2 * Copyright (c) 1994-1996 Søren Schmidt
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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 without specific prior written permission
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.
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/systm.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/imgact_aout.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
45 #include <sys/signalvar.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysent.h>
48 #include <sys/sysproto.h>
49 #include <sys/vnode.h>
50 #include <sys/eventhandler.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_param.h>
60 #include <machine/cpu.h>
61 #include <machine/cputypes.h>
62 #include <machine/md_var.h>
63 #include <machine/pcb.h>
65 #include <i386/linux/linux.h>
66 #include <i386/linux/linux_proto.h>
67 #include <compat/linux/linux_emul.h>
68 #include <compat/linux/linux_futex.h>
69 #include <compat/linux/linux_ioctl.h>
70 #include <compat/linux/linux_mib.h>
71 #include <compat/linux/linux_misc.h>
72 #include <compat/linux/linux_signal.h>
73 #include <compat/linux/linux_util.h>
75 MODULE_VERSION(linux, 1);
77 MALLOC_DEFINE(M_LINUX, "linux", "Linux mode structures");
79 #if BYTE_ORDER == LITTLE_ENDIAN
80 #define SHELLMAGIC 0x2123 /* #! */
82 #define SHELLMAGIC 0x2321
86 * Allow the sendsig functions to use the ldebug() facility
87 * even though they are not syscalls themselves. Map them
88 * to syscall 0. This is slightly less bogus than using
91 #define LINUX_SYS_linux_rt_sendsig 0
92 #define LINUX_SYS_linux_sendsig 0
94 #define LINUX_PS_STRINGS (LINUX_USRSTACK - sizeof(struct ps_strings))
96 extern char linux_sigcode[];
97 extern int linux_szsigcode;
99 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
101 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
102 SET_DECLARE(linux_device_handler_set, struct linux_device_handler);
104 static int linux_fixup(register_t **stack_base,
105 struct image_params *iparams);
106 static int elf_linux_fixup(register_t **stack_base,
107 struct image_params *iparams);
108 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
109 static void exec_linux_setregs(struct thread *td,
110 struct image_params *imgp, u_long stack);
111 static register_t *linux_copyout_strings(struct image_params *imgp);
112 static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
114 static int linux_szplatform;
115 const char *linux_platform;
117 static eventhandler_tag linux_exit_tag;
118 static eventhandler_tag linux_exec_tag;
121 * Linux syscalls return negative errno's, we do positive and map them
123 * FreeBSD: src/sys/sys/errno.h
124 * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h
125 * linux-2.6.17.8/include/asm-generic/errno.h
127 static int bsd_to_linux_errno[ELAST + 1] = {
128 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9,
129 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
130 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
131 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
132 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
133 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
134 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
135 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9,
136 -6, -6, -43, -42, -75,-125, -84, -95, -16, -74,
140 int bsd_to_linux_signal[LINUX_SIGTBLSZ] = {
141 LINUX_SIGHUP, LINUX_SIGINT, LINUX_SIGQUIT, LINUX_SIGILL,
142 LINUX_SIGTRAP, LINUX_SIGABRT, 0, LINUX_SIGFPE,
143 LINUX_SIGKILL, LINUX_SIGBUS, LINUX_SIGSEGV, LINUX_SIGSYS,
144 LINUX_SIGPIPE, LINUX_SIGALRM, LINUX_SIGTERM, LINUX_SIGURG,
145 LINUX_SIGSTOP, LINUX_SIGTSTP, LINUX_SIGCONT, LINUX_SIGCHLD,
146 LINUX_SIGTTIN, LINUX_SIGTTOU, LINUX_SIGIO, LINUX_SIGXCPU,
147 LINUX_SIGXFSZ, LINUX_SIGVTALRM, LINUX_SIGPROF, LINUX_SIGWINCH,
148 0, LINUX_SIGUSR1, LINUX_SIGUSR2
151 int linux_to_bsd_signal[LINUX_SIGTBLSZ] = {
152 SIGHUP, SIGINT, SIGQUIT, SIGILL,
153 SIGTRAP, SIGABRT, SIGBUS, SIGFPE,
154 SIGKILL, SIGUSR1, SIGSEGV, SIGUSR2,
155 SIGPIPE, SIGALRM, SIGTERM, SIGBUS,
156 SIGCHLD, SIGCONT, SIGSTOP, SIGTSTP,
157 SIGTTIN, SIGTTOU, SIGURG, SIGXCPU,
158 SIGXFSZ, SIGVTALRM, SIGPROF, SIGWINCH,
159 SIGIO, SIGURG, SIGSYS
162 #define LINUX_T_UNKNOWN 255
163 static int _bsd_to_linux_trapcode[] = {
164 LINUX_T_UNKNOWN, /* 0 */
165 6, /* 1 T_PRIVINFLT */
166 LINUX_T_UNKNOWN, /* 2 */
168 LINUX_T_UNKNOWN, /* 4 */
169 LINUX_T_UNKNOWN, /* 5 */
170 16, /* 6 T_ARITHTRAP */
171 254, /* 7 T_ASTFLT */
172 LINUX_T_UNKNOWN, /* 8 */
173 13, /* 9 T_PROTFLT */
174 1, /* 10 T_TRCTRAP */
175 LINUX_T_UNKNOWN, /* 11 */
176 14, /* 12 T_PAGEFLT */
177 LINUX_T_UNKNOWN, /* 13 */
178 17, /* 14 T_ALIGNFLT */
179 LINUX_T_UNKNOWN, /* 15 */
180 LINUX_T_UNKNOWN, /* 16 */
181 LINUX_T_UNKNOWN, /* 17 */
187 8, /* 23 T_DOUBLEFLT */
188 9, /* 24 T_FPOPFLT */
189 10, /* 25 T_TSSFLT */
190 11, /* 26 T_SEGNPFLT */
191 12, /* 27 T_STKFLT */
193 19, /* 29 T_XMMFLT */
194 15 /* 30 T_RESERVED */
196 #define bsd_to_linux_trapcode(code) \
197 ((code)<sizeof(_bsd_to_linux_trapcode)/sizeof(*_bsd_to_linux_trapcode)? \
198 _bsd_to_linux_trapcode[(code)]: \
202 * If FreeBSD & Linux have a difference of opinion about what a trap
203 * means, deal with it here.
208 translate_traps(int signal, int trap_code)
210 if (signal != SIGBUS)
224 linux_fixup(register_t **stack_base, struct image_params *imgp)
226 register_t *argv, *envp;
229 envp = *stack_base + (imgp->args->argc + 1);
231 suword(*stack_base, (intptr_t)(void *)envp);
233 suword(*stack_base, (intptr_t)(void *)argv);
235 suword(*stack_base, imgp->args->argc);
240 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
244 Elf32_Addr *uplatform;
245 struct ps_strings *arginfo;
248 KASSERT(curthread->td_proc == imgp->proc,
249 ("unsafe elf_linux_fixup(), should be curproc"));
252 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
253 uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
254 args = (Elf32_Auxargs *)imgp->auxargs;
255 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
257 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
260 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
261 * as it has appeared in the 2.4.0-rc7 first time.
262 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
263 * glibc falls back to the hard-coded CLK_TCK value when aux entry
265 * Also see linux_times() implementation.
267 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
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_FLAGS, args->flags);
274 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
275 AUXARGS_ENTRY(pos, AT_BASE, args->base);
276 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, 0);
277 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
278 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
279 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
280 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
281 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
282 if (args->execfd != -1)
283 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
284 AUXARGS_ENTRY(pos, AT_NULL, 0);
286 free(imgp->auxargs, M_TEMP);
287 imgp->auxargs = NULL;
290 suword(*stack_base, (register_t)imgp->args->argc);
295 * Copied from kern/kern_exec.c
298 linux_copyout_strings(struct image_params *imgp)
302 char *stringp, *destp;
303 register_t *stack_base;
304 struct ps_strings *arginfo;
308 * Calculate string base and vector table pointers.
309 * Also deal with signal trampoline code for this exec type.
312 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
313 destp = (caddr_t)arginfo - SPARE_USRSPACE - linux_szplatform -
314 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
317 * install LINUX_PLATFORM
319 copyout(linux_platform, ((caddr_t)arginfo - linux_szplatform),
323 * If we have a valid auxargs ptr, prepare some room
328 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
329 * lower compatibility.
331 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
332 (LINUX_AT_COUNT * 2);
334 * The '+ 2' is for the null pointers at the end of each of
335 * the arg and env vector sets,and imgp->auxarg_size is room
336 * for argument of Runtime loader.
338 vectp = (char **)(destp - (imgp->args->argc +
339 imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
342 * The '+ 2' is for the null pointers at the end of each of
343 * the arg and env vector sets
345 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
350 * vectp also becomes our initial stack base
352 stack_base = (register_t *)vectp;
354 stringp = imgp->args->begin_argv;
355 argc = imgp->args->argc;
356 envc = imgp->args->envc;
359 * Copy out strings - arguments and environment.
361 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
364 * Fill in "ps_strings" struct for ps, w, etc.
366 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
367 suword(&arginfo->ps_nargvstr, argc);
370 * Fill in argument portion of vector table.
372 for (; argc > 0; --argc) {
373 suword(vectp++, (long)(intptr_t)destp);
374 while (*stringp++ != 0)
379 /* a null vector table pointer separates the argp's from the envp's */
382 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
383 suword(&arginfo->ps_nenvstr, envc);
386 * Fill in environment portion of vector table.
388 for (; envc > 0; --envc) {
389 suword(vectp++, (long)(intptr_t)destp);
390 while (*stringp++ != 0)
395 /* end of vector table is a null pointer */
403 extern int _ucodesel, _udatasel;
404 extern unsigned long linux_sznonrtsigcode;
407 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
409 struct thread *td = curthread;
410 struct proc *p = td->td_proc;
412 struct trapframe *regs;
413 struct l_rt_sigframe *fp, frame;
417 sig = ksi->ksi_signo;
418 code = ksi->ksi_code;
419 PROC_LOCK_ASSERT(p, MA_OWNED);
421 mtx_assert(&psp->ps_mtx, MA_OWNED);
423 oonstack = sigonstack(regs->tf_esp);
426 if (ldebug(rt_sendsig))
427 printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
428 catcher, sig, (void*)mask, code);
431 * Allocate space for the signal handler context.
433 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
434 SIGISMEMBER(psp->ps_sigonstack, sig)) {
435 fp = (struct l_rt_sigframe *)(td->td_sigstk.ss_sp +
436 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
438 fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
439 mtx_unlock(&psp->ps_mtx);
442 * Build the argument list for the signal handler.
444 if (p->p_sysent->sv_sigtbl)
445 if (sig <= p->p_sysent->sv_sigsize)
446 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
448 bzero(&frame, sizeof(frame));
450 frame.sf_handler = catcher;
452 frame.sf_siginfo = &fp->sf_si;
453 frame.sf_ucontext = &fp->sf_sc;
455 /* Fill in POSIX parts */
456 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
459 * Build the signal context to be used by sigreturn.
461 frame.sf_sc.uc_flags = 0; /* XXX ??? */
462 frame.sf_sc.uc_link = NULL; /* XXX ??? */
464 frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
465 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
466 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
467 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
470 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
472 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__bits[0];
473 frame.sf_sc.uc_mcontext.sc_gs = rgs();
474 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
475 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
476 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
477 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_edi;
478 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_esi;
479 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_ebp;
480 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_ebx;
481 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_edx;
482 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_ecx;
483 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_eax;
484 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_eip;
485 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
486 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
487 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
488 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
489 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
490 frame.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
491 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
494 if (ldebug(rt_sendsig))
495 printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
496 frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
497 td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
500 if (copyout(&frame, fp, sizeof(frame)) != 0) {
502 * Process has trashed its stack; give it an illegal
503 * instruction to halt it in its tracks.
506 if (ldebug(rt_sendsig))
507 printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
515 * Build context to run handler in.
517 regs->tf_esp = (int)fp;
518 regs->tf_eip = p->p_sysent->sv_sigcode_base + linux_sznonrtsigcode;
519 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
520 regs->tf_cs = _ucodesel;
521 regs->tf_ds = _udatasel;
522 regs->tf_es = _udatasel;
523 regs->tf_fs = _udatasel;
524 regs->tf_ss = _udatasel;
526 mtx_lock(&psp->ps_mtx);
531 * Send an interrupt to process.
533 * Stack is set up to allow sigcode stored
534 * in u. to call routine, followed by kcall
535 * to sigreturn routine below. After sigreturn
536 * resets the signal mask, the stack, and the
537 * frame pointer, it returns to the user
541 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
543 struct thread *td = curthread;
544 struct proc *p = td->td_proc;
546 struct trapframe *regs;
547 struct l_sigframe *fp, frame;
552 PROC_LOCK_ASSERT(p, MA_OWNED);
554 sig = ksi->ksi_signo;
555 code = ksi->ksi_code;
556 mtx_assert(&psp->ps_mtx, MA_OWNED);
557 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
558 /* Signal handler installed with SA_SIGINFO. */
559 linux_rt_sendsig(catcher, ksi, mask);
563 oonstack = sigonstack(regs->tf_esp);
567 printf(ARGS(sendsig, "%p, %d, %p, %u"),
568 catcher, sig, (void*)mask, code);
572 * Allocate space for the signal handler context.
574 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
575 SIGISMEMBER(psp->ps_sigonstack, sig)) {
576 fp = (struct l_sigframe *)(td->td_sigstk.ss_sp +
577 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
579 fp = (struct l_sigframe *)regs->tf_esp - 1;
580 mtx_unlock(&psp->ps_mtx);
584 * Build the argument list for the signal handler.
586 if (p->p_sysent->sv_sigtbl)
587 if (sig <= p->p_sysent->sv_sigsize)
588 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
590 bzero(&frame, sizeof(frame));
592 frame.sf_handler = catcher;
595 bsd_to_linux_sigset(mask, &lmask);
598 * Build the signal context to be used by sigreturn.
600 frame.sf_sc.sc_mask = lmask.__bits[0];
601 frame.sf_sc.sc_gs = rgs();
602 frame.sf_sc.sc_fs = regs->tf_fs;
603 frame.sf_sc.sc_es = regs->tf_es;
604 frame.sf_sc.sc_ds = regs->tf_ds;
605 frame.sf_sc.sc_edi = regs->tf_edi;
606 frame.sf_sc.sc_esi = regs->tf_esi;
607 frame.sf_sc.sc_ebp = regs->tf_ebp;
608 frame.sf_sc.sc_ebx = regs->tf_ebx;
609 frame.sf_sc.sc_edx = regs->tf_edx;
610 frame.sf_sc.sc_ecx = regs->tf_ecx;
611 frame.sf_sc.sc_eax = regs->tf_eax;
612 frame.sf_sc.sc_eip = regs->tf_eip;
613 frame.sf_sc.sc_cs = regs->tf_cs;
614 frame.sf_sc.sc_eflags = regs->tf_eflags;
615 frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
616 frame.sf_sc.sc_ss = regs->tf_ss;
617 frame.sf_sc.sc_err = regs->tf_err;
618 frame.sf_sc.sc_cr2 = (register_t)ksi->ksi_addr;
619 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
621 for (i = 0; i < (LINUX_NSIG_WORDS-1); i++)
622 frame.sf_extramask[i] = lmask.__bits[i+1];
624 if (copyout(&frame, fp, sizeof(frame)) != 0) {
626 * Process has trashed its stack; give it an illegal
627 * instruction to halt it in its tracks.
634 * Build context to run handler in.
636 regs->tf_esp = (int)fp;
637 regs->tf_eip = p->p_sysent->sv_sigcode_base;
638 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
639 regs->tf_cs = _ucodesel;
640 regs->tf_ds = _udatasel;
641 regs->tf_es = _udatasel;
642 regs->tf_fs = _udatasel;
643 regs->tf_ss = _udatasel;
645 mtx_lock(&psp->ps_mtx);
649 * System call to cleanup state after a signal
650 * has been taken. Reset signal mask and
651 * stack state from context left by sendsig (above).
652 * Return to previous pc and psl as specified by
653 * context left by sendsig. Check carefully to
654 * make sure that the user has not modified the
655 * psl to gain improper privileges or to cause
659 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
661 struct l_sigframe frame;
662 struct trapframe *regs;
671 if (ldebug(sigreturn))
672 printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
675 * The trampoline code hands us the sigframe.
676 * It is unsafe to keep track of it ourselves, in the event that a
677 * program jumps out of a signal handler.
679 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
683 * Check for security violations.
685 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
686 eflags = frame.sf_sc.sc_eflags;
688 * XXX do allow users to change the privileged flag PSL_RF. The
689 * cpu sets PSL_RF in tf_eflags for faults. Debuggers should
690 * sometimes set it there too. tf_eflags is kept in the signal
691 * context during signal handling and there is no other place
692 * to remember it, so the PSL_RF bit may be corrupted by the
693 * signal handler without us knowing. Corruption of the PSL_RF
694 * bit at worst causes one more or one less debugger trap, so
695 * allowing it is fairly harmless.
697 if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF))
701 * Don't allow users to load a valid privileged %cs. Let the
702 * hardware check for invalid selectors, excess privilege in
703 * other selectors, invalid %eip's and invalid %esp's.
705 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
706 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
707 ksiginfo_init_trap(&ksi);
708 ksi.ksi_signo = SIGBUS;
709 ksi.ksi_code = BUS_OBJERR;
710 ksi.ksi_trapno = T_PROTFLT;
711 ksi.ksi_addr = (void *)regs->tf_eip;
712 trapsignal(td, &ksi);
716 lmask.__bits[0] = frame.sf_sc.sc_mask;
717 for (i = 0; i < (LINUX_NSIG_WORDS-1); i++)
718 lmask.__bits[i+1] = frame.sf_extramask[i];
719 linux_to_bsd_sigset(&lmask, &bmask);
720 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
723 * Restore signal context.
725 /* %gs was restored by the trampoline. */
726 regs->tf_fs = frame.sf_sc.sc_fs;
727 regs->tf_es = frame.sf_sc.sc_es;
728 regs->tf_ds = frame.sf_sc.sc_ds;
729 regs->tf_edi = frame.sf_sc.sc_edi;
730 regs->tf_esi = frame.sf_sc.sc_esi;
731 regs->tf_ebp = frame.sf_sc.sc_ebp;
732 regs->tf_ebx = frame.sf_sc.sc_ebx;
733 regs->tf_edx = frame.sf_sc.sc_edx;
734 regs->tf_ecx = frame.sf_sc.sc_ecx;
735 regs->tf_eax = frame.sf_sc.sc_eax;
736 regs->tf_eip = frame.sf_sc.sc_eip;
737 regs->tf_cs = frame.sf_sc.sc_cs;
738 regs->tf_eflags = eflags;
739 regs->tf_esp = frame.sf_sc.sc_esp_at_signal;
740 regs->tf_ss = frame.sf_sc.sc_ss;
742 return (EJUSTRETURN);
746 * System call to cleanup state after a signal
747 * has been taken. Reset signal mask and
748 * stack state from context left by rt_sendsig (above).
749 * Return to previous pc and psl as specified by
750 * context left by sendsig. Check carefully to
751 * make sure that the user has not modified the
752 * psl to gain improper privileges or to cause
756 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
758 struct l_ucontext uc;
759 struct l_sigcontext *context;
763 struct trapframe *regs;
770 if (ldebug(rt_sigreturn))
771 printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
774 * The trampoline code hands us the ucontext.
775 * It is unsafe to keep track of it ourselves, in the event that a
776 * program jumps out of a signal handler.
778 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
781 context = &uc.uc_mcontext;
784 * Check for security violations.
786 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
787 eflags = context->sc_eflags;
789 * XXX do allow users to change the privileged flag PSL_RF. The
790 * cpu sets PSL_RF in tf_eflags for faults. Debuggers should
791 * sometimes set it there too. tf_eflags is kept in the signal
792 * context during signal handling and there is no other place
793 * to remember it, so the PSL_RF bit may be corrupted by the
794 * signal handler without us knowing. Corruption of the PSL_RF
795 * bit at worst causes one more or one less debugger trap, so
796 * allowing it is fairly harmless.
798 if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF))
802 * Don't allow users to load a valid privileged %cs. Let the
803 * hardware check for invalid selectors, excess privilege in
804 * other selectors, invalid %eip's and invalid %esp's.
806 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
807 if (!CS_SECURE(context->sc_cs)) {
808 ksiginfo_init_trap(&ksi);
809 ksi.ksi_signo = SIGBUS;
810 ksi.ksi_code = BUS_OBJERR;
811 ksi.ksi_trapno = T_PROTFLT;
812 ksi.ksi_addr = (void *)regs->tf_eip;
813 trapsignal(td, &ksi);
817 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
818 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
821 * Restore signal context
823 /* %gs was restored by the trampoline. */
824 regs->tf_fs = context->sc_fs;
825 regs->tf_es = context->sc_es;
826 regs->tf_ds = context->sc_ds;
827 regs->tf_edi = context->sc_edi;
828 regs->tf_esi = context->sc_esi;
829 regs->tf_ebp = context->sc_ebp;
830 regs->tf_ebx = context->sc_ebx;
831 regs->tf_edx = context->sc_edx;
832 regs->tf_ecx = context->sc_ecx;
833 regs->tf_eax = context->sc_eax;
834 regs->tf_eip = context->sc_eip;
835 regs->tf_cs = context->sc_cs;
836 regs->tf_eflags = eflags;
837 regs->tf_esp = context->sc_esp_at_signal;
838 regs->tf_ss = context->sc_ss;
841 * call sigaltstack & ignore results..
844 ss.ss_sp = lss->ss_sp;
845 ss.ss_size = lss->ss_size;
846 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
849 if (ldebug(rt_sigreturn))
850 printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
851 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
853 (void)kern_sigaltstack(td, &ss, NULL);
855 return (EJUSTRETURN);
859 linux_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
862 struct trapframe *frame;
865 frame = td->td_frame;
867 sa->code = frame->tf_eax;
868 sa->args[0] = frame->tf_ebx;
869 sa->args[1] = frame->tf_ecx;
870 sa->args[2] = frame->tf_edx;
871 sa->args[3] = frame->tf_esi;
872 sa->args[4] = frame->tf_edi;
873 sa->args[5] = frame->tf_ebp; /* Unconfirmed */
875 if (sa->code >= p->p_sysent->sv_size)
876 sa->callp = &p->p_sysent->sv_table[0];
878 sa->callp = &p->p_sysent->sv_table[sa->code];
879 sa->narg = sa->callp->sy_narg;
881 td->td_retval[0] = 0;
882 td->td_retval[1] = frame->tf_edx;
888 * If a linux binary is exec'ing something, try this image activator
889 * first. We override standard shell script execution in order to
890 * be able to modify the interpreter path. We only do this if a linux
891 * binary is doing the exec, so we do not create an EXEC module for it.
893 static int exec_linux_imgact_try(struct image_params *iparams);
896 exec_linux_imgact_try(struct image_params *imgp)
898 const char *head = (const char *)imgp->image_header;
903 * The interpreter for shell scripts run from a linux binary needs
904 * to be located in /compat/linux if possible in order to recursively
905 * maintain linux path emulation.
907 if (((const short *)head)[0] == SHELLMAGIC) {
909 * Run our normal shell image activator. If it succeeds attempt
910 * to use the alternate path for the interpreter. If an alternate
911 * path is found, use our stringspace to store it.
913 if ((error = exec_shell_imgact(imgp)) == 0) {
914 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
915 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, AT_FDCWD);
917 imgp->args->fname_buf =
918 imgp->interpreter_name = rpath;
925 * exec_setregs may initialize some registers differently than Linux
926 * does, thus potentially confusing Linux binaries. If necessary, we
927 * override the exec_setregs default(s) here.
930 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack)
932 struct pcb *pcb = td->td_pcb;
934 exec_setregs(td, imgp, stack);
936 /* Linux sets %gs to 0, we default to _udatasel */
940 pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
944 linux_get_machine(const char **dst)
962 struct sysentvec linux_sysvec = {
963 .sv_size = LINUX_SYS_MAXSYSCALL,
964 .sv_table = linux_sysent,
966 .sv_sigsize = LINUX_SIGTBLSZ,
967 .sv_sigtbl = bsd_to_linux_signal,
968 .sv_errsize = ELAST + 1,
969 .sv_errtbl = bsd_to_linux_errno,
970 .sv_transtrap = translate_traps,
971 .sv_fixup = linux_fixup,
972 .sv_sendsig = linux_sendsig,
973 .sv_sigcode = linux_sigcode,
974 .sv_szsigcode = &linux_szsigcode,
975 .sv_prepsyscall = NULL,
976 .sv_name = "Linux a.out",
978 .sv_imgact_try = exec_linux_imgact_try,
979 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
980 .sv_pagesize = PAGE_SIZE,
981 .sv_minuser = VM_MIN_ADDRESS,
982 .sv_maxuser = VM_MAXUSER_ADDRESS,
983 .sv_usrstack = LINUX_USRSTACK,
984 .sv_psstrings = PS_STRINGS,
985 .sv_stackprot = VM_PROT_ALL,
986 .sv_copyout_strings = exec_copyout_strings,
987 .sv_setregs = exec_linux_setregs,
990 .sv_flags = SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32,
991 .sv_set_syscall_retval = cpu_set_syscall_retval,
992 .sv_fetch_syscall_args = linux_fetch_syscall_args,
993 .sv_syscallnames = NULL,
994 .sv_shared_page_base = LINUX_SHAREDPAGE,
995 .sv_shared_page_len = PAGE_SIZE,
996 .sv_schedtail = linux_schedtail,
998 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
1000 struct sysentvec elf_linux_sysvec = {
1001 .sv_size = LINUX_SYS_MAXSYSCALL,
1002 .sv_table = linux_sysent,
1004 .sv_sigsize = LINUX_SIGTBLSZ,
1005 .sv_sigtbl = bsd_to_linux_signal,
1006 .sv_errsize = ELAST + 1,
1007 .sv_errtbl = bsd_to_linux_errno,
1008 .sv_transtrap = translate_traps,
1009 .sv_fixup = elf_linux_fixup,
1010 .sv_sendsig = linux_sendsig,
1011 .sv_sigcode = linux_sigcode,
1012 .sv_szsigcode = &linux_szsigcode,
1013 .sv_prepsyscall = NULL,
1014 .sv_name = "Linux ELF",
1015 .sv_coredump = elf32_coredump,
1016 .sv_imgact_try = exec_linux_imgact_try,
1017 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
1018 .sv_pagesize = PAGE_SIZE,
1019 .sv_minuser = VM_MIN_ADDRESS,
1020 .sv_maxuser = VM_MAXUSER_ADDRESS,
1021 .sv_usrstack = LINUX_USRSTACK,
1022 .sv_psstrings = LINUX_PS_STRINGS,
1023 .sv_stackprot = VM_PROT_ALL,
1024 .sv_copyout_strings = linux_copyout_strings,
1025 .sv_setregs = exec_linux_setregs,
1026 .sv_fixlimit = NULL,
1028 .sv_flags = SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP,
1029 .sv_set_syscall_retval = cpu_set_syscall_retval,
1030 .sv_fetch_syscall_args = linux_fetch_syscall_args,
1031 .sv_syscallnames = NULL,
1032 .sv_shared_page_base = LINUX_SHAREDPAGE,
1033 .sv_shared_page_len = PAGE_SIZE,
1034 .sv_schedtail = linux_schedtail,
1036 INIT_SYSENTVEC(elf_sysvec, &elf_linux_sysvec);
1038 static char GNU_ABI_VENDOR[] = "GNU";
1039 static int GNULINUX_ABI_DESC = 0;
1042 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
1044 const Elf32_Word *desc;
1047 p = (uintptr_t)(note + 1);
1048 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1050 desc = (const Elf32_Word *)p;
1051 if (desc[0] != GNULINUX_ABI_DESC)
1055 * For linux we encode osrel as follows (see linux_mib.c):
1056 * VVVMMMIII (version, major, minor), see linux_mib.c.
1058 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1063 static Elf_Brandnote linux_brandnote = {
1064 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
1065 .hdr.n_descsz = 16, /* XXX at least 16 */
1067 .vendor = GNU_ABI_VENDOR,
1068 .flags = BN_TRANSLATE_OSREL,
1069 .trans_osrel = linux_trans_osrel
1072 static Elf32_Brandinfo linux_brand = {
1073 .brand = ELFOSABI_LINUX,
1075 .compat_3_brand = "Linux",
1076 .emul_path = "/compat/linux",
1077 .interp_path = "/lib/ld-linux.so.1",
1078 .sysvec = &elf_linux_sysvec,
1079 .interp_newpath = NULL,
1080 .brand_note = &linux_brandnote,
1081 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1084 static Elf32_Brandinfo linux_glibc2brand = {
1085 .brand = ELFOSABI_LINUX,
1087 .compat_3_brand = "Linux",
1088 .emul_path = "/compat/linux",
1089 .interp_path = "/lib/ld-linux.so.2",
1090 .sysvec = &elf_linux_sysvec,
1091 .interp_newpath = NULL,
1092 .brand_note = &linux_brandnote,
1093 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1096 Elf32_Brandinfo *linux_brandlist[] = {
1103 linux_elf_modevent(module_t mod, int type, void *data)
1105 Elf32_Brandinfo **brandinfo;
1107 struct linux_ioctl_handler **lihp;
1108 struct linux_device_handler **ldhp;
1114 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1116 if (elf32_insert_brand_entry(*brandinfo) < 0)
1119 SET_FOREACH(lihp, linux_ioctl_handler_set)
1120 linux_ioctl_register_handler(*lihp);
1121 SET_FOREACH(ldhp, linux_device_handler_set)
1122 linux_device_register_handler(*ldhp);
1123 mtx_init(&emul_lock, "emuldata lock", NULL, MTX_DEF);
1124 sx_init(&emul_shared_lock, "emuldata->shared lock");
1125 LIST_INIT(&futex_list);
1126 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1127 linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit,
1129 linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec,
1131 linux_get_machine(&linux_platform);
1132 linux_szplatform = roundup(strlen(linux_platform) + 1,
1134 linux_osd_jail_register();
1135 stclohz = (stathz ? stathz : hz);
1137 printf("Linux ELF exec handler installed\n");
1139 printf("cannot insert Linux ELF brand handler\n");
1142 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1144 if (elf32_brand_inuse(*brandinfo))
1147 for (brandinfo = &linux_brandlist[0];
1148 *brandinfo != NULL; ++brandinfo)
1149 if (elf32_remove_brand_entry(*brandinfo) < 0)
1153 SET_FOREACH(lihp, linux_ioctl_handler_set)
1154 linux_ioctl_unregister_handler(*lihp);
1155 SET_FOREACH(ldhp, linux_device_handler_set)
1156 linux_device_unregister_handler(*ldhp);
1157 mtx_destroy(&emul_lock);
1158 sx_destroy(&emul_shared_lock);
1159 mtx_destroy(&futex_mtx);
1160 EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
1161 EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
1162 linux_osd_jail_deregister();
1164 printf("Linux ELF exec handler removed\n");
1166 printf("Could not deinstall ELF interpreter entry\n");
1174 static moduledata_t linux_elf_mod = {
1180 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);