2 * Copyright (c) 2004 Tim J. Robbins
3 * Copyright (c) 2003 Peter Wemm
4 * Copyright (c) 2002 Doug Rabson
5 * Copyright (c) 1998-1999 Andrew Gallatin
6 * Copyright (c) 1994-1996 Søren Schmidt
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer
14 * in this position and unchanged.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. The name of the author may not be used to endorse or promote products
19 * derived from this software without specific prior written permission
21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
22 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
23 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
24 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
26 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
30 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 #include "opt_compat.h"
37 #ifndef COMPAT_FREEBSD32
38 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!"
41 #define __ELF_WORD_SIZE 32
43 #include <sys/param.h>
44 #include <sys/systm.h>
46 #include <sys/fcntl.h>
47 #include <sys/imgact.h>
48 #include <sys/imgact_elf.h>
49 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/module.h>
53 #include <sys/mutex.h>
55 #include <sys/resourcevar.h>
56 #include <sys/signalvar.h>
57 #include <sys/sysctl.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysent.h>
60 #include <sys/sysproto.h>
61 #include <sys/vnode.h>
62 #include <sys/eventhandler.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_map.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_param.h>
72 #include <machine/cpu.h>
73 #include <machine/md_var.h>
74 #include <machine/pcb.h>
75 #include <machine/specialreg.h>
77 #include <amd64/linux32/linux.h>
78 #include <amd64/linux32/linux32_proto.h>
79 #include <compat/linux/linux_emul.h>
80 #include <compat/linux/linux_futex.h>
81 #include <compat/linux/linux_ioctl.h>
82 #include <compat/linux/linux_mib.h>
83 #include <compat/linux/linux_misc.h>
84 #include <compat/linux/linux_signal.h>
85 #include <compat/linux/linux_util.h>
86 #include <compat/linux/linux_vdso.h>
88 MODULE_VERSION(linux, 1);
90 #define AUXARGS_ENTRY_32(pos, id, val) \
92 suword32(pos++, id); \
93 suword32(pos++, val); \
96 #if BYTE_ORDER == LITTLE_ENDIAN
97 #define SHELLMAGIC 0x2123 /* #! */
99 #define SHELLMAGIC 0x2321
103 * Allow the sendsig functions to use the ldebug() facility
104 * even though they are not syscalls themselves. Map them
105 * to syscall 0. This is slightly less bogus than using
108 #define LINUX32_SYS_linux_rt_sendsig 0
109 #define LINUX32_SYS_linux_sendsig 0
111 const char *linux_kplatform;
112 static int linux_szsigcode;
113 static vm_object_t linux_shared_page_obj;
114 static char *linux_shared_page_mapping;
115 extern char _binary_linux32_locore_o_start;
116 extern char _binary_linux32_locore_o_end;
118 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
120 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
122 static int elf_linux_fixup(register_t **stack_base,
123 struct image_params *iparams);
124 static register_t *linux_copyout_strings(struct image_params *imgp);
125 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
126 static void exec_linux_setregs(struct thread *td,
127 struct image_params *imgp, u_long stack);
128 static void linux32_fixlimit(struct rlimit *rl, int which);
129 static boolean_t linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
130 static void linux_vdso_install(void *param);
131 static void linux_vdso_deinstall(void *param);
134 * Linux syscalls return negative errno's, we do positive and map them
136 * FreeBSD: src/sys/sys/errno.h
137 * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h
138 * linux-2.6.17.8/include/asm-generic/errno.h
140 static int bsd_to_linux_errno[ELAST + 1] = {
141 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9,
142 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
143 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
144 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
145 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
146 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
147 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
148 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9,
149 -6, -6, -43, -42, -75,-125, -84, -95, -16, -74,
153 #define LINUX_T_UNKNOWN 255
154 static int _bsd_to_linux_trapcode[] = {
155 LINUX_T_UNKNOWN, /* 0 */
156 6, /* 1 T_PRIVINFLT */
157 LINUX_T_UNKNOWN, /* 2 */
159 LINUX_T_UNKNOWN, /* 4 */
160 LINUX_T_UNKNOWN, /* 5 */
161 16, /* 6 T_ARITHTRAP */
162 254, /* 7 T_ASTFLT */
163 LINUX_T_UNKNOWN, /* 8 */
164 13, /* 9 T_PROTFLT */
165 1, /* 10 T_TRCTRAP */
166 LINUX_T_UNKNOWN, /* 11 */
167 14, /* 12 T_PAGEFLT */
168 LINUX_T_UNKNOWN, /* 13 */
169 17, /* 14 T_ALIGNFLT */
170 LINUX_T_UNKNOWN, /* 15 */
171 LINUX_T_UNKNOWN, /* 16 */
172 LINUX_T_UNKNOWN, /* 17 */
178 8, /* 23 T_DOUBLEFLT */
179 9, /* 24 T_FPOPFLT */
180 10, /* 25 T_TSSFLT */
181 11, /* 26 T_SEGNPFLT */
182 12, /* 27 T_STKFLT */
184 19, /* 29 T_XMMFLT */
185 15 /* 30 T_RESERVED */
187 #define bsd_to_linux_trapcode(code) \
188 ((code)<nitems(_bsd_to_linux_trapcode)? \
189 _bsd_to_linux_trapcode[(code)]: \
192 struct linux32_ps_strings {
193 u_int32_t ps_argvstr; /* first of 0 or more argument strings */
194 u_int ps_nargvstr; /* the number of argument strings */
195 u_int32_t ps_envstr; /* first of 0 or more environment strings */
196 u_int ps_nenvstr; /* the number of environment strings */
199 LINUX_VDSO_SYM_INTPTR(linux32_sigcode);
200 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode);
201 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall);
202 LINUX_VDSO_SYM_CHAR(linux_platform);
205 * If FreeBSD & Linux have a difference of opinion about what a trap
206 * means, deal with it here.
211 translate_traps(int signal, int trap_code)
213 if (signal != SIGBUS)
227 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
232 struct linux32_ps_strings *arginfo;
235 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
237 KASSERT(curthread->td_proc == imgp->proc,
238 ("unsafe elf_linux_fixup(), should be curproc"));
239 base = (Elf32_Addr *)*stack_base;
240 args = (Elf32_Auxargs *)imgp->auxargs;
241 pos = base + (imgp->args->argc + imgp->args->envc + 2);
243 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
244 AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO_EHDR,
245 imgp->proc->p_sysent->sv_shared_page_base);
246 AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO, linux32_vsyscall);
247 AUXARGS_ENTRY_32(pos, LINUX_AT_HWCAP, cpu_feature);
250 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
251 * as it has appeared in the 2.4.0-rc7 first time.
252 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
253 * glibc falls back to the hard-coded CLK_TCK value when aux entry
255 * Also see linux_times() implementation.
257 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
258 AUXARGS_ENTRY_32(pos, LINUX_AT_CLKTCK, stclohz);
259 AUXARGS_ENTRY_32(pos, AT_PHDR, args->phdr);
260 AUXARGS_ENTRY_32(pos, AT_PHENT, args->phent);
261 AUXARGS_ENTRY_32(pos, AT_PHNUM, args->phnum);
262 AUXARGS_ENTRY_32(pos, AT_PAGESZ, args->pagesz);
263 AUXARGS_ENTRY_32(pos, AT_FLAGS, args->flags);
264 AUXARGS_ENTRY_32(pos, AT_ENTRY, args->entry);
265 AUXARGS_ENTRY_32(pos, AT_BASE, args->base);
266 AUXARGS_ENTRY_32(pos, LINUX_AT_SECURE, issetugid);
267 AUXARGS_ENTRY_32(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
268 AUXARGS_ENTRY_32(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
269 AUXARGS_ENTRY_32(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
270 AUXARGS_ENTRY_32(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
271 AUXARGS_ENTRY_32(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
272 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
273 if (imgp->execpathp != 0)
274 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
275 if (args->execfd != -1)
276 AUXARGS_ENTRY_32(pos, AT_EXECFD, args->execfd);
277 AUXARGS_ENTRY_32(pos, AT_NULL, 0);
279 free(imgp->auxargs, M_TEMP);
280 imgp->auxargs = NULL;
283 suword32(base, (uint32_t)imgp->args->argc);
284 *stack_base = (register_t *)base;
289 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
291 struct thread *td = curthread;
292 struct proc *p = td->td_proc;
294 struct trapframe *regs;
295 struct l_rt_sigframe *fp, frame;
300 sig = ksi->ksi_signo;
301 code = ksi->ksi_code;
302 PROC_LOCK_ASSERT(p, MA_OWNED);
304 mtx_assert(&psp->ps_mtx, MA_OWNED);
306 oonstack = sigonstack(regs->tf_rsp);
309 if (ldebug(rt_sendsig))
310 printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
311 catcher, sig, (void*)mask, code);
314 * Allocate space for the signal handler context.
316 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
317 SIGISMEMBER(psp->ps_sigonstack, sig)) {
318 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
319 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
321 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
322 mtx_unlock(&psp->ps_mtx);
325 * Build the argument list for the signal handler.
327 sig = bsd_to_linux_signal(sig);
329 bzero(&frame, sizeof(frame));
331 frame.sf_handler = PTROUT(catcher);
333 frame.sf_siginfo = PTROUT(&fp->sf_si);
334 frame.sf_ucontext = PTROUT(&fp->sf_sc);
336 /* Fill in POSIX parts */
337 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
340 * Build the signal context to be used by sigreturn
343 frame.sf_sc.uc_flags = 0; /* XXX ??? */
344 frame.sf_sc.uc_link = 0; /* XXX ??? */
346 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
347 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
348 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
349 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
352 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
354 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask;
355 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi;
356 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi;
357 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp;
358 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx;
359 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp;
360 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx;
361 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx;
362 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax;
363 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip;
364 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
365 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs;
366 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
367 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
368 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
369 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
370 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
371 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
372 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
373 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
374 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
377 if (ldebug(rt_sendsig))
378 printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"),
379 frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
380 td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
383 if (copyout(&frame, fp, sizeof(frame)) != 0) {
385 * Process has trashed its stack; give it an illegal
386 * instruction to halt it in its tracks.
389 if (ldebug(rt_sendsig))
390 printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
398 * Build context to run handler in.
400 regs->tf_rsp = PTROUT(fp);
401 regs->tf_rip = linux32_rt_sigcode;
402 regs->tf_rflags &= ~(PSL_T | PSL_D);
403 regs->tf_cs = _ucode32sel;
404 regs->tf_ss = _udatasel;
405 regs->tf_ds = _udatasel;
406 regs->tf_es = _udatasel;
407 regs->tf_fs = _ufssel;
408 regs->tf_gs = _ugssel;
409 regs->tf_flags = TF_HASSEGS;
410 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
412 mtx_lock(&psp->ps_mtx);
417 * Send an interrupt to process.
419 * Stack is set up to allow sigcode stored
420 * in u. to call routine, followed by kcall
421 * to sigreturn routine below. After sigreturn
422 * resets the signal mask, the stack, and the
423 * frame pointer, it returns to the user
427 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
429 struct thread *td = curthread;
430 struct proc *p = td->td_proc;
432 struct trapframe *regs;
433 struct l_sigframe *fp, frame;
438 sig = ksi->ksi_signo;
439 code = ksi->ksi_code;
440 PROC_LOCK_ASSERT(p, MA_OWNED);
442 mtx_assert(&psp->ps_mtx, MA_OWNED);
443 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
444 /* Signal handler installed with SA_SIGINFO. */
445 linux_rt_sendsig(catcher, ksi, mask);
450 oonstack = sigonstack(regs->tf_rsp);
454 printf(ARGS(sendsig, "%p, %d, %p, %u"),
455 catcher, sig, (void*)mask, code);
459 * Allocate space for the signal handler context.
461 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
462 SIGISMEMBER(psp->ps_sigonstack, sig)) {
463 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
464 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
466 fp = (struct l_sigframe *)regs->tf_rsp - 1;
467 mtx_unlock(&psp->ps_mtx);
471 * Build the argument list for the signal handler.
473 sig = bsd_to_linux_signal(sig);
475 bzero(&frame, sizeof(frame));
477 frame.sf_handler = PTROUT(catcher);
480 bsd_to_linux_sigset(mask, &lmask);
483 * Build the signal context to be used by sigreturn.
485 frame.sf_sc.sc_mask = lmask.__mask;
486 frame.sf_sc.sc_gs = regs->tf_gs;
487 frame.sf_sc.sc_fs = regs->tf_fs;
488 frame.sf_sc.sc_es = regs->tf_es;
489 frame.sf_sc.sc_ds = regs->tf_ds;
490 frame.sf_sc.sc_edi = regs->tf_rdi;
491 frame.sf_sc.sc_esi = regs->tf_rsi;
492 frame.sf_sc.sc_ebp = regs->tf_rbp;
493 frame.sf_sc.sc_ebx = regs->tf_rbx;
494 frame.sf_sc.sc_esp = regs->tf_rsp;
495 frame.sf_sc.sc_edx = regs->tf_rdx;
496 frame.sf_sc.sc_ecx = regs->tf_rcx;
497 frame.sf_sc.sc_eax = regs->tf_rax;
498 frame.sf_sc.sc_eip = regs->tf_rip;
499 frame.sf_sc.sc_cs = regs->tf_cs;
500 frame.sf_sc.sc_eflags = regs->tf_rflags;
501 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
502 frame.sf_sc.sc_ss = regs->tf_ss;
503 frame.sf_sc.sc_err = regs->tf_err;
504 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
505 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
507 frame.sf_extramask[0] = lmask.__mask;
509 if (copyout(&frame, fp, sizeof(frame)) != 0) {
511 * Process has trashed its stack; give it an illegal
512 * instruction to halt it in its tracks.
519 * Build context to run handler in.
521 regs->tf_rsp = PTROUT(fp);
522 regs->tf_rip = linux32_sigcode;
523 regs->tf_rflags &= ~(PSL_T | PSL_D);
524 regs->tf_cs = _ucode32sel;
525 regs->tf_ss = _udatasel;
526 regs->tf_ds = _udatasel;
527 regs->tf_es = _udatasel;
528 regs->tf_fs = _ufssel;
529 regs->tf_gs = _ugssel;
530 regs->tf_flags = TF_HASSEGS;
531 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
533 mtx_lock(&psp->ps_mtx);
537 * System call to cleanup state after a signal
538 * has been taken. Reset signal mask and
539 * stack state from context left by sendsig (above).
540 * Return to previous pc and psl as specified by
541 * context left by sendsig. Check carefully to
542 * make sure that the user has not modified the
543 * psl to gain improper privileges or to cause
547 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
549 struct l_sigframe frame;
550 struct trapframe *regs;
559 if (ldebug(sigreturn))
560 printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
563 * The trampoline code hands us the sigframe.
564 * It is unsafe to keep track of it ourselves, in the event that a
565 * program jumps out of a signal handler.
567 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
571 * Check for security violations.
573 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
574 eflags = frame.sf_sc.sc_eflags;
575 if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
579 * Don't allow users to load a valid privileged %cs. Let the
580 * hardware check for invalid selectors, excess privilege in
581 * other selectors, invalid %eip's and invalid %esp's.
583 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
584 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
585 ksiginfo_init_trap(&ksi);
586 ksi.ksi_signo = SIGBUS;
587 ksi.ksi_code = BUS_OBJERR;
588 ksi.ksi_trapno = T_PROTFLT;
589 ksi.ksi_addr = (void *)regs->tf_rip;
590 trapsignal(td, &ksi);
594 lmask.__mask = frame.sf_sc.sc_mask;
595 lmask.__mask = frame.sf_extramask[0];
596 linux_to_bsd_sigset(&lmask, &bmask);
597 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
600 * Restore signal context.
602 regs->tf_rdi = frame.sf_sc.sc_edi;
603 regs->tf_rsi = frame.sf_sc.sc_esi;
604 regs->tf_rbp = frame.sf_sc.sc_ebp;
605 regs->tf_rbx = frame.sf_sc.sc_ebx;
606 regs->tf_rdx = frame.sf_sc.sc_edx;
607 regs->tf_rcx = frame.sf_sc.sc_ecx;
608 regs->tf_rax = frame.sf_sc.sc_eax;
609 regs->tf_rip = frame.sf_sc.sc_eip;
610 regs->tf_cs = frame.sf_sc.sc_cs;
611 regs->tf_ds = frame.sf_sc.sc_ds;
612 regs->tf_es = frame.sf_sc.sc_es;
613 regs->tf_fs = frame.sf_sc.sc_fs;
614 regs->tf_gs = frame.sf_sc.sc_gs;
615 regs->tf_rflags = eflags;
616 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal;
617 regs->tf_ss = frame.sf_sc.sc_ss;
618 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
620 return (EJUSTRETURN);
624 * System call to cleanup state after a signal
625 * has been taken. Reset signal mask and
626 * stack state from context left by rt_sendsig (above).
627 * Return to previous pc and psl as specified by
628 * context left by sendsig. Check carefully to
629 * make sure that the user has not modified the
630 * psl to gain improper privileges or to cause
634 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
636 struct l_ucontext uc;
637 struct l_sigcontext *context;
641 struct trapframe *regs;
648 if (ldebug(rt_sigreturn))
649 printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
652 * The trampoline code hands us the ucontext.
653 * It is unsafe to keep track of it ourselves, in the event that a
654 * program jumps out of a signal handler.
656 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
659 context = &uc.uc_mcontext;
662 * Check for security violations.
664 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
665 eflags = context->sc_eflags;
666 if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
670 * Don't allow users to load a valid privileged %cs. Let the
671 * hardware check for invalid selectors, excess privilege in
672 * other selectors, invalid %eip's and invalid %esp's.
674 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
675 if (!CS_SECURE(context->sc_cs)) {
676 ksiginfo_init_trap(&ksi);
677 ksi.ksi_signo = SIGBUS;
678 ksi.ksi_code = BUS_OBJERR;
679 ksi.ksi_trapno = T_PROTFLT;
680 ksi.ksi_addr = (void *)regs->tf_rip;
681 trapsignal(td, &ksi);
685 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
686 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
689 * Restore signal context
691 regs->tf_gs = context->sc_gs;
692 regs->tf_fs = context->sc_fs;
693 regs->tf_es = context->sc_es;
694 regs->tf_ds = context->sc_ds;
695 regs->tf_rdi = context->sc_edi;
696 regs->tf_rsi = context->sc_esi;
697 regs->tf_rbp = context->sc_ebp;
698 regs->tf_rbx = context->sc_ebx;
699 regs->tf_rdx = context->sc_edx;
700 regs->tf_rcx = context->sc_ecx;
701 regs->tf_rax = context->sc_eax;
702 regs->tf_rip = context->sc_eip;
703 regs->tf_cs = context->sc_cs;
704 regs->tf_rflags = eflags;
705 regs->tf_rsp = context->sc_esp_at_signal;
706 regs->tf_ss = context->sc_ss;
707 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
710 * call sigaltstack & ignore results..
713 ss.ss_sp = PTRIN(lss->ss_sp);
714 ss.ss_size = lss->ss_size;
715 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
718 if (ldebug(rt_sigreturn))
719 printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"),
720 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
722 (void)kern_sigaltstack(td, &ss, NULL);
724 return (EJUSTRETURN);
728 linux32_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
731 struct trapframe *frame;
734 frame = td->td_frame;
736 sa->args[0] = frame->tf_rbx;
737 sa->args[1] = frame->tf_rcx;
738 sa->args[2] = frame->tf_rdx;
739 sa->args[3] = frame->tf_rsi;
740 sa->args[4] = frame->tf_rdi;
741 sa->args[5] = frame->tf_rbp; /* Unconfirmed */
742 sa->code = frame->tf_rax;
744 if (sa->code >= p->p_sysent->sv_size)
746 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
748 sa->callp = &p->p_sysent->sv_table[sa->code];
749 sa->narg = sa->callp->sy_narg;
751 td->td_retval[0] = 0;
752 td->td_retval[1] = frame->tf_rdx;
758 * If a linux binary is exec'ing something, try this image activator
759 * first. We override standard shell script execution in order to
760 * be able to modify the interpreter path. We only do this if a linux
761 * binary is doing the exec, so we do not create an EXEC module for it.
763 static int exec_linux_imgact_try(struct image_params *iparams);
766 exec_linux_imgact_try(struct image_params *imgp)
768 const char *head = (const char *)imgp->image_header;
773 * The interpreter for shell scripts run from a linux binary needs
774 * to be located in /compat/linux if possible in order to recursively
775 * maintain linux path emulation.
777 if (((const short *)head)[0] == SHELLMAGIC) {
779 * Run our normal shell image activator. If it succeeds attempt
780 * to use the alternate path for the interpreter. If an
781 * alternate * path is found, use our stringspace to store it.
783 if ((error = exec_shell_imgact(imgp)) == 0) {
784 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
785 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
788 imgp->args->fname_buf =
789 imgp->interpreter_name = rpath;
796 * Clear registers on exec
797 * XXX copied from ia32_signal.c.
800 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack)
802 struct trapframe *regs = td->td_frame;
803 struct pcb *pcb = td->td_pcb;
806 if (td->td_proc->p_md.md_ldt != NULL)
809 mtx_unlock(&dt_lock);
812 wrmsr(MSR_FSBASE, 0);
813 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */
817 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
819 bzero((char *)regs, sizeof(struct trapframe));
820 regs->tf_rip = imgp->entry_addr;
821 regs->tf_rsp = stack;
822 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
823 regs->tf_gs = _ugssel;
824 regs->tf_fs = _ufssel;
825 regs->tf_es = _udatasel;
826 regs->tf_ds = _udatasel;
827 regs->tf_ss = _udatasel;
828 regs->tf_flags = TF_HASSEGS;
829 regs->tf_cs = _ucode32sel;
830 regs->tf_rbx = imgp->ps_strings;
834 /* Do full restore on return so that we can change to a different %cs */
835 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
836 td->td_retval[1] = 0;
840 * XXX copied from ia32_sysvec.c.
843 linux_copyout_strings(struct image_params *imgp)
847 char *stringp, *destp;
848 u_int32_t *stack_base;
849 struct linux32_ps_strings *arginfo;
850 char canary[LINUX_AT_RANDOM_LEN];
854 * Calculate string base and vector table pointers.
856 if (imgp->execpath != NULL && imgp->auxargs != NULL)
857 execpath_len = strlen(imgp->execpath) + 1;
861 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
862 destp = (caddr_t)arginfo - SPARE_USRSPACE -
863 roundup(sizeof(canary), sizeof(char *)) -
864 roundup(execpath_len, sizeof(char *)) -
865 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
867 if (execpath_len != 0) {
868 imgp->execpathp = (uintptr_t)arginfo - execpath_len;
869 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
873 * Prepare the canary for SSP.
875 arc4rand(canary, sizeof(canary), 0);
876 imgp->canary = (uintptr_t)arginfo -
877 roundup(execpath_len, sizeof(char *)) -
878 roundup(sizeof(canary), sizeof(char *));
879 copyout(canary, (void *)imgp->canary, sizeof(canary));
882 * If we have a valid auxargs ptr, prepare some room
887 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
888 * lower compatibility.
890 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
891 (LINUX_AT_COUNT * 2);
893 * The '+ 2' is for the null pointers at the end of each of
894 * the arg and env vector sets,and imgp->auxarg_size is room
895 * for argument of Runtime loader.
897 vectp = (u_int32_t *) (destp - (imgp->args->argc +
898 imgp->args->envc + 2 + imgp->auxarg_size) *
903 * The '+ 2' is for the null pointers at the end of each of
904 * the arg and env vector sets
906 vectp = (u_int32_t *)(destp - (imgp->args->argc +
907 imgp->args->envc + 2) * sizeof(u_int32_t));
910 * vectp also becomes our initial stack base
914 stringp = imgp->args->begin_argv;
915 argc = imgp->args->argc;
916 envc = imgp->args->envc;
918 * Copy out strings - arguments and environment.
920 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
923 * Fill in "ps_strings" struct for ps, w, etc.
925 suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp);
926 suword32(&arginfo->ps_nargvstr, argc);
929 * Fill in argument portion of vector table.
931 for (; argc > 0; --argc) {
932 suword32(vectp++, (uint32_t)(intptr_t)destp);
933 while (*stringp++ != 0)
938 /* a null vector table pointer separates the argp's from the envp's */
939 suword32(vectp++, 0);
941 suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp);
942 suword32(&arginfo->ps_nenvstr, envc);
945 * Fill in environment portion of vector table.
947 for (; envc > 0; --envc) {
948 suword32(vectp++, (uint32_t)(intptr_t)destp);
949 while (*stringp++ != 0)
954 /* end of vector table is a null pointer */
957 return ((register_t *)stack_base);
960 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0,
961 "32-bit Linux emulation");
963 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ;
964 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
965 &linux32_maxdsiz, 0, "");
966 static u_long linux32_maxssiz = LINUX32_MAXSSIZ;
967 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
968 &linux32_maxssiz, 0, "");
969 static u_long linux32_maxvmem = LINUX32_MAXVMEM;
970 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
971 &linux32_maxvmem, 0, "");
974 SYSCTL_PROC(_compat_linux32, OID_AUTO, debug,
975 CTLTYPE_STRING | CTLFLAG_RW,
976 0, 0, linux_sysctl_debug, "A",
977 "Linux debugging control");
981 linux32_fixlimit(struct rlimit *rl, int which)
986 if (linux32_maxdsiz != 0) {
987 if (rl->rlim_cur > linux32_maxdsiz)
988 rl->rlim_cur = linux32_maxdsiz;
989 if (rl->rlim_max > linux32_maxdsiz)
990 rl->rlim_max = linux32_maxdsiz;
994 if (linux32_maxssiz != 0) {
995 if (rl->rlim_cur > linux32_maxssiz)
996 rl->rlim_cur = linux32_maxssiz;
997 if (rl->rlim_max > linux32_maxssiz)
998 rl->rlim_max = linux32_maxssiz;
1002 if (linux32_maxvmem != 0) {
1003 if (rl->rlim_cur > linux32_maxvmem)
1004 rl->rlim_cur = linux32_maxvmem;
1005 if (rl->rlim_max > linux32_maxvmem)
1006 rl->rlim_max = linux32_maxvmem;
1012 struct sysentvec elf_linux_sysvec = {
1013 .sv_size = LINUX32_SYS_MAXSYSCALL,
1014 .sv_table = linux32_sysent,
1016 .sv_errsize = ELAST + 1,
1017 .sv_errtbl = bsd_to_linux_errno,
1018 .sv_transtrap = translate_traps,
1019 .sv_fixup = elf_linux_fixup,
1020 .sv_sendsig = linux_sendsig,
1021 .sv_sigcode = &_binary_linux32_locore_o_start,
1022 .sv_szsigcode = &linux_szsigcode,
1023 .sv_name = "Linux ELF32",
1024 .sv_coredump = elf32_coredump,
1025 .sv_imgact_try = exec_linux_imgact_try,
1026 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
1027 .sv_pagesize = PAGE_SIZE,
1028 .sv_minuser = VM_MIN_ADDRESS,
1029 .sv_maxuser = LINUX32_MAXUSER,
1030 .sv_usrstack = LINUX32_USRSTACK,
1031 .sv_psstrings = LINUX32_PS_STRINGS,
1032 .sv_stackprot = VM_PROT_ALL,
1033 .sv_copyout_strings = linux_copyout_strings,
1034 .sv_setregs = exec_linux_setregs,
1035 .sv_fixlimit = linux32_fixlimit,
1036 .sv_maxssiz = &linux32_maxssiz,
1037 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP,
1038 .sv_set_syscall_retval = cpu_set_syscall_retval,
1039 .sv_fetch_syscall_args = linux32_fetch_syscall_args,
1040 .sv_syscallnames = NULL,
1041 .sv_shared_page_base = LINUX32_SHAREDPAGE,
1042 .sv_shared_page_len = PAGE_SIZE,
1043 .sv_schedtail = linux_schedtail,
1044 .sv_thread_detach = linux_thread_detach,
1049 linux_vdso_install(void *param)
1052 linux_szsigcode = (&_binary_linux32_locore_o_end -
1053 &_binary_linux32_locore_o_start);
1055 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
1056 panic("Linux invalid vdso size\n");
1058 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
1060 linux_shared_page_obj = __elfN(linux_shared_page_init)
1061 (&linux_shared_page_mapping);
1063 __elfN(linux_vdso_reloc)(&elf_linux_sysvec, LINUX32_SHAREDPAGE);
1065 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
1067 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
1069 linux_kplatform = linux_shared_page_mapping +
1070 (linux_platform - (caddr_t)LINUX32_SHAREDPAGE);
1072 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
1073 (sysinit_cfunc_t)linux_vdso_install, NULL);
1076 linux_vdso_deinstall(void *param)
1079 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
1081 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1082 (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
1084 static char GNU_ABI_VENDOR[] = "GNU";
1085 static int GNULINUX_ABI_DESC = 0;
1088 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
1090 const Elf32_Word *desc;
1093 p = (uintptr_t)(note + 1);
1094 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1096 desc = (const Elf32_Word *)p;
1097 if (desc[0] != GNULINUX_ABI_DESC)
1101 * For linux we encode osrel as follows (see linux_mib.c):
1102 * VVVMMMIII (version, major, minor), see linux_mib.c.
1104 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1109 static Elf_Brandnote linux32_brandnote = {
1110 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
1111 .hdr.n_descsz = 16, /* XXX at least 16 */
1113 .vendor = GNU_ABI_VENDOR,
1114 .flags = BN_TRANSLATE_OSREL,
1115 .trans_osrel = linux32_trans_osrel
1118 static Elf32_Brandinfo linux_brand = {
1119 .brand = ELFOSABI_LINUX,
1121 .compat_3_brand = "Linux",
1122 .emul_path = "/compat/linux",
1123 .interp_path = "/lib/ld-linux.so.1",
1124 .sysvec = &elf_linux_sysvec,
1125 .interp_newpath = NULL,
1126 .brand_note = &linux32_brandnote,
1127 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1130 static Elf32_Brandinfo linux_glibc2brand = {
1131 .brand = ELFOSABI_LINUX,
1133 .compat_3_brand = "Linux",
1134 .emul_path = "/compat/linux",
1135 .interp_path = "/lib/ld-linux.so.2",
1136 .sysvec = &elf_linux_sysvec,
1137 .interp_newpath = NULL,
1138 .brand_note = &linux32_brandnote,
1139 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1142 Elf32_Brandinfo *linux_brandlist[] = {
1149 linux_elf_modevent(module_t mod, int type, void *data)
1151 Elf32_Brandinfo **brandinfo;
1153 struct linux_ioctl_handler **lihp;
1159 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1161 if (elf32_insert_brand_entry(*brandinfo) < 0)
1164 SET_FOREACH(lihp, linux_ioctl_handler_set)
1165 linux_ioctl_register_handler(*lihp);
1166 LIST_INIT(&futex_list);
1167 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1168 stclohz = (stathz ? stathz : hz);
1170 printf("Linux ELF exec handler installed\n");
1172 printf("cannot insert Linux ELF brand handler\n");
1175 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1177 if (elf32_brand_inuse(*brandinfo))
1180 for (brandinfo = &linux_brandlist[0];
1181 *brandinfo != NULL; ++brandinfo)
1182 if (elf32_remove_brand_entry(*brandinfo) < 0)
1186 SET_FOREACH(lihp, linux_ioctl_handler_set)
1187 linux_ioctl_unregister_handler(*lihp);
1188 mtx_destroy(&futex_mtx);
1190 printf("Linux ELF exec handler removed\n");
1192 printf("Could not deinstall ELF interpreter entry\n");
1195 return (EOPNOTSUPP);
1200 static moduledata_t linux_elf_mod = {
1206 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1207 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1208 FEATURE(linux, "Linux 32bit support");