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 "opt_compat.h"
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 #include "opt_compat.h"
39 #ifndef COMPAT_FREEBSD32
40 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!"
43 #define __ELF_WORD_SIZE 32
45 #include <sys/param.h>
46 #include <sys/systm.h>
48 #include <sys/fcntl.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
51 #include <sys/kernel.h>
53 #include <sys/malloc.h>
54 #include <sys/module.h>
55 #include <sys/mutex.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
62 #include <sys/sysproto.h>
63 #include <sys/vnode.h>
64 #include <sys/eventhandler.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_param.h>
74 #include <machine/cpu.h>
75 #include <machine/md_var.h>
76 #include <machine/pcb.h>
77 #include <machine/specialreg.h>
79 #include <amd64/linux32/linux.h>
80 #include <amd64/linux32/linux32_proto.h>
81 #include <compat/linux/linux_emul.h>
82 #include <compat/linux/linux_futex.h>
83 #include <compat/linux/linux_ioctl.h>
84 #include <compat/linux/linux_mib.h>
85 #include <compat/linux/linux_misc.h>
86 #include <compat/linux/linux_signal.h>
87 #include <compat/linux/linux_util.h>
88 #include <compat/linux/linux_vdso.h>
90 MODULE_VERSION(linux, 1);
92 #define AUXARGS_ENTRY_32(pos, id, val) \
94 suword32(pos++, id); \
95 suword32(pos++, val); \
98 #if BYTE_ORDER == LITTLE_ENDIAN
99 #define SHELLMAGIC 0x2123 /* #! */
101 #define SHELLMAGIC 0x2321
105 * Allow the sendsig functions to use the ldebug() facility even though they
106 * are not syscalls themselves. Map them to syscall 0. This is slightly less
107 * bogus than using ldebug(sigreturn).
109 #define LINUX32_SYS_linux_rt_sendsig 0
110 #define LINUX32_SYS_linux_sendsig 0
112 const char *linux_kplatform;
113 static int linux_szsigcode;
114 static vm_object_t linux_shared_page_obj;
115 static char *linux_shared_page_mapping;
116 extern char _binary_linux32_locore_o_start;
117 extern char _binary_linux32_locore_o_end;
119 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
121 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
123 static int linux_fixup_elf(register_t **stack_base,
124 struct image_params *iparams);
125 static register_t *linux_copyout_strings(struct image_params *imgp);
126 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
127 static int linux_exec_imgact_try(struct image_params *iparams);
128 static void linux_exec_setregs(struct thread *td,
129 struct image_params *imgp, u_long stack);
130 static void linux32_fixlimit(struct rlimit *rl, int which);
131 static bool linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
132 static void linux_vdso_install(void *param);
133 static void linux_vdso_deinstall(void *param);
135 #define LINUX_T_UNKNOWN 255
136 static int _bsd_to_linux_trapcode[] = {
137 LINUX_T_UNKNOWN, /* 0 */
138 6, /* 1 T_PRIVINFLT */
139 LINUX_T_UNKNOWN, /* 2 */
141 LINUX_T_UNKNOWN, /* 4 */
142 LINUX_T_UNKNOWN, /* 5 */
143 16, /* 6 T_ARITHTRAP */
144 254, /* 7 T_ASTFLT */
145 LINUX_T_UNKNOWN, /* 8 */
146 13, /* 9 T_PROTFLT */
147 1, /* 10 T_TRCTRAP */
148 LINUX_T_UNKNOWN, /* 11 */
149 14, /* 12 T_PAGEFLT */
150 LINUX_T_UNKNOWN, /* 13 */
151 17, /* 14 T_ALIGNFLT */
152 LINUX_T_UNKNOWN, /* 15 */
153 LINUX_T_UNKNOWN, /* 16 */
154 LINUX_T_UNKNOWN, /* 17 */
160 8, /* 23 T_DOUBLEFLT */
161 9, /* 24 T_FPOPFLT */
162 10, /* 25 T_TSSFLT */
163 11, /* 26 T_SEGNPFLT */
164 12, /* 27 T_STKFLT */
166 19, /* 29 T_XMMFLT */
167 15 /* 30 T_RESERVED */
169 #define bsd_to_linux_trapcode(code) \
170 ((code)<nitems(_bsd_to_linux_trapcode)? \
171 _bsd_to_linux_trapcode[(code)]: \
174 struct linux32_ps_strings {
175 u_int32_t ps_argvstr; /* first of 0 or more argument strings */
176 u_int ps_nargvstr; /* the number of argument strings */
177 u_int32_t ps_envstr; /* first of 0 or more environment strings */
178 u_int ps_nenvstr; /* the number of environment strings */
181 LINUX_VDSO_SYM_INTPTR(linux32_sigcode);
182 LINUX_VDSO_SYM_INTPTR(linux32_rt_sigcode);
183 LINUX_VDSO_SYM_INTPTR(linux32_vsyscall);
184 LINUX_VDSO_SYM_CHAR(linux_platform);
187 * If FreeBSD & Linux have a difference of opinion about what a trap
188 * means, deal with it here.
193 linux_translate_traps(int signal, int trap_code)
195 if (signal != SIGBUS)
209 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
214 struct linux32_ps_strings *arginfo;
217 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
219 KASSERT(curthread->td_proc == imgp->proc,
220 ("unsafe linux_fixup_elf(), should be curproc"));
221 base = (Elf32_Addr *)*stack_base;
222 args = (Elf32_Auxargs *)imgp->auxargs;
223 pos = base + (imgp->args->argc + imgp->args->envc + 2);
225 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
226 AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO_EHDR,
227 imgp->proc->p_sysent->sv_shared_page_base);
228 AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO, linux32_vsyscall);
229 AUXARGS_ENTRY_32(pos, LINUX_AT_HWCAP, cpu_feature);
232 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
233 * as it has appeared in the 2.4.0-rc7 first time.
234 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
235 * glibc falls back to the hard-coded CLK_TCK value when aux entry
237 * Also see linux_times() implementation.
239 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
240 AUXARGS_ENTRY_32(pos, LINUX_AT_CLKTCK, stclohz);
241 AUXARGS_ENTRY_32(pos, AT_PHDR, args->phdr);
242 AUXARGS_ENTRY_32(pos, AT_PHENT, args->phent);
243 AUXARGS_ENTRY_32(pos, AT_PHNUM, args->phnum);
244 AUXARGS_ENTRY_32(pos, AT_PAGESZ, args->pagesz);
245 AUXARGS_ENTRY_32(pos, AT_FLAGS, args->flags);
246 AUXARGS_ENTRY_32(pos, AT_ENTRY, args->entry);
247 AUXARGS_ENTRY_32(pos, AT_BASE, args->base);
248 AUXARGS_ENTRY_32(pos, LINUX_AT_SECURE, issetugid);
249 AUXARGS_ENTRY_32(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
250 AUXARGS_ENTRY_32(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
251 AUXARGS_ENTRY_32(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
252 AUXARGS_ENTRY_32(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
253 AUXARGS_ENTRY_32(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
254 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
255 if (imgp->execpathp != 0)
256 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
257 if (args->execfd != -1)
258 AUXARGS_ENTRY_32(pos, AT_EXECFD, args->execfd);
259 AUXARGS_ENTRY_32(pos, AT_NULL, 0);
261 free(imgp->auxargs, M_TEMP);
262 imgp->auxargs = NULL;
265 suword32(base, (uint32_t)imgp->args->argc);
266 *stack_base = (register_t *)base;
271 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
273 struct thread *td = curthread;
274 struct proc *p = td->td_proc;
276 struct trapframe *regs;
277 struct l_rt_sigframe *fp, frame;
282 sig = ksi->ksi_signo;
283 code = ksi->ksi_code;
284 PROC_LOCK_ASSERT(p, MA_OWNED);
286 mtx_assert(&psp->ps_mtx, MA_OWNED);
288 oonstack = sigonstack(regs->tf_rsp);
291 if (ldebug(rt_sendsig))
292 printf(ARGS(rt_sendsig, "%p, %d, %p, %u"),
293 catcher, sig, (void*)mask, code);
295 /* Allocate space for the signal handler context. */
296 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
297 SIGISMEMBER(psp->ps_sigonstack, sig)) {
298 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
299 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
301 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
302 mtx_unlock(&psp->ps_mtx);
304 /* Build the argument list for the signal handler. */
305 sig = bsd_to_linux_signal(sig);
307 bzero(&frame, sizeof(frame));
309 frame.sf_handler = PTROUT(catcher);
311 frame.sf_siginfo = PTROUT(&fp->sf_si);
312 frame.sf_ucontext = PTROUT(&fp->sf_sc);
314 /* Fill in POSIX parts. */
315 ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig);
318 * Build the signal context to be used by sigreturn and libgcc unwind.
320 frame.sf_sc.uc_flags = 0; /* XXX ??? */
321 frame.sf_sc.uc_link = 0; /* XXX ??? */
323 frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
324 frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
325 frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
326 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
329 bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask);
331 frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask;
332 frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi;
333 frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi;
334 frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp;
335 frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx;
336 frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp;
337 frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx;
338 frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx;
339 frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax;
340 frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip;
341 frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
342 frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs;
343 frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs;
344 frame.sf_sc.uc_mcontext.sc_es = regs->tf_es;
345 frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds;
346 frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags;
347 frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
348 frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss;
349 frame.sf_sc.uc_mcontext.sc_err = regs->tf_err;
350 frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
351 frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
354 if (ldebug(rt_sendsig))
355 printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"),
356 frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp,
357 td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask);
360 if (copyout(&frame, fp, sizeof(frame)) != 0) {
362 * Process has trashed its stack; give it an illegal
363 * instruction to halt it in its tracks.
366 if (ldebug(rt_sendsig))
367 printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"),
374 /* Build context to run handler in. */
375 regs->tf_rsp = PTROUT(fp);
376 regs->tf_rip = linux32_rt_sigcode;
377 regs->tf_rflags &= ~(PSL_T | PSL_D);
378 regs->tf_cs = _ucode32sel;
379 regs->tf_ss = _udatasel;
380 regs->tf_ds = _udatasel;
381 regs->tf_es = _udatasel;
382 regs->tf_fs = _ufssel;
383 regs->tf_gs = _ugssel;
384 regs->tf_flags = TF_HASSEGS;
385 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
387 mtx_lock(&psp->ps_mtx);
392 * Send an interrupt to process.
394 * Stack is set up to allow sigcode stored
395 * in u. to call routine, followed by kcall
396 * to sigreturn routine below. After sigreturn
397 * resets the signal mask, the stack, and the
398 * frame pointer, it returns to the user
402 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
404 struct thread *td = curthread;
405 struct proc *p = td->td_proc;
407 struct trapframe *regs;
408 struct l_sigframe *fp, frame;
413 sig = ksi->ksi_signo;
414 code = ksi->ksi_code;
415 PROC_LOCK_ASSERT(p, MA_OWNED);
417 mtx_assert(&psp->ps_mtx, MA_OWNED);
418 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
419 /* Signal handler installed with SA_SIGINFO. */
420 linux_rt_sendsig(catcher, ksi, mask);
425 oonstack = sigonstack(regs->tf_rsp);
429 printf(ARGS(sendsig, "%p, %d, %p, %u"),
430 catcher, sig, (void*)mask, code);
433 /* Allocate space for the signal handler context. */
434 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
435 SIGISMEMBER(psp->ps_sigonstack, sig)) {
436 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
437 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
439 fp = (struct l_sigframe *)regs->tf_rsp - 1;
440 mtx_unlock(&psp->ps_mtx);
443 /* Build the argument list for the signal handler. */
444 sig = bsd_to_linux_signal(sig);
446 bzero(&frame, sizeof(frame));
448 frame.sf_handler = PTROUT(catcher);
451 bsd_to_linux_sigset(mask, &lmask);
453 /* Build the signal context to be used by sigreturn. */
454 frame.sf_sc.sc_mask = lmask.__mask;
455 frame.sf_sc.sc_gs = regs->tf_gs;
456 frame.sf_sc.sc_fs = regs->tf_fs;
457 frame.sf_sc.sc_es = regs->tf_es;
458 frame.sf_sc.sc_ds = regs->tf_ds;
459 frame.sf_sc.sc_edi = regs->tf_rdi;
460 frame.sf_sc.sc_esi = regs->tf_rsi;
461 frame.sf_sc.sc_ebp = regs->tf_rbp;
462 frame.sf_sc.sc_ebx = regs->tf_rbx;
463 frame.sf_sc.sc_esp = regs->tf_rsp;
464 frame.sf_sc.sc_edx = regs->tf_rdx;
465 frame.sf_sc.sc_ecx = regs->tf_rcx;
466 frame.sf_sc.sc_eax = regs->tf_rax;
467 frame.sf_sc.sc_eip = regs->tf_rip;
468 frame.sf_sc.sc_cs = regs->tf_cs;
469 frame.sf_sc.sc_eflags = regs->tf_rflags;
470 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
471 frame.sf_sc.sc_ss = regs->tf_ss;
472 frame.sf_sc.sc_err = regs->tf_err;
473 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
474 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
476 frame.sf_extramask[0] = lmask.__mask;
478 if (copyout(&frame, fp, sizeof(frame)) != 0) {
480 * Process has trashed its stack; give it an illegal
481 * instruction to halt it in its tracks.
487 /* Build context to run handler in. */
488 regs->tf_rsp = PTROUT(fp);
489 regs->tf_rip = linux32_sigcode;
490 regs->tf_rflags &= ~(PSL_T | PSL_D);
491 regs->tf_cs = _ucode32sel;
492 regs->tf_ss = _udatasel;
493 regs->tf_ds = _udatasel;
494 regs->tf_es = _udatasel;
495 regs->tf_fs = _ufssel;
496 regs->tf_gs = _ugssel;
497 regs->tf_flags = TF_HASSEGS;
498 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
500 mtx_lock(&psp->ps_mtx);
504 * System call to cleanup state after a signal
505 * has been taken. Reset signal mask and
506 * stack state from context left by sendsig (above).
507 * Return to previous pc and psl as specified by
508 * context left by sendsig. Check carefully to
509 * make sure that the user has not modified the
510 * psl to gain improper privileges or to cause
514 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
516 struct l_sigframe frame;
517 struct trapframe *regs;
526 if (ldebug(sigreturn))
527 printf(ARGS(sigreturn, "%p"), (void *)args->sfp);
530 * The trampoline code hands us the sigframe.
531 * It is unsafe to keep track of it ourselves, in the event that a
532 * program jumps out of a signal handler.
534 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
537 /* Check for security violations. */
538 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
539 eflags = frame.sf_sc.sc_eflags;
540 if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
544 * Don't allow users to load a valid privileged %cs. Let the
545 * hardware check for invalid selectors, excess privilege in
546 * other selectors, invalid %eip's and invalid %esp's.
548 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
549 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
550 ksiginfo_init_trap(&ksi);
551 ksi.ksi_signo = SIGBUS;
552 ksi.ksi_code = BUS_OBJERR;
553 ksi.ksi_trapno = T_PROTFLT;
554 ksi.ksi_addr = (void *)regs->tf_rip;
555 trapsignal(td, &ksi);
559 lmask.__mask = frame.sf_sc.sc_mask;
560 lmask.__mask = frame.sf_extramask[0];
561 linux_to_bsd_sigset(&lmask, &bmask);
562 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
564 /* Restore signal context. */
565 regs->tf_rdi = frame.sf_sc.sc_edi;
566 regs->tf_rsi = frame.sf_sc.sc_esi;
567 regs->tf_rbp = frame.sf_sc.sc_ebp;
568 regs->tf_rbx = frame.sf_sc.sc_ebx;
569 regs->tf_rdx = frame.sf_sc.sc_edx;
570 regs->tf_rcx = frame.sf_sc.sc_ecx;
571 regs->tf_rax = frame.sf_sc.sc_eax;
572 regs->tf_rip = frame.sf_sc.sc_eip;
573 regs->tf_cs = frame.sf_sc.sc_cs;
574 regs->tf_ds = frame.sf_sc.sc_ds;
575 regs->tf_es = frame.sf_sc.sc_es;
576 regs->tf_fs = frame.sf_sc.sc_fs;
577 regs->tf_gs = frame.sf_sc.sc_gs;
578 regs->tf_rflags = eflags;
579 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal;
580 regs->tf_ss = frame.sf_sc.sc_ss;
581 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
583 return (EJUSTRETURN);
587 * System call to cleanup state after a signal
588 * has been taken. Reset signal mask and
589 * stack state from context left by rt_sendsig (above).
590 * Return to previous pc and psl as specified by
591 * context left by sendsig. Check carefully to
592 * make sure that the user has not modified the
593 * psl to gain improper privileges or to cause
597 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
599 struct l_ucontext uc;
600 struct l_sigcontext *context;
604 struct trapframe *regs;
611 if (ldebug(rt_sigreturn))
612 printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp);
615 * The trampoline code hands us the ucontext.
616 * It is unsafe to keep track of it ourselves, in the event that a
617 * program jumps out of a signal handler.
619 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
622 context = &uc.uc_mcontext;
624 /* Check for security violations. */
625 #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
626 eflags = context->sc_eflags;
627 if (!EFLAGS_SECURE(eflags, regs->tf_rflags))
631 * Don't allow users to load a valid privileged %cs. Let the
632 * hardware check for invalid selectors, excess privilege in
633 * other selectors, invalid %eip's and invalid %esp's.
635 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
636 if (!CS_SECURE(context->sc_cs)) {
637 ksiginfo_init_trap(&ksi);
638 ksi.ksi_signo = SIGBUS;
639 ksi.ksi_code = BUS_OBJERR;
640 ksi.ksi_trapno = T_PROTFLT;
641 ksi.ksi_addr = (void *)regs->tf_rip;
642 trapsignal(td, &ksi);
646 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
647 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
650 * Restore signal context
652 regs->tf_gs = context->sc_gs;
653 regs->tf_fs = context->sc_fs;
654 regs->tf_es = context->sc_es;
655 regs->tf_ds = context->sc_ds;
656 regs->tf_rdi = context->sc_edi;
657 regs->tf_rsi = context->sc_esi;
658 regs->tf_rbp = context->sc_ebp;
659 regs->tf_rbx = context->sc_ebx;
660 regs->tf_rdx = context->sc_edx;
661 regs->tf_rcx = context->sc_ecx;
662 regs->tf_rax = context->sc_eax;
663 regs->tf_rip = context->sc_eip;
664 regs->tf_cs = context->sc_cs;
665 regs->tf_rflags = eflags;
666 regs->tf_rsp = context->sc_esp_at_signal;
667 regs->tf_ss = context->sc_ss;
668 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
671 * call sigaltstack & ignore results..
674 ss.ss_sp = PTRIN(lss->ss_sp);
675 ss.ss_size = lss->ss_size;
676 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
679 if (ldebug(rt_sigreturn))
680 printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"),
681 ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask);
683 (void)kern_sigaltstack(td, &ss, NULL);
685 return (EJUSTRETURN);
689 linux32_fetch_syscall_args(struct thread *td)
692 struct trapframe *frame;
693 struct syscall_args *sa;
696 frame = td->td_frame;
699 sa->args[0] = frame->tf_rbx;
700 sa->args[1] = frame->tf_rcx;
701 sa->args[2] = frame->tf_rdx;
702 sa->args[3] = frame->tf_rsi;
703 sa->args[4] = frame->tf_rdi;
704 sa->args[5] = frame->tf_rbp; /* Unconfirmed */
705 sa->code = frame->tf_rax;
707 if (sa->code >= p->p_sysent->sv_size)
709 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
711 sa->callp = &p->p_sysent->sv_table[sa->code];
712 sa->narg = sa->callp->sy_narg;
714 td->td_retval[0] = 0;
715 td->td_retval[1] = frame->tf_rdx;
721 * If a Linux binary is exec'ing something, try this image activator
722 * first. We override standard shell script execution in order to
723 * be able to modify the interpreter path. We only do this if a Linux
724 * binary is doing the exec, so we do not create an EXEC module for it.
727 linux_exec_imgact_try(struct image_params *imgp)
729 const char *head = (const char *)imgp->image_header;
734 * The interpreter for shell scripts run from a Linux binary needs
735 * to be located in /compat/linux if possible in order to recursively
736 * maintain Linux path emulation.
738 if (((const short *)head)[0] == SHELLMAGIC) {
740 * Run our normal shell image activator. If it succeeds then
741 * attempt to use the alternate path for the interpreter. If
742 * an alternate path is found, use our stringspace to store it.
744 if ((error = exec_shell_imgact(imgp)) == 0) {
745 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
746 imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0,
749 imgp->args->fname_buf =
750 imgp->interpreter_name = rpath;
757 * Clear registers on exec
758 * XXX copied from ia32_signal.c.
761 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
763 struct trapframe *regs = td->td_frame;
764 struct pcb *pcb = td->td_pcb;
765 register_t saved_rflags;
771 if (td->td_proc->p_md.md_ldt != NULL)
774 mtx_unlock(&dt_lock);
777 wrmsr(MSR_FSBASE, 0);
778 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */
782 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
784 saved_rflags = regs->tf_rflags & PSL_T;
785 bzero((char *)regs, sizeof(struct trapframe));
786 regs->tf_rip = imgp->entry_addr;
787 regs->tf_rsp = stack;
788 regs->tf_rflags = PSL_USER | saved_rflags;
789 regs->tf_gs = _ugssel;
790 regs->tf_fs = _ufssel;
791 regs->tf_es = _udatasel;
792 regs->tf_ds = _udatasel;
793 regs->tf_ss = _udatasel;
794 regs->tf_flags = TF_HASSEGS;
795 regs->tf_cs = _ucode32sel;
796 regs->tf_rbx = imgp->ps_strings;
800 /* Do full restore on return so that we can change to a different %cs */
801 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
802 td->td_retval[1] = 0;
806 * XXX copied from ia32_sysvec.c.
809 linux_copyout_strings(struct image_params *imgp)
813 char *stringp, *destp;
814 u_int32_t *stack_base;
815 struct linux32_ps_strings *arginfo;
816 char canary[LINUX_AT_RANDOM_LEN];
819 /* Calculate string base and vector table pointers. */
820 if (imgp->execpath != NULL && imgp->auxargs != NULL)
821 execpath_len = strlen(imgp->execpath) + 1;
825 arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS;
826 destp = (caddr_t)arginfo - SPARE_USRSPACE -
827 roundup(sizeof(canary), sizeof(char *)) -
828 roundup(execpath_len, sizeof(char *)) -
829 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
831 if (execpath_len != 0) {
832 imgp->execpathp = (uintptr_t)arginfo - execpath_len;
833 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
836 /* Prepare the canary for SSP. */
837 arc4rand(canary, sizeof(canary), 0);
838 imgp->canary = (uintptr_t)arginfo -
839 roundup(execpath_len, sizeof(char *)) -
840 roundup(sizeof(canary), sizeof(char *));
841 copyout(canary, (void *)imgp->canary, sizeof(canary));
843 vectp = (uint32_t *)destp;
846 * Allocate room on the stack for the ELF auxargs
847 * array. It has LINUX_AT_COUNT entries.
849 vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf32_Auxinfo),
854 * Allocate room for the argv[] and env vectors including the
855 * terminating NULL pointers.
857 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
859 /* vectp also becomes our initial stack base. */
862 stringp = imgp->args->begin_argv;
863 argc = imgp->args->argc;
864 envc = imgp->args->envc;
865 /* Copy out strings - arguments and environment. */
866 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
868 /* Fill in "ps_strings" struct for ps, w, etc. */
869 suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp);
870 suword32(&arginfo->ps_nargvstr, argc);
872 /* Fill in argument portion of vector table. */
873 for (; argc > 0; --argc) {
874 suword32(vectp++, (uint32_t)(intptr_t)destp);
875 while (*stringp++ != 0)
880 /* A null vector table pointer separates the argp's from the envp's. */
881 suword32(vectp++, 0);
883 suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp);
884 suword32(&arginfo->ps_nenvstr, envc);
886 /* Fill in environment portion of vector table. */
887 for (; envc > 0; --envc) {
888 suword32(vectp++, (uint32_t)(intptr_t)destp);
889 while (*stringp++ != 0)
894 /* The end of the vector table is a null pointer. */
897 return ((register_t *)stack_base);
900 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0,
901 "32-bit Linux emulation");
903 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ;
904 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
905 &linux32_maxdsiz, 0, "");
906 static u_long linux32_maxssiz = LINUX32_MAXSSIZ;
907 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
908 &linux32_maxssiz, 0, "");
909 static u_long linux32_maxvmem = LINUX32_MAXVMEM;
910 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
911 &linux32_maxvmem, 0, "");
914 SYSCTL_PROC(_compat_linux32, OID_AUTO, debug, CTLTYPE_STRING | CTLFLAG_RW, 0, 0,
915 linux_sysctl_debug, "A", "Linux debugging control");
919 linux32_fixlimit(struct rlimit *rl, int which)
924 if (linux32_maxdsiz != 0) {
925 if (rl->rlim_cur > linux32_maxdsiz)
926 rl->rlim_cur = linux32_maxdsiz;
927 if (rl->rlim_max > linux32_maxdsiz)
928 rl->rlim_max = linux32_maxdsiz;
932 if (linux32_maxssiz != 0) {
933 if (rl->rlim_cur > linux32_maxssiz)
934 rl->rlim_cur = linux32_maxssiz;
935 if (rl->rlim_max > linux32_maxssiz)
936 rl->rlim_max = linux32_maxssiz;
940 if (linux32_maxvmem != 0) {
941 if (rl->rlim_cur > linux32_maxvmem)
942 rl->rlim_cur = linux32_maxvmem;
943 if (rl->rlim_max > linux32_maxvmem)
944 rl->rlim_max = linux32_maxvmem;
950 struct sysentvec elf_linux_sysvec = {
951 .sv_size = LINUX32_SYS_MAXSYSCALL,
952 .sv_table = linux32_sysent,
954 .sv_errsize = ELAST + 1,
955 .sv_errtbl = linux_errtbl,
956 .sv_transtrap = linux_translate_traps,
957 .sv_fixup = linux_fixup_elf,
958 .sv_sendsig = linux_sendsig,
959 .sv_sigcode = &_binary_linux32_locore_o_start,
960 .sv_szsigcode = &linux_szsigcode,
961 .sv_name = "Linux ELF32",
962 .sv_coredump = elf32_coredump,
963 .sv_imgact_try = linux_exec_imgact_try,
964 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
965 .sv_pagesize = PAGE_SIZE,
966 .sv_minuser = VM_MIN_ADDRESS,
967 .sv_maxuser = LINUX32_MAXUSER,
968 .sv_usrstack = LINUX32_USRSTACK,
969 .sv_psstrings = LINUX32_PS_STRINGS,
970 .sv_stackprot = VM_PROT_ALL,
971 .sv_copyout_strings = linux_copyout_strings,
972 .sv_setregs = linux_exec_setregs,
973 .sv_fixlimit = linux32_fixlimit,
974 .sv_maxssiz = &linux32_maxssiz,
975 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP,
976 .sv_set_syscall_retval = cpu_set_syscall_retval,
977 .sv_fetch_syscall_args = linux32_fetch_syscall_args,
978 .sv_syscallnames = NULL,
979 .sv_shared_page_base = LINUX32_SHAREDPAGE,
980 .sv_shared_page_len = PAGE_SIZE,
981 .sv_schedtail = linux_schedtail,
982 .sv_thread_detach = linux_thread_detach,
987 linux_vdso_install(void *param)
990 linux_szsigcode = (&_binary_linux32_locore_o_end -
991 &_binary_linux32_locore_o_start);
993 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
994 panic("Linux invalid vdso size\n");
996 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
998 linux_shared_page_obj = __elfN(linux_shared_page_init)
999 (&linux_shared_page_mapping);
1001 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
1003 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
1005 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
1007 linux_kplatform = linux_shared_page_mapping +
1008 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
1010 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
1011 (sysinit_cfunc_t)linux_vdso_install, NULL);
1014 linux_vdso_deinstall(void *param)
1017 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
1019 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1020 (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
1022 static char GNU_ABI_VENDOR[] = "GNU";
1023 static int GNULINUX_ABI_DESC = 0;
1026 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
1028 const Elf32_Word *desc;
1031 p = (uintptr_t)(note + 1);
1032 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1034 desc = (const Elf32_Word *)p;
1035 if (desc[0] != GNULINUX_ABI_DESC)
1039 * For Linux we encode osrel as follows (see linux_mib.c):
1040 * VVVMMMIII (version, major, minor), see linux_mib.c.
1042 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
1047 static Elf_Brandnote linux32_brandnote = {
1048 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
1049 .hdr.n_descsz = 16, /* XXX at least 16 */
1051 .vendor = GNU_ABI_VENDOR,
1052 .flags = BN_TRANSLATE_OSREL,
1053 .trans_osrel = linux32_trans_osrel
1056 static Elf32_Brandinfo linux_brand = {
1057 .brand = ELFOSABI_LINUX,
1059 .compat_3_brand = "Linux",
1060 .emul_path = "/compat/linux",
1061 .interp_path = "/lib/ld-linux.so.1",
1062 .sysvec = &elf_linux_sysvec,
1063 .interp_newpath = NULL,
1064 .brand_note = &linux32_brandnote,
1065 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1068 static Elf32_Brandinfo linux_glibc2brand = {
1069 .brand = ELFOSABI_LINUX,
1071 .compat_3_brand = "Linux",
1072 .emul_path = "/compat/linux",
1073 .interp_path = "/lib/ld-linux.so.2",
1074 .sysvec = &elf_linux_sysvec,
1075 .interp_newpath = NULL,
1076 .brand_note = &linux32_brandnote,
1077 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1080 static Elf32_Brandinfo linux_muslbrand = {
1081 .brand = ELFOSABI_LINUX,
1083 .compat_3_brand = "Linux",
1084 .emul_path = "/compat/linux",
1085 .interp_path = "/lib/ld-musl-i386.so.1",
1086 .sysvec = &elf_linux_sysvec,
1087 .interp_newpath = NULL,
1088 .brand_note = &linux32_brandnote,
1089 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1092 Elf32_Brandinfo *linux_brandlist[] = {
1100 linux_elf_modevent(module_t mod, int type, void *data)
1102 Elf32_Brandinfo **brandinfo;
1104 struct linux_ioctl_handler **lihp;
1110 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1112 if (elf32_insert_brand_entry(*brandinfo) < 0)
1115 SET_FOREACH(lihp, linux_ioctl_handler_set)
1116 linux32_ioctl_register_handler(*lihp);
1117 LIST_INIT(&futex_list);
1118 mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF);
1119 stclohz = (stathz ? stathz : hz);
1121 printf("Linux ELF exec handler installed\n");
1123 printf("cannot insert Linux ELF brand handler\n");
1126 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1128 if (elf32_brand_inuse(*brandinfo))
1131 for (brandinfo = &linux_brandlist[0];
1132 *brandinfo != NULL; ++brandinfo)
1133 if (elf32_remove_brand_entry(*brandinfo) < 0)
1137 SET_FOREACH(lihp, linux_ioctl_handler_set)
1138 linux32_ioctl_unregister_handler(*lihp);
1139 mtx_destroy(&futex_mtx);
1141 printf("Linux ELF exec handler removed\n");
1143 printf("Could not deinstall ELF interpreter entry\n");
1146 return (EOPNOTSUPP);
1151 static moduledata_t linux_elf_mod = {
1157 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1158 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1159 FEATURE(linux, "Linux 32bit support");