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
8 * Copyright (c) 2013, 2021 Dmitry Chagin <dchagin@FreeBSD.org>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #define __ELF_WORD_SIZE 64
39 #include <sys/param.h>
40 #include <sys/systm.h>
42 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/module.h>
50 #include <sys/mutex.h>
52 #include <sys/resourcevar.h>
53 #include <sys/stddef.h>
54 #include <sys/signalvar.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysent.h>
58 #include <sys/sysproto.h>
59 #include <sys/vnode.h>
60 #include <sys/eventhandler.h>
64 #include <vm/vm_extern.h>
65 #include <vm/vm_map.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_param.h>
70 #include <machine/cpu.h>
71 #include <machine/md_var.h>
72 #include <machine/pcb.h>
73 #include <machine/specialreg.h>
74 #include <machine/trap.h>
76 #include <x86/linux/linux_x86.h>
77 #include <amd64/linux/linux.h>
78 #include <amd64/linux/linux_proto.h>
79 #include <compat/linux/linux_emul.h>
80 #include <compat/linux/linux_ioctl.h>
81 #include <compat/linux/linux_mib.h>
82 #include <compat/linux/linux_misc.h>
83 #include <compat/linux/linux_signal.h>
84 #include <compat/linux/linux_sysproto.h>
85 #include <compat/linux/linux_util.h>
86 #include <compat/linux/linux_vdso.h>
88 MODULE_VERSION(linux64, 1);
90 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2
91 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \
93 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE)
95 * PAGE_SIZE - the size
96 * of the native SHAREDPAGE
98 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48
99 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \
100 sizeof(struct ps_strings))
102 static int linux_szsigcode;
103 static vm_object_t linux_vdso_obj;
104 static char *linux_vdso_mapping;
105 extern char _binary_linux_vdso_so_o_start;
106 extern char _binary_linux_vdso_so_o_end;
107 static vm_offset_t linux_vdso_base;
109 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
111 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
113 static int linux_copyout_strings(struct image_params *imgp,
114 uintptr_t *stack_base);
115 static int linux_fixup_elf(uintptr_t *stack_base,
116 struct image_params *iparams);
117 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
118 static void linux_vdso_install(const void *param);
119 static void linux_vdso_deinstall(const void *param);
120 static void linux_vdso_reloc(char *mapping, Elf_Addr offset);
121 static void linux_set_syscall_retval(struct thread *td, int error);
122 static int linux_fetch_syscall_args(struct thread *td);
123 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
125 static void linux_exec_sysvec_init(void *param);
126 static int linux_on_exec_vmspace(struct proc *p,
127 struct image_params *imgp);
128 static int linux_vsyscall(struct thread *td);
130 #define LINUX_T_UNKNOWN 255
131 static int _bsd_to_linux_trapcode[] = {
132 LINUX_T_UNKNOWN, /* 0 */
133 6, /* 1 T_PRIVINFLT */
134 LINUX_T_UNKNOWN, /* 2 */
136 LINUX_T_UNKNOWN, /* 4 */
137 LINUX_T_UNKNOWN, /* 5 */
138 16, /* 6 T_ARITHTRAP */
139 254, /* 7 T_ASTFLT */
140 LINUX_T_UNKNOWN, /* 8 */
141 13, /* 9 T_PROTFLT */
142 1, /* 10 T_TRCTRAP */
143 LINUX_T_UNKNOWN, /* 11 */
144 14, /* 12 T_PAGEFLT */
145 LINUX_T_UNKNOWN, /* 13 */
146 17, /* 14 T_ALIGNFLT */
147 LINUX_T_UNKNOWN, /* 15 */
148 LINUX_T_UNKNOWN, /* 16 */
149 LINUX_T_UNKNOWN, /* 17 */
155 8, /* 23 T_DOUBLEFLT */
156 9, /* 24 T_FPOPFLT */
157 10, /* 25 T_TSSFLT */
158 11, /* 26 T_SEGNPFLT */
159 12, /* 27 T_STKFLT */
161 19, /* 29 T_XMMFLT */
162 15 /* 30 T_RESERVED */
164 #define bsd_to_linux_trapcode(code) \
165 ((code)<nitems(_bsd_to_linux_trapcode)? \
166 _bsd_to_linux_trapcode[(code)]: \
169 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
170 LINUX_VDSO_SYM_CHAR(linux_platform);
171 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
172 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
175 * If FreeBSD & Linux have a difference of opinion about what a trap
176 * means, deal with it here.
181 linux_translate_traps(int signal, int trap_code)
184 if (signal != SIGBUS)
198 linux_fetch_syscall_args(struct thread *td)
201 struct trapframe *frame;
202 struct syscall_args *sa;
205 frame = td->td_frame;
208 sa->args[0] = frame->tf_rdi;
209 sa->args[1] = frame->tf_rsi;
210 sa->args[2] = frame->tf_rdx;
211 sa->args[3] = frame->tf_rcx;
212 sa->args[4] = frame->tf_r8;
213 sa->args[5] = frame->tf_r9;
214 sa->code = frame->tf_rax;
216 if (sa->code >= p->p_sysent->sv_size)
218 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
220 sa->callp = &p->p_sysent->sv_table[sa->code];
222 td->td_retval[0] = 0;
227 linux_set_syscall_retval(struct thread *td, int error)
229 struct trapframe *frame;
231 frame = td->td_frame;
235 frame->tf_rax = td->td_retval[0];
236 frame->tf_r10 = frame->tf_rcx;
241 * Reconstruct pc, we know that 'syscall' is 2 bytes,
242 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
243 * We saved this in tf_err.
246 frame->tf_rip -= frame->tf_err;
247 frame->tf_r10 = frame->tf_rcx;
254 frame->tf_rax = bsd_to_linux_errno(error);
255 frame->tf_r10 = frame->tf_rcx;
260 * Differently from FreeBSD native ABI, on Linux only %rcx
261 * and %r11 values are not preserved across the syscall.
262 * Require full context restore to get all registers except
263 * those two restored at return to usermode.
265 * XXX: Would be great to be able to avoid PCB_FULL_IRET
266 * for the error == 0 case.
268 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
272 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
275 Elf_Auxinfo *argarray, *pos;
277 int error, issetugid;
280 args = (Elf64_Auxargs *)imgp->auxargs;
281 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
284 issetugid = p->p_flag & P_SUGID ? 1 : 0;
285 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
286 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
287 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
288 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
289 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
290 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
291 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
292 AUXARGS_ENTRY(pos, AT_BASE, args->base);
293 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
294 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
295 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
296 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
297 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
298 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
299 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
300 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
301 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
302 if (imgp->execpathp != 0)
303 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
304 if (args->execfd != -1)
305 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
306 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
307 AUXARGS_ENTRY(pos, AT_NULL, 0);
309 free(imgp->auxargs, M_TEMP);
310 imgp->auxargs = NULL;
311 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
313 error = copyout(argarray, (void *)base,
314 sizeof(*argarray) * LINUX_AT_COUNT);
315 free(argarray, M_TEMP);
320 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
324 base = (Elf64_Addr *)*stack_base;
326 if (suword(base, (uint64_t)imgp->args->argc) == -1)
329 *stack_base = (uintptr_t)base;
334 * Copy strings out to the new process address space, constructing new arg
335 * and env vector tables. Return a pointer to the base so that it can be used
336 * as the initial stack pointer.
339 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
341 int argc, envc, error;
344 uintptr_t destp, ustringp;
345 struct ps_strings *arginfo;
346 char canary[LINUX_AT_RANDOM_LEN];
351 arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
352 destp = (uintptr_t)arginfo;
354 if (imgp->execpath != NULL && imgp->auxargs != NULL) {
355 execpath_len = strlen(imgp->execpath) + 1;
356 destp -= execpath_len;
357 destp = rounddown2(destp, sizeof(void *));
358 imgp->execpathp = (void *)destp;
359 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
364 /* Prepare the canary for SSP. */
365 arc4rand(canary, sizeof(canary), 0);
366 destp -= roundup(sizeof(canary), sizeof(void *));
367 imgp->canary = (void *)destp;
368 error = copyout(canary, imgp->canary, sizeof(canary));
372 /* Allocate room for the argument and environment strings. */
373 destp -= ARG_MAX - imgp->args->stringspace;
374 destp = rounddown2(destp, sizeof(void *));
379 * Allocate room on the stack for the ELF auxargs
380 * array. It has LINUX_AT_COUNT entries.
382 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
383 destp = rounddown2(destp, sizeof(void *));
386 vectp = (char **)destp;
389 * Allocate room for the argv[] and env vectors including the
390 * terminating NULL pointers.
392 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
395 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
396 * One "long argc" will be prepended later.
398 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
400 /* vectp also becomes our initial stack base. */
401 *stack_base = (uintptr_t)vectp;
403 stringp = imgp->args->begin_argv;
404 argc = imgp->args->argc;
405 envc = imgp->args->envc;
407 /* Copy out strings - arguments and environment. */
408 error = copyout(stringp, (void *)ustringp,
409 ARG_MAX - imgp->args->stringspace);
413 /* Fill in "ps_strings" struct for ps, w, etc. */
414 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
415 suword(&arginfo->ps_nargvstr, argc) != 0)
418 /* Fill in argument portion of vector table. */
419 for (; argc > 0; --argc) {
420 if (suword(vectp++, ustringp) != 0)
422 while (*stringp++ != 0)
427 /* A null vector table pointer separates the argp's from the envp's. */
428 if (suword(vectp++, 0) != 0)
431 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
432 suword(&arginfo->ps_nenvstr, envc) != 0)
435 /* Fill in environment portion of vector table. */
436 for (; envc > 0; --envc) {
437 if (suword(vectp++, ustringp) != 0)
439 while (*stringp++ != 0)
444 /* The end of the vector table is a null pointer. */
445 if (suword(vectp, 0) != 0)
450 error = imgp->sysent->sv_copyout_auxargs(imgp,
460 * Reset registers to default values on exec.
463 linux_exec_setregs(struct thread *td, struct image_params *imgp,
466 struct trapframe *regs;
468 register_t saved_rflags;
473 if (td->td_proc->p_md.md_ldt != NULL)
478 clear_pcb_flags(pcb, PCB_32BIT);
479 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
480 set_pcb_flags(pcb, PCB_FULL_IRET);
482 saved_rflags = regs->tf_rflags & PSL_T;
483 bzero((char *)regs, sizeof(struct trapframe));
484 regs->tf_rip = imgp->entry_addr;
485 regs->tf_rsp = stack;
486 regs->tf_rflags = PSL_USER | saved_rflags;
487 regs->tf_ss = _udatasel;
488 regs->tf_cs = _ucodesel;
489 regs->tf_ds = _udatasel;
490 regs->tf_es = _udatasel;
491 regs->tf_fs = _ufssel;
492 regs->tf_gs = _ugssel;
493 regs->tf_flags = TF_HASSEGS;
495 x86_clear_dbregs(pcb);
498 * Drop the FP state if we hold it, so that the process gets a
499 * clean FP state if it uses the FPU again.
505 * Copied from amd64/amd64/machdep.c
507 * XXX fpu state need? don't think so
510 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
513 struct l_ucontext uc;
514 struct l_sigcontext *context;
515 struct trapframe *regs;
516 unsigned long rflags;
521 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
526 context = &uc.uc_mcontext;
527 rflags = context->sc_rflags;
530 * Don't allow users to change privileged or reserved flags.
533 * XXX do allow users to change the privileged flag PSL_RF.
534 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
535 * should sometimes set it there too. tf_rflags is kept in
536 * the signal context during signal handling and there is no
537 * other place to remember it, so the PSL_RF bit may be
538 * corrupted by the signal handler without us knowing.
539 * Corruption of the PSL_RF bit at worst causes one more or
540 * one less debugger trap, so allowing it is fairly harmless.
542 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
543 uprintf("pid %d comm %s linux mangled rflags %#lx\n",
544 p->p_pid, p->p_comm, rflags);
549 * Don't allow users to load a valid privileged %cs. Let the
550 * hardware check for invalid selectors, excess privilege in
551 * other selectors, invalid %eip's and invalid %esp's.
553 if (!CS_SECURE(context->sc_cs)) {
554 uprintf("pid %d comm %s linux mangled cs %#x\n",
555 p->p_pid, p->p_comm, context->sc_cs);
556 ksiginfo_init_trap(&ksi);
557 ksi.ksi_signo = SIGBUS;
558 ksi.ksi_code = BUS_OBJERR;
559 ksi.ksi_trapno = T_PROTFLT;
560 ksi.ksi_addr = (void *)regs->tf_rip;
561 trapsignal(td, &ksi);
566 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
567 SIG_CANTMASK(td->td_sigmask);
571 regs->tf_rdi = context->sc_rdi;
572 regs->tf_rsi = context->sc_rsi;
573 regs->tf_rdx = context->sc_rdx;
574 regs->tf_rbp = context->sc_rbp;
575 regs->tf_rbx = context->sc_rbx;
576 regs->tf_rcx = context->sc_rcx;
577 regs->tf_rax = context->sc_rax;
578 regs->tf_rip = context->sc_rip;
579 regs->tf_rsp = context->sc_rsp;
580 regs->tf_r8 = context->sc_r8;
581 regs->tf_r9 = context->sc_r9;
582 regs->tf_r10 = context->sc_r10;
583 regs->tf_r11 = context->sc_r11;
584 regs->tf_r12 = context->sc_r12;
585 regs->tf_r13 = context->sc_r13;
586 regs->tf_r14 = context->sc_r14;
587 regs->tf_r15 = context->sc_r15;
588 regs->tf_cs = context->sc_cs;
589 regs->tf_err = context->sc_err;
590 regs->tf_rflags = rflags;
592 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
593 return (EJUSTRETURN);
597 * copied from amd64/amd64/machdep.c
599 * Send an interrupt to process.
602 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
604 struct l_rt_sigframe sf, *sfp;
609 struct trapframe *regs;
615 PROC_LOCK_ASSERT(p, MA_OWNED);
616 sig = ksi->ksi_signo;
618 code = ksi->ksi_code;
619 mtx_assert(&psp->ps_mtx, MA_OWNED);
621 oonstack = sigonstack(regs->tf_rsp);
623 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
624 catcher, sig, mask, code);
626 /* Save user context. */
627 bzero(&sf, sizeof(sf));
628 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
629 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
631 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
632 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
633 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
634 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
636 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi;
637 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi;
638 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx;
639 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp;
640 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx;
641 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx;
642 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax;
643 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip;
644 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp;
645 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8;
646 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9;
647 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10;
648 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11;
649 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12;
650 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13;
651 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14;
652 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15;
653 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
654 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
655 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err;
656 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
657 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
659 /* Allocate space for the signal handler context. */
660 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
661 SIGISMEMBER(psp->ps_sigonstack, sig)) {
662 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
663 sizeof(struct l_rt_sigframe);
665 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
666 /* Align to 16 bytes. */
667 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
669 /* Translate the signal. */
670 sig = bsd_to_linux_signal(sig);
672 /* Build the argument list for the signal handler. */
673 regs->tf_rdi = sig; /* arg 1 in %rdi */
675 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
676 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */
678 /* Fill in POSIX parts. */
679 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig);
680 sf.sf_handler = catcher;
682 mtx_unlock(&psp->ps_mtx);
685 /* Copy the sigframe out to the user's stack. */
686 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
691 regs->tf_rsp = (long)sfp;
692 regs->tf_rip = linux_rt_sigcode;
693 regs->tf_rflags &= ~(PSL_T | PSL_D);
694 regs->tf_cs = _ucodesel;
695 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
697 mtx_lock(&psp->ps_mtx);
700 #define LINUX_VSYSCALL_START (-10UL << 20)
701 #define LINUX_VSYSCALL_SZ 1024
703 const unsigned long linux_vsyscall_vector[] = {
704 LINUX_SYS_gettimeofday,
705 LINUX_SYS_linux_time,
706 LINUX_SYS_linux_getcpu,
710 linux_vsyscall(struct thread *td)
712 struct trapframe *frame;
717 frame = td->td_frame;
719 /* Check %rip for vsyscall area. */
720 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
722 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
724 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
725 if (code >= nitems(linux_vsyscall_vector))
729 * vsyscall called as callq *(%rax), so we must
730 * use return address from %rsp and also fixup %rsp.
732 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
736 frame->tf_rip = retqaddr;
737 frame->tf_rax = linux_vsyscall_vector[code];
740 traced = (frame->tf_flags & PSL_T);
742 amd64_syscall(td, traced);
747 struct sysentvec elf_linux_sysvec = {
748 .sv_size = LINUX_SYS_MAXSYSCALL,
749 .sv_table = linux_sysent,
750 .sv_transtrap = linux_translate_traps,
751 .sv_fixup = linux_fixup_elf,
752 .sv_sendsig = linux_rt_sendsig,
753 .sv_sigcode = &_binary_linux_vdso_so_o_start,
754 .sv_szsigcode = &linux_szsigcode,
755 .sv_name = "Linux ELF64",
756 .sv_coredump = elf64_coredump,
757 .sv_elf_core_osabi = ELFOSABI_NONE,
758 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
759 .sv_elf_core_prepare_notes = linux64_prepare_notes,
760 .sv_imgact_try = linux_exec_imgact_try,
761 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
762 .sv_minuser = VM_MIN_ADDRESS,
763 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48,
764 .sv_usrstack = LINUX_USRSTACK_LA48,
765 .sv_psstrings = LINUX_PS_STRINGS_LA48,
766 .sv_psstringssz = sizeof(struct ps_strings),
767 .sv_stackprot = VM_PROT_ALL,
768 .sv_copyout_auxargs = linux_copyout_auxargs,
769 .sv_copyout_strings = linux_copyout_strings,
770 .sv_setregs = linux_exec_setregs,
773 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
774 SV_SIG_WAITNDQ | SV_TIMEKEEP,
775 .sv_set_syscall_retval = linux_set_syscall_retval,
776 .sv_fetch_syscall_args = linux_fetch_syscall_args,
777 .sv_syscallnames = NULL,
778 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48,
779 .sv_shared_page_len = PAGE_SIZE,
780 .sv_schedtail = linux_schedtail,
781 .sv_thread_detach = linux_thread_detach,
782 .sv_trap = linux_vsyscall,
783 .sv_onexec = linux_on_exec_vmspace,
784 .sv_onexit = linux_on_exit,
785 .sv_ontdexit = linux_thread_dtor,
786 .sv_setid_allowed = &linux_setid_allowed_query,
790 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
794 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
795 LINUX_VDSOPAGE_SIZE, imgp);
797 linux_on_exec(p, imgp);
802 * linux_vdso_install() and linux_exec_sysvec_init() must be called
803 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
806 linux_exec_sysvec_init(void *param)
808 l_uintptr_t *ktimekeep_base, *ktsc_selector;
809 struct sysentvec *sv;
813 amd64_lower_shared_page(sv);
814 /* Fill timekeep_base */
815 exec_sysvec_init(sv);
817 tkoff = kern_timekeep_base - linux_vdso_base;
818 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
819 *ktimekeep_base = sv->sv_timekeep_base;
821 tkoff = kern_tsc_selector - linux_vdso_base;
822 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
823 *ktsc_selector = linux_vdso_tsc_selector_idx();
825 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector);
827 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
828 linux_exec_sysvec_init, &elf_linux_sysvec);
831 linux_vdso_install(const void *param)
833 char *vdso_start = &_binary_linux_vdso_so_o_start;
834 char *vdso_end = &_binary_linux_vdso_so_o_end;
836 linux_szsigcode = vdso_end - vdso_start;
837 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
839 linux_vdso_base = LINUX_VDSOPAGE_LA48;
840 if (hw_lower_amd64_sharedpage != 0)
841 linux_vdso_base -= PAGE_SIZE;
843 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
845 linux_vdso_obj = __elfN(linux_shared_page_init)
846 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
847 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
849 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
851 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
852 linux_vdso_install, NULL);
855 linux_vdso_deinstall(const void *param)
858 __elfN(linux_shared_page_fini)(linux_vdso_obj,
859 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
861 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
862 linux_vdso_deinstall, NULL);
865 linux_vdso_reloc(char *mapping, Elf_Addr offset)
867 const Elf_Ehdr *ehdr;
868 const Elf_Shdr *shdr;
869 Elf64_Addr *where, val;
870 Elf_Size rtype, symidx;
871 const Elf_Rela *rela;
872 Elf_Addr addr, addend;
879 ehdr = (const Elf_Ehdr *)mapping;
880 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
881 for (i = 0; i < ehdr->e_shnum; i++)
883 switch (shdr[i].sh_type) {
885 printf("Linux x86_64 vDSO: unexpected Rel section\n");
888 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
889 relacnt = shdr[i].sh_size / sizeof(*rela);
893 for (j = 0; j < relacnt; j++, rela++) {
894 where = (Elf_Addr *)(mapping + rela->r_offset);
895 addend = rela->r_addend;
896 rtype = ELF_R_TYPE(rela->r_info);
897 symidx = ELF_R_SYM(rela->r_info);
900 case R_X86_64_NONE: /* none */
903 case R_X86_64_RELATIVE: /* B + A */
904 addr = (Elf_Addr)(offset + addend);
909 case R_X86_64_IRELATIVE:
910 printf("Linux x86_64 vDSO: unexpected ifunc relocation, "
911 "symbol index %ld\n", symidx);
914 printf("Linux x86_64 vDSO: unexpected relocation type %ld, "
915 "symbol index %ld\n", rtype, symidx);
920 static char GNULINUX_ABI_VENDOR[] = "GNU";
921 static int GNULINUX_ABI_DESC = 0;
924 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
926 const Elf32_Word *desc;
929 p = (uintptr_t)(note + 1);
930 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
932 desc = (const Elf32_Word *)p;
933 if (desc[0] != GNULINUX_ABI_DESC)
937 * For Linux we encode osrel using the Linux convention of
938 * (version << 16) | (major << 8) | (minor)
939 * See macro in linux_mib.h
941 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
946 static Elf_Brandnote linux64_brandnote = {
947 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR),
950 .vendor = GNULINUX_ABI_VENDOR,
951 .flags = BN_TRANSLATE_OSREL,
952 .trans_osrel = linux_trans_osrel
955 static Elf64_Brandinfo linux_glibc2brand = {
956 .brand = ELFOSABI_LINUX,
957 .machine = EM_X86_64,
958 .compat_3_brand = "Linux",
959 .emul_path = linux_emul_path,
960 .interp_path = "/lib64/ld-linux-x86-64.so.2",
961 .sysvec = &elf_linux_sysvec,
962 .interp_newpath = NULL,
963 .brand_note = &linux64_brandnote,
964 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
967 static Elf64_Brandinfo linux_glibc2brandshort = {
968 .brand = ELFOSABI_LINUX,
969 .machine = EM_X86_64,
970 .compat_3_brand = "Linux",
971 .emul_path = linux_emul_path,
972 .interp_path = "/lib64/ld-linux.so.2",
973 .sysvec = &elf_linux_sysvec,
974 .interp_newpath = NULL,
975 .brand_note = &linux64_brandnote,
976 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
979 static Elf64_Brandinfo linux_muslbrand = {
980 .brand = ELFOSABI_LINUX,
981 .machine = EM_X86_64,
982 .compat_3_brand = "Linux",
983 .emul_path = linux_emul_path,
984 .interp_path = "/lib/ld-musl-x86_64.so.1",
985 .sysvec = &elf_linux_sysvec,
986 .interp_newpath = NULL,
987 .brand_note = &linux64_brandnote,
988 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
991 Elf64_Brandinfo *linux_brandlist[] = {
993 &linux_glibc2brandshort,
999 linux64_elf_modevent(module_t mod, int type, void *data)
1001 Elf64_Brandinfo **brandinfo;
1003 struct linux_ioctl_handler **lihp;
1009 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1011 if (elf64_insert_brand_entry(*brandinfo) < 0)
1014 SET_FOREACH(lihp, linux_ioctl_handler_set)
1015 linux_ioctl_register_handler(*lihp);
1016 stclohz = (stathz ? stathz : hz);
1018 printf("Linux x86-64 ELF exec handler installed\n");
1020 printf("cannot insert Linux x86-64 ELF brand handler\n");
1023 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
1025 if (elf64_brand_inuse(*brandinfo))
1028 for (brandinfo = &linux_brandlist[0];
1029 *brandinfo != NULL; ++brandinfo)
1030 if (elf64_remove_brand_entry(*brandinfo) < 0)
1034 SET_FOREACH(lihp, linux_ioctl_handler_set)
1035 linux_ioctl_unregister_handler(*lihp);
1037 printf("Linux x86_64 ELF exec handler removed\n");
1039 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n");
1042 return (EOPNOTSUPP);
1047 static moduledata_t linux64_elf_mod = {
1049 linux64_elf_modevent,
1053 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1054 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
1055 FEATURE(linux64, "Linux 64bit support");