linux_copyout_strings,
771 .sv_setregs = linux_exec_setregs,
772 .sv_fixlimit = NULL,
773 .sv_maxssiz = NULL,
774 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP,
775 .sv_set_syscall_retval = linux_set_syscall_retval,
776 .sv_fetch_syscall_args = linux_fetch_syscall_args,
777 .sv_syscallnames = NULL,
778 .sv_shared_page_base = SHAREDPAGE_LA48,
779 .sv_shared_page_len = PAGE_SIZE,
780 .sv_schedtail = linux_schedtail,
781 .sv_thread_detach = linux_thread_detach,
782 .sv_trap = linux_vsyscall,
783 .sv_onexec = linux_on_exec,
784 .sv_onexit = linux_on_exit,
785 .sv_ontdexit = linux_thread_dtor,
786 };
787
788 static void
linux_vdso_install(void * param)789 linux_vdso_install(void *param)
790 {
791
792 amd64_lower_shared_page(&elf_linux_sysvec);
793
794 linux_szsigcode = (&_binary_linux_locore_o_end -
795 &_binary_linux_locore_o_start);
796
797 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
798 panic("Linux invalid vdso size\n");
799
800 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
801
802 linux_shared_page_obj = __elfN(linux_shared_page_init)
803 (&linux_shared_page_mapping);
804
805 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
806
807 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
808 linux_szsigcode);
809 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
810
811 linux_kplatform = linux_shared_page_mapping +
812 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
813 }
814 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
815 linux_vdso_install, NULL);
816
817 static void
linux_vdso_deinstall(void * param)818 linux_vdso_deinstall(void *param)
819 {
820
821 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
822 }
823 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
824 linux_vdso_deinstall, NULL);
825
826 static char GNULINUX_ABI_VENDOR[] = "GNU";
827 static int GNULINUX_ABI_DESC = 0;
828
829 static bool
linux_trans_osrel(const Elf_Note * note,int32_t * osrel)830 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
831 {
832 const Elf32_Word *desc;
833 uintptr_t p;
834
835 p = (uintptr_t)(note + 1);
836 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
837
838 desc = (const Elf32_Word *)p;
839 if (desc[0] != GNULINUX_ABI_DESC)
840 return (false);
841
842 /*
843 * For Linux we encode osrel using the Linux convention of
844 * (version << 16) | (major << 8) | (minor)
845 * See macro in linux_mib.h
846 */
847 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
848
849 return (true);
850 }
851
852 static Elf_Brandnote linux64_brandnote = {
853 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR),
854 .hdr.n_descsz = 16,
855 .hdr.n_type = 1,
856 .vendor = GNULINUX_ABI_VENDOR,
857 .flags = BN_TRANSLATE_OSREL,
858 .trans_osrel = linux_trans_osrel
859 };
860
861 static Elf64_Brandinfo linux_glibc2brand = {
862 .brand = ELFOSABI_LINUX,
863 .machine = EM_X86_64,
864 .compat_3_brand = "Linux",
865 .emul_path = linux_emul_path,
866 .interp_path = "/lib64/ld-linux-x86-64.so.2",
867 .sysvec = &elf_linux_sysvec,
868 .interp_newpath = NULL,
869 .brand_note = &linux64_brandnote,
870 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
871 };
872
873 static Elf64_Brandinfo linux_glibc2brandshort = {
874 .brand = ELFOSABI_LINUX,
875 .machine = EM_X86_64,
876 .compat_3_brand = "Linux",
877 .emul_path = linux_emul_path,
878 .interp_path = "/lib64/ld-linux.so.2",
879 .sysvec = &elf_linux_sysvec,
880 .interp_newpath = NULL,
881 .brand_note = &linux64_brandnote,
882 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
883 };
884
885 static Elf64_Brandinfo linux_muslbrand = {
886 .brand = ELFOSABI_LINUX,
887 .machine = EM_X86_64,
888 .compat_3_brand = "Linux",
889 .emul_path = linux_emul_path,
890 .interp_path = "/lib/ld-musl-x86_64.so.1",
891 .sysvec = &elf_linux_sysvec,
892 .interp_newpath = NULL,
893 .brand_note = &linux64_brandnote,
894 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
895 };
896
897 Elf64_Brandinfo *linux_brandlist[] = {
898 &linux_glibc2brand,
899 &linux_glibc2brandshort,
900 &linux_muslbrand,
901 NULL
902 };
903
904 static int
linux64_elf_modevent(module_t mod,int type,void * data)905 linux64_elf_modevent(module_t mod, int type, void *data)
906 {
907 Elf64_Brandinfo **brandinfo;
908 int error;
909 struct linux_ioctl_handler **lihp;
910
911 error = 0;
912
913 switch(type) {
914 case MOD_LOAD:
915 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
916 ++brandinfo)
917 if (elf64_insert_brand_entry(*brandinfo) < 0)
918 error = EINVAL;
919 if (error == 0) {
920 SET_FOREACH(lihp, linux_ioctl_handler_set)
921 linux_ioctl_register_handler(*lihp);
922 stclohz = (stathz ? stathz : hz);
923 if (bootverbose)
924 printf("Linux x86-64 ELF exec handler installed\n");
925 } else
926 printf("cannot insert Linux x86-64 ELF brand handler\n");
927 break;
928 case MOD_UNLOAD:
929 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
930 ++brandinfo)
931 if (elf64_brand_inuse(*brandinfo))
932 error = EBUSY;
933 if (error == 0) {
934 for (brandinfo = &linux_brandlist[0];
935 *brandinfo != NULL; ++brandinfo)
936 if (elf64_remove_brand_entry(*brandinfo) < 0)
937 error = EINVAL;
938 }
939 if (error == 0) {
940 SET_FOREACH(lihp, linux_ioctl_handler_set)
941 linux_ioctl_unregister_handler(*lihp);
942 if (bootverbose)
943 printf("Linux ELF exec handler removed\n");
944 } else
945 printf("Could not deinstall ELF interpreter entry\n");
946 break;
947 default:
948 return (EOPNOTSUPP);
949 }
950 return (error);
951 }
952
953 static moduledata_t linux64_elf_mod = {
954 "linux64elf",
955 linux64_elf_modevent,
956 0
957 };
958
959 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
960 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
961 FEATURE(linux64, "Linux 64bit support");
962