2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2004 Tim J. Robbins
5 * Copyright (c) 2003 Peter Wemm
6 * Copyright (c) 2002 Doug Rabson
7 * Copyright (c) 1998-1999 Andrew Gallatin
8 * Copyright (c) 1994-1996 Søren Schmidt
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer
16 * in this position and unchanged.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. The name of the author may not be used to endorse or promote products
21 * derived from this software without specific prior written permission
23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include "opt_compat.h"
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #ifndef COMPAT_FREEBSD32
41 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!"
44 #define __ELF_WORD_SIZE 32
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/fcntl.h>
50 #include <sys/imgact.h>
51 #include <sys/imgact_elf.h>
52 #include <sys/kernel.h>
54 #include <sys/malloc.h>
55 #include <sys/module.h>
56 #include <sys/mutex.h>
58 #include <sys/resourcevar.h>
59 #include <sys/stddef.h>
60 #include <sys/signalvar.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysent.h>
64 #include <sys/sysproto.h>
65 #include <sys/vnode.h>
66 #include <sys/eventhandler.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_param.h>
76 #include <machine/cpu.h>
77 #include <machine/md_var.h>
78 #include <machine/pcb.h>
79 #include <machine/specialreg.h>
80 #include <machine/trap.h>
82 #include <x86/linux/linux_x86.h>
83 #include <amd64/linux32/linux.h>
84 #include <amd64/linux32/linux32_proto.h>
85 #include <compat/linux/linux_emul.h>
86 #include <compat/linux/linux_fork.h>
87 #include <compat/linux/linux_ioctl.h>
88 #include <compat/linux/linux_mib.h>
89 #include <compat/linux/linux_misc.h>
90 #include <compat/linux/linux_signal.h>
91 #include <compat/linux/linux_util.h>
92 #include <compat/linux/linux_vdso.h>
94 #include <x86/linux/linux_x86_sigframe.h>
96 MODULE_VERSION(linux, 1);
98 #define LINUX32_MAXUSER ((1ul << 32) - PAGE_SIZE)
99 #define LINUX32_VDSOPAGE_SIZE PAGE_SIZE * 2
100 #define LINUX32_VDSOPAGE (LINUX32_MAXUSER - LINUX32_VDSOPAGE_SIZE)
101 #define LINUX32_SHAREDPAGE (LINUX32_VDSOPAGE - PAGE_SIZE)
103 * PAGE_SIZE - the size
104 * of the native SHAREDPAGE
106 #define LINUX32_USRSTACK LINUX32_SHAREDPAGE
108 static int linux_szsigcode;
109 static vm_object_t linux_vdso_obj;
110 static char *linux_vdso_mapping;
111 extern char _binary_linux32_vdso_so_o_start;
112 extern char _binary_linux32_vdso_so_o_end;
113 static vm_offset_t linux_vdso_base;
115 extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL];
117 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
119 static int linux_fixup_elf(uintptr_t *stack_base,
120 struct image_params *iparams);
121 static int linux_copyout_strings(struct image_params *imgp,
122 uintptr_t *stack_base);
123 static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask);
124 static void linux_exec_setregs(struct thread *td,
125 struct image_params *imgp, uintptr_t stack);
126 static void linux_exec_sysvec_init(void *param);
127 static int linux_on_exec_vmspace(struct proc *p,
128 struct image_params *imgp);
129 static void linux32_fixlimit(struct rlimit *rl, int which);
130 static bool linux32_trans_osrel(const Elf_Note *note, int32_t *osrel);
131 static void linux_vdso_install(const void *param);
132 static void linux_vdso_deinstall(const void *param);
133 static void linux_vdso_reloc(char *mapping, Elf_Addr offset);
134 static void linux32_set_fork_retval(struct thread *td);
135 static void linux32_set_syscall_retval(struct thread *td, int error);
137 #define LINUX_T_UNKNOWN 255
138 static int _bsd_to_linux_trapcode[] = {
139 LINUX_T_UNKNOWN, /* 0 */
140 6, /* 1 T_PRIVINFLT */
141 LINUX_T_UNKNOWN, /* 2 */
143 LINUX_T_UNKNOWN, /* 4 */
144 LINUX_T_UNKNOWN, /* 5 */
145 16, /* 6 T_ARITHTRAP */
146 254, /* 7 T_ASTFLT */
147 LINUX_T_UNKNOWN, /* 8 */
148 13, /* 9 T_PROTFLT */
149 1, /* 10 T_TRCTRAP */
150 LINUX_T_UNKNOWN, /* 11 */
151 14, /* 12 T_PAGEFLT */
152 LINUX_T_UNKNOWN, /* 13 */
153 17, /* 14 T_ALIGNFLT */
154 LINUX_T_UNKNOWN, /* 15 */
155 LINUX_T_UNKNOWN, /* 16 */
156 LINUX_T_UNKNOWN, /* 17 */
162 8, /* 23 T_DOUBLEFLT */
163 9, /* 24 T_FPOPFLT */
164 10, /* 25 T_TSSFLT */
165 11, /* 26 T_SEGNPFLT */
166 12, /* 27 T_STKFLT */
168 19, /* 29 T_XMMFLT */
169 15 /* 30 T_RESERVED */
171 #define bsd_to_linux_trapcode(code) \
172 ((code)<nitems(_bsd_to_linux_trapcode)? \
173 _bsd_to_linux_trapcode[(code)]: \
176 struct linux32_ps_strings {
177 u_int32_t ps_argvstr; /* first of 0 or more argument strings */
178 u_int ps_nargvstr; /* the number of argument strings */
179 u_int32_t ps_envstr; /* first of 0 or more environment strings */
180 u_int ps_nenvstr; /* the number of environment strings */
182 #define LINUX32_PS_STRINGS (LINUX32_USRSTACK - \
183 sizeof(struct linux32_ps_strings))
185 LINUX_VDSO_SYM_INTPTR(__kernel_vsyscall);
186 LINUX_VDSO_SYM_INTPTR(linux32_vdso_sigcode);
187 LINUX_VDSO_SYM_INTPTR(linux32_vdso_rt_sigcode);
188 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
189 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
190 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
191 LINUX_VDSO_SYM_CHAR(linux_platform);
194 * If FreeBSD & Linux have a difference of opinion about what a trap
195 * means, deal with it here.
200 linux_translate_traps(int signal, int trap_code)
202 if (signal != SIGBUS)
216 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
219 Elf32_Auxinfo *argarray, *pos;
220 int error, issetugid;
222 args = (Elf32_Auxargs *)imgp->auxargs;
223 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
226 issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0;
227 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, __kernel_vsyscall);
228 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
229 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
230 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
233 * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0,
234 * as it has appeared in the 2.4.0-rc7 first time.
235 * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK),
236 * glibc falls back to the hard-coded CLK_TCK value when aux entry
238 * Also see linux_times() implementation.
240 if (linux_kernver(curthread) >= LINUX_KERNVER_2004000)
241 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
242 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
243 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
244 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
245 AUXARGS_ENTRY(pos, AT_BASE, args->base);
246 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
247 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
248 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
249 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
250 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
251 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
252 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
253 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary));
254 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
255 if (imgp->execpathp != 0)
256 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp));
257 if (args->execfd != -1)
258 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
259 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
260 AUXARGS_ENTRY(pos, AT_NULL, 0);
262 free(imgp->auxargs, M_TEMP);
263 imgp->auxargs = NULL;
264 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
266 error = copyout(argarray, (void *)base,
267 sizeof(*argarray) * LINUX_AT_COUNT);
268 free(argarray, M_TEMP);
273 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
277 base = (Elf32_Addr *)*stack_base;
279 if (suword32(base, (uint32_t)imgp->args->argc) == -1)
281 *stack_base = (uintptr_t)base;
286 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
288 struct thread *td = curthread;
289 struct proc *p = td->td_proc;
291 struct trapframe *regs;
292 struct l_rt_sigframe *fp, frame;
297 sig = ksi->ksi_signo;
298 code = ksi->ksi_code;
299 PROC_LOCK_ASSERT(p, MA_OWNED);
301 mtx_assert(&psp->ps_mtx, MA_OWNED);
303 oonstack = sigonstack(regs->tf_rsp);
305 /* Allocate space for the signal handler context. */
306 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
307 SIGISMEMBER(psp->ps_sigonstack, sig)) {
308 fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
309 td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe));
311 fp = (struct l_rt_sigframe *)regs->tf_rsp - 1;
312 mtx_unlock(&psp->ps_mtx);
314 /* Build the argument list for the signal handler. */
315 sig = bsd_to_linux_signal(sig);
317 bzero(&frame, sizeof(frame));
320 frame.sf_siginfo = PTROUT(&fp->sf_si);
321 frame.sf_ucontext = PTROUT(&fp->sf_uc);
323 /* Fill in POSIX parts. */
324 siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
327 * Build the signal context to be used by sigreturn and libgcc unwind.
329 frame.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
330 frame.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
331 frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
332 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
335 bsd_to_linux_sigset(mask, &frame.sf_uc.uc_sigmask);
337 frame.sf_uc.uc_mcontext.sc_mask = frame.sf_uc.uc_sigmask.__mask;
338 frame.sf_uc.uc_mcontext.sc_edi = regs->tf_rdi;
339 frame.sf_uc.uc_mcontext.sc_esi = regs->tf_rsi;
340 frame.sf_uc.uc_mcontext.sc_ebp = regs->tf_rbp;
341 frame.sf_uc.uc_mcontext.sc_ebx = regs->tf_rbx;
342 frame.sf_uc.uc_mcontext.sc_esp = regs->tf_rsp;
343 frame.sf_uc.uc_mcontext.sc_edx = regs->tf_rdx;
344 frame.sf_uc.uc_mcontext.sc_ecx = regs->tf_rcx;
345 frame.sf_uc.uc_mcontext.sc_eax = regs->tf_rax;
346 frame.sf_uc.uc_mcontext.sc_eip = regs->tf_rip;
347 frame.sf_uc.uc_mcontext.sc_cs = regs->tf_cs;
348 frame.sf_uc.uc_mcontext.sc_gs = regs->tf_gs;
349 frame.sf_uc.uc_mcontext.sc_fs = regs->tf_fs;
350 frame.sf_uc.uc_mcontext.sc_es = regs->tf_es;
351 frame.sf_uc.uc_mcontext.sc_ds = regs->tf_ds;
352 frame.sf_uc.uc_mcontext.sc_eflags = regs->tf_rflags;
353 frame.sf_uc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp;
354 frame.sf_uc.uc_mcontext.sc_ss = regs->tf_ss;
355 frame.sf_uc.uc_mcontext.sc_err = regs->tf_err;
356 frame.sf_uc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
357 frame.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
359 if (copyout(&frame, fp, sizeof(frame)) != 0) {
361 * Process has trashed its stack; give it an illegal
362 * instruction to halt it in its tracks.
368 /* Build context to run handler in. */
369 regs->tf_rsp = PTROUT(fp);
370 regs->tf_rip = linux32_vdso_rt_sigcode;
371 regs->tf_rdi = PTROUT(catcher);
372 regs->tf_rflags &= ~(PSL_T | PSL_D);
373 regs->tf_cs = _ucode32sel;
374 regs->tf_ss = _udatasel;
375 regs->tf_ds = _udatasel;
376 regs->tf_es = _udatasel;
377 regs->tf_fs = _ufssel;
378 regs->tf_gs = _ugssel;
379 regs->tf_flags = TF_HASSEGS;
380 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
382 mtx_lock(&psp->ps_mtx);
386 * Send an interrupt to process.
388 * Stack is set up to allow sigcode stored
389 * in u. to call routine, followed by kcall
390 * to sigreturn routine below. After sigreturn
391 * resets the signal mask, the stack, and the
392 * frame pointer, it returns to the user
396 linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
398 struct thread *td = curthread;
399 struct proc *p = td->td_proc;
401 struct trapframe *regs;
402 struct l_sigframe *fp, frame;
407 sig = ksi->ksi_signo;
408 code = ksi->ksi_code;
409 PROC_LOCK_ASSERT(p, MA_OWNED);
411 mtx_assert(&psp->ps_mtx, MA_OWNED);
412 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
413 /* Signal handler installed with SA_SIGINFO. */
414 linux_rt_sendsig(catcher, ksi, mask);
419 oonstack = sigonstack(regs->tf_rsp);
421 /* Allocate space for the signal handler context. */
422 if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
423 SIGISMEMBER(psp->ps_sigonstack, sig)) {
424 fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
425 td->td_sigstk.ss_size - sizeof(struct l_sigframe));
427 fp = (struct l_sigframe *)regs->tf_rsp - 1;
428 mtx_unlock(&psp->ps_mtx);
431 /* Build the argument list for the signal handler. */
432 sig = bsd_to_linux_signal(sig);
434 bzero(&frame, sizeof(frame));
438 bsd_to_linux_sigset(mask, &lmask);
440 /* Build the signal context to be used by sigreturn. */
441 frame.sf_sc.sc_mask = lmask.__mask;
442 frame.sf_sc.sc_gs = regs->tf_gs;
443 frame.sf_sc.sc_fs = regs->tf_fs;
444 frame.sf_sc.sc_es = regs->tf_es;
445 frame.sf_sc.sc_ds = regs->tf_ds;
446 frame.sf_sc.sc_edi = regs->tf_rdi;
447 frame.sf_sc.sc_esi = regs->tf_rsi;
448 frame.sf_sc.sc_ebp = regs->tf_rbp;
449 frame.sf_sc.sc_ebx = regs->tf_rbx;
450 frame.sf_sc.sc_esp = regs->tf_rsp;
451 frame.sf_sc.sc_edx = regs->tf_rdx;
452 frame.sf_sc.sc_ecx = regs->tf_rcx;
453 frame.sf_sc.sc_eax = regs->tf_rax;
454 frame.sf_sc.sc_eip = regs->tf_rip;
455 frame.sf_sc.sc_cs = regs->tf_cs;
456 frame.sf_sc.sc_eflags = regs->tf_rflags;
457 frame.sf_sc.sc_esp_at_signal = regs->tf_rsp;
458 frame.sf_sc.sc_ss = regs->tf_ss;
459 frame.sf_sc.sc_err = regs->tf_err;
460 frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr;
461 frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code);
463 frame.sf_extramask[0] = lmask.__mask;
465 if (copyout(&frame, fp, sizeof(frame)) != 0) {
467 * Process has trashed its stack; give it an illegal
468 * instruction to halt it in its tracks.
474 /* Build context to run handler in. */
475 regs->tf_rsp = PTROUT(fp);
476 regs->tf_rip = linux32_vdso_sigcode;
477 regs->tf_rdi = PTROUT(catcher);
478 regs->tf_rflags &= ~(PSL_T | PSL_D);
479 regs->tf_cs = _ucode32sel;
480 regs->tf_ss = _udatasel;
481 regs->tf_ds = _udatasel;
482 regs->tf_es = _udatasel;
483 regs->tf_fs = _ufssel;
484 regs->tf_gs = _ugssel;
485 regs->tf_flags = TF_HASSEGS;
486 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
488 mtx_lock(&psp->ps_mtx);
492 * System call to cleanup state after a signal
493 * has been taken. Reset signal mask and
494 * stack state from context left by sendsig (above).
495 * Return to previous pc and psl as specified by
496 * context left by sendsig. Check carefully to
497 * make sure that the user has not modified the
498 * psl to gain improper privileges or to cause
502 linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args)
504 struct l_sigframe frame;
505 struct trapframe *regs;
514 * The trampoline code hands us the sigframe.
515 * It is unsafe to keep track of it ourselves, in the event that a
516 * program jumps out of a signal handler.
518 if (copyin(args->sfp, &frame, sizeof(frame)) != 0)
521 /* Check for security violations. */
522 eflags = frame.sf_sc.sc_eflags;
523 if (!EFL_SECURE(eflags, regs->tf_rflags))
527 * Don't allow users to load a valid privileged %cs. Let the
528 * hardware check for invalid selectors, excess privilege in
529 * other selectors, invalid %eip's and invalid %esp's.
531 if (!CS_SECURE(frame.sf_sc.sc_cs)) {
532 ksiginfo_init_trap(&ksi);
533 ksi.ksi_signo = SIGBUS;
534 ksi.ksi_code = BUS_OBJERR;
535 ksi.ksi_trapno = T_PROTFLT;
536 ksi.ksi_addr = (void *)regs->tf_rip;
537 trapsignal(td, &ksi);
541 lmask.__mask = frame.sf_sc.sc_mask;
542 lmask.__mask = frame.sf_extramask[0];
543 linux_to_bsd_sigset(&lmask, &bmask);
544 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
546 /* Restore signal context. */
547 regs->tf_rdi = frame.sf_sc.sc_edi;
548 regs->tf_rsi = frame.sf_sc.sc_esi;
549 regs->tf_rbp = frame.sf_sc.sc_ebp;
550 regs->tf_rbx = frame.sf_sc.sc_ebx;
551 regs->tf_rdx = frame.sf_sc.sc_edx;
552 regs->tf_rcx = frame.sf_sc.sc_ecx;
553 regs->tf_rax = frame.sf_sc.sc_eax;
554 regs->tf_rip = frame.sf_sc.sc_eip;
555 regs->tf_cs = frame.sf_sc.sc_cs;
556 regs->tf_ds = frame.sf_sc.sc_ds;
557 regs->tf_es = frame.sf_sc.sc_es;
558 regs->tf_fs = frame.sf_sc.sc_fs;
559 regs->tf_gs = frame.sf_sc.sc_gs;
560 regs->tf_rflags = eflags;
561 regs->tf_rsp = frame.sf_sc.sc_esp_at_signal;
562 regs->tf_ss = frame.sf_sc.sc_ss;
563 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
565 return (EJUSTRETURN);
569 * System call to cleanup state after a signal
570 * has been taken. Reset signal mask and
571 * stack state from context left by rt_sendsig (above).
572 * Return to previous pc and psl as specified by
573 * context left by sendsig. Check carefully to
574 * make sure that the user has not modified the
575 * psl to gain improper privileges or to cause
579 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
581 struct l_ucontext uc;
582 struct l_sigcontext *context;
586 struct trapframe *regs;
593 * The trampoline code hands us the ucontext.
594 * It is unsafe to keep track of it ourselves, in the event that a
595 * program jumps out of a signal handler.
597 if (copyin(args->ucp, &uc, sizeof(uc)) != 0)
600 context = &uc.uc_mcontext;
602 /* Check for security violations. */
603 eflags = context->sc_eflags;
604 if (!EFL_SECURE(eflags, regs->tf_rflags))
608 * Don't allow users to load a valid privileged %cs. Let the
609 * hardware check for invalid selectors, excess privilege in
610 * other selectors, invalid %eip's and invalid %esp's.
612 if (!CS_SECURE(context->sc_cs)) {
613 ksiginfo_init_trap(&ksi);
614 ksi.ksi_signo = SIGBUS;
615 ksi.ksi_code = BUS_OBJERR;
616 ksi.ksi_trapno = T_PROTFLT;
617 ksi.ksi_addr = (void *)regs->tf_rip;
618 trapsignal(td, &ksi);
622 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
623 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
626 * Restore signal context
628 regs->tf_gs = context->sc_gs;
629 regs->tf_fs = context->sc_fs;
630 regs->tf_es = context->sc_es;
631 regs->tf_ds = context->sc_ds;
632 regs->tf_rdi = context->sc_edi;
633 regs->tf_rsi = context->sc_esi;
634 regs->tf_rbp = context->sc_ebp;
635 regs->tf_rbx = context->sc_ebx;
636 regs->tf_rdx = context->sc_edx;
637 regs->tf_rcx = context->sc_ecx;
638 regs->tf_rax = context->sc_eax;
639 regs->tf_rip = context->sc_eip;
640 regs->tf_cs = context->sc_cs;
641 regs->tf_rflags = eflags;
642 regs->tf_rsp = context->sc_esp_at_signal;
643 regs->tf_ss = context->sc_ss;
644 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
647 * call sigaltstack & ignore results..
650 ss.ss_sp = PTRIN(lss->ss_sp);
651 ss.ss_size = lss->ss_size;
652 ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags);
654 (void)kern_sigaltstack(td, &ss, NULL);
656 return (EJUSTRETURN);
660 linux32_fetch_syscall_args(struct thread *td)
663 struct trapframe *frame;
664 struct syscall_args *sa;
667 frame = td->td_frame;
670 sa->args[0] = frame->tf_rbx;
671 sa->args[1] = frame->tf_rcx;
672 sa->args[2] = frame->tf_rdx;
673 sa->args[3] = frame->tf_rsi;
674 sa->args[4] = frame->tf_rdi;
675 sa->args[5] = frame->tf_rbp; /* Unconfirmed */
676 sa->code = frame->tf_rax;
678 if (sa->code >= p->p_sysent->sv_size)
680 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
682 sa->callp = &p->p_sysent->sv_table[sa->code];
684 td->td_retval[0] = 0;
685 td->td_retval[1] = frame->tf_rdx;
691 linux32_set_syscall_retval(struct thread *td, int error)
693 struct trapframe *frame = td->td_frame;
695 cpu_set_syscall_retval(td, error);
697 if (__predict_false(error != 0)) {
698 if (error != ERESTART && error != EJUSTRETURN)
699 frame->tf_rax = bsd_to_linux_errno(error);
704 linux32_set_fork_retval(struct thread *td)
706 struct trapframe *frame = td->td_frame;
712 * Clear registers on exec
713 * XXX copied from ia32_signal.c.
716 linux_exec_setregs(struct thread *td, struct image_params *imgp,
719 struct trapframe *regs = td->td_frame;
720 struct pcb *pcb = td->td_pcb;
721 register_t saved_rflags;
726 if (td->td_proc->p_md.md_ldt != NULL)
730 wrmsr(MSR_FSBASE, 0);
731 wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */
735 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
737 saved_rflags = regs->tf_rflags & PSL_T;
738 bzero((char *)regs, sizeof(struct trapframe));
739 regs->tf_rip = imgp->entry_addr;
740 regs->tf_rsp = stack;
741 regs->tf_rflags = PSL_USER | saved_rflags;
742 regs->tf_gs = _ugssel;
743 regs->tf_fs = _ufssel;
744 regs->tf_es = _udatasel;
745 regs->tf_ds = _udatasel;
746 regs->tf_ss = _udatasel;
747 regs->tf_flags = TF_HASSEGS;
748 regs->tf_cs = _ucode32sel;
749 regs->tf_rbx = (register_t)imgp->ps_strings;
751 x86_clear_dbregs(pcb);
755 /* Do full restore on return so that we can change to a different %cs */
756 set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
760 * XXX copied from ia32_sysvec.c.
763 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
765 int argc, envc, error;
768 uintptr_t destp, ustringp;
769 struct linux32_ps_strings *arginfo;
770 char canary[LINUX_AT_RANDOM_LEN];
773 arginfo = (struct linux32_ps_strings *)PROC_PS_STRINGS(imgp->proc);
774 destp = (uintptr_t)arginfo;
776 if (imgp->execpath != NULL && imgp->auxargs != NULL) {
777 execpath_len = strlen(imgp->execpath) + 1;
778 destp -= execpath_len;
779 destp = rounddown2(destp, sizeof(uint32_t));
780 imgp->execpathp = (void *)destp;
781 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
786 /* Prepare the canary for SSP. */
787 arc4rand(canary, sizeof(canary), 0);
788 destp -= roundup(sizeof(canary), sizeof(uint32_t));
789 imgp->canary = (void *)destp;
790 error = copyout(canary, imgp->canary, sizeof(canary));
794 /* Allocate room for the argument and environment strings. */
795 destp -= ARG_MAX - imgp->args->stringspace;
796 destp = rounddown2(destp, sizeof(uint32_t));
801 * Allocate room on the stack for the ELF auxargs
802 * array. It has LINUX_AT_COUNT entries.
804 destp -= LINUX_AT_COUNT * sizeof(Elf32_Auxinfo);
805 destp = rounddown2(destp, sizeof(uint32_t));
808 vectp = (uint32_t *)destp;
811 * Allocate room for the argv[] and env vectors including the
812 * terminating NULL pointers.
814 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
816 /* vectp also becomes our initial stack base. */
817 *stack_base = (uintptr_t)vectp;
819 stringp = imgp->args->begin_argv;
820 argc = imgp->args->argc;
821 envc = imgp->args->envc;
823 /* Copy out strings - arguments and environment. */
824 error = copyout(stringp, (void *)ustringp,
825 ARG_MAX - imgp->args->stringspace);
829 /* Fill in "ps_strings" struct for ps, w, etc. */
830 if (suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp) != 0 ||
831 suword32(&arginfo->ps_nargvstr, argc) != 0)
834 /* Fill in argument portion of vector table. */
835 for (; argc > 0; --argc) {
836 if (suword32(vectp++, ustringp) != 0)
838 while (*stringp++ != 0)
843 /* A null vector table pointer separates the argp's from the envp's. */
844 if (suword32(vectp++, 0) != 0)
847 if (suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp) != 0 ||
848 suword32(&arginfo->ps_nenvstr, envc) != 0)
851 /* Fill in environment portion of vector table. */
852 for (; envc > 0; --envc) {
853 if (suword32(vectp++, ustringp) != 0)
855 while (*stringp++ != 0)
860 /* The end of the vector table is a null pointer. */
861 if (suword32(vectp, 0) != 0)
866 error = imgp->sysent->sv_copyout_auxargs(imgp,
875 static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
876 "32-bit Linux emulation");
878 static u_long linux32_maxdsiz = LINUX32_MAXDSIZ;
879 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW,
880 &linux32_maxdsiz, 0, "");
881 static u_long linux32_maxssiz = LINUX32_MAXSSIZ;
882 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW,
883 &linux32_maxssiz, 0, "");
884 static u_long linux32_maxvmem = LINUX32_MAXVMEM;
885 SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW,
886 &linux32_maxvmem, 0, "");
887 bool linux32_emulate_i386 = false;
888 SYSCTL_BOOL(_compat_linux32, OID_AUTO, emulate_i386, CTLFLAG_RWTUN,
889 &linux32_emulate_i386, 0, "Emulate the real i386");
892 linux32_fixlimit(struct rlimit *rl, int which)
897 if (linux32_maxdsiz != 0) {
898 if (rl->rlim_cur > linux32_maxdsiz)
899 rl->rlim_cur = linux32_maxdsiz;
900 if (rl->rlim_max > linux32_maxdsiz)
901 rl->rlim_max = linux32_maxdsiz;
905 if (linux32_maxssiz != 0) {
906 if (rl->rlim_cur > linux32_maxssiz)
907 rl->rlim_cur = linux32_maxssiz;
908 if (rl->rlim_max > linux32_maxssiz)
909 rl->rlim_max = linux32_maxssiz;
913 if (linux32_maxvmem != 0) {
914 if (rl->rlim_cur > linux32_maxvmem)
915 rl->rlim_cur = linux32_maxvmem;
916 if (rl->rlim_max > linux32_maxvmem)
917 rl->rlim_max = linux32_maxvmem;
923 struct sysentvec elf_linux_sysvec = {
924 .sv_size = LINUX32_SYS_MAXSYSCALL,
925 .sv_table = linux32_sysent,
926 .sv_transtrap = linux_translate_traps,
927 .sv_fixup = linux_fixup_elf,
928 .sv_sendsig = linux_sendsig,
929 .sv_sigcode = &_binary_linux32_vdso_so_o_start,
930 .sv_szsigcode = &linux_szsigcode,
931 .sv_name = "Linux ELF32",
932 .sv_coredump = elf32_coredump,
933 .sv_elf_core_osabi = ELFOSABI_NONE,
934 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
935 .sv_elf_core_prepare_notes = linux32_prepare_notes,
936 .sv_imgact_try = linux_exec_imgact_try,
937 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
938 .sv_minuser = VM_MIN_ADDRESS,
939 .sv_maxuser = LINUX32_MAXUSER,
940 .sv_usrstack = LINUX32_USRSTACK,
941 .sv_psstrings = LINUX32_PS_STRINGS,
942 .sv_psstringssz = sizeof(struct linux32_ps_strings),
943 .sv_stackprot = VM_PROT_ALL,
944 .sv_copyout_auxargs = linux_copyout_auxargs,
945 .sv_copyout_strings = linux_copyout_strings,
946 .sv_setregs = linux_exec_setregs,
947 .sv_fixlimit = linux32_fixlimit,
948 .sv_maxssiz = &linux32_maxssiz,
949 .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP |
950 SV_SIG_DISCIGN | SV_SIG_WAITNDQ | SV_TIMEKEEP,
951 .sv_set_syscall_retval = linux32_set_syscall_retval,
952 .sv_fetch_syscall_args = linux32_fetch_syscall_args,
953 .sv_syscallnames = NULL,
954 .sv_shared_page_base = LINUX32_SHAREDPAGE,
955 .sv_shared_page_len = PAGE_SIZE,
956 .sv_schedtail = linux_schedtail,
957 .sv_thread_detach = linux_thread_detach,
959 .sv_onexec = linux_on_exec_vmspace,
960 .sv_onexit = linux_on_exit,
961 .sv_ontdexit = linux_thread_dtor,
962 .sv_setid_allowed = &linux_setid_allowed_query,
963 .sv_set_fork_retval = linux32_set_fork_retval,
967 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
971 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
972 LINUX32_VDSOPAGE_SIZE, imgp);
974 linux_on_exec(p, imgp);
979 * linux_vdso_install() and linux_exec_sysvec_init() must be called
980 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
983 linux_exec_sysvec_init(void *param)
985 l_uintptr_t *ktimekeep_base, *ktsc_selector;
986 struct sysentvec *sv;
990 /* Fill timekeep_base */
991 exec_sysvec_init(sv);
993 tkoff = kern_timekeep_base - linux_vdso_base;
994 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
995 *ktimekeep_base = sv->sv_timekeep_base;
997 tkoff = kern_tsc_selector - linux_vdso_base;
998 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
999 *ktsc_selector = linux_vdso_tsc_selector_idx();
1001 printf("Linux i386 vDSO tsc_selector: %u\n", *ktsc_selector);
1003 tkoff = kern_cpu_selector - linux_vdso_base;
1004 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
1005 *ktsc_selector = linux_vdso_cpu_selector_idx();
1007 printf("Linux i386 vDSO cpu_selector: %u\n", *ktsc_selector);
1009 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
1010 linux_exec_sysvec_init, &elf_linux_sysvec);
1013 linux_vdso_install(const void *param)
1015 char *vdso_start = &_binary_linux32_vdso_so_o_start;
1016 char *vdso_end = &_binary_linux32_vdso_so_o_end;
1018 linux_szsigcode = vdso_end - vdso_start;
1019 MPASS(linux_szsigcode <= LINUX32_VDSOPAGE_SIZE);
1021 linux_vdso_base = LINUX32_VDSOPAGE;
1023 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
1025 linux_vdso_obj = __elfN(linux_shared_page_init)
1026 (&linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
1027 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
1029 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
1031 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
1032 linux_vdso_install, NULL);
1035 linux_vdso_deinstall(const void *param)
1038 __elfN(linux_shared_page_fini)(linux_vdso_obj,
1039 linux_vdso_mapping, LINUX32_VDSOPAGE_SIZE);
1041 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
1042 linux_vdso_deinstall, NULL);
1045 linux_vdso_reloc(char *mapping, Elf_Addr offset)
1047 const Elf_Shdr *shdr;
1049 const Elf_Ehdr *ehdr;
1051 Elf_Size rtype, symidx;
1052 Elf32_Addr addr, addend;
1058 ehdr = (const Elf_Ehdr *)mapping;
1059 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
1060 for (i = 0; i < ehdr->e_shnum; i++)
1062 switch (shdr[i].sh_type) {
1064 rel = (const Elf_Rel *)(mapping + shdr[i].sh_offset);
1065 relcnt = shdr[i].sh_size / sizeof(*rel);
1068 printf("Linux i386 vDSO: unexpected Rela section\n");
1073 for (i = 0; i < relcnt; i++, rel++) {
1074 where = (Elf32_Addr *)(mapping + rel->r_offset);
1076 rtype = ELF_R_TYPE(rel->r_info);
1077 symidx = ELF_R_SYM(rel->r_info);
1080 case R_386_NONE: /* none */
1083 case R_386_RELATIVE: /* B + A */
1084 addr = (Elf32_Addr)PTROUT(offset + addend);
1089 case R_386_IRELATIVE:
1090 printf("Linux i386 vDSO: unexpected ifunc relocation, "
1091 "symbol index %ld\n", (intmax_t)symidx);
1094 printf("Linux i386 vDSO: unexpected relocation type %ld, "
1095 "symbol index %ld\n", (intmax_t)rtype, (intmax_t)symidx);
1100 static char GNU_ABI_VENDOR[] = "GNU";
1101 static int GNULINUX_ABI_DESC = 0;
1104 linux32_trans_osrel(const Elf_Note *note, int32_t *osrel)
1106 const Elf32_Word *desc;
1109 p = (uintptr_t)(note + 1);
1110 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1112 desc = (const Elf32_Word *)p;
1113 if (desc[0] != GNULINUX_ABI_DESC)
1117 * For Linux we encode osrel using the Linux convention of
1118 * (version << 16) | (major << 8) | (minor)
1119 * See macro in linux_mib.h
1121 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
1126 static Elf_Brandnote linux32_brandnote = {
1127 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
1128 .hdr.n_descsz = 16, /* XXX at least 16 */
1130 .vendor = GNU_ABI_VENDOR,
1131 .flags = BN_TRANSLATE_OSREL,
1132 .trans_osrel = linux32_trans_osrel
1135 static Elf32_Brandinfo linux_brand = {
1136 .brand = ELFOSABI_LINUX,
1138 .compat_3_brand = "Linux",
1139 .emul_path = linux_emul_path,
1140 .interp_path = "/lib/ld-linux.so.1",
1141 .sysvec = &elf_linux_sysvec,
1142 .interp_newpath = NULL,
1143 .brand_note = &linux32_brandnote,
1144 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1147 static Elf32_Brandinfo linux_glibc2brand = {
1148 .brand = ELFOSABI_LINUX,
1150 .compat_3_brand = "Linux",
1151 .emul_path = linux_emul_path,
1152 .interp_path = "/lib/ld-linux.so.2",
1153 .sysvec = &elf_linux_sysvec,
1154 .interp_newpath = NULL,
1155 .brand_note = &linux32_brandnote,
1156 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
1159 static Elf32_Brandinfo linux_muslbrand = {
1160 .brand = ELFOSABI_LINUX,
1162 .compat_3_brand = "Linux",
1163 .emul_path = linux_emul_path,
1164 .interp_path = "/lib/ld-musl-i386.so.1",
1165 .sysvec = &elf_linux_sysvec,
1166 .interp_newpath = NULL,
1167 .brand_note = &linux32_brandnote,
1168 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
1169 LINUX_BI_FUTEX_REQUEUE
1172 Elf32_Brandinfo *linux_brandlist[] = {
1180 linux_elf_modevent(module_t mod, int type, void *data)
1182 Elf32_Brandinfo **brandinfo;
1184 struct linux_ioctl_handler **lihp;
1190 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1192 if (elf32_insert_brand_entry(*brandinfo) < 0)
1195 SET_FOREACH(lihp, linux_ioctl_handler_set)
1196 linux32_ioctl_register_handler(*lihp);
1197 stclohz = (stathz ? stathz : hz);
1199 printf("Linux i386 ELF exec handler installed\n");
1201 printf("cannot insert Linux i386 ELF brand handler\n");
1204 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1206 if (elf32_brand_inuse(*brandinfo))
1209 for (brandinfo = &linux_brandlist[0];
1210 *brandinfo != NULL; ++brandinfo)
1211 if (elf32_remove_brand_entry(*brandinfo) < 0)
1215 SET_FOREACH(lihp, linux_ioctl_handler_set)
1216 linux32_ioctl_unregister_handler(*lihp);
1218 printf("Linux i386 ELF exec handler removed\n");
1220 printf("Could not deinstall Linux i386 ELF interpreter entry\n");
1223 return (EOPNOTSUPP);
1228 static moduledata_t linux_elf_mod = {
1234 DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1235 MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1);
1236 FEATURE(linux, "Linux 32bit support");