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>
87 MODULE_VERSION(linux, 1);
89 MALLOC_DEFINE(M_LINUX, "linux", "Linux mode structures");
91 #define AUXARGS_ENTRY_32(pos, id, val) \
93 suword32(pos++, id); \
94 suword32(pos++, val); \
97 #if BYTE_ORDER == LITTLE_ENDIAN
98 #define SHELLMAGIC 0x2123 /* #! */
100 #define SHELLMAGIC 0x2321
104 * Allow the sendsig functions to use the ldebug() facility
105 * even though they are not syscalls themselves. Map them
106 * to syscall 0. This is slightly less bogus than using
109 #define LINUX_SYS_linux_rt_sendsig 0
110 #define LINUX_SYS_linux_sendsig 0
112 const char *linux_platform = "i686";
113 static int linux_szplatform;
114 extern char linux_sigcode[];
115 extern int linux_szsigcode;
117 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
119 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
120 SET_DECLARE(linux_device_handler_set, struct linux_device_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);
131 static eventhandler_tag linux_exit_tag;
132 static eventhandler_tag linux_exec_tag;
135 * Linux syscalls return negative errno's, we do positive and map them
137 * FreeBSD: src/sys/sys/errno.h
138 * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h
139 * linux-2.6.17.8/include/asm-generic/errno.h
141 static int bsd_to_linux_errno[ELAST + 1] = {
142 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9,
143 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19,
144 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29,
145 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89,
146 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99,
147 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109,
148 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122,
149 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9,
150 -6, -6, -43, -42, -75,-125, -84, -95, -16, -74,
154 int bsd_to_linux_signal[LINUX_SIGTBLSZ] = {
155 LINUX_SIGHUP, LINUX_SIGINT, LINUX_SIGQUIT, LINUX_SIGILL,
156 LINUX_SIGTRAP, LINUX_SIGABRT, 0, LINUX_SIGFPE,
157 LINUX_SIGKILL, LINUX_SIGBUS, LINUX_SIGSEGV, LINUX_SIGSYS,
158 LINUX_SIGPIPE, LINUX_SIGALRM, LINUX_SIGTERM, LINUX_SIGURG,
159 LINUX_SIGSTOP, LINUX_SIGTSTP, LINUX_SIGCONT, LINUX_SIGCHLD,
160 LINUX_SIGTTIN, LINUX_SIGTTOU, LINUX_SIGIO, LINUX_SIGXCPU,
161 LINUX_SIGXFSZ, LINUX_SIGVTALRM, LINUX_SIGPROF, LINUX_SIGWINCH,
162 0, LINUX_SIGUSR1, LINUX_SIGUSR2
165 int linux_to_bsd_signal[LINUX_SIGTBLSZ] = {
166 SIGHUP, SIGINT, SIGQUIT, SIGILL,
167 SIGTRAP, SIGABRT, SIGBUS, SIGFPE,
168 SIGKILL, SIGUSR1, SIGSEGV, SIGUSR2,
169 SIGPIPE, SIGALRM, SIGTERM, SIGBUS,
170 SIGCHLD, SIGCONT, SIGSTOP, SIGTSTP,
171 SIGTTIN, SIGTTOU, SIGURG, SIGXCPU,
172 SIGXFSZ, SIGVTALRM, SIGPROF, SIGWINCH,
173 SIGIO, SIGURG, SIGSYS
176 #define LINUX_T_UNKNOWN 255
177 static int _bsd_to_linux_trapcode[] = {
178 LINUX_T_UNKNOWN, /* 0 */
179 6, /* 1 T_PRIVINFLT */
180 LINUX_T_UNKNOWN, /* 2 */
182 LINUX_T_UNKNOWN, /* 4 */
183 LINUX_T_UNKNOWN, /* 5 */
184 16, /* 6 T_ARITHTRAP */
185 254, /* 7 T_ASTFLT */
186 LINUX_T_UNKNOWN, /* 8 */
187 13, /* 9 T_PROTFLT */
188 1, /* 10 T_TRCTRAP */
189 LINUX_T_UNKNOWN, /* 11 */
190 14, /* 12 T_PAGEFLT */
191 LINUX_T_UNKNOWN, /* 13 */
192 17, /* 14 T_ALIGNFLT */
193 LINUX_T_UNKNOWN, /* 15 */
194 LINUX_T_UNKNOWN, /* 16 */
195 LINUX_T_UNKNOWN, /* 17 */
201 8, /* 23 T_DOUBLEFLT */
202 9, /* 24 T_FPOPFLT */
203 10, /* 25 T_TSSFLT */
204 11, /* 26 T_SEGNPFLT */
205 12, /* 27 T_STKFLT */
207 19, /* 29 T_XMMFLT */
208 15 /* 30 T_RESERVED */
210 #define bsd_to_linux_trapcode(code) \
211 ((code)<sizeof(_bsd_to_linux_trapcode)/sizeof(*_bsd_to_linux_trapcode)? \
212 _bsd_to_linux_trapcode[(code)]: \
215 struct linux32_ps_strings {
216 u_int32_t ps_argvstr; /* first of 0 or more argument strings */
217 u_int ps_nargvstr; /* the number of argument strings */
218 u_int32_t ps_envstr; /* first of 0 or more environment strings */
219 u_int ps_nenvstr; /* the number of environment strings */
223 * If FreeBSD & Linux have a difference of opinion about what a trap
224 * means, deal with it here.
229 translate_traps(int signal, int trap_code)
231 if (signal != SIGBUS)
245 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
249 Elf32_Addr *pos, *uplatform;
250 struct linux32_ps_strings *arginfo;
252 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
253 uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform);
255 KASSERT(curthread->td_proc == imgp->proc,
256 ("unsafe elf_linux_fixup(), should be curproc"));
257 base = (Elf32_Addr *)*stack_base;
258 args = (Elf32_Auxargs *)imgp->auxargs;
259 pos = base + (imgp->args->argc + imgp->args->envc + 2);
261 AUXARGS_ENTRY_32(pos, LINUX_AT_HWCAP, cpu_feature);
264 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
265 * as it has appeared in the 2.4.0-rc7 first time.
266 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
267 * glibc falls back to the hard-coded CLK_TCK value when aux entry
269 * Also see linux_times() implementation.
271 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
272 AUXARGS_ENTRY_32(pos, LINUX_AT_CLKTCK, stclohz);
273 AUXARGS_ENTRY_32(pos, AT_PHDR, args->phdr);
274 AUXARGS_ENTRY_32(pos, AT_PHENT, args->phent);
275 AUXARGS_ENTRY_32(pos, AT_PHNUM, args->phnum);
276 AUXARGS_ENTRY_32(pos, AT_PAGESZ, args->pagesz);
277 AUXARGS_ENTRY_32(pos, AT_FLAGS, args->flags);
278 AUXARGS_ENTRY_32(pos, AT_ENTRY, args->entry);
279 AUXARGS_ENTRY_32(pos, AT_BASE, args->base);
280 AUXARGS_ENTRY_32(pos, LINUX_AT_SECURE, 0);
281 AUXARGS_ENTRY_32(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
282 AUXARGS_ENTRY_32(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
283 AUXARGS_ENTRY_32(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
284 AUXARGS_ENTRY_32(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
285 AUXARGS_ENTRY_32(pos, LINUX_AT_PLATFORM, PTROUT(uplatform));
286 if (args->execfd != -1)
287 AUXARGS_ENTRY_32(pos, AT_EXECFD, args->execfd);
288 AUXARGS_ENTRY_32(pos, AT_NULL, 0);
290 free(imgp->auxargs, M_TEMP);
291 imgp->auxargs = NULL;
294 suword32(base, (uint32_t)imgp->args->argc);
295 *stack_base = (register_t *)base;
299 extern unsigned long linux_sznonrtsigcode;
302 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
304 struct thread *td = curthread;
305 struct proc *p = td->td_proc;
307 struct trapframe *regs;
308 struct l_rt_sigframe *fp, frame;
313 sig = ksi->ksi_signo;
314 code = ksi->ksi_code;
315 PROC_LOCK_ASSERT(p, MA_OWNED);
317 mtx_assert(&psp->ps_mtx, MA_OWNED);
319 oonstack = sigonstack(regs->tf_rsp);
322 if (ldebug(rt_sendsig))
323 printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
324 catcher, sig, (void*)mask, code);
327 * Allocate space for the signal handler context.
329 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
330 SIGISMEMBER(psp->ps_sigonstack, sig)) {
331 fp = (struct l_rt_sigframe *)(td->td_sigstk.ss_sp +
332 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
334 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
335 mtx_unlock(&psp->ps_mtx);
338 * Build the argument list for the signal handler.
340 if (p->p_sysent->sv_sigtbl)
341 if (sig <= p->p_sysent->sv_sigsize)
342 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
344 bzero(&frame, sizeof(frame));
346 frame.sf_handler = PTROUT(catcher);
348 frame.sf_siginfo = PTROUT(&fp->sf_si);
349 frame.sf_ucontext = PTROUT(&fp->sf_sc);
351 /* Fill in POSIX parts */
352 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
355 * Build the signal context to be used by sigreturn.
357 frame.sf_sc.uc_flags = 0; /* XXX ??? */
358 frame.sf_sc.uc_link = 0; /* XXX ??? */
360 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
361 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
362 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
363 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
366 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
368 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__bits[0];
369 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi;
370 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi;
371 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp;
372 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx;
373 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx;
374 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx;
375 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax;
376 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip;
377 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
378 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs;
379 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
380 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
381 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
382 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
383 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
384 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
385 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
386 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
387 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
390 if (ldebug(rt_sendsig))
391 printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"),
392 frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
393 td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
396 if (copyout(&frame, fp, sizeof(frame)) != 0) {
398 * Process has trashed its stack; give it an illegal
399 * instruction to halt it in its tracks.
402 if (ldebug(rt_sendsig))
403 printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
411 * Build context to run handler in.
413 regs->tf_rsp = PTROUT(fp);
414 regs->tf_rip = p->p_sysent->sv_sigcode_base + linux_sznonrtsigcode;
415 regs->tf_rflags &= ~(PSL_T | PSL_D);
416 regs->tf_cs = _ucode32sel;
417 regs->tf_ss = _udatasel;
418 regs->tf_ds = _udatasel;
419 regs->tf_es = _udatasel;
420 regs->tf_fs = _ufssel;
421 regs->tf_gs = _ugssel;
422 regs->tf_flags = TF_HASSEGS;
423 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
425 mtx_lock(&psp->ps_mtx);
430 * Send an interrupt to process.
432 * Stack is set up to allow sigcode stored
433 * in u. to call routine, followed by kcall
434 * to sigreturn routine below. After sigreturn
435 * resets the signal mask, the stack, and the
436 * frame pointer, it returns to the user
440 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
442 struct thread *td = curthread;
443 struct proc *p = td->td_proc;
445 struct trapframe *regs;
446 struct l_sigframe *fp, frame;
451 sig = ksi->ksi_signo;
452 code = ksi->ksi_code;
453 PROC_LOCK_ASSERT(p, MA_OWNED);
455 mtx_assert(&psp->ps_mtx, MA_OWNED);
456 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
457 /* Signal handler installed with SA_SIGINFO. */
458 linux_rt_sendsig(catcher, ksi, mask);
463 oonstack = sigonstack(regs->tf_rsp);
467 printf(ARGS(sendsig, "%p, %d, %p, %u"),
468 catcher, sig, (void*)mask, code);
472 * Allocate space for the signal handler context.
474 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
475 SIGISMEMBER(psp->ps_sigonstack, sig)) {
476 fp = (struct l_sigframe *)(td->td_sigstk.ss_sp +
477 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
479 fp = (struct l_sigframe *)regs->tf_rsp - 1;
480 mtx_unlock(&psp->ps_mtx);
484 * Build the argument list for the signal handler.
486 if (p->p_sysent->sv_sigtbl)
487 if (sig <= p->p_sysent->sv_sigsize)
488 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
490 bzero(&frame, sizeof(frame));
492 frame.sf_handler = PTROUT(catcher);
495 bsd_to_linux_sigset(mask, &lmask);
498 * Build the signal context to be used by sigreturn.
500 frame.sf_sc.sc_mask = lmask.__bits[0];
501 frame.sf_sc.sc_gs = regs->tf_gs;
502 frame.sf_sc.sc_fs = regs->tf_fs;
503 frame.sf_sc.sc_es = regs->tf_es;
504 frame.sf_sc.sc_ds = regs->tf_ds;
505 frame.sf_sc.sc_edi = regs->tf_rdi;
506 frame.sf_sc.sc_esi = regs->tf_rsi;
507 frame.sf_sc.sc_ebp = regs->tf_rbp;
508 frame.sf_sc.sc_ebx = regs->tf_rbx;
509 frame.sf_sc.sc_edx = regs->tf_rdx;
510 frame.sf_sc.sc_ecx = regs->tf_rcx;
511 frame.sf_sc.sc_eax = regs->tf_rax;
512 frame.sf_sc.sc_eip = regs->tf_rip;
513 frame.sf_sc.sc_cs = regs->tf_cs;
514 frame.sf_sc.sc_eflags = regs->tf_rflags;
515 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
516 frame.sf_sc.sc_ss = regs->tf_ss;
517 frame.sf_sc.sc_err = regs->tf_err;
518 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
519 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
521 for (i = 0; i < (LINUX_NSIG_WORDS-1); i++)
522 frame.sf_extramask[i] = lmask.__bits[i+1];
524 if (copyout(&frame, fp, sizeof(frame)) != 0) {
526 * Process has trashed its stack; give it an illegal
527 * instruction to halt it in its tracks.
534 * Build context to run handler in.
536 regs->tf_rsp = PTROUT(fp);
537 regs->tf_rip = p->p_sysent->sv_sigcode_base;
538 regs->tf_rflags &= ~(PSL_T | PSL_D);
539 regs->tf_cs = _ucode32sel;
540 regs->tf_ss = _udatasel;
541 regs->tf_ds = _udatasel;
542 regs->tf_es = _udatasel;
543 regs->tf_fs = _ufssel;
544 regs->tf_gs = _ugssel;
545 regs->tf_flags = TF_HASSEGS;
546 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
548 mtx_lock(&psp->ps_mtx);
552 * System call to cleanup state after a signal
553 * has been taken. Reset signal mask and
554 * stack state from context left by sendsig (above).
555 * Return to previous pc and psl as specified by
556 * context left by sendsig. Check carefully to
557 * make sure that the user has not modified the
558 * psl to gain improper privileges or to cause
562 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
564 struct l_sigframe frame;
565 struct trapframe *regs;
574 if (ldebug(sigreturn))
575 printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
578 * The trampoline code hands us the sigframe.
579 * It is unsafe to keep track of it ourselves, in the event that a
580 * program jumps out of a signal handler.
582 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
586 * Check for security violations.
588 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
589 eflags = frame.sf_sc.sc_eflags;
591 * XXX do allow users to change the privileged flag PSL_RF. The
592 * cpu sets PSL_RF in tf_eflags for faults. Debuggers should
593 * sometimes set it there too. tf_eflags is kept in the signal
594 * context during signal handling and there is no other place
595 * to remember it, so the PSL_RF bit may be corrupted by the
596 * signal handler without us knowing. Corruption of the PSL_RF
597 * bit at worst causes one more or one less debugger trap, so
598 * allowing it is fairly harmless.
600 if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF))
604 * Don't allow users to load a valid privileged %cs. Let the
605 * hardware check for invalid selectors, excess privilege in
606 * other selectors, invalid %eip's and invalid %esp's.
608 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
609 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
610 ksiginfo_init_trap(&ksi);
611 ksi.ksi_signo = SIGBUS;
612 ksi.ksi_code = BUS_OBJERR;
613 ksi.ksi_trapno = T_PROTFLT;
614 ksi.ksi_addr = (void *)regs->tf_rip;
615 trapsignal(td, &ksi);
619 lmask.__bits[0] = frame.sf_sc.sc_mask;
620 for (i = 0; i < (LINUX_NSIG_WORDS-1); i++)
621 lmask.__bits[i+1] = frame.sf_extramask[i];
622 linux_to_bsd_sigset(&lmask, &bmask);
623 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
626 * Restore signal context.
628 regs->tf_rdi = frame.sf_sc.sc_edi;
629 regs->tf_rsi = frame.sf_sc.sc_esi;
630 regs->tf_rbp = frame.sf_sc.sc_ebp;
631 regs->tf_rbx = frame.sf_sc.sc_ebx;
632 regs->tf_rdx = frame.sf_sc.sc_edx;
633 regs->tf_rcx = frame.sf_sc.sc_ecx;
634 regs->tf_rax = frame.sf_sc.sc_eax;
635 regs->tf_rip = frame.sf_sc.sc_eip;
636 regs->tf_cs = frame.sf_sc.sc_cs;
637 regs->tf_ds = frame.sf_sc.sc_ds;
638 regs->tf_es = frame.sf_sc.sc_es;
639 regs->tf_fs = frame.sf_sc.sc_fs;
640 regs->tf_gs = frame.sf_sc.sc_gs;
641 regs->tf_rflags = eflags;
642 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal;
643 regs->tf_ss = frame.sf_sc.sc_ss;
644 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
646 return (EJUSTRETURN);
650 * System call to cleanup state after a signal
651 * has been taken. Reset signal mask and
652 * stack state from context left by rt_sendsig (above).
653 * Return to previous pc and psl as specified by
654 * context left by sendsig. Check carefully to
655 * make sure that the user has not modified the
656 * psl to gain improper privileges or to cause
660 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
662 struct l_ucontext uc;
663 struct l_sigcontext *context;
667 struct trapframe *regs;
674 if (ldebug(rt_sigreturn))
675 printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
678 * The trampoline code hands us the ucontext.
679 * It is unsafe to keep track of it ourselves, in the event that a
680 * program jumps out of a signal handler.
682 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
685 context = &uc.uc_mcontext;
688 * Check for security violations.
690 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
691 eflags = context->sc_eflags;
693 * XXX do allow users to change the privileged flag PSL_RF. The
694 * cpu sets PSL_RF in tf_eflags for faults. Debuggers should
695 * sometimes set it there too. tf_eflags is kept in the signal
696 * context during signal handling and there is no other place
697 * to remember it, so the PSL_RF bit may be corrupted by the
698 * signal handler without us knowing. Corruption of the PSL_RF
699 * bit at worst causes one more or one less debugger trap, so
700 * allowing it is fairly harmless.
702 if (!EFLAGS_SECURE(eflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF))
706 * Don't allow users to load a valid privileged %cs. Let the
707 * hardware check for invalid selectors, excess privilege in
708 * other selectors, invalid %eip's and invalid %esp's.
710 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
711 if (!CS_SECURE(context->sc_cs)) {
712 ksiginfo_init_trap(&ksi);
713 ksi.ksi_signo = SIGBUS;
714 ksi.ksi_code = BUS_OBJERR;
715 ksi.ksi_trapno = T_PROTFLT;
716 ksi.ksi_addr = (void *)regs->tf_rip;
717 trapsignal(td, &ksi);
721 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
722 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
725 * Restore signal context
727 regs->tf_gs = context->sc_gs;
728 regs->tf_fs = context->sc_fs;
729 regs->tf_es = context->sc_es;
730 regs->tf_ds = context->sc_ds;
731 regs->tf_rdi = context->sc_edi;
732 regs->tf_rsi = context->sc_esi;
733 regs->tf_rbp = context->sc_ebp;
734 regs->tf_rbx = context->sc_ebx;
735 regs->tf_rdx = context->sc_edx;
736 regs->tf_rcx = context->sc_ecx;
737 regs->tf_rax = context->sc_eax;
738 regs->tf_rip = context->sc_eip;
739 regs->tf_cs = context->sc_cs;
740 regs->tf_rflags = eflags;
741 regs->tf_rsp = context->sc_esp_at_signal;
742 regs->tf_ss = context->sc_ss;
743 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
746 * call sigaltstack & ignore results..
749 ss.ss_sp = PTRIN(lss->ss_sp);
750 ss.ss_size = lss->ss_size;
751 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
754 if (ldebug(rt_sigreturn))
755 printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"),
756 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
758 (void)kern_sigaltstack(td, &ss, NULL);
760 return (EJUSTRETURN);
764 linux32_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
767 struct trapframe *frame;
770 frame = td->td_frame;
772 sa->args[0] = frame->tf_rbx;
773 sa->args[1] = frame->tf_rcx;
774 sa->args[2] = frame->tf_rdx;
775 sa->args[3] = frame->tf_rsi;
776 sa->args[4] = frame->tf_rdi;
777 sa->args[5] = frame->tf_rbp; /* Unconfirmed */
778 sa->code = frame->tf_rax;
780 if (sa->code >= p->p_sysent->sv_size)
781 sa->callp = &p->p_sysent->sv_table[0];
783 sa->callp = &p->p_sysent->sv_table[sa->code];
784 sa->narg = sa->callp->sy_narg;
786 td->td_retval[0] = 0;
787 td->td_retval[1] = frame->tf_rdx;
793 * If a linux binary is exec'ing something, try this image activator
794 * first. We override standard shell script execution in order to
795 * be able to modify the interpreter path. We only do this if a linux
796 * binary is doing the exec, so we do not create an EXEC module for it.
798 static int exec_linux_imgact_try(struct image_params *iparams);
801 exec_linux_imgact_try(struct image_params *imgp)
803 const char *head = (const char *)imgp->image_header;
808 * The interpreter for shell scripts run from a linux binary needs
809 * to be located in /compat/linux if possible in order to recursively
810 * maintain linux path emulation.
812 if (((const short *)head)[0] == SHELLMAGIC) {
814 * Run our normal shell image activator. If it succeeds attempt
815 * to use the alternate path for the interpreter. If an
816 * alternate * path is found, use our stringspace to store it.
818 if ((error = exec_shell_imgact(imgp)) == 0) {
819 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
820 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
823 imgp->args->fname_buf =
824 imgp->interpreter_name = rpath;
831 * Clear registers on exec
832 * XXX copied from ia32_signal.c.
835 exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack)
837 struct trapframe *regs = td->td_frame;
838 struct pcb *pcb = td->td_pcb;
841 if (td->td_proc->p_md.md_ldt != NULL)
844 mtx_unlock(&dt_lock);
847 wrmsr(MSR_FSBASE, 0);
848 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */
852 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
854 bzero((char *)regs, sizeof(struct trapframe));
855 regs->tf_rip = imgp->entry_addr;
856 regs->tf_rsp = stack;
857 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
858 regs->tf_gs = _ugssel;
859 regs->tf_fs = _ufssel;
860 regs->tf_es = _udatasel;
861 regs->tf_ds = _udatasel;
862 regs->tf_ss = _udatasel;
863 regs->tf_flags = TF_HASSEGS;
864 regs->tf_cs = _ucode32sel;
865 regs->tf_rbx = imgp->ps_strings;
869 /* Do full restore on return so that we can change to a different %cs */
870 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
871 td->td_retval[1] = 0;
875 * XXX copied from ia32_sysvec.c.
878 linux_copyout_strings(struct image_params *imgp)
882 char *stringp, *destp;
883 u_int32_t *stack_base;
884 struct linux32_ps_strings *arginfo;
887 * Calculate string base and vector table pointers.
888 * Also deal with signal trampoline code for this exec type.
890 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
891 destp = (caddr_t)arginfo - SPARE_USRSPACE - linux_szplatform -
892 roundup((ARG_MAX - imgp->args->stringspace),
896 * Install LINUX_PLATFORM
898 copyout(linux_platform, ((caddr_t)arginfo - linux_szplatform),
902 * If we have a valid auxargs ptr, prepare some room
907 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
908 * lower compatibility.
910 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
911 (LINUX_AT_COUNT * 2);
913 * The '+ 2' is for the null pointers at the end of each of
914 * the arg and env vector sets,and imgp->auxarg_size is room
915 * for argument of Runtime loader.
917 vectp = (u_int32_t *) (destp - (imgp->args->argc +
918 imgp->args->envc + 2 + imgp->auxarg_size) *
923 * The '+ 2' is for the null pointers at the end of each of
924 * the arg and env vector sets
926 vectp = (u_int32_t *)(destp - (imgp->args->argc +
927 imgp->args->envc + 2) * sizeof(u_int32_t));
930 * vectp also becomes our initial stack base
934 stringp = imgp->args->begin_argv;
935 argc = imgp->args->argc;
936 envc = imgp->args->envc;
938 * Copy out strings - arguments and environment.
940 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
943 * Fill in "ps_strings" struct for ps, w, etc.
945 suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp);
946 suword32(&arginfo->ps_nargvstr, argc);
949 * Fill in argument portion of vector table.
951 for (; argc > 0; --argc) {
952 suword32(vectp++, (uint32_t)(intptr_t)destp);
953 while (*stringp++ != 0)
958 /* a null vector table pointer separates the argp's from the envp's */
959 suword32(vectp++, 0);
961 suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp);
962 suword32(&arginfo->ps_nenvstr, envc);
965 * Fill in environment portion of vector table.
967 for (; envc > 0; --envc) {
968 suword32(vectp++, (uint32_t)(intptr_t)destp);
969 while (*stringp++ != 0)
974 /* end of vector table is a null pointer */
977 return ((register_t *)stack_base);
980 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0,
981 "32-bit Linux emulation");
983 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ;
984 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
985 &linux32_maxdsiz, 0, "");
986 static u_long linux32_maxssiz = LINUX32_MAXSSIZ;
987 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
988 &linux32_maxssiz, 0, "");
989 static u_long linux32_maxvmem = LINUX32_MAXVMEM;
990 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
991 &linux32_maxvmem, 0, "");
994 linux32_fixlimit(struct rlimit *rl, int which)
999 if (linux32_maxdsiz != 0) {
1000 if (rl->rlim_cur > linux32_maxdsiz)
1001 rl->rlim_cur = linux32_maxdsiz;
1002 if (rl->rlim_max > linux32_maxdsiz)
1003 rl->rlim_max = linux32_maxdsiz;
1007 if (linux32_maxssiz != 0) {
1008 if (rl->rlim_cur > linux32_maxssiz)
1009 rl->rlim_cur = linux32_maxssiz;
1010 if (rl->rlim_max > linux32_maxssiz)
1011 rl->rlim_max = linux32_maxssiz;
1015 if (linux32_maxvmem != 0) {
1016 if (rl->rlim_cur > linux32_maxvmem)
1017 rl->rlim_cur = linux32_maxvmem;
1018 if (rl->rlim_max > linux32_maxvmem)
1019 rl->rlim_max = linux32_maxvmem;
1025 struct sysentvec elf_linux_sysvec = {
1026 .sv_size = LINUX_SYS_MAXSYSCALL,
1027 .sv_table = linux_sysent,
1029 .sv_sigsize = LINUX_SIGTBLSZ,
1030 .sv_sigtbl = bsd_to_linux_signal,
1031 .sv_errsize = ELAST + 1,
1032 .sv_errtbl = bsd_to_linux_errno,
1033 .sv_transtrap = translate_traps,
1034 .sv_fixup = elf_linux_fixup,
1035 .sv_sendsig = linux_sendsig,
1036 .sv_sigcode = linux_sigcode,
1037 .sv_szsigcode = &linux_szsigcode,
1038 .sv_prepsyscall = NULL,
1039 .sv_name = "Linux ELF32",
1040 .sv_coredump = elf32_coredump,
1041 .sv_imgact_try = exec_linux_imgact_try,
1042 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
1043 .sv_pagesize = PAGE_SIZE,
1044 .sv_minuser = VM_MIN_ADDRESS,
1045 .sv_maxuser = LINUX32_MAXUSER,
1046 .sv_usrstack = LINUX32_USRSTACK,
1047 .sv_psstrings = LINUX32_PS_STRINGS,
1048 .sv_stackprot = VM_PROT_ALL,
1049 .sv_copyout_strings = linux_copyout_strings,
1050 .sv_setregs = exec_linux_setregs,
1051 .sv_fixlimit = linux32_fixlimit,
1052 .sv_maxssiz = &linux32_maxssiz,
1053 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP,
1054 .sv_set_syscall_retval = cpu_set_syscall_retval,
1055 .sv_fetch_syscall_args = linux32_fetch_syscall_args,
1056 .sv_syscallnames = NULL,
1057 .sv_shared_page_base = LINUX32_SHAREDPAGE,
1058 .sv_shared_page_len = PAGE_SIZE,
1059 .sv_schedtail = linux_schedtail,
1061 INIT_SYSENTVEC(elf_sysvec, &elf_linux_sysvec);
1063 static char GNU_ABI_VENDOR[] = "GNU";
1064 static int GNULINUX_ABI_DESC = 0;
1067 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
1069 const Elf32_Word *desc;
1072 p = (uintptr_t)(note + 1);
1073 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1075 desc = (const Elf32_Word *)p;
1076 if (desc[0] != GNULINUX_ABI_DESC)
1080 * For linux we encode osrel as follows (see linux_mib.c):
1081 * VVVMMMIII (version, major, minor), see linux_mib.c.
1083 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1088 static Elf_Brandnote linux32_brandnote = {
1089 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
1090 .hdr.n_descsz = 16, /* XXX at least 16 */
1092 .vendor = GNU_ABI_VENDOR,
1093 .flags = BN_TRANSLATE_OSREL,
1094 .trans_osrel = linux32_trans_osrel
1097 static Elf32_Brandinfo linux_brand = {
1098 .brand = ELFOSABI_LINUX,
1100 .compat_3_brand = "Linux",
1101 .emul_path = "/compat/linux",
1102 .interp_path = "/lib/ld-linux.so.1",
1103 .sysvec = &elf_linux_sysvec,
1104 .interp_newpath = NULL,
1105 .brand_note = &linux32_brandnote,
1106 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1109 static Elf32_Brandinfo linux_glibc2brand = {
1110 .brand = ELFOSABI_LINUX,
1112 .compat_3_brand = "Linux",
1113 .emul_path = "/compat/linux",
1114 .interp_path = "/lib/ld-linux.so.2",
1115 .sysvec = &elf_linux_sysvec,
1116 .interp_newpath = NULL,
1117 .brand_note = &linux32_brandnote,
1118 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1121 Elf32_Brandinfo *linux_brandlist[] = {
1128 linux_elf_modevent(module_t mod, int type, void *data)
1130 Elf32_Brandinfo **brandinfo;
1132 struct linux_ioctl_handler **lihp;
1133 struct linux_device_handler **ldhp;
1139 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1141 if (elf32_insert_brand_entry(*brandinfo) < 0)
1144 SET_FOREACH(lihp, linux_ioctl_handler_set)
1145 linux_ioctl_register_handler(*lihp);
1146 SET_FOREACH(ldhp, linux_device_handler_set)
1147 linux_device_register_handler(*ldhp);
1148 mtx_init(&emul_lock, "emuldata lock", NULL, MTX_DEF);
1149 sx_init(&emul_shared_lock, "emuldata->shared lock");
1150 LIST_INIT(&futex_list);
1151 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1152 linux_exit_tag = EVENTHANDLER_REGISTER(process_exit,
1153 linux_proc_exit, NULL, 1000);
1154 linux_exec_tag = EVENTHANDLER_REGISTER(process_exec,
1155 linux_proc_exec, NULL, 1000);
1156 linux_szplatform = roundup(strlen(linux_platform) + 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 SET_FOREACH(ldhp, linux_device_handler_set)
1180 linux_device_unregister_handler(*ldhp);
1181 mtx_destroy(&emul_lock);
1182 sx_destroy(&emul_shared_lock);
1183 mtx_destroy(&futex_mtx);
1184 EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
1185 EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
1186 linux_osd_jail_deregister();
1188 printf("Linux ELF exec handler removed\n");
1190 printf("Could not deinstall ELF interpreter entry\n");
1198 static moduledata_t linux_elf_mod = {
1204 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);