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 <dchagin@FreeBSD.org> 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/exec.h> 41 #include <sys/imgact.h> 42 #include <sys/imgact_elf.h> 43 #include <sys/kernel.h> 44 #include <sys/ktr.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/module.h> 48 #include <sys/mutex.h> 49 #include <sys/proc.h> 50 #include <sys/stddef.h> 51 #include <sys/syscallsubr.h> 52 #include <sys/sysctl.h> 53 #include <sys/sysent.h> 54 55 #include <vm/pmap.h> 56 #include <vm/vm.h> 57 #include <vm/vm_map.h> 58 #include <vm/vm_page.h> 59 60 #include <machine/cpu.h> 61 #include <machine/md_var.h> 62 #include <machine/pcb.h> 63 #include <machine/specialreg.h> 64 #include <machine/trap.h> 65 66 #include <x86/linux/linux_x86.h> 67 #include <amd64/linux/linux.h> 68 #include <amd64/linux/linux_proto.h> 69 #include <compat/linux/linux_elf.h> 70 #include <compat/linux/linux_emul.h> 71 #include <compat/linux/linux_fork.h> 72 #include <compat/linux/linux_ioctl.h> 73 #include <compat/linux/linux_mib.h> 74 #include <compat/linux/linux_misc.h> 75 #include <compat/linux/linux_signal.h> 76 #include <compat/linux/linux_sysproto.h> 77 #include <compat/linux/linux_util.h> 78 #include <compat/linux/linux_vdso.h> 79 80 #include <x86/linux/linux_x86_sigframe.h> 81 82 MODULE_VERSION(linux64, 1); 83 84 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2 85 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \ 86 LINUX_VDSOPAGE_SIZE) 87 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE) 88 /* 89 * PAGE_SIZE - the size 90 * of the native SHAREDPAGE 91 */ 92 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48 93 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \ 94 sizeof(struct ps_strings)) 95 96 static int linux_szsigcode; 97 static vm_object_t linux_vdso_obj; 98 static char *linux_vdso_mapping; 99 extern char _binary_linux_vdso_so_o_start; 100 extern char _binary_linux_vdso_so_o_end; 101 static vm_offset_t linux_vdso_base; 102 103 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 104 extern const char *linux_syscallnames[]; 105 106 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 107 108 static void linux_vdso_install(const void *param); 109 static void linux_vdso_deinstall(const void *param); 110 static void linux_vdso_reloc(char *mapping, Elf_Addr offset); 111 static void linux_set_syscall_retval(struct thread *td, int error); 112 static int linux_fetch_syscall_args(struct thread *td); 113 static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 114 uintptr_t stack); 115 static void linux_exec_sysvec_init(void *param); 116 static int linux_on_exec_vmspace(struct proc *p, 117 struct image_params *imgp); 118 static void linux_set_fork_retval(struct thread *td); 119 static int linux_vsyscall(struct thread *td); 120 121 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 122 LINUX_VDSO_SYM_CHAR(linux_platform); 123 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base); 124 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector); 125 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector); 126 127 /* 128 * According to the Intel x86 ISA 64-bit syscall 129 * saves %rip to %rcx and rflags to %r11. Registers on syscall entry: 130 * %rax system call number 131 * %rcx return address 132 * %r11 saved rflags 133 * %rdi arg1 134 * %rsi arg2 135 * %rdx arg3 136 * %r10 arg4 137 * %r8 arg5 138 * %r9 arg6 139 * 140 * Then FreeBSD fast_syscall() move registers: 141 * %rcx -> trapframe.tf_rip 142 * %r10 -> trapframe.tf_rcx 143 */ 144 static int 145 linux_fetch_syscall_args(struct thread *td) 146 { 147 struct proc *p; 148 struct trapframe *frame; 149 struct syscall_args *sa; 150 151 p = td->td_proc; 152 frame = td->td_frame; 153 sa = &td->td_sa; 154 155 sa->args[0] = frame->tf_rdi; 156 sa->args[1] = frame->tf_rsi; 157 sa->args[2] = frame->tf_rdx; 158 sa->args[3] = frame->tf_rcx; 159 sa->args[4] = frame->tf_r8; 160 sa->args[5] = frame->tf_r9; 161 sa->code = frame->tf_rax; 162 sa->original_code = sa->code; 163 164 if (sa->code >= p->p_sysent->sv_size) 165 /* nosys */ 166 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 167 else 168 sa->callp = &p->p_sysent->sv_table[sa->code]; 169 170 td->td_retval[0] = 0; 171 return (0); 172 } 173 174 static void 175 linux_set_syscall_retval(struct thread *td, int error) 176 { 177 struct trapframe *frame; 178 179 frame = td->td_frame; 180 181 switch (error) { 182 case 0: 183 frame->tf_rax = td->td_retval[0]; 184 frame->tf_r10 = frame->tf_rcx; 185 break; 186 187 case ERESTART: 188 /* 189 * Reconstruct pc, we know that 'syscall' is 2 bytes, 190 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes. 191 * We saved this in tf_err. 192 * 193 */ 194 frame->tf_rip -= frame->tf_err; 195 frame->tf_r10 = frame->tf_rcx; 196 break; 197 198 case EJUSTRETURN: 199 break; 200 201 default: 202 frame->tf_rax = bsd_to_linux_errno(error); 203 frame->tf_r10 = frame->tf_rcx; 204 break; 205 } 206 207 /* 208 * Differently from FreeBSD native ABI, on Linux only %rcx 209 * and %r11 values are not preserved across the syscall. 210 * Require full context restore to get all registers except 211 * those two restored at return to usermode. 212 */ 213 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 214 } 215 216 static void 217 linux_set_fork_retval(struct thread *td) 218 { 219 struct trapframe *frame = td->td_frame; 220 221 frame->tf_rax = 0; 222 } 223 224 void 225 linux64_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos) 226 { 227 228 AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base); 229 AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature); 230 AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, 0); 231 AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform)); 232 } 233 234 /* 235 * Reset registers to default values on exec. 236 */ 237 static void 238 linux_exec_setregs(struct thread *td, struct image_params *imgp, 239 uintptr_t stack) 240 { 241 struct trapframe *regs; 242 struct pcb *pcb; 243 register_t saved_rflags; 244 245 regs = td->td_frame; 246 pcb = td->td_pcb; 247 248 if (td->td_proc->p_md.md_ldt != NULL) 249 user_ldt_free(td); 250 251 pcb->pcb_fsbase = 0; 252 pcb->pcb_gsbase = 0; 253 clear_pcb_flags(pcb, PCB_32BIT); 254 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 255 set_pcb_flags(pcb, PCB_FULL_IRET); 256 257 saved_rflags = regs->tf_rflags & PSL_T; 258 bzero((char *)regs, sizeof(struct trapframe)); 259 regs->tf_rip = imgp->entry_addr; 260 regs->tf_rsp = stack; 261 regs->tf_rflags = PSL_USER | saved_rflags; 262 regs->tf_ss = _udatasel; 263 regs->tf_cs = _ucodesel; 264 regs->tf_ds = _udatasel; 265 regs->tf_es = _udatasel; 266 regs->tf_fs = _ufssel; 267 regs->tf_gs = _ugssel; 268 regs->tf_flags = TF_HASSEGS; 269 270 x86_clear_dbregs(pcb); 271 272 /* 273 * Drop the FP state if we hold it, so that the process gets a 274 * clean FP state if it uses the FPU again. 275 */ 276 fpstate_drop(td); 277 } 278 279 /* 280 * Copied from amd64/amd64/machdep.c 281 */ 282 int 283 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 284 { 285 struct proc *p; 286 struct l_rt_sigframe sf; 287 struct l_sigcontext *context; 288 struct trapframe *regs; 289 mcontext_t mc; 290 unsigned long rflags; 291 sigset_t bmask; 292 int error, i; 293 ksiginfo_t ksi; 294 295 regs = td->td_frame; 296 error = copyin((void *)regs->tf_rbx, &sf, sizeof(sf)); 297 if (error != 0) 298 return (error); 299 300 p = td->td_proc; 301 context = &sf.sf_uc.uc_mcontext; 302 rflags = context->sc_rflags; 303 304 /* 305 * Don't allow users to change privileged or reserved flags. 306 */ 307 /* 308 * XXX do allow users to change the privileged flag PSL_RF. 309 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 310 * should sometimes set it there too. tf_rflags is kept in 311 * the signal context during signal handling and there is no 312 * other place to remember it, so the PSL_RF bit may be 313 * corrupted by the signal handler without us knowing. 314 * Corruption of the PSL_RF bit at worst causes one more or 315 * one less debugger trap, so allowing it is fairly harmless. 316 */ 317 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 318 uprintf("pid %d comm %s linux mangled rflags %#lx\n", 319 p->p_pid, p->p_comm, rflags); 320 return (EINVAL); 321 } 322 323 /* 324 * Don't allow users to load a valid privileged %cs. Let the 325 * hardware check for invalid selectors, excess privilege in 326 * other selectors, invalid %eip's and invalid %esp's. 327 */ 328 if (!CS_SECURE(context->sc_cs)) { 329 uprintf("pid %d comm %s linux mangled cs %#x\n", 330 p->p_pid, p->p_comm, context->sc_cs); 331 ksiginfo_init_trap(&ksi); 332 ksi.ksi_signo = SIGBUS; 333 ksi.ksi_code = BUS_OBJERR; 334 ksi.ksi_trapno = T_PROTFLT; 335 ksi.ksi_addr = (void *)regs->tf_rip; 336 trapsignal(td, &ksi); 337 return (EINVAL); 338 } 339 340 linux_to_bsd_sigset(&sf.sf_uc.uc_sigmask, &bmask); 341 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); 342 343 regs->tf_rdi = context->sc_rdi; 344 regs->tf_rsi = context->sc_rsi; 345 regs->tf_rdx = context->sc_rdx; 346 regs->tf_rbp = context->sc_rbp; 347 regs->tf_rbx = context->sc_rbx; 348 regs->tf_rcx = context->sc_rcx; 349 regs->tf_rax = context->sc_rax; 350 regs->tf_rip = context->sc_rip; 351 regs->tf_rsp = context->sc_rsp; 352 regs->tf_r8 = context->sc_r8; 353 regs->tf_r9 = context->sc_r9; 354 regs->tf_r10 = context->sc_r10; 355 regs->tf_r11 = context->sc_r11; 356 regs->tf_r12 = context->sc_r12; 357 regs->tf_r13 = context->sc_r13; 358 regs->tf_r14 = context->sc_r14; 359 regs->tf_r15 = context->sc_r15; 360 regs->tf_cs = context->sc_cs; 361 regs->tf_err = context->sc_err; 362 regs->tf_rflags = rflags; 363 364 if (sf.sf_uc.uc_mcontext.sc_fpstate != NULL) { 365 struct savefpu *svfp = (struct savefpu *)mc.mc_fpstate; 366 367 bzero(&mc, sizeof(mc)); 368 mc.mc_ownedfp = _MC_FPOWNED_FPU; 369 mc.mc_fpformat = _MC_FPFMT_XMM; 370 371 svfp->sv_env.en_cw = sf.sf_fs.cwd; 372 svfp->sv_env.en_sw = sf.sf_fs.swd; 373 svfp->sv_env.en_tw = sf.sf_fs.twd; 374 svfp->sv_env.en_opcode = sf.sf_fs.fop; 375 svfp->sv_env.en_rip = sf.sf_fs.rip; 376 svfp->sv_env.en_rdp = sf.sf_fs.rdp; 377 svfp->sv_env.en_mxcsr = sf.sf_fs.mxcsr; 378 svfp->sv_env.en_mxcsr_mask = sf.sf_fs.mxcsr_mask; 379 /* FPU registers */ 380 for (i = 0; i < nitems(svfp->sv_fp); ++i) 381 bcopy(&sf.sf_fs.st[i], svfp->sv_fp[i].fp_acc.fp_bytes, 382 sizeof(svfp->sv_fp[i].fp_acc.fp_bytes)); 383 /* SSE registers */ 384 for (i = 0; i < nitems(svfp->sv_xmm); ++i) 385 bcopy(&sf.sf_fs.xmm[i], svfp->sv_xmm[i].xmm_bytes, 386 sizeof(svfp->sv_xmm[i].xmm_bytes)); 387 error = set_fpcontext(td, &mc, NULL, 0); 388 if (error != 0) { 389 uprintf("pid %d comm %s linux can't restore fpu state %d\n", 390 p->p_pid, p->p_comm, error); 391 return (error); 392 } 393 } 394 395 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 396 return (EJUSTRETURN); 397 } 398 399 /* 400 * copied from amd64/amd64/machdep.c 401 * 402 * Send an interrupt to process. 403 */ 404 static void 405 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 406 { 407 struct l_rt_sigframe sf, *sfp; 408 struct proc *p; 409 struct thread *td; 410 struct sigacts *psp; 411 caddr_t sp; 412 struct trapframe *regs; 413 struct savefpu *svfp; 414 mcontext_t mc; 415 int sig, code; 416 int oonstack, issiginfo, i; 417 418 td = curthread; 419 p = td->td_proc; 420 PROC_LOCK_ASSERT(p, MA_OWNED); 421 sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno); 422 psp = p->p_sigacts; 423 issiginfo = SIGISMEMBER(psp->ps_siginfo, sig); 424 code = ksi->ksi_code; 425 mtx_assert(&psp->ps_mtx, MA_OWNED); 426 regs = td->td_frame; 427 oonstack = sigonstack(regs->tf_rsp); 428 429 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 430 catcher, sig, mask, code); 431 432 bzero(&sf, sizeof(sf)); 433 sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 434 sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size; 435 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 436 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 437 438 /* Allocate space for the signal handler context. */ 439 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 440 SIGISMEMBER(psp->ps_sigonstack, sig)) { 441 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size; 442 } else 443 sp = (caddr_t)regs->tf_rsp - 128; 444 445 mtx_unlock(&psp->ps_mtx); 446 PROC_UNLOCK(p); 447 448 /* Make room, keeping the stack aligned. */ 449 sp -= sizeof(struct l_rt_sigframe); 450 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 451 452 /* Save user context. */ 453 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask); 454 sf.sf_uc.uc_mcontext.sc_mask = sf.sf_uc.uc_sigmask; 455 sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi; 456 sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi; 457 sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx; 458 sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp; 459 sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx; 460 sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx; 461 sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax; 462 sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip; 463 sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp; 464 sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8; 465 sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9; 466 sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10; 467 sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11; 468 sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12; 469 sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13; 470 sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14; 471 sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15; 472 sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs; 473 sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags; 474 sf.sf_uc.uc_mcontext.sc_err = regs->tf_err; 475 sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 476 sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 477 478 get_fpcontext(td, &mc, NULL, NULL); 479 KASSERT(mc.mc_fpformat != _MC_FPFMT_NODEV, ("fpu not present")); 480 svfp = (struct savefpu *)mc.mc_fpstate; 481 482 sf.sf_fs.cwd = svfp->sv_env.en_cw; 483 sf.sf_fs.swd = svfp->sv_env.en_sw; 484 sf.sf_fs.twd = svfp->sv_env.en_tw; 485 sf.sf_fs.fop = svfp->sv_env.en_opcode; 486 sf.sf_fs.rip = svfp->sv_env.en_rip; 487 sf.sf_fs.rdp = svfp->sv_env.en_rdp; 488 sf.sf_fs.mxcsr = svfp->sv_env.en_mxcsr; 489 sf.sf_fs.mxcsr_mask = svfp->sv_env.en_mxcsr_mask; 490 /* FPU registers */ 491 for (i = 0; i < nitems(svfp->sv_fp); ++i) 492 bcopy(svfp->sv_fp[i].fp_acc.fp_bytes, &sf.sf_fs.st[i], 493 sizeof(svfp->sv_fp[i].fp_acc.fp_bytes)); 494 /* SSE registers */ 495 for (i = 0; i < nitems(svfp->sv_xmm); ++i) 496 bcopy(svfp->sv_xmm[i].xmm_bytes, &sf.sf_fs.xmm[i], 497 sizeof(svfp->sv_xmm[i].xmm_bytes)); 498 sf.sf_uc.uc_mcontext.sc_fpstate = (struct l_fpstate *)((caddr_t)sfp + 499 offsetof(struct l_rt_sigframe, sf_fs)); 500 501 /* Translate the signal. */ 502 sig = bsd_to_linux_signal(sig); 503 /* Fill in POSIX parts. */ 504 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig); 505 506 /* Copy the sigframe out to the user's stack. */ 507 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 508 uprintf("pid %d comm %s has trashed its stack, killing\n", 509 p->p_pid, p->p_comm); 510 PROC_LOCK(p); 511 sigexit(td, SIGILL); 512 } 513 514 fpstate_drop(td); 515 /* Build the argument list for the signal handler. */ 516 regs->tf_rdi = sig; /* arg 1 in %rdi */ 517 regs->tf_rax = 0; 518 if (issiginfo) { 519 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 520 regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */ 521 } else { 522 regs->tf_rsi = 0; 523 regs->tf_rdx = 0; 524 } 525 regs->tf_rcx = (register_t)catcher; 526 regs->tf_rsp = (long)sfp; 527 regs->tf_rip = linux_rt_sigcode; 528 regs->tf_rflags &= ~(PSL_T | PSL_D); 529 regs->tf_cs = _ucodesel; 530 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 531 PROC_LOCK(p); 532 mtx_lock(&psp->ps_mtx); 533 } 534 535 #define LINUX_VSYSCALL_START (-10UL << 20) 536 #define LINUX_VSYSCALL_SZ 1024 537 538 const unsigned long linux_vsyscall_vector[] = { 539 LINUX_SYS_gettimeofday, 540 LINUX_SYS_linux_time, 541 LINUX_SYS_linux_getcpu, 542 }; 543 544 static int 545 linux_vsyscall(struct thread *td) 546 { 547 struct trapframe *frame; 548 uint64_t retqaddr; 549 int code, traced; 550 int error; 551 552 frame = td->td_frame; 553 554 /* Check %rip for vsyscall area. */ 555 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 556 return (EINVAL); 557 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 558 return (EINVAL); 559 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 560 if (code >= nitems(linux_vsyscall_vector)) 561 return (EINVAL); 562 563 /* 564 * vsyscall called as callq *(%rax), so we must 565 * use return address from %rsp and also fixup %rsp. 566 */ 567 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 568 if (error) 569 return (error); 570 571 frame->tf_rip = retqaddr; 572 frame->tf_rax = linux_vsyscall_vector[code]; 573 frame->tf_rsp += 8; 574 575 traced = (frame->tf_flags & PSL_T); 576 577 amd64_syscall(td, traced); 578 579 return (0); 580 } 581 582 struct sysentvec elf_linux_sysvec = { 583 .sv_size = LINUX_SYS_MAXSYSCALL, 584 .sv_table = linux_sysent, 585 .sv_fixup = __elfN(freebsd_fixup), 586 .sv_sendsig = linux_rt_sendsig, 587 .sv_sigcode = &_binary_linux_vdso_so_o_start, 588 .sv_szsigcode = &linux_szsigcode, 589 .sv_name = "Linux ELF64", 590 .sv_coredump = elf64_coredump, 591 .sv_elf_core_osabi = ELFOSABI_NONE, 592 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR, 593 .sv_elf_core_prepare_notes = linux64_prepare_notes, 594 .sv_imgact_try = linux_exec_imgact_try, 595 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 596 .sv_minuser = VM_MIN_ADDRESS, 597 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48, 598 .sv_usrstack = LINUX_USRSTACK_LA48, 599 .sv_psstrings = LINUX_PS_STRINGS_LA48, 600 .sv_psstringssz = sizeof(struct ps_strings), 601 .sv_stackprot = VM_PROT_ALL, 602 .sv_copyout_auxargs = __linuxN(copyout_auxargs), 603 .sv_copyout_strings = __linuxN(copyout_strings), 604 .sv_setregs = linux_exec_setregs, 605 .sv_fixlimit = NULL, 606 .sv_maxssiz = NULL, 607 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN | 608 SV_SIG_WAITNDQ | SV_TIMEKEEP, 609 .sv_set_syscall_retval = linux_set_syscall_retval, 610 .sv_fetch_syscall_args = linux_fetch_syscall_args, 611 .sv_syscallnames = linux_syscallnames, 612 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48, 613 .sv_shared_page_len = PAGE_SIZE, 614 .sv_schedtail = linux_schedtail, 615 .sv_thread_detach = linux_thread_detach, 616 .sv_trap = linux_vsyscall, 617 .sv_hwcap = NULL, 618 .sv_hwcap2 = NULL, 619 .sv_onexec = linux_on_exec_vmspace, 620 .sv_onexit = linux_on_exit, 621 .sv_ontdexit = linux_thread_dtor, 622 .sv_setid_allowed = &linux_setid_allowed_query, 623 .sv_set_fork_retval = linux_set_fork_retval, 624 }; 625 626 static int 627 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp) 628 { 629 int error; 630 631 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base, 632 LINUX_VDSOPAGE_SIZE, imgp); 633 if (error == 0) 634 linux_on_exec(p, imgp); 635 return (error); 636 } 637 638 /* 639 * linux_vdso_install() and linux_exec_sysvec_init() must be called 640 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY). 641 */ 642 static void 643 linux_exec_sysvec_init(void *param) 644 { 645 l_uintptr_t *ktimekeep_base, *ktsc_selector; 646 struct sysentvec *sv; 647 ptrdiff_t tkoff; 648 649 sv = param; 650 amd64_lower_shared_page(sv); 651 /* Fill timekeep_base */ 652 exec_sysvec_init(sv); 653 654 tkoff = kern_timekeep_base - linux_vdso_base; 655 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 656 *ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset; 657 658 tkoff = kern_tsc_selector - linux_vdso_base; 659 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 660 *ktsc_selector = linux_vdso_tsc_selector_idx(); 661 if (bootverbose) 662 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector); 663 664 tkoff = kern_cpu_selector - linux_vdso_base; 665 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff); 666 *ktsc_selector = linux_vdso_cpu_selector_idx(); 667 if (bootverbose) 668 printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector); 669 } 670 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY, 671 linux_exec_sysvec_init, &elf_linux_sysvec); 672 673 static void 674 linux_vdso_install(const void *param) 675 { 676 char *vdso_start = &_binary_linux_vdso_so_o_start; 677 char *vdso_end = &_binary_linux_vdso_so_o_end; 678 679 linux_szsigcode = vdso_end - vdso_start; 680 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE); 681 682 linux_vdso_base = LINUX_VDSOPAGE_LA48; 683 if (hw_lower_amd64_sharedpage != 0) 684 linux_vdso_base -= PAGE_SIZE; 685 686 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base); 687 688 linux_vdso_obj = __elfN(linux_shared_page_init) 689 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 690 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode); 691 692 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base); 693 } 694 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST, 695 linux_vdso_install, NULL); 696 697 static void 698 linux_vdso_deinstall(const void *param) 699 { 700 701 __elfN(linux_shared_page_fini)(linux_vdso_obj, 702 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE); 703 } 704 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 705 linux_vdso_deinstall, NULL); 706 707 static void 708 linux_vdso_reloc(char *mapping, Elf_Addr offset) 709 { 710 const Elf_Ehdr *ehdr; 711 const Elf_Shdr *shdr; 712 Elf64_Addr *where, val; 713 Elf_Size rtype, symidx; 714 const Elf_Rela *rela; 715 Elf_Addr addr, addend; 716 int relacnt; 717 int i, j; 718 719 MPASS(offset != 0); 720 721 relacnt = 0; 722 ehdr = (const Elf_Ehdr *)mapping; 723 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff); 724 for (i = 0; i < ehdr->e_shnum; i++) 725 { 726 switch (shdr[i].sh_type) { 727 case SHT_REL: 728 printf("Linux x86_64 vDSO: unexpected Rel section\n"); 729 break; 730 case SHT_RELA: 731 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset); 732 relacnt = shdr[i].sh_size / sizeof(*rela); 733 } 734 } 735 736 for (j = 0; j < relacnt; j++, rela++) { 737 where = (Elf_Addr *)(mapping + rela->r_offset); 738 addend = rela->r_addend; 739 rtype = ELF_R_TYPE(rela->r_info); 740 symidx = ELF_R_SYM(rela->r_info); 741 742 switch (rtype) { 743 case R_X86_64_NONE: /* none */ 744 break; 745 746 case R_X86_64_RELATIVE: /* B + A */ 747 addr = (Elf_Addr)(offset + addend); 748 val = addr; 749 if (*where != val) 750 *where = val; 751 break; 752 case R_X86_64_IRELATIVE: 753 printf("Linux x86_64 vDSO: unexpected ifunc relocation, " 754 "symbol index %ld\n", symidx); 755 break; 756 default: 757 printf("Linux x86_64 vDSO: unexpected relocation type %ld, " 758 "symbol index %ld\n", rtype, symidx); 759 } 760 } 761 } 762 763 static Elf_Brandnote linux64_brandnote = { 764 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 765 .hdr.n_descsz = 16, 766 .hdr.n_type = 1, 767 .vendor = GNU_ABI_VENDOR, 768 .flags = BN_TRANSLATE_OSREL, 769 .trans_osrel = linux_trans_osrel 770 }; 771 772 static Elf64_Brandinfo linux_glibc2brand = { 773 .brand = ELFOSABI_LINUX, 774 .machine = EM_X86_64, 775 .compat_3_brand = "Linux", 776 .emul_path = linux_emul_path, 777 .interp_path = "/lib64/ld-linux-x86-64.so.2", 778 .sysvec = &elf_linux_sysvec, 779 .interp_newpath = NULL, 780 .brand_note = &linux64_brandnote, 781 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 782 }; 783 784 static Elf64_Brandinfo linux_glibc2brandshort = { 785 .brand = ELFOSABI_LINUX, 786 .machine = EM_X86_64, 787 .compat_3_brand = "Linux", 788 .emul_path = linux_emul_path, 789 .interp_path = "/lib64/ld-linux.so.2", 790 .sysvec = &elf_linux_sysvec, 791 .interp_newpath = NULL, 792 .brand_note = &linux64_brandnote, 793 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 794 }; 795 796 static Elf64_Brandinfo linux_muslbrand = { 797 .brand = ELFOSABI_LINUX, 798 .machine = EM_X86_64, 799 .compat_3_brand = "Linux", 800 .emul_path = linux_emul_path, 801 .interp_path = "/lib/ld-musl-x86_64.so.1", 802 .sysvec = &elf_linux_sysvec, 803 .interp_newpath = NULL, 804 .brand_note = &linux64_brandnote, 805 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE | 806 LINUX_BI_FUTEX_REQUEUE 807 }; 808 809 Elf64_Brandinfo *linux_brandlist[] = { 810 &linux_glibc2brand, 811 &linux_glibc2brandshort, 812 &linux_muslbrand, 813 NULL 814 }; 815 816 static int 817 linux64_elf_modevent(module_t mod, int type, void *data) 818 { 819 Elf64_Brandinfo **brandinfo; 820 int error; 821 struct linux_ioctl_handler **lihp; 822 823 error = 0; 824 825 switch(type) { 826 case MOD_LOAD: 827 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 828 ++brandinfo) 829 if (elf64_insert_brand_entry(*brandinfo) < 0) 830 error = EINVAL; 831 if (error == 0) { 832 SET_FOREACH(lihp, linux_ioctl_handler_set) 833 linux_ioctl_register_handler(*lihp); 834 stclohz = (stathz ? stathz : hz); 835 if (bootverbose) 836 printf("Linux x86-64 ELF exec handler installed\n"); 837 } else 838 printf("cannot insert Linux x86-64 ELF brand handler\n"); 839 break; 840 case MOD_UNLOAD: 841 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 842 ++brandinfo) 843 if (elf64_brand_inuse(*brandinfo)) 844 error = EBUSY; 845 if (error == 0) { 846 for (brandinfo = &linux_brandlist[0]; 847 *brandinfo != NULL; ++brandinfo) 848 if (elf64_remove_brand_entry(*brandinfo) < 0) 849 error = EINVAL; 850 } 851 if (error == 0) { 852 SET_FOREACH(lihp, linux_ioctl_handler_set) 853 linux_ioctl_unregister_handler(*lihp); 854 if (bootverbose) 855 printf("Linux x86_64 ELF exec handler removed\n"); 856 } else 857 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n"); 858 break; 859 default: 860 return (EOPNOTSUPP); 861 } 862 return (error); 863 } 864 865 static moduledata_t linux64_elf_mod = { 866 "linux64elf", 867 linux64_elf_modevent, 868 0 869 }; 870 871 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 872 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 873 FEATURE(linux64, "Linux 64bit support"); 874