?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
linux emulation - Major update
[dragonfly.git] / sys / emulation / linux / i386 / linux_sysvec.c
blobabe895b96a50627b7b47366d29ba4e4e91dd9370
1 /*-
2 * Copyright (c) 1994-1996 Søren Schmidt
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer
10 * in this position and unchanged.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote products
15 * derived from this software withough specific prior written permission
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * $FreeBSD: src/sys/i386/linux/linux_sysvec.c,v 1.55.2.9 2002/01/12 11:03:30 bde Exp $
29 * $DragonFly: src/sys/emulation/linux/i386/linux_sysvec.c,v 1.31 2008/04/21 15:47:53 dillon Exp $
30 */
31
32 /* XXX we use functions that might not exist. */
33 #include "opt_compat.h"
34
35 #ifndef COMPAT_43
36 #error "Unable to compile Linux-emulator due to missing COMPAT_43 option!"
37 #endif
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/imgact.h>
42 #include <sys/imgact_aout.h>
43 #include <sys/imgact_elf.h>
44 #include <sys/kern_syscall.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/proc.h>
48 #include <sys/signalvar.h>
49 #include <sys/sysent.h>
50 #include <sys/sysproto.h>
51 #include <sys/eventhandler.h>
52
53 #include <vm/vm.h>
54 #include <vm/vm_param.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_extern.h>
57 #include <sys/exec.h>
58 #include <sys/kernel.h>
59 #include <sys/module.h>
60 #include <machine/cpu.h>
61
62 #include "linux.h"
63 #include "linux_proto.h"
64 #include "../linux_signal.h"
65 #include "../linux_util.h"
66 #include "../linux_futex.h"
67 #include "../linux_emuldata.h"
68
69 MODULE_VERSION(linux, 1);
70
71 MALLOC_DEFINE(M_LINUX, "linux", "Linux mode structures");
72
73 #if BYTE_ORDER == LITTLE_ENDIAN
74 #define SHELLMAGIC 0x2123 /* #! */
75 #else
76 #define SHELLMAGIC 0x2321
77 #endif
78
79 /*
80 * Allow the sendsig functions to use the ldebug() facility
81 * even though they are not syscalls themselves. Map them
82 * to syscall 0. This is slightly less bogus than using
83 * ldebug(sigreturn).
84 */
85 #define LINUX_SYS_linux_rt_sendsig 0
86 #define LINUX_SYS_linux_sendsig 0
87
88 extern char linux_sigcode[];
89 extern int linux_szsigcode;
90
91 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
92
93 static int linux_fixup (register_t **stack_base,
94 struct image_params *iparams);
95 static int elf_linux_fixup (register_t **stack_base,
96 struct image_params *iparams);
97 static void linux_prepsyscall (struct trapframe *tf, int *args,
98 u_int *code, caddr_t *params);
99 static void linux_sendsig (sig_t catcher, int sig, sigset_t *mask,
100 u_long code);
101
102 static eventhandler_tag linux_exec_tag;
103 static eventhandler_tag linux_exit_tag;
104
105 /*
106 * Linux syscalls return negative errno's, we do positive and map them
107 */
108 static int bsd_to_linux_errno[ELAST + 1] = {
109 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9,
110 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
111 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
112 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
113 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
114 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
115 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
116 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9,
117 -6, -6, -43, -42, -75, -6, -84
118 };
119
120 int bsd_to_linux_signal[LINUX_SIGTBLSZ] = {
121 LINUX_SIGHUP, LINUX_SIGINT, LINUX_SIGQUIT, LINUX_SIGILL,
122 LINUX_SIGTRAP, LINUX_SIGABRT, 0, LINUX_SIGFPE,
123 LINUX_SIGKILL, LINUX_SIGBUS, LINUX_SIGSEGV, LINUX_SIGSYS,
124 LINUX_SIGPIPE, LINUX_SIGALRM, LINUX_SIGTERM, LINUX_SIGURG,
125 LINUX_SIGSTOP, LINUX_SIGTSTP, LINUX_SIGCONT, LINUX_SIGCHLD,
126 LINUX_SIGTTIN, LINUX_SIGTTOU, LINUX_SIGIO, LINUX_SIGXCPU,
127 LINUX_SIGXFSZ, LINUX_SIGVTALRM, LINUX_SIGPROF, LINUX_SIGWINCH,
128 0, LINUX_SIGUSR1, LINUX_SIGUSR2
129 };
130
131 int linux_to_bsd_signal[LINUX_SIGTBLSZ] = {
132 SIGHUP, SIGINT, SIGQUIT, SIGILL,
133 SIGTRAP, SIGABRT, SIGBUS, SIGFPE,
134 SIGKILL, SIGUSR1, SIGSEGV, SIGUSR2,
135 SIGPIPE, SIGALRM, SIGTERM, SIGBUS,
136 SIGCHLD, SIGCONT, SIGSTOP, SIGTSTP,
137 SIGTTIN, SIGTTOU, SIGURG, SIGXCPU,
138 SIGXFSZ, SIGVTALRM, SIGPROF, SIGWINCH,
139 SIGIO, SIGURG, SIGSYS
140 };
141
142 #define LINUX_T_UNKNOWN 255
143 static int _bsd_to_linux_trapcode[] = {
144 LINUX_T_UNKNOWN, /* 0 */
145 6, /* 1 T_PRIVINFLT */
146 LINUX_T_UNKNOWN, /* 2 */
147 3, /* 3 T_BPTFLT */
148 LINUX_T_UNKNOWN, /* 4 */
149 LINUX_T_UNKNOWN, /* 5 */
150 16, /* 6 T_ARITHTRAP */
151 254, /* 7 T_ASTFLT */
152 LINUX_T_UNKNOWN, /* 8 */
153 13, /* 9 T_PROTFLT */
154 1, /* 10 T_TRCTRAP */
155 LINUX_T_UNKNOWN, /* 11 */
156 14, /* 12 T_PAGEFLT */
157 LINUX_T_UNKNOWN, /* 13 */
158 17, /* 14 T_ALIGNFLT */
159 LINUX_T_UNKNOWN, /* 15 */
160 LINUX_T_UNKNOWN, /* 16 */
161 LINUX_T_UNKNOWN, /* 17 */
162 0, /* 18 T_DIVIDE */
163 2, /* 19 T_NMI */
164 4, /* 20 T_OFLOW */
165 5, /* 21 T_BOUND */
166 7, /* 22 T_DNA */
167 8, /* 23 T_DOUBLEFLT */
168 9, /* 24 T_FPOPFLT */
169 10, /* 25 T_TSSFLT */
170 11, /* 26 T_SEGNPFLT */
171 12, /* 27 T_STKFLT */
172 18, /* 28 T_MCHK */
173 19, /* 29 T_XMMFLT */
174 15 /* 30 T_RESERVED */
175 };
176 #define bsd_to_linux_trapcode(code) \
177 ((code)<sizeof(_bsd_to_linux_trapcode)/sizeof(*_bsd_to_linux_trapcode)? \
178 _bsd_to_linux_trapcode[(code)]: \
179 LINUX_T_UNKNOWN)
180
181 /*
182 * If FreeBSD & Linux have a difference of opinion about what a trap
183 * means, deal with it here.
184 */
185 static int
186 translate_traps(int signal, int trap_code)
187 {
188 if (signal != SIGBUS)
189 return signal;
190 switch (trap_code) {
191 case T_PROTFLT:
192 case T_TSSFLT:
193 case T_DOUBLEFLT:
194 case T_PAGEFLT:
195 return SIGSEGV;
196 default:
197 return signal;
198 }
199 }
200
201 static int
202 linux_fixup(register_t **stack_base, struct image_params *imgp)
203 {
204 register_t *argv, *envp;
205
206 argv = *stack_base;
207 envp = *stack_base + (imgp->args->argc + 1);
208 (*stack_base)--;
209 **stack_base = (intptr_t)(void *)envp;
210 (*stack_base)--;
211 **stack_base = (intptr_t)(void *)argv;
212 (*stack_base)--;
213 **stack_base = imgp->args->argc;
214 return 0;
215 }
216
217 static int
218 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
219 {
220 Elf32_Auxargs *args = (Elf32_Auxargs *)imgp->auxargs;
221 register_t *pos;
222
223 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
224
225 if (args->trace) {
226 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
227 }
228 if (args->execfd != -1) {
229 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
230 }
231 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
232 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
233 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
234 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
235 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
236 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
237 AUXARGS_ENTRY(pos, AT_BASE, args->base);
238 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
239 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
240 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
241 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
242 AUXARGS_ENTRY(pos, AT_NULL, 0);
243
244 kfree(imgp->auxargs, M_TEMP);
245 imgp->auxargs = NULL;
246
247 (*stack_base)--;
248 **stack_base = (long)imgp->args->argc;
249 return 0;
250 }
251
252 extern int _ucodesel, _udatasel;
253 extern unsigned long linux_sznonrtsigcode;
254
255 static void
256 linux_rt_sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
257 {
258 struct proc *p = curproc;
259 struct lwp *lp = curthread->td_lwp;
260 struct trapframe *regs;
261 struct l_rt_sigframe *fp, frame;
262 int oonstack;
263
264 regs = lp->lwp_md.md_regs;
265 oonstack = lp->lwp_sigstk.ss_flags & SS_ONSTACK;
266
267 #ifdef DEBUG
268 if (ldebug(rt_sendsig))
269 kprintf(ARGS(rt_sendsig, "%p, %d, %p, %lu"),
270 catcher, sig, (void*)mask, code);
271 #endif
272 /*
273 * Allocate space for the signal handler context.
274 */
275 if ((lp->lwp_flag & LWP_ALTSTACK) && !oonstack &&
276 SIGISMEMBER(p->p_sigacts->ps_sigonstack, sig)) {
277 fp = (struct l_rt_sigframe *)(lp->lwp_sigstk.ss_sp +
278 lp->lwp_sigstk.ss_size - sizeof(struct l_rt_sigframe));
279 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
280 } else
281 fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
282
283 /*
284 * grow() will return FALSE if the fp will not fit inside the stack
285 * and the stack can not be grown. useracc will return FALSE
286 * if access is denied.
287 */
288 if ((vm_map_growstack(p, (vm_offset_t)fp) != KERN_SUCCESS) ||
289 !useracc((caddr_t)fp, sizeof (struct l_rt_sigframe),
290 VM_PROT_WRITE)) {
291 /*
292 * Process has trashed its stack; give it an illegal
293 * instruction to halt it in its tracks.
294 */
295 SIGACTION(p, SIGILL) = SIG_DFL;
296 SIGDELSET(p->p_sigignore, SIGILL);
297 SIGDELSET(p->p_sigcatch, SIGILL);
298 SIGDELSET(lp->lwp_sigmask, SIGILL);
299 #ifdef DEBUG
300 if (ldebug(rt_sendsig))
301 kprintf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
302 fp, oonstack);
303 #endif
304 lwpsignal(p, lp, SIGILL);
305 return;
306 }
307
308 /*
309 * Build the argument list for the signal handler.
310 */
311 if (p->p_sysent->sv_sigtbl)
312 if (sig <= p->p_sysent->sv_sigsize)
313 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
314
315 frame.sf_handler = catcher;
316 frame.sf_sig = sig;
317 frame.sf_siginfo = &fp->sf_si;
318 frame.sf_ucontext = &fp->sf_sc;
319
320 /* Fill siginfo structure. */
321 frame.sf_si.lsi_signo = sig;
322 frame.sf_si.lsi_code = code;
323 frame.sf_si.lsi_addr = (void *)regs->tf_err;
324
325 /*
326 * Build the signal context to be used by sigreturn.
327 */
328 frame.sf_sc.uc_flags = 0; /* XXX ??? */
329 frame.sf_sc.uc_link = NULL; /* XXX ??? */
330
331 frame.sf_sc.uc_stack.ss_sp = lp->lwp_sigstk.ss_sp;
332 frame.sf_sc.uc_stack.ss_size = lp->lwp_sigstk.ss_size;
333 frame.sf_sc.uc_stack.ss_flags = (lp->lwp_flag & LWP_ALTSTACK)
334 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
335
336 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
337
338 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__bits[0];
339 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs;
340 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
341 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
342 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
343 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_edi;
344 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_esi;
345 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_ebp;
346 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_ebx;
347 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_edx;
348 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_ecx;
349 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_eax;
350 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_eip;
351 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
352 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
353 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
354 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
355 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
356 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
357
358 #ifdef DEBUG
359 if (ldebug(rt_sendsig))
360 kprintf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
361 frame.sf_sc.uc_stack.ss_flags, lp->lwp_sigstk.ss_sp,
362 lp->lwp_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
363 #endif
364
365 if (copyout(&frame, fp, sizeof(frame)) != 0) {
366 /*
367 * Process has trashed its stack; give it an illegal
368 * instruction to halt it in its tracks.
369 */
370 sigexit(lp, SIGILL);
371 /* NOTREACHED */
372 }
373
374 /*
375 * Build context to run handler in.
376 */
377 regs->tf_esp = (int)fp;
378 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode) +
379 linux_sznonrtsigcode;
380
381 /*
382 * i386 abi specifies that the direction flag must be cleared
383 * on function entry
384 */
385 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
386
387 regs->tf_cs = _ucodesel;
388 regs->tf_ds = _udatasel;
389 regs->tf_es = _udatasel;
390 /* allow %fs and %gs to be inherited by the signal handler */
391 /*
392 regs->tf_fs = _udatasel;
393 regs->tf_gs = _udatasel;
394 */
395 regs->tf_ss = _udatasel;
396 }
397
398
399 /*
400 * Send an interrupt to process.
401 *
402 * Stack is set up to allow sigcode stored
403 * in u. to call routine, followed by kcall
404 * to sigreturn routine below. After sigreturn
405 * resets the signal mask, the stack, and the
406 * frame pointer, it returns to the user
407 * specified pc, psl.
408 */
409
410 static void
411 linux_sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
412 {
413 struct proc *p = curproc;
414 struct lwp *lp = curthread->td_lwp;
415 struct trapframe *regs;
416 struct l_sigframe *fp, frame;
417 l_sigset_t lmask;
418 int oonstack, i;
419
420 if (SIGISMEMBER(p->p_sigacts->ps_siginfo, sig)) {
421 /* Signal handler installed with SA_SIGINFO. */
422 linux_rt_sendsig(catcher, sig, mask, code);
423 return;
424 }
425
426 regs = lp->lwp_md.md_regs;
427 oonstack = lp->lwp_sigstk.ss_flags & SS_ONSTACK;
428
429 #ifdef DEBUG
430 if (ldebug(sendsig))
431 kprintf(ARGS(sendsig, "%p, %d, %p, %lu"),
432 catcher, sig, (void*)mask, code);
433 #endif
434
435 /*
436 * Allocate space for the signal handler context.
437 */
438 if ((lp->lwp_flag & LWP_ALTSTACK) && !oonstack &&
439 SIGISMEMBER(p->p_sigacts->ps_sigonstack, sig)) {
440 fp = (struct l_sigframe *)(lp->lwp_sigstk.ss_sp +
441 lp->lwp_sigstk.ss_size - sizeof(struct l_sigframe));
442 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
443 } else
444 fp = (struct l_sigframe *)regs->tf_esp - 1;
445
446 /*
447 * grow() will return FALSE if the fp will not fit inside the stack
448 * and the stack can not be grown. useracc will return FALSE
449 * if access is denied.
450 */
451 if ((vm_map_growstack(p, (vm_offset_t)fp) != KERN_SUCCESS) ||
452 !useracc((caddr_t)fp, sizeof (struct l_sigframe),
453 VM_PROT_WRITE)) {
454 /*
455 * Process has trashed its stack; give it an illegal
456 * instruction to halt it in its tracks.
457 */
458 SIGACTION(p, SIGILL) = SIG_DFL;
459 SIGDELSET(p->p_sigignore, SIGILL);
460 SIGDELSET(p->p_sigcatch, SIGILL);
461 SIGDELSET(lp->lwp_sigmask, SIGILL);
462 lwpsignal(p, lp, SIGILL);
463 return;
464 }
465
466 /*
467 * Build the argument list for the signal handler.
468 */
469 if (p->p_sysent->sv_sigtbl)
470 if (sig <= p->p_sysent->sv_sigsize)
471 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
472
473 frame.sf_handler = catcher;
474 frame.sf_sig = sig;
475
476 bsd_to_linux_sigset(mask, &lmask);
477
478 /*
479 * Build the signal context to be used by sigreturn.
480 */
481 frame.sf_sc.sc_mask = lmask.__bits[0];
482 frame.sf_sc.sc_gs = regs->tf_gs;
483 frame.sf_sc.sc_fs = regs->tf_fs;
484 frame.sf_sc.sc_es = regs->tf_es;
485 frame.sf_sc.sc_ds = regs->tf_ds;
486 frame.sf_sc.sc_edi = regs->tf_edi;
487 frame.sf_sc.sc_esi = regs->tf_esi;
488 frame.sf_sc.sc_ebp = regs->tf_ebp;
489 frame.sf_sc.sc_ebx = regs->tf_ebx;
490 frame.sf_sc.sc_edx = regs->tf_edx;
491 frame.sf_sc.sc_ecx = regs->tf_ecx;
492 frame.sf_sc.sc_eax = regs->tf_eax;
493 frame.sf_sc.sc_eip = regs->tf_eip;
494 frame.sf_sc.sc_cs = regs->tf_cs;
495 frame.sf_sc.sc_eflags = regs->tf_eflags;
496 frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
497 frame.sf_sc.sc_ss = regs->tf_ss;
498 frame.sf_sc.sc_err = regs->tf_err;
499 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
500
501 bzero(&frame.sf_fpstate, sizeof(struct l_fpstate));
502
503 for (i = 0; i < (LINUX_NSIG_WORDS-1); i++)
504 frame.sf_extramask[i] = lmask.__bits[i+1];
505
506 if (copyout(&frame, fp, sizeof(frame)) != 0) {
507 /*
508 * Process has trashed its stack; give it an illegal
509 * instruction to halt it in its tracks.
510 */
511 sigexit(lp, SIGILL);
512 /* NOTREACHED */
513 }
514
515 /*
516 * Build context to run handler in.
517 */
518 regs->tf_esp = (int)fp;
519 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
520
521 /*
522 * i386 abi specifies that the direction flag must be cleared
523 * on function entry
524 */
525 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
526
527 regs->tf_cs = _ucodesel;
528 regs->tf_ds = _udatasel;
529 regs->tf_es = _udatasel;
530 /* Allow %fs and %gs to be inherited by the signal handler */
531 /*
532 regs->tf_fs = _udatasel;
533 regs->tf_gs = _udatasel;
534 */
535 regs->tf_ss = _udatasel;
536 }
537
538 /*
539 * System call to cleanup state after a signal
540 * has been taken. Reset signal mask and
541 * stack state from context left by sendsig (above).
542 * Return to previous pc and psl as specified by
543 * context left by sendsig. Check carefully to
544 * make sure that the user has not modified the
545 * psl to gain improper privileges or to cause
546 * a machine fault.
547 *
548 * MPSAFE
549 */
550 int
551 sys_linux_sigreturn(struct linux_sigreturn_args *args)
552 {
553 struct lwp *lp = curthread->td_lwp;
554 struct l_sigframe frame;
555 struct trapframe *regs;
556 l_sigset_t lmask;
557 int eflags, i;
558
559 regs = lp->lwp_md.md_regs;
560
561 #ifdef DEBUG
562 if (ldebug(sigreturn))
563 kprintf(ARGS(sigreturn, "%p"), (void *)args->sfp);
564 #endif
565 /*
566 * The trampoline code hands us the sigframe.
567 * It is unsafe to keep track of it ourselves, in the event that a
568 * program jumps out of a signal handler.
569 */
570 if (copyin((caddr_t)args->sfp, &frame, sizeof(frame)) != 0)
571 return (EFAULT);
572
573 /*
574 * Check for security violations.
575 */
576 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
577 eflags = frame.sf_sc.sc_eflags;
578 /*
579 * XXX do allow users to change the privileged flag PSL_RF. The
580 * cpu sets PSL_RF in tf_eflags for faults. Debuggers should
581 * sometimes set it there too. tf_eflags is kept in the signal
582 * context during signal handling and there is no other place
583 * to remember it, so the PSL_RF bit may be corrupted by the
584 * signal handler without us knowing. Corruption of the PSL_RF
585 * bit at worst causes one more or one less debugger trap, so
586 * allowing it is fairly harmless.
587 */
588 if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
589 return(EINVAL);
590 }
591
592 /*
593 * Don't allow users to load a valid privileged %cs. Let the
594 * hardware check for invalid selectors, excess privilege in
595 * other selectors, invalid %eip's and invalid %esp's.
596 */
597 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
598 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
599 trapsignal(lp, SIGBUS, T_PROTFLT);
600 return(EINVAL);
601 }
602
603 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
604 lmask.__bits[0] = frame.sf_sc.sc_mask;
605 for (i = 0; i < (LINUX_NSIG_WORDS-1); i++)
606 lmask.__bits[i+1] = frame.sf_extramask[i];
607 linux_to_bsd_sigset(&lmask, &lp->lwp_sigmask);
608 SIG_CANTMASK(lp->lwp_sigmask);
609
610 /*
611 * Restore signal context.
612 */
613 /* %gs was restored by the trampoline. */
614 regs->tf_fs = frame.sf_sc.sc_fs;
615 regs->tf_es = frame.sf_sc.sc_es;
616 regs->tf_ds = frame.sf_sc.sc_ds;
617 regs->tf_edi = frame.sf_sc.sc_edi;
618 regs->tf_esi = frame.sf_sc.sc_esi;
619 regs->tf_ebp = frame.sf_sc.sc_ebp;
620 regs->tf_ebx = frame.sf_sc.sc_ebx;
621 regs->tf_edx = frame.sf_sc.sc_edx;
622 regs->tf_ecx = frame.sf_sc.sc_ecx;
623 regs->tf_eax = frame.sf_sc.sc_eax;
624 regs->tf_eip = frame.sf_sc.sc_eip;
625 regs->tf_cs = frame.sf_sc.sc_cs;
626 regs->tf_eflags = eflags;
627 regs->tf_esp = frame.sf_sc.sc_esp_at_signal;
628 regs->tf_ss = frame.sf_sc.sc_ss;
629
630 return (EJUSTRETURN);
631 }
632
633 /*
634 * System call to cleanup state after a signal
635 * has been taken. Reset signal mask and
636 * stack state from context left by rt_sendsig (above).
637 * Return to previous pc and psl as specified by
638 * context left by sendsig. Check carefully to
639 * make sure that the user has not modified the
640 * psl to gain improper privileges or to cause
641 * a machine fault.
642 *
643 * MPSAFE
644 */
645 int
646 sys_linux_rt_sigreturn(struct linux_rt_sigreturn_args *args)
647 {
648 struct lwp *lp = curthread->td_lwp;
649 struct l_ucontext uc;
650 struct l_sigcontext *context;
651 l_stack_t *lss;
652 stack_t ss;
653 struct trapframe *regs;
654 int eflags;
655
656 regs = lp->lwp_md.md_regs;
657
658 #ifdef DEBUG
659 if (ldebug(rt_sigreturn))
660 kprintf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
661 #endif
662 /*
663 * The trampoline code hands us the ucontext.
664 * It is unsafe to keep track of it ourselves, in the event that a
665 * program jumps out of a signal handler.
666 */
667 if (copyin((caddr_t)args->ucp, &uc, sizeof(uc)) != 0)
668 return (EFAULT);
669
670 context = &uc.uc_mcontext;
671
672 /*
673 * Check for security violations.
674 */
675 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
676 eflags = context->sc_eflags;
677 /*
678 * XXX do allow users to change the privileged flag PSL_RF. The
679 * cpu sets PSL_RF in tf_eflags for faults. Debuggers should
680 * sometimes set it there too. tf_eflags is kept in the signal
681 * context during signal handling and there is no other place
682 * to remember it, so the PSL_RF bit may be corrupted by the
683 * signal handler without us knowing. Corruption of the PSL_RF
684 * bit at worst causes one more or one less debugger trap, so
685 * allowing it is fairly harmless.
686 */
687 if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
688 return(EINVAL);
689 }
690
691 /*
692 * Don't allow users to load a valid privileged %cs. Let the
693 * hardware check for invalid selectors, excess privilege in
694 * other selectors, invalid %eip's and invalid %esp's.
695 */
696 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
697 if (!CS_SECURE(context->sc_cs)) {
698 trapsignal(lp, SIGBUS, T_PROTFLT);
699 return(EINVAL);
700 }
701
702 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
703 linux_to_bsd_sigset(&uc.uc_sigmask, &lp->lwp_sigmask);
704 SIG_CANTMASK(lp->lwp_sigmask);
705
706 /*
707 * Restore signal context
708 */
709 /* %gs was restored by the trampoline. */
710 regs->tf_fs = context->sc_fs;
711 regs->tf_es = context->sc_es;
712 regs->tf_ds = context->sc_ds;
713 regs->tf_edi = context->sc_edi;
714 regs->tf_esi = context->sc_esi;
715 regs->tf_ebp = context->sc_ebp;
716 regs->tf_ebx = context->sc_ebx;
717 regs->tf_edx = context->sc_edx;
718 regs->tf_ecx = context->sc_ecx;
719 regs->tf_eax = context->sc_eax;
720 regs->tf_eip = context->sc_eip;
721 regs->tf_cs = context->sc_cs;
722 regs->tf_eflags = eflags;
723 regs->tf_esp = context->sc_esp_at_signal;
724 regs->tf_ss = context->sc_ss;
725
726 /*
727 * call sigaltstack & ignore results..
728 */
729 lss = &uc.uc_stack;
730 ss.ss_sp = lss->ss_sp;
731 ss.ss_size = lss->ss_size;
732 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
733
734 #ifdef DEBUG
735 if (ldebug(rt_sigreturn))
736 kprintf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
737 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
738 #endif
739 kern_sigaltstack(&ss, NULL);
740
741 return (EJUSTRETURN);
742 }
743
744 /*
745 * Prep arguments.
746 *
747 * MUST BE MPSAFE
748 */
749 static void
750 linux_prepsyscall(struct trapframe *tf, int *args, u_int *code, caddr_t *params)
751 {
752 args[0] = tf->tf_ebx;
753 args[1] = tf->tf_ecx;
754 args[2] = tf->tf_edx;
755 args[3] = tf->tf_esi;
756 args[4] = tf->tf_edi;
757 args[5] = tf->tf_ebp;
758 *params = NULL; /* no copyin */
759 }
760
761 /*
762 * If a linux binary is exec'ing something, try this image activator
763 * first. We override standard shell script execution in order to
764 * be able to modify the interpreter path. We only do this if a linux
765 * binary is doing the exec, so we do not create an EXEC module for it.
766 */
767 static int exec_linux_imgact_try (struct image_params *iparams);
768
769 static int
770 exec_linux_imgact_try(struct image_params *imgp)
771 {
772 const char *head = (const char *)imgp->image_header;
773 int error = -1;
774
775 /*
776 * The interpreter for shell scripts run from a linux binary needs
777 * to be located in /compat/linux if possible in order to recursively
778 * maintain linux path emulation.
779 */
780 if (((const short *)head)[0] == SHELLMAGIC) {
781 /*
782 * Run our normal shell image activator. If it succeeds attempt
783 * to use the alternate path for the interpreter. If an alternate
784 * path is found, use our stringspace to store it.
785 */
786 if ((error = exec_shell_imgact(imgp)) == 0) {
787 linux_translate_path(imgp->interpreter_name,
788 MAXSHELLCMDLEN);
789 }
790 }
791 return(error);
792 }
793
794 struct sysentvec linux_sysvec = {
795 LINUX_SYS_MAXSYSCALL,
796 linux_sysent,
797 0xffffffff,
798 LINUX_SIGTBLSZ,
799 bsd_to_linux_signal,
800 ELAST + 1,
801 bsd_to_linux_errno,
802 translate_traps,
803 linux_fixup,
804 linux_sendsig,
805 linux_sigcode,
806 &linux_szsigcode,
807 linux_prepsyscall,
808 "Linux a.out",
809 NULL,
810 exec_linux_imgact_try,
811 LINUX_MINSIGSTKSZ
812 };
813
814 struct sysentvec elf_linux_sysvec = {
815 LINUX_SYS_MAXSYSCALL,
816 linux_sysent,
817 0xffffffff,
818 LINUX_SIGTBLSZ,
819 bsd_to_linux_signal,
820 ELAST + 1,
821 bsd_to_linux_errno,
822 translate_traps,
823 elf_linux_fixup,
824 linux_sendsig,
825 linux_sigcode,
826 &linux_szsigcode,
827 linux_prepsyscall,
828 "Linux ELF",
829 elf_coredump,
830 exec_linux_imgact_try,
831 LINUX_MINSIGSTKSZ
832 };
833
834 static int linux_match_abi_note(const Elf_Note *abi_note);
835 static int linux_match_suse_note(const Elf_Note *abi_note);
836
837 static Elf32_Brandinfo linux_brand = {
838 ELFOSABI_LINUX,
839 "Linux",
840 linux_match_abi_note,
841 "/compat/linux",
842 "/lib/ld-linux.so.1",
843 &elf_linux_sysvec
844 };
845
846 static Elf32_Brandinfo linux_glibc2brand = {
847 ELFOSABI_LINUX,
848 "Linux",
849 linux_match_abi_note,
850 "/compat/linux",
851 "/lib/ld-linux.so.2",
852 &elf_linux_sysvec
853 };
854
855 static Elf32_Brandinfo linux_suse_brand = {
856 ELFOSABI_LINUX,
857 "Linux",
858 linux_match_suse_note,
859 "/compat/linux",
860 "/lib/ld-linux.so.2",
861 &elf_linux_sysvec
862 };
863
864 Elf32_Brandinfo *linux_brandlist[] = {
865 &linux_brand,
866 &linux_glibc2brand,
867 &linux_suse_brand,
868 NULL
869 };
870
871 static int
872 linux_match_abi_note(const Elf_Note *abi_note)
873 {
874 const char *abi_name = (const char *)
875 ((const uint8_t *)abi_note + sizeof(*abi_note));
876 const uint32_t *descr = (const uint32_t *)
877 ((const uint8_t *)abi_name + abi_note->n_namesz);
878
879 if (abi_note->n_namesz != sizeof("GNU"))
880 return(FALSE);
881 if (memcmp(abi_name, "GNU", sizeof("GNU")))
882 return(FALSE);
883 if (abi_note->n_descsz < sizeof(uint32_t))
884 return(FALSE);
885
886 if (*descr != 0)
887 return(FALSE);
888 return(TRUE);
889 }
890
891 static int
892 linux_match_suse_note(const Elf_Note *abi_note)
893 {
894 const char *abi_name = (const char *)
895 ((const uint8_t *)abi_note + sizeof(*abi_note));
896
897 if (abi_note->n_namesz != sizeof("SuSE"))
898 return(FALSE);
899 if (memcmp(abi_name, "SuSE", sizeof("SuSE")))
900 return(FALSE);
901
902 return(TRUE);
903 }
904
905 static int
906 linux_elf_modevent(module_t mod, int type, void *data)
907 {
908 Elf32_Brandinfo **brandinfo;
909 int error;
910
911 error = 0;
912
913 switch(type) {
914 case MOD_LOAD:
915 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
916 ++brandinfo)
917 if (elf_insert_brand_entry(*brandinfo) < 0)
918 error = EINVAL;
919 if (error == 0) {
920 if (bootverbose)
921 kprintf("Linux ELF exec handler installed\n");
922 } else {
923 kprintf("cannot insert Linux ELF brand handler\n");
924 }
925 EMUL_LOCKINIT();
926 lockinit(&futex_mtx, "linftxs", 0, LK_CANRECURSE);
927 linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_transition,
928 NULL, 1000);
929 linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, emuldata_exit,
930 NULL, 1000);
931 break;
932 case MOD_UNLOAD:
933 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
934 ++brandinfo)
935 if (elf_brand_inuse(*brandinfo))
936 error = EBUSY;
937 if (error == 0) {
938 for (brandinfo = &linux_brandlist[0];
939 *brandinfo != NULL; ++brandinfo)
940 if (elf_remove_brand_entry(*brandinfo) < 0)
941 error = EINVAL;
942 }
943 if (error == 0) {
944 if (bootverbose)
945 kprintf("Linux ELF exec handler removed\n");
946 } else {
947 kprintf("Could not deinstall ELF interpreter entry\n");
948 }
949 EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
950 EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
951 lockuninit(&futex_mtx);
952 EMUL_LOCKUNINIT();
953 break;
954 default:
955 break;
956 }
957 return error;
958 }
959
960 static moduledata_t linux_elf_mod = {
961 "linuxelf",
962 linux_elf_modevent,
963 0
964 };
965
966 DECLARE_MODULE(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
DragonFly BSD