1 /*- 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 7 * All rights reserved. 8 * Copyright (c) 2013, 2021 Dmitry Chagin <[email protected]> 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 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 21 * 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. 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #define __ELF_WORD_SIZE 64 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/exec.h> 42 #include <sys/fcntl.h> 43 #include <sys/imgact.h> 44 #include <sys/imgact_elf.h> 45 #include <sys/kernel.h> 46 #include <sys/ktr.h> 47 #include <sys/lock.h> 48 #include <sys/malloc.h> 49 #include <sys/module.h> 50 #include <sys/mutex.h> 51 #include <sys/proc.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> 61 62 #include <vm/vm.h> 63 #include <vm/pmap.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> 69 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> 75 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_elf.h> 80 #include <compat/linux/linux_emul.h> 81 #include <compat/linux/linux_fork.h> 82 #include <compat/linux/linux_ioctl.h> 83 #include <compat/linux/linux_mib.h> 84 #include <compat/linux/linux_misc.h> 85 #include <compat/linux/linux_signal.h> 86 #include <compat/linux/linux_sysproto.h> 87 #include <compat/linux/linux_util.h> 88 #include <compat/linux/linux_vdso.h> 89 90 #include <x86/linux/linux_x86_sigframe.h> 91 92 MODULE_VERSION(linux64, 1); 93 94 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 95 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \ 96 LINUX_VDSOPAGE_SIZE) 97 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE) 98 /* 99 * PAGE_SIZE - the size 100 * of the native SHAREDPAGE 101 */ 102 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48 103 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \ 104 sizeof(struct ps_strings)) 105 106 static int linux_szsigcode; 107 static vm_object_t linux_vdso_obj; 108 static char *linux_vdso_mapping; 109 extern char _binary_linux_vdso_so_o_start; 110 extern char _binary_linux_vdso_so_o_end; 111 static vm_offset_t linux_vdso_base; 112 113 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 114 extern const char *linux_syscallnames[]; 115 116 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 117 118 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, 124 uintptr_t stack); 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 void linux_set_fork_retval(struct thread *td); 129 static int linux_vsyscall(struct thread *td); 130 131 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 132 LINUX_VDSO_SYM_CHAR(linux_platform); 133 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); 134 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector); 135 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector); 136 137 static int 138 linux_fetch_syscall_args(struct thread *td) 139 { 140 struct proc *p; 141 struct trapframe *frame; 142 struct syscall_args *sa; 143 144 p = td->td_proc; 145 frame = td->td_frame; 146 sa = &td->td_sa; 147 148 sa->args[0] = frame->tf_rdi; 149 sa->args[1] = frame->tf_rsi; 150 sa->args[2] = frame->tf_rdx; 151 sa->args[3] = frame->tf_rcx; 152 sa->args[4] = frame->tf_r8; 153 sa->args[5] = frame->tf_r9; 154 sa->code = frame->tf_rax; 155 sa->original_code = sa->code; 156 157 if (sa->code >= p->p_sysent->sv_size) 158 /* nosys */ 159 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 160 else 161 sa->callp = &p->p_sysent->sv_table[sa->code]; 162 163 td->td_retval[0] = 0; 164 return (0); 165 } 166 167 static void 168 linux_set_syscall_retval(struct thread *td, int error) 169 { 170 struct trapframe *frame; 171 172 frame = td->td_frame; 173 174 switch (error) { 175 case 0: 176 frame->tf_rax = td->td_retval[0]; 177 frame->tf_r10 = frame->tf_rcx; 178 break; 179 180 case ERESTART: 181 /* 182 * Reconstruct pc, we know that 'syscall' is 2 bytes, 183 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. 184 * We saved this in tf_err. 185 * 186 */ 187 frame->tf_rip -= frame->tf_err; 188 frame->tf_r10 = frame->tf_rcx; 189 break; 190 191 case EJUSTRETURN: 192 break; 193 194 default: 195 frame->tf_rax = bsd_to_linux_errno(error); 196 frame->tf_r10 = frame->tf_rcx; 197 break; 198 } 199 200 /* 201 * Differently from FreeBSD native ABI, on Linux only %rcx 202 * and %r11 values are not preserved across the syscall. 203 * Require full context restore to get all registers except 204 * those two restored at return to usermode. 205 * 206 * XXX: Would be great to be able to avoid PCB_FULL_IRET 207 * for the error == 0 case. 208 */ 209 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 210 } 211 212 static void 213 linux_set_fork_retval(struct thread *td) 214 { 215 struct trapframe *frame = td->td_frame; 216 217 frame->tf_rax = 0; 218 } 219 220 static int 221 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base) 222 { 223 Elf_Auxargs *args; 224 Elf_Auxinfo *argarray, *pos; 225 struct proc *p; 226 int error, issetugid; 227 228 p = imgp->proc; 229 args = (Elf64_Auxargs *)imgp->auxargs; 230 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP, 231 M_WAITOK | M_ZERO); 232 233 issetugid = p->p_flag & P_SUGID ? 1 : 0; 234 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base); 235 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 236 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 237 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 238 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 239 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 240 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 241 AUXARGS_ENTRY(pos, AT_BASE, args->base); 242 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 243 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 244 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 245 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 246 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 247 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 248 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 249 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary); 250 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0); 251 if (imgp->execpathp != 0) 252 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp); 253 if (args->execfd != -1) 254 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 255 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 256 AUXARGS_ENTRY(pos, AT_NULL, 0); 257 258 free(imgp->auxargs, M_TEMP); 259 imgp->auxargs = NULL; 260 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs")); 261 262 error = copyout(argarray, (void *)base, 263 sizeof(*argarray) * LINUX_AT_COUNT); 264 free(argarray, M_TEMP); 265 return (error); 266 } 267 268 /* 269 * Reset registers to default values on exec. 270 */ 271 static void 272 linux_exec_setregs(struct thread *td, struct image_params *imgp, 273 uintptr_t stack) 274 { 275 struct trapframe *regs; 276 struct pcb *pcb; 277 register_t saved_rflags; 278 279 regs = td->td_frame; 280 pcb = td->td_pcb; 281 282 if (td->td_proc->p_md.md_ldt != NULL) 283 user_ldt_free(td); 284 285 pcb->pcb_fsbase = 0; 286 pcb->pcb_gsbase = 0; 287 clear_pcb_flags(pcb, PCB_32BIT); 288 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 289 set_pcb_flags(pcb, PCB_FULL_IRET); 290 291 saved_rflags = regs->tf_rflags & PSL_T; 292 bzero((char *)regs, sizeof(struct trapframe)); 293 regs->tf_rip = imgp->entry_addr; 294 regs->tf_rsp = stack; 295 regs->tf_rflags = PSL_USER | saved_rflags; 296 regs->tf_ss = _udatasel; 297 regs->tf_cs = _ucodesel; 298 regs->tf_ds = _udatasel; 299 regs->tf_es = _udatasel; 300 regs->tf_fs = _ufssel; 301 regs->tf_gs = _ugssel; 302 regs->tf_flags = TF_HASSEGS; 303 304 x86_clear_dbregs(pcb); 305 306 /* 307 * Drop the FP state if we hold it, so that the process gets a 308 * clean FP state if it uses the FPU again. 309 */ 310 fpstate_drop(td); 311 } 312 313 /* 314 * Copied from amd64/amd64/machdep.c 315 * 316 * XXX fpu state need? don't think so 317 */ 318 int 319 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 320 { 321 struct proc *p; 322 struct l_rt_sigframe sf; 323 struct l_sigcontext *context; 324 struct trapframe *regs; 325 mcontext_t mc; 326 unsigned long rflags; 327 sigset_t bmask; 328 int error, i; 329 ksiginfo_t ksi; 330 331 regs = td->td_frame; 332 error = copyin((void *)regs->tf_rbx, &sf, sizeof(sf)); 333 if (error != 0) 334 return (error); 335 336 p = td->td_proc; 337 context = &sf.sf_uc.uc_mcontext; 338 rflags = context->sc_rflags; 339 340 /* 341 * Don't allow users to change privileged or reserved flags. 342 */ 343 /* 344 * XXX do allow users to change the privileged flag PSL_RF. 345 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 346 * should sometimes set it there too. tf_rflags is kept in 347 * the signal context during signal handling and there is no 348 * other place to remember it, so the PSL_RF bit may be 349 * corrupted by the signal handler without us knowing. 350 * Corruption of the PSL_RF bit at worst causes one more or 351 * one less debugger trap, so allowing it is fairly harmless. 352 */ 353 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 354 uprintf("pid %d comm %s linux mangled rflags %#lx\n", 355 p->p_pid, p->p_comm, rflags); 356 return (EINVAL); 357 } 358 359 /* 360 * Don't allow users to load a valid privileged %cs. Let the 361 * hardware check for invalid selectors, excess privilege in 362 * other selectors, invalid %eip's and invalid %esp's. 363 */ 364 if (!CS_SECURE(context->sc_cs)) { 365 uprintf("pid %d comm %s linux mangled cs %#x\n", 366 p->p_pid, p->p_comm, context->sc_cs); 367 ksiginfo_init_trap(&ksi); 368 ksi.ksi_signo = SIGBUS; 369 ksi.ksi_code = BUS_OBJERR; 370 ksi.ksi_trapno = T_PROTFLT; 371 ksi.ksi_addr = (void *)regs->tf_rip; 372 trapsignal(td, &ksi); 373 return (EINVAL); 374 } 375 376 linux_to_bsd_sigset(&sf.sf_uc.uc_sigmask, &bmask); 377 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 378 379 regs->tf_rdi = context->sc_rdi; 380 regs->tf_rsi = context->sc_rsi; 381 regs->tf_rdx = context->sc_rdx; 382 regs->tf_rbp = context->sc_rbp; 383 regs->tf_rbx = context->sc_rbx; 384 regs->tf_rcx = context->sc_rcx; 385 regs->tf_rax = context->sc_rax; 386 regs->tf_rip = context->sc_rip; 387 regs->tf_rsp = context->sc_rsp; 388 regs->tf_r8 = context->sc_r8; 389 regs->tf_r9 = context->sc_r9; 390 regs->tf_r10 = context->sc_r10; 391 regs->tf_r11 = context->sc_r11; 392 regs->tf_r12 = context->sc_r12; 393 regs->tf_r13 = context->sc_r13; 394 regs->tf_r14 = context->sc_r14; 395 regs->tf_r15 = context->sc_r15; 396 regs->tf_cs = context->sc_cs; 397 regs->tf_err = context->sc_err; 398 regs->tf_rflags = rflags; 399 400 if (sf.sf_uc.uc_mcontext.sc_fpstate != NULL) { 401 struct savefpu *svfp = (struct savefpu *)mc.mc_fpstate; 402 403 bzero(&mc, sizeof(mc)); 404 mc.mc_ownedfp = _MC_FPOWNED_FPU; 405 mc.mc_fpformat = _MC_FPFMT_XMM; 406 407 svfp->sv_env.en_cw = sf.sf_fs.cwd; 408 svfp->sv_env.en_sw = sf.sf_fs.swd; 409 svfp->sv_env.en_tw = sf.sf_fs.twd; 410 svfp->sv_env.en_opcode = sf.sf_fs.fop; 411 svfp->sv_env.en_rip = sf.sf_fs.rip; 412 svfp->sv_env.en_rdp = sf.sf_fs.rdp; 413 svfp->sv_env.en_mxcsr = sf.sf_fs.mxcsr; 414 svfp->sv_env.en_mxcsr_mask = sf.sf_fs.mxcsr_mask; 415 /* FPU registers */ 416 for (i = 0; i < nitems(svfp->sv_fp); ++i) 417 bcopy(&sf.sf_fs.st[i], svfp->sv_fp[i].fp_acc.fp_bytes, 418 sizeof(svfp->sv_fp[i].fp_acc.fp_bytes)); 419 /* SSE registers */ 420 for (i = 0; i < nitems(svfp->sv_xmm); ++i) 421 bcopy(&sf.sf_fs.xmm[i], svfp->sv_xmm[i].xmm_bytes, 422 sizeof(svfp->sv_xmm[i].xmm_bytes)); 423 error = set_fpcontext(td, &mc, NULL, 0); 424 if (error != 0) { 425 uprintf("pid %d comm %s linux can't restore fpu state %d\n", 426 p->p_pid, p->p_comm, error); 427 return (error); 428 } 429 } 430 431 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 432 return (EJUSTRETURN); 433 } 434 435 /* 436 * copied from amd64/amd64/machdep.c 437 * 438 * Send an interrupt to process. 439 */ 440 static void 441 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 442 { 443 struct l_rt_sigframe sf, *sfp; 444 struct proc *p; 445 struct thread *td; 446 struct sigacts *psp; 447 caddr_t sp; 448 struct trapframe *regs; 449 struct savefpu *svfp; 450 mcontext_t mc; 451 int sig, code; 452 int oonstack, issiginfo, i; 453 454 td = curthread; 455 p = td->td_proc; 456 PROC_LOCK_ASSERT(p, MA_OWNED); 457 sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno); 458 psp = p->p_sigacts; 459 issiginfo = SIGISMEMBER(psp->ps_siginfo, sig); 460 code = ksi->ksi_code; 461 mtx_assert(&psp->ps_mtx, MA_OWNED); 462 regs = td->td_frame; 463 oonstack = sigonstack(regs->tf_rsp); 464 465 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 466 catcher, sig, mask, code); 467 468 /* Save user context. */ 469 bzero(&sf, sizeof(sf)); 470 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask); 471 sf.sf_uc.uc_mcontext.sc_mask = sf.sf_uc.uc_sigmask; 472 473 sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 474 sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; 475 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 476 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 477 478 sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi; 479 sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi; 480 sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx; 481 sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp; 482 sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx; 483 sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx; 484 sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax; 485 sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip; 486 sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp; 487 sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8; 488 sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9; 489 sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10; 490 sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11; 491 sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12; 492 sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13; 493 sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14; 494 sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15; 495 sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs; 496 sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags; 497 sf.sf_uc.uc_mcontext.sc_err = regs->tf_err; 498 sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 499 sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 500 501 /* Allocate space for the signal handler context. */ 502 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 503 SIGISMEMBER(psp->ps_sigonstack, sig)) { 504 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size; 505 } else 506 sp = (caddr_t)regs->tf_rsp - 128; 507 sp -= sizeof(struct l_rt_sigframe); 508 /* Align to 16 bytes. */ 509 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 510 511 mtx_unlock(&psp->ps_mtx); 512 PROC_UNLOCK(p); 513 514 get_fpcontext(td, &mc, NULL, NULL); 515 KASSERT(mc.mc_fpformat != _MC_FPFMT_NODEV, ("fpu not present")); 516 svfp = (struct savefpu *)mc.mc_fpstate; 517 518 sf.sf_fs.cwd = svfp->sv_env.en_cw; 519 sf.sf_fs.swd = svfp->sv_env.en_sw; 520 sf.sf_fs.twd = svfp->sv_env.en_tw; 521 sf.sf_fs.fop = svfp->sv_env.en_opcode; 522 sf.sf_fs.rip = svfp->sv_env.en_rip; 523 sf.sf_fs.rdp = svfp->sv_env.en_rdp; 524 sf.sf_fs.mxcsr = svfp->sv_env.en_mxcsr; 525 sf.sf_fs.mxcsr_mask = svfp->sv_env.en_mxcsr_mask; 526 /* FPU registers */ 527 for (i = 0; i < nitems(svfp->sv_fp); ++i) 528 bcopy(svfp->sv_fp[i].fp_acc.fp_bytes, &sf.sf_fs.st[i], 529 sizeof(svfp->sv_fp[i].fp_acc.fp_bytes)); 530 /* SSE registers */ 531 for (i = 0; i < nitems(svfp->sv_xmm); ++i) 532 bcopy(svfp->sv_xmm[i].xmm_bytes, &sf.sf_fs.xmm[i], 533 sizeof(svfp->sv_xmm[i].xmm_bytes)); 534 sf.sf_uc.uc_mcontext.sc_fpstate = (struct l_fpstate *)((caddr_t)sfp + 535 offsetof(struct l_rt_sigframe, sf_fs)); 536 537 /* Translate the signal. */ 538 sig = bsd_to_linux_signal(sig); 539 /* Fill in POSIX parts. */ 540 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig); 541 542 /* Copy the sigframe out to the user's stack. */ 543 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 544 uprintf("pid %d comm %s has trashed its stack, killing\n", 545 p->p_pid, p->p_comm); 546 PROC_LOCK(p); 547 sigexit(td, SIGILL); 548 } 549 550 fpstate_drop(td); 551 /* Build the argument list for the signal handler. */ 552 regs->tf_rdi = sig; /* arg 1 in %rdi */ 553 regs->tf_rax = 0; 554 if (issiginfo) { 555 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 556 regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */ 557 } else { 558 regs->tf_rsi = 0; 559 regs->tf_rdx = 0; 560 } 561 regs->tf_rcx = (register_t)catcher; 562 regs->tf_rsp = (long)sfp; 563 regs->tf_rip = linux_rt_sigcode; 564 regs->tf_rflags &= ~(PSL_T | PSL_D); 565 regs->tf_cs = _ucodesel; 566 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 567 PROC_LOCK(p); 568 mtx_lock(&psp->ps_mtx); 569 } 570 571 #define LINUX_VSYSCALL_START (-10UL << 20) 572 #define LINUX_VSYSCALL_SZ 1024 573 574 const unsigned long linux_vsyscall_vector[] = { 575 LINUX_SYS_gettimeofday, 576 LINUX_SYS_linux_time, 577 LINUX_SYS_linux_getcpu, 578 }; 579 580 static int 581 linux_vsyscall(struct thread *td) 582 { 583 struct trapframe *frame; 584 uint64_t retqaddr; 585 int code, traced; 586 int error; 587 588 frame = td->td_frame; 589 590 /* Check %rip for vsyscall area. */ 591 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 592 return (EINVAL); 593 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 594 return (EINVAL); 595 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 596 if (code >= nitems(linux_vsyscall_vector)) 597 return (EINVAL); 598 599 /* 600 * vsyscall called as callq *(%rax), so we must 601 * use return address from %rsp and also fixup %rsp. 602 */ 603 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 604 if (error) 605 return (error); 606 607 frame->tf_rip = retqaddr; 608 frame->tf_rax = linux_vsyscall_vector[code]; 609 frame->tf_rsp += 8; 610 611 traced = (frame->tf_flags & PSL_T); 612 613 amd64_syscall(td, traced); 614 615 return (0); 616 } 617 618 struct sysentvec elf_linux_sysvec = { 619 .sv_size = LINUX_SYS_MAXSYSCALL, 620 .sv_table = linux_sysent, 621 .sv_fixup = __elfN(freebsd_fixup), 622 .sv_sendsig = linux_rt_sendsig, 623 .sv_sigcode = &_binary_linux_vdso_so_o_start, 624 .sv_szsigcode = &linux_szsigcode, 625 .sv_name = "Linux ELF64", 626 .sv_coredump = elf64_coredump, 627 .sv_elf_core_osabi = ELFOSABI_NONE, 628 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, 629 .sv_elf_core_prepare_notes = linux64_prepare_notes, 630 .sv_imgact_try = linux_exec_imgact_try, 631 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 632 .sv_minuser = VM_MIN_ADDRESS, 633 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, 634 .sv_usrstack = LINUX_USRSTACK_LA48, 635 .sv_psstrings = LINUX_PS_STRINGS_LA48, 636 .sv_psstringssz = sizeof(struct ps_strings), 637 .sv_stackprot = VM_PROT_ALL, 638 .sv_copyout_auxargs = linux_copyout_auxargs, 639 .sv_copyout_strings = __linuxN(copyout_strings), 640 .sv_setregs = linux_exec_setregs, 641 .sv_fixlimit = NULL, 642 .sv_maxssiz = NULL, 643 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | 644 SV_SIG_WAITNDQ | SV_TIMEKEEP, 645 .sv_set_syscall_retval = linux_set_syscall_retval, 646 .sv_fetch_syscall_args = linux_fetch_syscall_args, 647 .sv_syscallnames = linux_syscallnames, 648 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48, 649 .sv_shared_page_len = PAGE_SIZE, 650 .sv_schedtail = linux_schedtail, 651 .sv_thread_detach = linux_thread_detach, 652 .sv_trap = linux_vsyscall, 653 .sv_onexec = linux_on_exec_vmspace, 654 .sv_onexit = linux_on_exit, 655 .sv_ontdexit = linux_thread_dtor, 656 .sv_setid_allowed = &linux_setid_allowed_query, 657 .sv_set_fork_retval = linux_set_fork_retval, 658 }; 659 660 static int 661 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) 662 { 663 int error; 664 665 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, 666 LINUX_VDSOPAGE_SIZE, imgp); 667 if (error == 0) 668 linux_on_exec(p, imgp); 669 return (error); 670 } 671 672 /* 673 * linux_vdso_install() and linux_exec_sysvec_init() must be called 674 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). 675 */ 676 static void 677 linux_exec_sysvec_init(void *param) 678 { 679 l_uintptr_t *ktimekeep_base, *ktsc_selector; 680 struct sysentvec *sv; 681 ptrdiff_t tkoff; 682 683 sv = param; 684 amd64_lower_shared_page(sv); 685 /* Fill timekeep_base */ 686 exec_sysvec_init(sv); 687 688 tkoff = kern_timekeep_base - linux_vdso_base; 689 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 690 *ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset; 691 692 tkoff = kern_tsc_selector - linux_vdso_base; 693 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 694 *ktsc_selector = linux_vdso_tsc_selector_idx(); 695 if (bootverbose) 696 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector); 697 698 tkoff = kern_cpu_selector - linux_vdso_base; 699 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 700 *ktsc_selector = linux_vdso_cpu_selector_idx(); 701 if (bootverbose) 702 printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector); 703 } 704 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, 705 linux_exec_sysvec_init, &elf_linux_sysvec); 706 707 static void 708 linux_vdso_install(const void *param) 709 { 710 char *vdso_start = &_binary_linux_vdso_so_o_start; 711 char *vdso_end = &_binary_linux_vdso_so_o_end; 712 713 linux_szsigcode = vdso_end - vdso_start; 714 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE); 715 716 linux_vdso_base = LINUX_VDSOPAGE_LA48; 717 if (hw_lower_amd64_sharedpage != 0) 718 linux_vdso_base -= PAGE_SIZE; 719 720 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); 721 722 linux_vdso_obj = __elfN(linux_shared_page_init) 723 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 724 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); 725 726 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); 727 } 728 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, 729 linux_vdso_install, NULL); 730 731 static void 732 linux_vdso_deinstall(const void *param) 733 { 734 735 __elfN(linux_shared_page_fini)(linux_vdso_obj, 736 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 737 } 738 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 739 linux_vdso_deinstall, NULL); 740 741 static void 742 linux_vdso_reloc(char *mapping, Elf_Addr offset) 743 { 744 const Elf_Ehdr *ehdr; 745 const Elf_Shdr *shdr; 746 Elf64_Addr *where, val; 747 Elf_Size rtype, symidx; 748 const Elf_Rela *rela; 749 Elf_Addr addr, addend; 750 int relacnt; 751 int i, j; 752 753 MPASS(offset != 0); 754 755 relacnt = 0; 756 ehdr = (const Elf_Ehdr *)mapping; 757 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); 758 for (i = 0; i < ehdr->e_shnum; i++) 759 { 760 switch (shdr[i].sh_type) { 761 case SHT_REL: 762 printf("Linux x86_64 vDSO: unexpected Rel section\n"); 763 break; 764 case SHT_RELA: 765 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset); 766 relacnt = shdr[i].sh_size / sizeof(*rela); 767 } 768 } 769 770 for (j = 0; j < relacnt; j++, rela++) { 771 where = (Elf_Addr *)(mapping + rela->r_offset); 772 addend = rela->r_addend; 773 rtype = ELF_R_TYPE(rela->r_info); 774 symidx = ELF_R_SYM(rela->r_info); 775 776 switch (rtype) { 777 case R_X86_64_NONE: /* none */ 778 break; 779 780 case R_X86_64_RELATIVE: /* B + A */ 781 addr = (Elf_Addr)(offset + addend); 782 val = addr; 783 if (*where != val) 784 *where = val; 785 break; 786 case R_X86_64_IRELATIVE: 787 printf("Linux x86_64 vDSO: unexpected ifunc relocation, " 788 "symbol index %ld\n", symidx); 789 break; 790 default: 791 printf("Linux x86_64 vDSO: unexpected relocation type %ld, " 792 "symbol index %ld\n", rtype, symidx); 793 } 794 } 795 } 796 797 static Elf_Brandnote linux64_brandnote = { 798 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 799 .hdr.n_descsz = 16, 800 .hdr.n_type = 1, 801 .vendor = GNU_ABI_VENDOR, 802 .flags = BN_TRANSLATE_OSREL, 803 .trans_osrel = linux_trans_osrel 804 }; 805 806 static Elf64_Brandinfo linux_glibc2brand = { 807 .brand = ELFOSABI_LINUX, 808 .machine = EM_X86_64, 809 .compat_3_brand = "Linux", 810 .emul_path = linux_emul_path, 811 .interp_path = "/lib64/ld-linux-x86-64.so.2", 812 .sysvec = &elf_linux_sysvec, 813 .interp_newpath = NULL, 814 .brand_note = &linux64_brandnote, 815 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 816 }; 817 818 static Elf64_Brandinfo linux_glibc2brandshort = { 819 .brand = ELFOSABI_LINUX, 820 .machine = EM_X86_64, 821 .compat_3_brand = "Linux", 822 .emul_path = linux_emul_path, 823 .interp_path = "/lib64/ld-linux.so.2", 824 .sysvec = &elf_linux_sysvec, 825 .interp_newpath = NULL, 826 .brand_note = &linux64_brandnote, 827 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 828 }; 829 830 static Elf64_Brandinfo linux_muslbrand = { 831 .brand = ELFOSABI_LINUX, 832 .machine = EM_X86_64, 833 .compat_3_brand = "Linux", 834 .emul_path = linux_emul_path, 835 .interp_path = "/lib/ld-musl-x86_64.so.1", 836 .sysvec = &elf_linux_sysvec, 837 .interp_newpath = NULL, 838 .brand_note = &linux64_brandnote, 839 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | 840 LINUX_BI_FUTEX_REQUEUE 841 }; 842 843 Elf64_Brandinfo *linux_brandlist[] = { 844 &linux_glibc2brand, 845 &linux_glibc2brandshort, 846 &linux_muslbrand, 847 NULL 848 }; 849 850 static int 851 linux64_elf_modevent(module_t mod, int type, void *data) 852 { 853 Elf64_Brandinfo **brandinfo; 854 int error; 855 struct linux_ioctl_handler **lihp; 856 857 error = 0; 858 859 switch(type) { 860 case MOD_LOAD: 861 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 862 ++brandinfo) 863 if (elf64_insert_brand_entry(*brandinfo) < 0) 864 error = EINVAL; 865 if (error == 0) { 866 SET_FOREACH(lihp, linux_ioctl_handler_set) 867 linux_ioctl_register_handler(*lihp); 868 stclohz = (stathz ? stathz : hz); 869 if (bootverbose) 870 printf("Linux x86-64 ELF exec handler installed\n"); 871 } else 872 printf("cannot insert Linux x86-64 ELF brand handler\n"); 873 break; 874 case MOD_UNLOAD: 875 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 876 ++brandinfo) 877 if (elf64_brand_inuse(*brandinfo)) 878 error = EBUSY; 879 if (error == 0) { 880 for (brandinfo = &linux_brandlist[0]; 881 *brandinfo != NULL; ++brandinfo) 882 if (elf64_remove_brand_entry(*brandinfo) < 0) 883 error = EINVAL; 884 } 885 if (error == 0) { 886 SET_FOREACH(lihp, linux_ioctl_handler_set) 887 linux_ioctl_unregister_handler(*lihp); 888 if (bootverbose) 889 printf("Linux x86_64 ELF exec handler removed\n"); 890 } else 891 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n"); 892 break; 893 default: 894 return (EOPNOTSUPP); 895 } 896 return (error); 897 } 898 899 static moduledata_t linux64_elf_mod = { 900 "linux64elf", 901 linux64_elf_modevent, 902 0 903 }; 904 905 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 906 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 907 FEATURE(linux64, "Linux 64bit support"); 908