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/sysctl.h>
48 #include <sys/sysent.h>
49 #include <sys/sysproto.h>
50 #include <sys/vnode.h>
51 #include <sys/eventhandler.h>
55 #include <vm/vm_extern.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_param.h>
61 #include <machine/cpu.h>
62 #include <machine/cputypes.h>
63 #include <machine/md_var.h>
64 #include <machine/pcb.h>
66 #include <i386/linux/linux.h>
67 #include <i386/linux/linux_proto.h>
68 #include <compat/linux/linux_emul.h>
69 #include <compat/linux/linux_futex.h>
70 #include <compat/linux/linux_ioctl.h>
71 #include <compat/linux/linux_mib.h>
72 #include <compat/linux/linux_misc.h>
73 #include <compat/linux/linux_signal.h>
74 #include <compat/linux/linux_util.h>
75 #include <compat/linux/linux_vdso.h>
77 MODULE_VERSION(linux, 1);
79 #if BYTE_ORDER == LITTLE_ENDIAN
80 #define SHELLMAGIC 0x2123 /* #! */
82 #define SHELLMAGIC 0x2321
86 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
87 CTLTYPE_STRING | CTLFLAG_RW,
88 0, 0, linux_sysctl_debug, "A",
89 "Linux debugging control");
93 * Allow the sendsig functions to use the ldebug() facility
94 * even though they are not syscalls themselves. Map them
95 * to syscall 0. This is slightly less bogus than using
98 #define LINUX_SYS_linux_rt_sendsig 0
99 #define LINUX_SYS_linux_sendsig 0
101 #define LINUX_PS_STRINGS (LINUX_USRSTACK - sizeof(struct ps_strings))
103 static int linux_szsigcode;
104 static vm_object_t linux_shared_page_obj;
105 static char *linux_shared_page_mapping;
106 extern char _binary_linux_locore_o_start;
107 extern char _binary_linux_locore_o_end;
109 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
111 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
113 static int linux_fixup(register_t **stack_base,
114 struct image_params *iparams);
115 static int elf_linux_fixup(register_t **stack_base,
116 struct image_params *iparams);
117 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
118 static void exec_linux_setregs(struct thread *td,
119 struct image_params *imgp, u_long stack);
120 static register_t *linux_copyout_strings(struct image_params *imgp);
121 static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
122 static void linux_vdso_install(void *param);
123 static void linux_vdso_deinstall(void *param);
125 static int linux_szplatform;
126 const char *linux_kplatform;
128 static eventhandler_tag linux_exit_tag;
129 static eventhandler_tag linux_exec_tag;
130 static eventhandler_tag linux_thread_dtor_tag;
133 * Linux syscalls return negative errno's, we do positive and map them
135 * FreeBSD: src/sys/sys/errno.h
136 * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h
137 * linux-2.6.17.8/include/asm-generic/errno.h
139 static int bsd_to_linux_errno[ELAST + 1] = {
140 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9,
141 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
142 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
143 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
144 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
145 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
146 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
147 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9,
148 -6, -6, -43, -42, -75,-125, -84, -95, -16, -74,
152 #define LINUX_T_UNKNOWN 255
153 static int _bsd_to_linux_trapcode[] = {
154 LINUX_T_UNKNOWN, /* 0 */
155 6, /* 1 T_PRIVINFLT */
156 LINUX_T_UNKNOWN, /* 2 */
158 LINUX_T_UNKNOWN, /* 4 */
159 LINUX_T_UNKNOWN, /* 5 */
160 16, /* 6 T_ARITHTRAP */
161 254, /* 7 T_ASTFLT */
162 LINUX_T_UNKNOWN, /* 8 */
163 13, /* 9 T_PROTFLT */
164 1, /* 10 T_TRCTRAP */
165 LINUX_T_UNKNOWN, /* 11 */
166 14, /* 12 T_PAGEFLT */
167 LINUX_T_UNKNOWN, /* 13 */
168 17, /* 14 T_ALIGNFLT */
169 LINUX_T_UNKNOWN, /* 15 */
170 LINUX_T_UNKNOWN, /* 16 */
171 LINUX_T_UNKNOWN, /* 17 */
177 8, /* 23 T_DOUBLEFLT */
178 9, /* 24 T_FPOPFLT */
179 10, /* 25 T_TSSFLT */
180 11, /* 26 T_SEGNPFLT */
181 12, /* 27 T_STKFLT */
183 19, /* 29 T_XMMFLT */
184 15 /* 30 T_RESERVED */
186 #define bsd_to_linux_trapcode(code) \
187 ((code)<sizeof(_bsd_to_linux_trapcode)/sizeof(*_bsd_to_linux_trapcode)? \
188 _bsd_to_linux_trapcode[(code)]: \
191 LINUX_VDSO_SYM_INTPTR(linux_sigcode);
192 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
193 LINUX_VDSO_SYM_INTPTR(linux_vsyscall);
196 * If FreeBSD & Linux have a difference of opinion about what a trap
197 * means, deal with it here.
202 translate_traps(int signal, int trap_code)
204 if (signal != SIGBUS)
218 linux_fixup(register_t **stack_base, struct image_params *imgp)
220 register_t *argv, *envp;
223 envp = *stack_base + (imgp->args->argc + 1);
225 suword(*stack_base, (intptr_t)(void *)envp);
227 suword(*stack_base, (intptr_t)(void *)argv);
229 suword(*stack_base, imgp->args->argc);
234 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
238 Elf32_Addr *uplatform;
239 struct ps_strings *arginfo;
242 KASSERT(curthread->td_proc == imgp->proc,
243 ("unsafe elf_linux_fixup(), should be curproc"));
246 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
247 uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
248 args = (Elf32_Auxargs *)imgp->auxargs;
249 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
251 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
252 imgp->proc->p_sysent->sv_shared_page_base);
253 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall);
254 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
257 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
258 * as it has appeared in the 2.4.0-rc7 first time.
259 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
260 * glibc falls back to the hard-coded CLK_TCK value when aux entry
262 * Also see linux_times() implementation.
264 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
265 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
266 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
267 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
268 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
269 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
270 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
271 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
272 AUXARGS_ENTRY(pos, AT_BASE, args->base);
273 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, 0);
274 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
275 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
276 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
277 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
278 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
279 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
280 if (imgp->execpathp != 0)
281 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
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;
305 char canary[LINUX_AT_RANDOM_LEN];
310 * Calculate string base and vector table pointers.
313 if (imgp->execpath != NULL && imgp->auxargs != NULL)
314 execpath_len = strlen(imgp->execpath) + 1;
317 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
318 destp = (caddr_t)arginfo - SPARE_USRSPACE - linux_szplatform -
319 roundup(sizeof(canary), sizeof(char *)) -
320 roundup(execpath_len, sizeof(char *)) -
321 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
324 * install LINUX_PLATFORM
326 copyout(linux_kplatform, ((caddr_t)arginfo - linux_szplatform),
329 if (execpath_len != 0) {
330 imgp->execpathp = (uintptr_t)arginfo -
331 linux_szplatform - execpath_len;
332 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
336 * Prepare the canary for SSP.
338 arc4rand(canary, sizeof(canary), 0);
339 imgp->canary = (uintptr_t)arginfo - linux_szplatform -
340 roundup(execpath_len, sizeof(char *)) -
341 roundup(sizeof(canary), sizeof(char *));
342 copyout(canary, (void *)imgp->canary, sizeof(canary));
345 * If we have a valid auxargs ptr, prepare some room
350 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
351 * lower compatibility.
353 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
354 (LINUX_AT_COUNT * 2);
356 * The '+ 2' is for the null pointers at the end of each of
357 * the arg and env vector sets,and imgp->auxarg_size is room
358 * for argument of Runtime loader.
360 vectp = (char **)(destp - (imgp->args->argc +
361 imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *));
364 * The '+ 2' is for the null pointers at the end of each of
365 * the arg and env vector sets
367 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
372 * vectp also becomes our initial stack base
374 stack_base = (register_t *)vectp;
376 stringp = imgp->args->begin_argv;
377 argc = imgp->args->argc;
378 envc = imgp->args->envc;
381 * Copy out strings - arguments and environment.
383 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
386 * Fill in "ps_strings" struct for ps, w, etc.
388 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
389 suword(&arginfo->ps_nargvstr, argc);
392 * Fill in argument portion of vector table.
394 for (; argc > 0; --argc) {
395 suword(vectp++, (long)(intptr_t)destp);
396 while (*stringp++ != 0)
401 /* a null vector table pointer separates the argp's from the envp's */
404 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
405 suword(&arginfo->ps_nenvstr, envc);
408 * Fill in environment portion of vector table.
410 for (; envc > 0; --envc) {
411 suword(vectp++, (long)(intptr_t)destp);
412 while (*stringp++ != 0)
417 /* end of vector table is a null pointer */
424 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
426 struct thread *td = curthread;
427 struct proc *p = td->td_proc;
429 struct trapframe *regs;
430 struct l_rt_sigframe *fp, frame;
434 sig = ksi->ksi_signo;
435 code = ksi->ksi_code;
436 PROC_LOCK_ASSERT(p, MA_OWNED);
438 mtx_assert(&psp->ps_mtx, MA_OWNED);
440 oonstack = sigonstack(regs->tf_esp);
443 if (ldebug(rt_sendsig))
444 printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
445 catcher, sig, (void*)mask, code);
448 * Allocate space for the signal handler context.
450 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
451 SIGISMEMBER(psp->ps_sigonstack, sig)) {
452 fp = (struct l_rt_sigframe *)(td->td_sigstk.ss_sp +
453 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
455 fp = (struct l_rt_sigframe *)regs->tf_esp - 1;
456 mtx_unlock(&psp->ps_mtx);
459 * Build the argument list for the signal handler.
461 sig = bsd_to_linux_signal(sig);
463 bzero(&frame, sizeof(frame));
465 frame.sf_handler = catcher;
467 frame.sf_siginfo = &fp->sf_si;
468 frame.sf_ucontext = &fp->sf_sc;
470 /* Fill in POSIX parts */
471 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
474 * Build the signal context to be used by sigreturn.
476 frame.sf_sc.uc_flags = 0; /* XXX ??? */
477 frame.sf_sc.uc_link = NULL; /* XXX ??? */
479 frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp;
480 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
481 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
482 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
485 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
487 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask;
488 frame.sf_sc.uc_mcontext.sc_gs = rgs();
489 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
490 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
491 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
492 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_edi;
493 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_esi;
494 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_ebp;
495 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_ebx;
496 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_esp;
497 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_edx;
498 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_ecx;
499 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_eax;
500 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_eip;
501 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
502 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags;
503 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp;
504 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
505 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
506 frame.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
507 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
510 if (ldebug(rt_sendsig))
511 printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
512 frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
513 td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
516 if (copyout(&frame, fp, sizeof(frame)) != 0) {
518 * Process has trashed its stack; give it an illegal
519 * instruction to halt it in its tracks.
522 if (ldebug(rt_sendsig))
523 printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
531 * Build context to run handler in.
533 regs->tf_esp = (int)fp;
534 regs->tf_eip = linux_rt_sigcode;
535 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
536 regs->tf_cs = _ucodesel;
537 regs->tf_ds = _udatasel;
538 regs->tf_es = _udatasel;
539 regs->tf_fs = _udatasel;
540 regs->tf_ss = _udatasel;
542 mtx_lock(&psp->ps_mtx);
547 * Send an interrupt to process.
549 * Stack is set up to allow sigcode stored
550 * in u. to call routine, followed by kcall
551 * to sigreturn routine below. After sigreturn
552 * resets the signal mask, the stack, and the
553 * frame pointer, it returns to the user
557 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
559 struct thread *td = curthread;
560 struct proc *p = td->td_proc;
562 struct trapframe *regs;
563 struct l_sigframe *fp, frame;
568 PROC_LOCK_ASSERT(p, MA_OWNED);
570 sig = ksi->ksi_signo;
571 code = ksi->ksi_code;
572 mtx_assert(&psp->ps_mtx, MA_OWNED);
573 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
574 /* Signal handler installed with SA_SIGINFO. */
575 linux_rt_sendsig(catcher, ksi, mask);
579 oonstack = sigonstack(regs->tf_esp);
583 printf(ARGS(sendsig, "%p, %d, %p, %u"),
584 catcher, sig, (void*)mask, code);
588 * Allocate space for the signal handler context.
590 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
591 SIGISMEMBER(psp->ps_sigonstack, sig)) {
592 fp = (struct l_sigframe *)(td->td_sigstk.ss_sp +
593 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
595 fp = (struct l_sigframe *)regs->tf_esp - 1;
596 mtx_unlock(&psp->ps_mtx);
600 * Build the argument list for the signal handler.
602 sig = bsd_to_linux_signal(sig);
604 bzero(&frame, sizeof(frame));
606 frame.sf_handler = catcher;
609 bsd_to_linux_sigset(mask, &lmask);
612 * Build the signal context to be used by sigreturn.
614 frame.sf_sc.sc_mask = lmask.__mask;
615 frame.sf_sc.sc_gs = rgs();
616 frame.sf_sc.sc_fs = regs->tf_fs;
617 frame.sf_sc.sc_es = regs->tf_es;
618 frame.sf_sc.sc_ds = regs->tf_ds;
619 frame.sf_sc.sc_edi = regs->tf_edi;
620 frame.sf_sc.sc_esi = regs->tf_esi;
621 frame.sf_sc.sc_ebp = regs->tf_ebp;
622 frame.sf_sc.sc_ebx = regs->tf_ebx;
623 frame.sf_sc.sc_esp = regs->tf_esp;
624 frame.sf_sc.sc_edx = regs->tf_edx;
625 frame.sf_sc.sc_ecx = regs->tf_ecx;
626 frame.sf_sc.sc_eax = regs->tf_eax;
627 frame.sf_sc.sc_eip = regs->tf_eip;
628 frame.sf_sc.sc_cs = regs->tf_cs;
629 frame.sf_sc.sc_eflags = regs->tf_eflags;
630 frame.sf_sc.sc_esp_at_signal = regs->tf_esp;
631 frame.sf_sc.sc_ss = regs->tf_ss;
632 frame.sf_sc.sc_err = regs->tf_err;
633 frame.sf_sc.sc_cr2 = (register_t)ksi->ksi_addr;
634 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno);
636 frame.sf_extramask[0] = lmask.__mask;
638 if (copyout(&frame, fp, sizeof(frame)) != 0) {
640 * Process has trashed its stack; give it an illegal
641 * instruction to halt it in its tracks.
648 * Build context to run handler in.
650 regs->tf_esp = (int)fp;
651 regs->tf_eip = linux_sigcode;
652 regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D);
653 regs->tf_cs = _ucodesel;
654 regs->tf_ds = _udatasel;
655 regs->tf_es = _udatasel;
656 regs->tf_fs = _udatasel;
657 regs->tf_ss = _udatasel;
659 mtx_lock(&psp->ps_mtx);
663 * System call to cleanup state after a signal
664 * has been taken. Reset signal mask and
665 * stack state from context left by sendsig (above).
666 * Return to previous pc and psl as specified by
667 * context left by sendsig. Check carefully to
668 * make sure that the user has not modified the
669 * psl to gain improper privileges or to cause
673 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
675 struct l_sigframe frame;
676 struct trapframe *regs;
685 if (ldebug(sigreturn))
686 printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
689 * The trampoline code hands us the sigframe.
690 * It is unsafe to keep track of it ourselves, in the event that a
691 * program jumps out of a signal handler.
693 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
697 * Check for security violations.
699 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
700 eflags = frame.sf_sc.sc_eflags;
701 if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
705 * Don't allow users to load a valid privileged %cs. Let the
706 * hardware check for invalid selectors, excess privilege in
707 * other selectors, invalid %eip's and invalid %esp's.
709 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
710 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
711 ksiginfo_init_trap(&ksi);
712 ksi.ksi_signo = SIGBUS;
713 ksi.ksi_code = BUS_OBJERR;
714 ksi.ksi_trapno = T_PROTFLT;
715 ksi.ksi_addr = (void *)regs->tf_eip;
716 trapsignal(td, &ksi);
720 lmask.__mask = frame.sf_sc.sc_mask;
721 linux_to_bsd_sigset(&lmask, &bmask);
722 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
725 * Restore signal context.
727 /* %gs was restored by the trampoline. */
728 regs->tf_fs = frame.sf_sc.sc_fs;
729 regs->tf_es = frame.sf_sc.sc_es;
730 regs->tf_ds = frame.sf_sc.sc_ds;
731 regs->tf_edi = frame.sf_sc.sc_edi;
732 regs->tf_esi = frame.sf_sc.sc_esi;
733 regs->tf_ebp = frame.sf_sc.sc_ebp;
734 regs->tf_ebx = frame.sf_sc.sc_ebx;
735 regs->tf_edx = frame.sf_sc.sc_edx;
736 regs->tf_ecx = frame.sf_sc.sc_ecx;
737 regs->tf_eax = frame.sf_sc.sc_eax;
738 regs->tf_eip = frame.sf_sc.sc_eip;
739 regs->tf_cs = frame.sf_sc.sc_cs;
740 regs->tf_eflags = eflags;
741 regs->tf_esp = frame.sf_sc.sc_esp_at_signal;
742 regs->tf_ss = frame.sf_sc.sc_ss;
744 return (EJUSTRETURN);
748 * System call to cleanup state after a signal
749 * has been taken. Reset signal mask and
750 * stack state from context left by rt_sendsig (above).
751 * Return to previous pc and psl as specified by
752 * context left by sendsig. Check carefully to
753 * make sure that the user has not modified the
754 * psl to gain improper privileges or to cause
758 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
760 struct l_ucontext uc;
761 struct l_sigcontext *context;
765 struct trapframe *regs;
772 if (ldebug(rt_sigreturn))
773 printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
776 * The trampoline code hands us the ucontext.
777 * It is unsafe to keep track of it ourselves, in the event that a
778 * program jumps out of a signal handler.
780 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
783 context = &uc.uc_mcontext;
786 * Check for security violations.
788 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
789 eflags = context->sc_eflags;
790 if (!EFLAGS_SECURE(eflags, regs->tf_eflags))
794 * Don't allow users to load a valid privileged %cs. Let the
795 * hardware check for invalid selectors, excess privilege in
796 * other selectors, invalid %eip's and invalid %esp's.
798 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
799 if (!CS_SECURE(context->sc_cs)) {
800 ksiginfo_init_trap(&ksi);
801 ksi.ksi_signo = SIGBUS;
802 ksi.ksi_code = BUS_OBJERR;
803 ksi.ksi_trapno = T_PROTFLT;
804 ksi.ksi_addr = (void *)regs->tf_eip;
805 trapsignal(td, &ksi);
809 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
810 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
813 * Restore signal context
815 /* %gs was restored by the trampoline. */
816 regs->tf_fs = context->sc_fs;
817 regs->tf_es = context->sc_es;
818 regs->tf_ds = context->sc_ds;
819 regs->tf_edi = context->sc_edi;
820 regs->tf_esi = context->sc_esi;
821 regs->tf_ebp = context->sc_ebp;
822 regs->tf_ebx = context->sc_ebx;
823 regs->tf_edx = context->sc_edx;
824 regs->tf_ecx = context->sc_ecx;
825 regs->tf_eax = context->sc_eax;
826 regs->tf_eip = context->sc_eip;
827 regs->tf_cs = context->sc_cs;
828 regs->tf_eflags = eflags;
829 regs->tf_esp = context->sc_esp_at_signal;
830 regs->tf_ss = context->sc_ss;
833 * call sigaltstack & ignore results..
836 ss.ss_sp = lss->ss_sp;
837 ss.ss_size = lss->ss_size;
838 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
841 if (ldebug(rt_sigreturn))
842 printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"),
843 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
845 (void)kern_sigaltstack(td, &ss, NULL);
847 return (EJUSTRETURN);
851 linux_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
854 struct trapframe *frame;
857 frame = td->td_frame;
859 sa->code = frame->tf_eax;
860 sa->args[0] = frame->tf_ebx;
861 sa->args[1] = frame->tf_ecx;
862 sa->args[2] = frame->tf_edx;
863 sa->args[3] = frame->tf_esi;
864 sa->args[4] = frame->tf_edi;
865 sa->args[5] = frame->tf_ebp; /* Unconfirmed */
867 if (sa->code >= p->p_sysent->sv_size)
869 sa->callp = &p->p_sysent->sv_table[LINUX_SYS_MAXSYSCALL];
871 sa->callp = &p->p_sysent->sv_table[sa->code];
872 sa->narg = sa->callp->sy_narg;
874 td->td_retval[0] = 0;
875 td->td_retval[1] = frame->tf_edx;
881 * If a linux binary is exec'ing something, try this image activator
882 * first. We override standard shell script execution in order to
883 * be able to modify the interpreter path. We only do this if a linux
884 * binary is doing the exec, so we do not create an EXEC module for it.
886 static int exec_linux_imgact_try(struct image_params *iparams);
889 exec_linux_imgact_try(struct image_params *imgp)
891 const char *head = (const char *)imgp->image_header;
896 * The interpreter for shell scripts run from a linux binary needs
897 * to be located in /compat/linux if possible in order to recursively
898 * maintain linux path emulation.
900 if (((const short *)head)[0] == SHELLMAGIC) {
902 * Run our normal shell image activator. If it succeeds attempt
903 * to use the alternate path for the interpreter. If an alternate
904 * path is found, use our stringspace to store it.
906 if ((error = exec_shell_imgact(imgp)) == 0) {
907 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
908 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, AT_FDCWD);
910 imgp->args->fname_buf =
911 imgp->interpreter_name = rpath;
918 * exec_setregs may initialize some registers differently than Linux
919 * does, thus potentially confusing Linux binaries. If necessary, we
920 * override the exec_setregs default(s) here.
923 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack)
925 struct pcb *pcb = td->td_pcb;
927 exec_setregs(td, imgp, stack);
929 /* Linux sets %gs to 0, we default to _udatasel */
933 pcb->pcb_initial_npxcw = __LINUX_NPXCW__;
937 linux_get_machine(const char **dst)
955 struct sysentvec linux_sysvec = {
956 .sv_size = LINUX_SYS_MAXSYSCALL,
957 .sv_table = linux_sysent,
961 .sv_errsize = ELAST + 1,
962 .sv_errtbl = bsd_to_linux_errno,
963 .sv_transtrap = translate_traps,
964 .sv_fixup = linux_fixup,
965 .sv_sendsig = linux_sendsig,
966 .sv_sigcode = &_binary_linux_locore_o_start,
967 .sv_szsigcode = &linux_szsigcode,
968 .sv_prepsyscall = NULL,
969 .sv_name = "Linux a.out",
971 .sv_imgact_try = exec_linux_imgact_try,
972 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
973 .sv_pagesize = PAGE_SIZE,
974 .sv_minuser = VM_MIN_ADDRESS,
975 .sv_maxuser = VM_MAXUSER_ADDRESS,
976 .sv_usrstack = LINUX_USRSTACK,
977 .sv_psstrings = PS_STRINGS,
978 .sv_stackprot = VM_PROT_ALL,
979 .sv_copyout_strings = exec_copyout_strings,
980 .sv_setregs = exec_linux_setregs,
983 .sv_flags = SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32,
984 .sv_set_syscall_retval = cpu_set_syscall_retval,
985 .sv_fetch_syscall_args = linux_fetch_syscall_args,
986 .sv_syscallnames = NULL,
987 .sv_shared_page_base = LINUX_SHAREDPAGE,
988 .sv_shared_page_len = PAGE_SIZE,
989 .sv_schedtail = linux_schedtail,
990 .sv_thread_detach = linux_thread_detach,
992 INIT_SYSENTVEC(aout_sysvec, &linux_sysvec);
994 struct sysentvec elf_linux_sysvec = {
995 .sv_size = LINUX_SYS_MAXSYSCALL,
996 .sv_table = linux_sysent,
1000 .sv_errsize = ELAST + 1,
1001 .sv_errtbl = bsd_to_linux_errno,
1002 .sv_transtrap = translate_traps,
1003 .sv_fixup = elf_linux_fixup,
1004 .sv_sendsig = linux_sendsig,
1005 .sv_sigcode = &_binary_linux_locore_o_start,
1006 .sv_szsigcode = &linux_szsigcode,
1007 .sv_prepsyscall = NULL,
1008 .sv_name = "Linux ELF",
1009 .sv_coredump = elf32_coredump,
1010 .sv_imgact_try = exec_linux_imgact_try,
1011 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
1012 .sv_pagesize = PAGE_SIZE,
1013 .sv_minuser = VM_MIN_ADDRESS,
1014 .sv_maxuser = VM_MAXUSER_ADDRESS,
1015 .sv_usrstack = LINUX_USRSTACK,
1016 .sv_psstrings = LINUX_PS_STRINGS,
1017 .sv_stackprot = VM_PROT_ALL,
1018 .sv_copyout_strings = linux_copyout_strings,
1019 .sv_setregs = exec_linux_setregs,
1020 .sv_fixlimit = NULL,
1022 .sv_flags = SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP,
1023 .sv_set_syscall_retval = cpu_set_syscall_retval,
1024 .sv_fetch_syscall_args = linux_fetch_syscall_args,
1025 .sv_syscallnames = NULL,
1026 .sv_shared_page_base = LINUX_SHAREDPAGE,
1027 .sv_shared_page_len = PAGE_SIZE,
1028 .sv_schedtail = linux_schedtail,
1029 .sv_thread_detach = linux_thread_detach,
1033 linux_vdso_install(void *param)
1036 linux_szsigcode = (&_binary_linux_locore_o_end -
1037 &_binary_linux_locore_o_start);
1039 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
1040 panic("Linux invalid vdso size\n");
1042 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
1044 linux_shared_page_obj = __elfN(linux_shared_page_init)
1045 (&linux_shared_page_mapping);
1047 __elfN(linux_vdso_reloc)(&elf_linux_sysvec, LINUX_SHAREDPAGE);
1049 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
1051 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
1053 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
1054 (sysinit_cfunc_t)linux_vdso_install, NULL);
1057 linux_vdso_deinstall(void *param)
1060 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
1062 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1063 (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
1065 static char GNU_ABI_VENDOR[] = "GNU";
1066 static int GNULINUX_ABI_DESC = 0;
1069 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
1071 const Elf32_Word *desc;
1074 p = (uintptr_t)(note + 1);
1075 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1077 desc = (const Elf32_Word *)p;
1078 if (desc[0] != GNULINUX_ABI_DESC)
1082 * For linux we encode osrel as follows (see linux_mib.c):
1083 * VVVMMMIII (version, major, minor), see linux_mib.c.
1085 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1090 static Elf_Brandnote linux_brandnote = {
1091 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
1092 .hdr.n_descsz = 16, /* XXX at least 16 */
1094 .vendor = GNU_ABI_VENDOR,
1095 .flags = BN_TRANSLATE_OSREL,
1096 .trans_osrel = linux_trans_osrel
1099 static Elf32_Brandinfo linux_brand = {
1100 .brand = ELFOSABI_LINUX,
1102 .compat_3_brand = "Linux",
1103 .emul_path = "/compat/linux",
1104 .interp_path = "/lib/ld-linux.so.1",
1105 .sysvec = &elf_linux_sysvec,
1106 .interp_newpath = NULL,
1107 .brand_note = &linux_brandnote,
1108 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1111 static Elf32_Brandinfo linux_glibc2brand = {
1112 .brand = ELFOSABI_LINUX,
1114 .compat_3_brand = "Linux",
1115 .emul_path = "/compat/linux",
1116 .interp_path = "/lib/ld-linux.so.2",
1117 .sysvec = &elf_linux_sysvec,
1118 .interp_newpath = NULL,
1119 .brand_note = &linux_brandnote,
1120 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1123 Elf32_Brandinfo *linux_brandlist[] = {
1130 linux_elf_modevent(module_t mod, int type, void *data)
1132 Elf32_Brandinfo **brandinfo;
1134 struct linux_ioctl_handler **lihp;
1140 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1142 if (elf32_insert_brand_entry(*brandinfo) < 0)
1145 SET_FOREACH(lihp, linux_ioctl_handler_set)
1146 linux_ioctl_register_handler(*lihp);
1147 LIST_INIT(&futex_list);
1148 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1149 linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit,
1151 linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec,
1153 linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor,
1154 linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY);
1155 linux_get_machine(&linux_kplatform);
1156 linux_szplatform = roundup(strlen(linux_kplatform) + 1,
1158 linux_osd_jail_register();
1159 stclohz = (stathz ? stathz : hz);
1161 printf("Linux ELF exec handler installed\n");
1163 printf("cannot insert Linux ELF brand handler\n");
1166 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1168 if (elf32_brand_inuse(*brandinfo))
1171 for (brandinfo = &linux_brandlist[0];
1172 *brandinfo != NULL; ++brandinfo)
1173 if (elf32_remove_brand_entry(*brandinfo) < 0)
1177 SET_FOREACH(lihp, linux_ioctl_handler_set)
1178 linux_ioctl_unregister_handler(*lihp);
1179 mtx_destroy(&futex_mtx);
1180 EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
1181 EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
1182 EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag);
1183 linux_osd_jail_deregister();
1185 printf("Linux ELF exec handler removed\n");
1187 printf("Could not deinstall ELF interpreter entry\n");
1190 return (EOPNOTSUPP);
1195 static moduledata_t linux_elf_mod = {
1201 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);