linux_sysvec.c revision 338691
1/*- 2 * Copyright (c) 2013 Dmitry Chagin 3 * Copyright (c) 2004 Tim J. Robbins 4 * Copyright (c) 2003 Peter Wemm 5 * Copyright (c) 2002 Doug Rabson 6 * Copyright (c) 1998-1999 Andrew Gallatin 7 * Copyright (c) 1994-1996 S��ren Schmidt 8 * All rights reserved. 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: stable/11/sys/amd64/linux/linux_sysvec.c 338691 2018-09-14 23:21:52Z jhb $"); 36 37#include "opt_compat.h" 38 39#define __ELF_WORD_SIZE 64 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/exec.h> 44#include <sys/fcntl.h> 45#include <sys/imgact.h> 46#include <sys/imgact_elf.h> 47#include <sys/kernel.h> 48#include <sys/ktr.h> 49#include <sys/lock.h> 50#include <sys/malloc.h> 51#include <sys/module.h> 52#include <sys/mutex.h> 53#include <sys/proc.h> 54#include <sys/resourcevar.h> 55#include <sys/signalvar.h> 56#include <sys/syscallsubr.h> 57#include <sys/sysctl.h> 58#include <sys/sysent.h> 59#include <sys/sysproto.h> 60#include <sys/vnode.h> 61#include <sys/eventhandler.h> 62 63#include <vm/vm.h> 64#include <vm/pmap.h> 65#include <vm/vm_extern.h> 66#include <vm/vm_map.h> 67#include <vm/vm_object.h> 68#include <vm/vm_page.h> 69#include <vm/vm_param.h> 70 71#include <machine/cpu.h> 72#include <machine/md_var.h> 73#include <machine/pcb.h> 74#include <machine/specialreg.h> 75 76#include <amd64/linux/linux.h> 77#include <amd64/linux/linux_proto.h> 78#include <compat/linux/linux_emul.h> 79#include <compat/linux/linux_futex.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> 87 88MODULE_VERSION(linux64, 1); 89 90#if BYTE_ORDER == LITTLE_ENDIAN 91#define SHELLMAGIC 0x2123 /* #! */ 92#else 93#define SHELLMAGIC 0x2321 94#endif 95 96#if defined(DEBUG) 97SYSCTL_PROC(_compat_linux, OID_AUTO, debug, 98 CTLTYPE_STRING | CTLFLAG_RW, 99 0, 0, linux_sysctl_debug, "A", 100 "Linux 64 debugging control"); 101#endif 102 103/* 104 * Allow the sendsig functions to use the ldebug() facility even though they 105 * are not syscalls themselves. Map them to syscall 0. This is slightly less 106 * bogus than using ldebug(sigreturn). 107 */ 108#define LINUX_SYS_linux_rt_sendsig 0 109 110const char *linux_kplatform; 111static int linux_szsigcode; 112static vm_object_t linux_shared_page_obj; 113static char *linux_shared_page_mapping; 114extern char _binary_linux_locore_o_start; 115extern char _binary_linux_locore_o_end; 116 117extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; 118 119SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); 120 121static register_t * linux_copyout_strings(struct image_params *imgp); 122static int elf_linux_fixup(register_t **stack_base, 123 struct image_params *iparams); 124static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel); 125static void linux_vdso_install(void *param); 126static void linux_vdso_deinstall(void *param); 127static void linux_set_syscall_retval(struct thread *td, int error); 128static int linux_fetch_syscall_args(struct thread *td); 129static void linux_exec_setregs(struct thread *td, struct image_params *imgp, 130 u_long stack); 131static int linux_vsyscall(struct thread *td); 132 133/* 134 * Linux syscalls return negative errno's, we do positive and map them 135 * Reference: 136 * FreeBSD: src/sys/sys/errno.h 137 * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h 138 * linux-2.6.17.8/include/asm-generic/errno.h 139 */ 140static int bsd_to_linux_errno[ELAST + 1] = { 141 -0, -1, -2, -3, -4, -5, -6, -7, -8, -9, 142 -10, -35, -12, -13, -14, -15, -16, -17, -18, -19, 143 -20, -21, -22, -23, -24, -25, -26, -27, -28, -29, 144 -30, -31, -32, -33, -34, -11,-115,-114, -88, -89, 145 -90, -91, -92, -93, -94, -95, -96, -97, -98, -99, 146 -100,-101,-102,-103,-104,-105,-106,-107,-108,-109, 147 -110,-111, -40, -36,-112,-113, -39, -11, -87,-122, 148 -116, -66, -6, -6, -6, -6, -6, -37, -38, -9, 149 -6, -6, -43, -42, -75,-125, -84, -61, -16, -74, 150 -72, -67, -71 151}; 152 153#define LINUX_T_UNKNOWN 255 154static int _bsd_to_linux_trapcode[] = { 155 LINUX_T_UNKNOWN, /* 0 */ 156 6, /* 1 T_PRIVINFLT */ 157 LINUX_T_UNKNOWN, /* 2 */ 158 3, /* 3 T_BPTFLT */ 159 LINUX_T_UNKNOWN, /* 4 */ 160 LINUX_T_UNKNOWN, /* 5 */ 161 16, /* 6 T_ARITHTRAP */ 162 254, /* 7 T_ASTFLT */ 163 LINUX_T_UNKNOWN, /* 8 */ 164 13, /* 9 T_PROTFLT */ 165 1, /* 10 T_TRCTRAP */ 166 LINUX_T_UNKNOWN, /* 11 */ 167 14, /* 12 T_PAGEFLT */ 168 LINUX_T_UNKNOWN, /* 13 */ 169 17, /* 14 T_ALIGNFLT */ 170 LINUX_T_UNKNOWN, /* 15 */ 171 LINUX_T_UNKNOWN, /* 16 */ 172 LINUX_T_UNKNOWN, /* 17 */ 173 0, /* 18 T_DIVIDE */ 174 2, /* 19 T_NMI */ 175 4, /* 20 T_OFLOW */ 176 5, /* 21 T_BOUND */ 177 7, /* 22 T_DNA */ 178 8, /* 23 T_DOUBLEFLT */ 179 9, /* 24 T_FPOPFLT */ 180 10, /* 25 T_TSSFLT */ 181 11, /* 26 T_SEGNPFLT */ 182 12, /* 27 T_STKFLT */ 183 18, /* 28 T_MCHK */ 184 19, /* 29 T_XMMFLT */ 185 15 /* 30 T_RESERVED */ 186}; 187#define bsd_to_linux_trapcode(code) \ 188 ((code)<nitems(_bsd_to_linux_trapcode)? \ 189 _bsd_to_linux_trapcode[(code)]: \ 190 LINUX_T_UNKNOWN) 191 192LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode); 193LINUX_VDSO_SYM_CHAR(linux_platform); 194 195/* 196 * If FreeBSD & Linux have a difference of opinion about what a trap 197 * means, deal with it here. 198 * 199 * MPSAFE 200 */ 201static int 202translate_traps(int signal, int trap_code) 203{ 204 205 if (signal != SIGBUS) 206 return (signal); 207 switch (trap_code) { 208 case T_PROTFLT: 209 case T_TSSFLT: 210 case T_DOUBLEFLT: 211 case T_PAGEFLT: 212 return (SIGSEGV); 213 default: 214 return (signal); 215 } 216} 217 218static int 219linux_fetch_syscall_args(struct thread *td) 220{ 221 struct proc *p; 222 struct trapframe *frame; 223 struct syscall_args *sa; 224 225 p = td->td_proc; 226 frame = td->td_frame; 227 sa = &td->td_sa; 228 229 sa->args[0] = frame->tf_rdi; 230 sa->args[1] = frame->tf_rsi; 231 sa->args[2] = frame->tf_rdx; 232 sa->args[3] = frame->tf_rcx; 233 sa->args[4] = frame->tf_r8; 234 sa->args[5] = frame->tf_r9; 235 sa->code = frame->tf_rax; 236 237 if (sa->code >= p->p_sysent->sv_size) 238 /* nosys */ 239 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; 240 else 241 sa->callp = &p->p_sysent->sv_table[sa->code]; 242 sa->narg = sa->callp->sy_narg; 243 244 td->td_retval[0] = 0; 245 return (0); 246} 247 248static void 249linux_set_syscall_retval(struct thread *td, int error) 250{ 251 struct trapframe *frame = td->td_frame; 252 253 /* 254 * On Linux only %rcx and %r11 values are not preserved across 255 * the syscall. So, do not clobber %rdx and %r10. 256 */ 257 td->td_retval[1] = frame->tf_rdx; 258 if (error != EJUSTRETURN) 259 frame->tf_r10 = frame->tf_rcx; 260 261 cpu_set_syscall_retval(td, error); 262 263 /* Restore all registers. */ 264 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 265} 266 267static int 268elf_linux_fixup(register_t **stack_base, struct image_params *imgp) 269{ 270 Elf_Auxargs *args; 271 Elf_Addr *base; 272 Elf_Addr *pos; 273 struct ps_strings *arginfo; 274 struct proc *p; 275 int issetugid; 276 277 p = imgp->proc; 278 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 279 280 KASSERT(curthread->td_proc == imgp->proc, 281 ("unsafe elf_linux_fixup(), should be curproc")); 282 base = (Elf64_Addr *)*stack_base; 283 args = (Elf64_Auxargs *)imgp->auxargs; 284 pos = base + (imgp->args->argc + imgp->args->envc + 2); 285 286 issetugid = p->p_flag & P_SUGID ? 1 : 0; 287 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, 288 imgp->proc->p_sysent->sv_shared_page_base); 289 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); 290 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); 291 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 292 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 293 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 294 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 295 AUXARGS_ENTRY(pos, AT_BASE, args->base); 296 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 297 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 298 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); 299 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); 300 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); 301 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); 302 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); 303 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); 304 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary); 305 if (imgp->execpathp != 0) 306 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp); 307 if (args->execfd != -1) 308 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 309 AUXARGS_ENTRY(pos, AT_NULL, 0); 310 free(imgp->auxargs, M_TEMP); 311 imgp->auxargs = NULL; 312 313 base--; 314 suword(base, (uint64_t)imgp->args->argc); 315 316 *stack_base = (register_t *)base; 317 return (0); 318} 319 320/* 321 * Copy strings out to the new process address space, constructing new arg 322 * and env vector tables. Return a pointer to the base so that it can be used 323 * as the initial stack pointer. 324 */ 325static register_t * 326linux_copyout_strings(struct image_params *imgp) 327{ 328 int argc, envc; 329 char **vectp; 330 char *stringp, *destp; 331 register_t *stack_base; 332 struct ps_strings *arginfo; 333 char canary[LINUX_AT_RANDOM_LEN]; 334 size_t execpath_len; 335 struct proc *p; 336 337 /* Calculate string base and vector table pointers. */ 338 if (imgp->execpath != NULL && imgp->auxargs != NULL) 339 execpath_len = strlen(imgp->execpath) + 1; 340 else 341 execpath_len = 0; 342 343 p = imgp->proc; 344 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 345 destp = (caddr_t)arginfo - SPARE_USRSPACE - 346 roundup(sizeof(canary), sizeof(char *)) - 347 roundup(execpath_len, sizeof(char *)) - 348 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); 349 350 if (execpath_len != 0) { 351 imgp->execpathp = (uintptr_t)arginfo - execpath_len; 352 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 353 } 354 355 /* Prepare the canary for SSP. */ 356 arc4rand(canary, sizeof(canary), 0); 357 imgp->canary = (uintptr_t)arginfo - 358 roundup(execpath_len, sizeof(char *)) - 359 roundup(sizeof(canary), sizeof(char *)); 360 copyout(canary, (void *)imgp->canary, sizeof(canary)); 361 362 vectp = (char **)destp; 363 if (imgp->auxargs) { 364 /* 365 * Allocate room on the stack for the ELF auxargs 366 * array. It has LINUX_AT_COUNT entries. 367 */ 368 vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf64_Auxinfo), 369 sizeof(*vectp)); 370 } 371 372 /* 373 * Allocate room for the argv[] and env vectors including the 374 * terminating NULL pointers. 375 */ 376 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 377 378 /* vectp also becomes our initial stack base. */ 379 stack_base = (register_t *)vectp; 380 381 stringp = imgp->args->begin_argv; 382 argc = imgp->args->argc; 383 envc = imgp->args->envc; 384 385 /* Copy out strings - arguments and environment. */ 386 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 387 388 /* Fill in "ps_strings" struct for ps, w, etc. */ 389 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 390 suword(&arginfo->ps_nargvstr, argc); 391 392 /* Fill in argument portion of vector table. */ 393 for (; argc > 0; --argc) { 394 suword(vectp++, (long)(intptr_t)destp); 395 while (*stringp++ != 0) 396 destp++; 397 destp++; 398 } 399 400 /* A null vector table pointer separates the argp's from the envp's. */ 401 suword(vectp++, 0); 402 403 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 404 suword(&arginfo->ps_nenvstr, envc); 405 406 /* Fill in environment portion of vector table. */ 407 for (; envc > 0; --envc) { 408 suword(vectp++, (long)(intptr_t)destp); 409 while (*stringp++ != 0) 410 destp++; 411 destp++; 412 } 413 414 /* The end of the vector table is a null pointer. */ 415 suword(vectp, 0); 416 return (stack_base); 417} 418 419/* 420 * Reset registers to default values on exec. 421 */ 422static void 423linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack) 424{ 425 struct trapframe *regs; 426 struct pcb *pcb; 427 register_t saved_rflags; 428 429 regs = td->td_frame; 430 pcb = td->td_pcb; 431 432 mtx_lock(&dt_lock); 433 if (td->td_proc->p_md.md_ldt != NULL) 434 user_ldt_free(td); 435 else 436 mtx_unlock(&dt_lock); 437 438 pcb->pcb_fsbase = 0; 439 pcb->pcb_gsbase = 0; 440 clear_pcb_flags(pcb, PCB_32BIT); 441 pcb->pcb_initial_fpucw = __LINUX_NPXCW__; 442 set_pcb_flags(pcb, PCB_FULL_IRET); 443 444 saved_rflags = regs->tf_rflags & PSL_T; 445 bzero((char *)regs, sizeof(struct trapframe)); 446 regs->tf_rip = imgp->entry_addr; 447 regs->tf_rsp = stack; 448 regs->tf_rflags = PSL_USER | saved_rflags; 449 regs->tf_ss = _udatasel; 450 regs->tf_cs = _ucodesel; 451 regs->tf_ds = _udatasel; 452 regs->tf_es = _udatasel; 453 regs->tf_fs = _ufssel; 454 regs->tf_gs = _ugssel; 455 regs->tf_flags = TF_HASSEGS; 456 457 /* 458 * Reset the hardware debug registers if they were in use. 459 * They won't have any meaning for the newly exec'd process. 460 */ 461 if (pcb->pcb_flags & PCB_DBREGS) { 462 pcb->pcb_dr0 = 0; 463 pcb->pcb_dr1 = 0; 464 pcb->pcb_dr2 = 0; 465 pcb->pcb_dr3 = 0; 466 pcb->pcb_dr6 = 0; 467 pcb->pcb_dr7 = 0; 468 if (pcb == curpcb) { 469 /* 470 * Clear the debug registers on the running 471 * CPU, otherwise they will end up affecting 472 * the next process we switch to. 473 */ 474 reset_dbregs(); 475 } 476 clear_pcb_flags(pcb, PCB_DBREGS); 477 } 478 479 /* 480 * Drop the FP state if we hold it, so that the process gets a 481 * clean FP state if it uses the FPU again. 482 */ 483 fpstate_drop(td); 484} 485 486/* 487 * Copied from amd64/amd64/machdep.c 488 * 489 * XXX fpu state need? don't think so 490 */ 491int 492linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) 493{ 494 struct proc *p; 495 struct l_ucontext uc; 496 struct l_sigcontext *context; 497 struct trapframe *regs; 498 unsigned long rflags; 499 int error; 500 ksiginfo_t ksi; 501 502 regs = td->td_frame; 503 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); 504 if (error != 0) 505 return (error); 506 507 p = td->td_proc; 508 context = &uc.uc_mcontext; 509 rflags = context->sc_rflags; 510 511 /* 512 * Don't allow users to change privileged or reserved flags. 513 */ 514 /* 515 * XXX do allow users to change the privileged flag PSL_RF. 516 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers 517 * should sometimes set it there too. tf_rflags is kept in 518 * the signal context during signal handling and there is no 519 * other place to remember it, so the PSL_RF bit may be 520 * corrupted by the signal handler without us knowing. 521 * Corruption of the PSL_RF bit at worst causes one more or 522 * one less debugger trap, so allowing it is fairly harmless. 523 */ 524 525#define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 526 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 527 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags); 528 return (EINVAL); 529 } 530 531 /* 532 * Don't allow users to load a valid privileged %cs. Let the 533 * hardware check for invalid selectors, excess privilege in 534 * other selectors, invalid %eip's and invalid %esp's. 535 */ 536#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 537 if (!CS_SECURE(context->sc_cs)) { 538 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs); 539 ksiginfo_init_trap(&ksi); 540 ksi.ksi_signo = SIGBUS; 541 ksi.ksi_code = BUS_OBJERR; 542 ksi.ksi_trapno = T_PROTFLT; 543 ksi.ksi_addr = (void *)regs->tf_rip; 544 trapsignal(td, &ksi); 545 return (EINVAL); 546 } 547 548 PROC_LOCK(p); 549 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); 550 SIG_CANTMASK(td->td_sigmask); 551 signotify(td); 552 PROC_UNLOCK(p); 553 554 regs->tf_rdi = context->sc_rdi; 555 regs->tf_rsi = context->sc_rsi; 556 regs->tf_rdx = context->sc_rdx; 557 regs->tf_rbp = context->sc_rbp; 558 regs->tf_rbx = context->sc_rbx; 559 regs->tf_rcx = context->sc_rcx; 560 regs->tf_rax = context->sc_rax; 561 regs->tf_rip = context->sc_rip; 562 regs->tf_rsp = context->sc_rsp; 563 regs->tf_r8 = context->sc_r8; 564 regs->tf_r9 = context->sc_r9; 565 regs->tf_r10 = context->sc_r10; 566 regs->tf_r11 = context->sc_r11; 567 regs->tf_r12 = context->sc_r12; 568 regs->tf_r13 = context->sc_r13; 569 regs->tf_r14 = context->sc_r14; 570 regs->tf_r15 = context->sc_r15; 571 regs->tf_cs = context->sc_cs; 572 regs->tf_err = context->sc_err; 573 regs->tf_rflags = rflags; 574 575 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 576 return (EJUSTRETURN); 577} 578 579/* 580 * copied from amd64/amd64/machdep.c 581 * 582 * Send an interrupt to process. 583 */ 584static void 585linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) 586{ 587 struct l_rt_sigframe sf, *sfp; 588 struct proc *p; 589 struct thread *td; 590 struct sigacts *psp; 591 caddr_t sp; 592 struct trapframe *regs; 593 int sig, code; 594 int oonstack; 595 596 td = curthread; 597 p = td->td_proc; 598 PROC_LOCK_ASSERT(p, MA_OWNED); 599 sig = ksi->ksi_signo; 600 psp = p->p_sigacts; 601 code = ksi->ksi_code; 602 mtx_assert(&psp->ps_mtx, MA_OWNED); 603 regs = td->td_frame; 604 oonstack = sigonstack(regs->tf_rsp); 605 606 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", 607 catcher, sig, mask, code); 608 609 /* Allocate space for the signal handler context. */ 610 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && 611 SIGISMEMBER(psp->ps_sigonstack, sig)) { 612 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - 613 sizeof(struct l_rt_sigframe); 614 } else 615 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; 616 /* Align to 16 bytes. */ 617 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); 618 mtx_unlock(&psp->ps_mtx); 619 620 /* Translate the signal. */ 621 sig = bsd_to_linux_signal(sig); 622 623 /* Save user context. */ 624 bzero(&sf, sizeof(sf)); 625 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); 626 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); 627 628 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); 629 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; 630 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) 631 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; 632 PROC_UNLOCK(p); 633 634 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; 635 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; 636 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; 637 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; 638 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; 639 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; 640 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; 641 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; 642 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; 643 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; 644 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; 645 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; 646 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; 647 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; 648 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; 649 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; 650 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; 651 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; 652 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; 653 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; 654 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); 655 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; 656 657 /* Build the argument list for the signal handler. */ 658 regs->tf_rdi = sig; /* arg 1 in %rdi */ 659 regs->tf_rax = 0; 660 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ 661 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ 662 663 sf.sf_handler = catcher; 664 /* Fill in POSIX parts. */ 665 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig); 666 667 /* Copy the sigframe out to the user's stack. */ 668 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { 669#ifdef DEBUG 670 printf("process %ld has trashed its stack\n", (long)p->p_pid); 671#endif 672 PROC_LOCK(p); 673 sigexit(td, SIGILL); 674 } 675 676 regs->tf_rsp = (long)sfp; 677 regs->tf_rip = linux_rt_sigcode; 678 regs->tf_rflags &= ~(PSL_T | PSL_D); 679 regs->tf_cs = _ucodesel; 680 set_pcb_flags(td->td_pcb, PCB_FULL_IRET); 681 PROC_LOCK(p); 682 mtx_lock(&psp->ps_mtx); 683} 684 685/* 686 * If a linux binary is exec'ing something, try this image activator 687 * first. We override standard shell script execution in order to 688 * be able to modify the interpreter path. We only do this if a linux 689 * binary is doing the exec, so we do not create an EXEC module for it. 690 */ 691static int exec_linux_imgact_try(struct image_params *iparams); 692 693static int 694exec_linux_imgact_try(struct image_params *imgp) 695{ 696 const char *head = (const char *)imgp->image_header; 697 char *rpath; 698 int error = -1; 699 700 /* 701 * The interpreter for shell scripts run from a linux binary needs 702 * to be located in /compat/linux if possible in order to recursively 703 * maintain linux path emulation. 704 */ 705 if (((const short *)head)[0] == SHELLMAGIC) { 706 /* 707 * Run our normal shell image activator. If it succeeds 708 * attempt to use the alternate path for the interpreter. 709 * If an alternate path is found, use our stringspace 710 * to store it. 711 */ 712 if ((error = exec_shell_imgact(imgp)) == 0) { 713 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), 714 imgp->interpreter_name, UIO_SYSSPACE, 715 &rpath, 0, AT_FDCWD); 716 if (rpath != NULL) 717 imgp->args->fname_buf = 718 imgp->interpreter_name = rpath; 719 } 720 } 721 return (error); 722} 723 724#define LINUX_VSYSCALL_START (-10UL << 20) 725#define LINUX_VSYSCALL_SZ 1024 726 727const unsigned long linux_vsyscall_vector[] = { 728 LINUX_SYS_gettimeofday, 729 LINUX_SYS_linux_time, 730 /* getcpu not implemented */ 731}; 732 733static int 734linux_vsyscall(struct thread *td) 735{ 736 struct trapframe *frame; 737 uint64_t retqaddr; 738 int code, traced; 739 int error; 740 741 frame = td->td_frame; 742 743 /* Check %rip for vsyscall area. */ 744 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) 745 return (EINVAL); 746 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) 747 return (EINVAL); 748 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; 749 if (code >= nitems(linux_vsyscall_vector)) 750 return (EINVAL); 751 752 /* 753 * vsyscall called as callq *(%rax), so we must 754 * use return address from %rsp and also fixup %rsp. 755 */ 756 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); 757 if (error) 758 return (error); 759 760 frame->tf_rip = retqaddr; 761 frame->tf_rax = linux_vsyscall_vector[code]; 762 frame->tf_rsp += 8; 763 764 traced = (frame->tf_flags & PSL_T); 765 766 amd64_syscall(td, traced); 767 768 return (0); 769} 770 771struct sysentvec elf_linux_sysvec = { 772 .sv_size = LINUX_SYS_MAXSYSCALL, 773 .sv_table = linux_sysent, 774 .sv_mask = 0, 775 .sv_errsize = ELAST + 1, 776 .sv_errtbl = bsd_to_linux_errno, 777 .sv_transtrap = translate_traps, 778 .sv_fixup = elf_linux_fixup, 779 .sv_sendsig = linux_rt_sendsig, 780 .sv_sigcode = &_binary_linux_locore_o_start, 781 .sv_szsigcode = &linux_szsigcode, 782 .sv_name = "Linux ELF64", 783 .sv_coredump = elf64_coredump, 784 .sv_imgact_try = exec_linux_imgact_try, 785 .sv_minsigstksz = LINUX_MINSIGSTKSZ, 786 .sv_pagesize = PAGE_SIZE, 787 .sv_minuser = VM_MIN_ADDRESS, 788 .sv_maxuser = VM_MAXUSER_ADDRESS, 789 .sv_usrstack = USRSTACK, 790 .sv_psstrings = PS_STRINGS, 791 .sv_stackprot = VM_PROT_ALL, 792 .sv_copyout_strings = linux_copyout_strings, 793 .sv_setregs = linux_exec_setregs, 794 .sv_fixlimit = NULL, 795 .sv_maxssiz = NULL, 796 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP, 797 .sv_set_syscall_retval = linux_set_syscall_retval, 798 .sv_fetch_syscall_args = linux_fetch_syscall_args, 799 .sv_syscallnames = NULL, 800 .sv_shared_page_base = SHAREDPAGE, 801 .sv_shared_page_len = PAGE_SIZE, 802 .sv_schedtail = linux_schedtail, 803 .sv_thread_detach = linux_thread_detach, 804 .sv_trap = linux_vsyscall, 805}; 806 807static void 808linux_vdso_install(void *param) 809{ 810 811 amd64_lower_shared_page(&elf_linux_sysvec); 812 813 linux_szsigcode = (&_binary_linux_locore_o_end - 814 &_binary_linux_locore_o_start); 815 816 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) 817 panic("Linux invalid vdso size\n"); 818 819 __elfN(linux_vdso_fixup)(&elf_linux_sysvec); 820 821 linux_shared_page_obj = __elfN(linux_shared_page_init) 822 (&linux_shared_page_mapping); 823 824 __elfN(linux_vdso_reloc)(&elf_linux_sysvec, SHAREDPAGE); 825 826 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, 827 linux_szsigcode); 828 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; 829 830 linux_kplatform = linux_shared_page_mapping + 831 (linux_platform - (caddr_t)SHAREDPAGE); 832} 833SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, 834 (sysinit_cfunc_t)linux_vdso_install, NULL); 835 836static void 837linux_vdso_deinstall(void *param) 838{ 839 840 __elfN(linux_shared_page_fini)(linux_shared_page_obj); 841}; 842SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, 843 (sysinit_cfunc_t)linux_vdso_deinstall, NULL); 844 845static char GNULINUX_ABI_VENDOR[] = "GNU"; 846static int GNULINUX_ABI_DESC = 0; 847 848static boolean_t 849linux_trans_osrel(const Elf_Note *note, int32_t *osrel) 850{ 851 const Elf32_Word *desc; 852 uintptr_t p; 853 854 p = (uintptr_t)(note + 1); 855 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 856 857 desc = (const Elf32_Word *)p; 858 if (desc[0] != GNULINUX_ABI_DESC) 859 return (FALSE); 860 861 /* 862 * For linux we encode osrel as follows (see linux_mib.c): 863 * VVVMMMIII (version, major, minor), see linux_mib.c. 864 */ 865 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3]; 866 867 return (TRUE); 868} 869 870static Elf_Brandnote linux64_brandnote = { 871 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), 872 .hdr.n_descsz = 16, 873 .hdr.n_type = 1, 874 .vendor = GNULINUX_ABI_VENDOR, 875 .flags = BN_TRANSLATE_OSREL, 876 .trans_osrel = linux_trans_osrel 877}; 878 879static Elf64_Brandinfo linux_glibc2brand = { 880 .brand = ELFOSABI_LINUX, 881 .machine = EM_X86_64, 882 .compat_3_brand = "Linux", 883 .emul_path = "/compat/linux", 884 .interp_path = "/lib64/ld-linux-x86-64.so.2", 885 .sysvec = &elf_linux_sysvec, 886 .interp_newpath = NULL, 887 .brand_note = &linux64_brandnote, 888 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 889}; 890 891static Elf64_Brandinfo linux_glibc2brandshort = { 892 .brand = ELFOSABI_LINUX, 893 .machine = EM_X86_64, 894 .compat_3_brand = "Linux", 895 .emul_path = "/compat/linux", 896 .interp_path = "/lib64/ld-linux.so.2", 897 .sysvec = &elf_linux_sysvec, 898 .interp_newpath = NULL, 899 .brand_note = &linux64_brandnote, 900 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 901}; 902 903static Elf64_Brandinfo linux_muslbrand = { 904 .brand = ELFOSABI_LINUX, 905 .machine = EM_X86_64, 906 .compat_3_brand = "Linux", 907 .emul_path = "/compat/linux", 908 .interp_path = "/lib/ld-musl-x86_64.so.1", 909 .sysvec = &elf_linux_sysvec, 910 .interp_newpath = NULL, 911 .brand_note = &linux64_brandnote, 912 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE 913}; 914 915Elf64_Brandinfo *linux_brandlist[] = { 916 &linux_glibc2brand, 917 &linux_glibc2brandshort, 918 &linux_muslbrand, 919 NULL 920}; 921 922static int 923linux64_elf_modevent(module_t mod, int type, void *data) 924{ 925 Elf64_Brandinfo **brandinfo; 926 int error; 927 struct linux_ioctl_handler **lihp; 928 929 error = 0; 930 931 switch(type) { 932 case MOD_LOAD: 933 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 934 ++brandinfo) 935 if (elf64_insert_brand_entry(*brandinfo) < 0) 936 error = EINVAL; 937 if (error == 0) { 938 SET_FOREACH(lihp, linux_ioctl_handler_set) 939 linux_ioctl_register_handler(*lihp); 940 LIST_INIT(&futex_list); 941 mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF); 942 stclohz = (stathz ? stathz : hz); 943 if (bootverbose) 944 printf("Linux x86-64 ELF exec handler installed\n"); 945 } else 946 printf("cannot insert Linux x86-64 ELF brand handler\n"); 947 break; 948 case MOD_UNLOAD: 949 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; 950 ++brandinfo) 951 if (elf64_brand_inuse(*brandinfo)) 952 error = EBUSY; 953 if (error == 0) { 954 for (brandinfo = &linux_brandlist[0]; 955 *brandinfo != NULL; ++brandinfo) 956 if (elf64_remove_brand_entry(*brandinfo) < 0) 957 error = EINVAL; 958 } 959 if (error == 0) { 960 SET_FOREACH(lihp, linux_ioctl_handler_set) 961 linux_ioctl_unregister_handler(*lihp); 962 mtx_destroy(&futex_mtx); 963 if (bootverbose) 964 printf("Linux ELF exec handler removed\n"); 965 } else 966 printf("Could not deinstall ELF interpreter entry\n"); 967 break; 968 default: 969 return (EOPNOTSUPP); 970 } 971 return (error); 972} 973 974static moduledata_t linux64_elf_mod = { 975 "linux64elf", 976 linux64_elf_modevent, 977 0 978}; 979 980DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); 981MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); 982FEATURE(linux64, "Linux 64bit support"); 983