1 /*- 2 * Copyright (c) 1990 The Regents of the University of California. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_capsicum.h" 36 #include "opt_kstack_pages.h" 37 38 #include <sys/param.h> 39 #include <sys/capsicum.h> 40 #include <sys/systm.h> 41 #include <sys/lock.h> 42 #include <sys/malloc.h> 43 #include <sys/mutex.h> 44 #include <sys/priv.h> 45 #include <sys/proc.h> 46 #include <sys/smp.h> 47 #include <sys/sysproto.h> 48 49 #include <vm/vm.h> 50 #include <vm/pmap.h> 51 #include <vm/vm_map.h> 52 #include <vm/vm_extern.h> 53 54 #include <machine/cpu.h> 55 #include <machine/pcb.h> 56 #include <machine/pcb_ext.h> 57 #include <machine/proc.h> 58 #include <machine/sysarch.h> 59 60 #include <security/audit/audit.h> 61 62 #include <vm/vm_kern.h> /* for kernel_map */ 63 64 #define MAX_LD 8192 65 #define LD_PER_PAGE 512 66 #define NEW_MAX_LD(num) rounddown2(num + LD_PER_PAGE, LD_PER_PAGE) 67 #define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3) 68 #define NULL_LDT_BASE ((caddr_t)NULL) 69 70 #ifdef SMP 71 static void set_user_ldt_rv(struct vmspace *vmsp); 72 #endif 73 static int i386_set_ldt_data(struct thread *, int start, int num, 74 union descriptor *descs); 75 static int i386_ldt_grow(struct thread *td, int len); 76 77 void 78 fill_based_sd(struct segment_descriptor *sdp, uint32_t base) 79 { 80 81 sdp->sd_lobase = base & 0xffffff; 82 sdp->sd_hibase = (base >> 24) & 0xff; 83 sdp->sd_lolimit = 0xffff; /* 4GB limit, wraps around */ 84 sdp->sd_hilimit = 0xf; 85 sdp->sd_type = SDT_MEMRWA; 86 sdp->sd_dpl = SEL_UPL; 87 sdp->sd_p = 1; 88 sdp->sd_xx = 0; 89 sdp->sd_def32 = 1; 90 sdp->sd_gran = 1; 91 } 92 93 #ifndef _SYS_SYSPROTO_H_ 94 struct sysarch_args { 95 int op; 96 char *parms; 97 }; 98 #endif 99 100 int 101 sysarch(td, uap) 102 struct thread *td; 103 register struct sysarch_args *uap; 104 { 105 int error; 106 union descriptor *lp; 107 union { 108 struct i386_ldt_args largs; 109 struct i386_ioperm_args iargs; 110 struct i386_get_xfpustate xfpu; 111 } kargs; 112 uint32_t base; 113 struct segment_descriptor sd, *sdp; 114 115 AUDIT_ARG_CMD(uap->op); 116 117 #ifdef CAPABILITY_MODE 118 /* 119 * When adding new operations, add a new case statement here to 120 * explicitly indicate whether or not the operation is safe to 121 * perform in capability mode. 122 */ 123 if (IN_CAPABILITY_MODE(td)) { 124 switch (uap->op) { 125 case I386_GET_LDT: 126 case I386_SET_LDT: 127 case I386_GET_IOPERM: 128 case I386_GET_FSBASE: 129 case I386_SET_FSBASE: 130 case I386_GET_GSBASE: 131 case I386_SET_GSBASE: 132 case I386_GET_XFPUSTATE: 133 break; 134 135 case I386_SET_IOPERM: 136 default: 137 #ifdef KTRACE 138 if (KTRPOINT(td, KTR_CAPFAIL)) 139 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL); 140 #endif 141 return (ECAPMODE); 142 } 143 } 144 #endif 145 146 switch (uap->op) { 147 case I386_GET_IOPERM: 148 case I386_SET_IOPERM: 149 if ((error = copyin(uap->parms, &kargs.iargs, 150 sizeof(struct i386_ioperm_args))) != 0) 151 return (error); 152 break; 153 case I386_GET_LDT: 154 case I386_SET_LDT: 155 if ((error = copyin(uap->parms, &kargs.largs, 156 sizeof(struct i386_ldt_args))) != 0) 157 return (error); 158 if (kargs.largs.num > MAX_LD || kargs.largs.num <= 0) 159 return (EINVAL); 160 break; 161 case I386_GET_XFPUSTATE: 162 if ((error = copyin(uap->parms, &kargs.xfpu, 163 sizeof(struct i386_get_xfpustate))) != 0) 164 return (error); 165 break; 166 default: 167 break; 168 } 169 170 switch(uap->op) { 171 case I386_GET_LDT: 172 error = i386_get_ldt(td, &kargs.largs); 173 break; 174 case I386_SET_LDT: 175 if (kargs.largs.descs != NULL) { 176 lp = (union descriptor *)malloc( 177 kargs.largs.num * sizeof(union descriptor), 178 M_TEMP, M_WAITOK); 179 error = copyin(kargs.largs.descs, lp, 180 kargs.largs.num * sizeof(union descriptor)); 181 if (error == 0) 182 error = i386_set_ldt(td, &kargs.largs, lp); 183 free(lp, M_TEMP); 184 } else { 185 error = i386_set_ldt(td, &kargs.largs, NULL); 186 } 187 break; 188 case I386_GET_IOPERM: 189 error = i386_get_ioperm(td, &kargs.iargs); 190 if (error == 0) 191 error = copyout(&kargs.iargs, uap->parms, 192 sizeof(struct i386_ioperm_args)); 193 break; 194 case I386_SET_IOPERM: 195 error = i386_set_ioperm(td, &kargs.iargs); 196 break; 197 case I386_VM86: 198 error = vm86_sysarch(td, uap->parms); 199 break; 200 case I386_GET_FSBASE: 201 sdp = &td->td_pcb->pcb_fsd; 202 base = sdp->sd_hibase << 24 | sdp->sd_lobase; 203 error = copyout(&base, uap->parms, sizeof(base)); 204 break; 205 case I386_SET_FSBASE: 206 error = copyin(uap->parms, &base, sizeof(base)); 207 if (error == 0) { 208 /* 209 * Construct a descriptor and store it in the pcb for 210 * the next context switch. Also store it in the gdt 211 * so that the load of tf_fs into %fs will activate it 212 * at return to userland. 213 */ 214 fill_based_sd(&sd, base); 215 critical_enter(); 216 td->td_pcb->pcb_fsd = sd; 217 PCPU_GET(fsgs_gdt)[0] = sd; 218 critical_exit(); 219 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL); 220 } 221 break; 222 case I386_GET_GSBASE: 223 sdp = &td->td_pcb->pcb_gsd; 224 base = sdp->sd_hibase << 24 | sdp->sd_lobase; 225 error = copyout(&base, uap->parms, sizeof(base)); 226 break; 227 case I386_SET_GSBASE: 228 error = copyin(uap->parms, &base, sizeof(base)); 229 if (error == 0) { 230 /* 231 * Construct a descriptor and store it in the pcb for 232 * the next context switch. Also store it in the gdt 233 * because we have to do a load_gs() right now. 234 */ 235 fill_based_sd(&sd, base); 236 critical_enter(); 237 td->td_pcb->pcb_gsd = sd; 238 PCPU_GET(fsgs_gdt)[1] = sd; 239 critical_exit(); 240 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 241 } 242 break; 243 case I386_GET_XFPUSTATE: 244 if (kargs.xfpu.len > cpu_max_ext_state_size - 245 sizeof(union savefpu)) 246 return (EINVAL); 247 npxgetregs(td); 248 error = copyout((char *)(get_pcb_user_save_td(td) + 1), 249 kargs.xfpu.addr, kargs.xfpu.len); 250 break; 251 default: 252 error = EINVAL; 253 break; 254 } 255 return (error); 256 } 257 258 int 259 i386_extend_pcb(struct thread *td) 260 { 261 int i, offset; 262 u_long *addr; 263 struct pcb_ext *ext; 264 struct soft_segment_descriptor ssd = { 265 0, /* segment base address (overwritten) */ 266 ctob(IOPAGES + 1) - 1, /* length */ 267 SDT_SYS386TSS, /* segment type */ 268 0, /* priority level */ 269 1, /* descriptor present */ 270 0, 0, 271 0, /* default 32 size */ 272 0 /* granularity */ 273 }; 274 275 ext = (struct pcb_ext *)kmem_malloc(kernel_arena, ctob(IOPAGES+1), 276 M_WAITOK | M_ZERO); 277 /* -16 is so we can convert a trapframe into vm86trapframe inplace */ 278 ext->ext_tss.tss_esp0 = (vm_offset_t)td->td_pcb - 16; 279 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL); 280 /* 281 * The last byte of the i/o map must be followed by an 0xff byte. 282 * We arbitrarily allocate 16 bytes here, to keep the starting 283 * address on a doubleword boundary. 284 */ 285 offset = PAGE_SIZE - 16; 286 ext->ext_tss.tss_ioopt = 287 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16; 288 ext->ext_iomap = (caddr_t)ext + offset; 289 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32; 290 291 addr = (u_long *)ext->ext_vm86.vm86_intmap; 292 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++) 293 *addr++ = ~0; 294 295 ssd.ssd_base = (unsigned)&ext->ext_tss; 296 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext); 297 ssdtosd(&ssd, &ext->ext_tssd); 298 299 KASSERT(td == curthread, ("giving TSS to !curthread")); 300 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!")); 301 302 /* Switch to the new TSS. */ 303 critical_enter(); 304 td->td_pcb->pcb_ext = ext; 305 PCPU_SET(private_tss, 1); 306 *PCPU_GET(tss_gdt) = ext->ext_tssd; 307 ltr(GSEL(GPROC0_SEL, SEL_KPL)); 308 critical_exit(); 309 310 return 0; 311 } 312 313 int 314 i386_set_ioperm(td, uap) 315 struct thread *td; 316 struct i386_ioperm_args *uap; 317 { 318 char *iomap; 319 u_int i; 320 int error; 321 322 if ((error = priv_check(td, PRIV_IO)) != 0) 323 return (error); 324 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 325 return (error); 326 /* 327 * XXX 328 * While this is restricted to root, we should probably figure out 329 * whether any other driver is using this i/o address, as so not to 330 * cause confusion. This probably requires a global 'usage registry'. 331 */ 332 333 if (td->td_pcb->pcb_ext == 0) 334 if ((error = i386_extend_pcb(td)) != 0) 335 return (error); 336 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; 337 338 if (uap->start > uap->start + uap->length || 339 uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY) 340 return (EINVAL); 341 342 for (i = uap->start; i < uap->start + uap->length; i++) { 343 if (uap->enable) 344 iomap[i >> 3] &= ~(1 << (i & 7)); 345 else 346 iomap[i >> 3] |= (1 << (i & 7)); 347 } 348 return (error); 349 } 350 351 int 352 i386_get_ioperm(td, uap) 353 struct thread *td; 354 struct i386_ioperm_args *uap; 355 { 356 int i, state; 357 char *iomap; 358 359 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY) 360 return (EINVAL); 361 362 if (td->td_pcb->pcb_ext == 0) { 363 uap->length = 0; 364 goto done; 365 } 366 367 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; 368 369 i = uap->start; 370 state = (iomap[i >> 3] >> (i & 7)) & 1; 371 uap->enable = !state; 372 uap->length = 1; 373 374 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) { 375 if (state != ((iomap[i >> 3] >> (i & 7)) & 1)) 376 break; 377 uap->length++; 378 } 379 380 done: 381 return (0); 382 } 383 384 /* 385 * Update the GDT entry pointing to the LDT to point to the LDT of the 386 * current process. Manage dt_lock holding/unholding autonomously. 387 */ 388 void 389 set_user_ldt(struct mdproc *mdp) 390 { 391 struct proc_ldt *pldt; 392 int dtlocked; 393 394 dtlocked = 0; 395 if (!mtx_owned(&dt_lock)) { 396 mtx_lock_spin(&dt_lock); 397 dtlocked = 1; 398 } 399 400 pldt = mdp->md_ldt; 401 #ifdef SMP 402 gdt[PCPU_GET(cpuid) * NGDT + GUSERLDT_SEL].sd = pldt->ldt_sd; 403 #else 404 gdt[GUSERLDT_SEL].sd = pldt->ldt_sd; 405 #endif 406 lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); 407 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL)); 408 if (dtlocked) 409 mtx_unlock_spin(&dt_lock); 410 } 411 412 #ifdef SMP 413 static void 414 set_user_ldt_rv(struct vmspace *vmsp) 415 { 416 struct thread *td; 417 418 td = curthread; 419 if (vmsp != td->td_proc->p_vmspace) 420 return; 421 422 set_user_ldt(&td->td_proc->p_md); 423 } 424 #endif 425 426 /* 427 * dt_lock must be held. Returns with dt_lock held. 428 */ 429 struct proc_ldt * 430 user_ldt_alloc(struct mdproc *mdp, int len) 431 { 432 struct proc_ldt *pldt, *new_ldt; 433 434 mtx_assert(&dt_lock, MA_OWNED); 435 mtx_unlock_spin(&dt_lock); 436 new_ldt = malloc(sizeof(struct proc_ldt), 437 M_SUBPROC, M_WAITOK); 438 439 new_ldt->ldt_len = len = NEW_MAX_LD(len); 440 new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena, 441 len * sizeof(union descriptor), M_WAITOK | M_ZERO); 442 new_ldt->ldt_refcnt = 1; 443 new_ldt->ldt_active = 0; 444 445 mtx_lock_spin(&dt_lock); 446 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base; 447 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1; 448 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd); 449 450 if ((pldt = mdp->md_ldt) != NULL) { 451 if (len > pldt->ldt_len) 452 len = pldt->ldt_len; 453 bcopy(pldt->ldt_base, new_ldt->ldt_base, 454 len * sizeof(union descriptor)); 455 } else 456 bcopy(ldt, new_ldt->ldt_base, sizeof(ldt)); 457 458 return (new_ldt); 459 } 460 461 /* 462 * Must be called with dt_lock held. Returns with dt_lock unheld. 463 */ 464 void 465 user_ldt_free(struct thread *td) 466 { 467 struct mdproc *mdp = &td->td_proc->p_md; 468 struct proc_ldt *pldt; 469 470 mtx_assert(&dt_lock, MA_OWNED); 471 if ((pldt = mdp->md_ldt) == NULL) { 472 mtx_unlock_spin(&dt_lock); 473 return; 474 } 475 476 if (td == curthread) { 477 lldt(_default_ldt); 478 PCPU_SET(currentldt, _default_ldt); 479 } 480 481 mdp->md_ldt = NULL; 482 user_ldt_deref(pldt); 483 } 484 485 void 486 user_ldt_deref(struct proc_ldt *pldt) 487 { 488 489 mtx_assert(&dt_lock, MA_OWNED); 490 if (--pldt->ldt_refcnt == 0) { 491 mtx_unlock_spin(&dt_lock); 492 kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base, 493 pldt->ldt_len * sizeof(union descriptor)); 494 free(pldt, M_SUBPROC); 495 } else 496 mtx_unlock_spin(&dt_lock); 497 } 498 499 /* 500 * Note for the authors of compat layers (linux, etc): copyout() in 501 * the function below is not a problem since it presents data in 502 * arch-specific format (i.e. i386-specific in this case), not in 503 * the OS-specific one. 504 */ 505 int 506 i386_get_ldt(td, uap) 507 struct thread *td; 508 struct i386_ldt_args *uap; 509 { 510 int error = 0; 511 struct proc_ldt *pldt; 512 int nldt, num; 513 union descriptor *lp; 514 515 #ifdef DEBUG 516 printf("i386_get_ldt: start=%d num=%d descs=%p\n", 517 uap->start, uap->num, (void *)uap->descs); 518 #endif 519 520 mtx_lock_spin(&dt_lock); 521 if ((pldt = td->td_proc->p_md.md_ldt) != NULL) { 522 nldt = pldt->ldt_len; 523 lp = &((union descriptor *)(pldt->ldt_base))[uap->start]; 524 mtx_unlock_spin(&dt_lock); 525 num = min(uap->num, nldt); 526 } else { 527 mtx_unlock_spin(&dt_lock); 528 nldt = sizeof(ldt)/sizeof(ldt[0]); 529 num = min(uap->num, nldt); 530 lp = &ldt[uap->start]; 531 } 532 533 if ((uap->start > (unsigned int)nldt) || 534 ((unsigned int)num > (unsigned int)nldt) || 535 ((unsigned int)(uap->start + num) > (unsigned int)nldt)) 536 return(EINVAL); 537 538 error = copyout(lp, uap->descs, num * sizeof(union descriptor)); 539 if (!error) 540 td->td_retval[0] = num; 541 542 return(error); 543 } 544 545 int 546 i386_set_ldt(td, uap, descs) 547 struct thread *td; 548 struct i386_ldt_args *uap; 549 union descriptor *descs; 550 { 551 int error = 0, i; 552 int largest_ld; 553 struct mdproc *mdp = &td->td_proc->p_md; 554 struct proc_ldt *pldt; 555 union descriptor *dp; 556 557 #ifdef DEBUG 558 printf("i386_set_ldt: start=%d num=%d descs=%p\n", 559 uap->start, uap->num, (void *)uap->descs); 560 #endif 561 562 if (descs == NULL) { 563 /* Free descriptors */ 564 if (uap->start == 0 && uap->num == 0) { 565 /* 566 * Treat this as a special case, so userland needn't 567 * know magic number NLDT. 568 */ 569 uap->start = NLDT; 570 uap->num = MAX_LD - NLDT; 571 } 572 if (uap->num == 0) 573 return (EINVAL); 574 mtx_lock_spin(&dt_lock); 575 if ((pldt = mdp->md_ldt) == NULL || 576 uap->start >= pldt->ldt_len) { 577 mtx_unlock_spin(&dt_lock); 578 return (0); 579 } 580 largest_ld = uap->start + uap->num; 581 if (largest_ld > pldt->ldt_len) 582 largest_ld = pldt->ldt_len; 583 i = largest_ld - uap->start; 584 bzero(&((union descriptor *)(pldt->ldt_base))[uap->start], 585 sizeof(union descriptor) * i); 586 mtx_unlock_spin(&dt_lock); 587 return (0); 588 } 589 590 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) { 591 /* verify range of descriptors to modify */ 592 largest_ld = uap->start + uap->num; 593 if (uap->start >= MAX_LD || largest_ld > MAX_LD) { 594 return (EINVAL); 595 } 596 } 597 598 /* Check descriptors for access violations */ 599 for (i = 0; i < uap->num; i++) { 600 dp = &descs[i]; 601 602 switch (dp->sd.sd_type) { 603 case SDT_SYSNULL: /* system null */ 604 dp->sd.sd_p = 0; 605 break; 606 case SDT_SYS286TSS: /* system 286 TSS available */ 607 case SDT_SYSLDT: /* system local descriptor table */ 608 case SDT_SYS286BSY: /* system 286 TSS busy */ 609 case SDT_SYSTASKGT: /* system task gate */ 610 case SDT_SYS286IGT: /* system 286 interrupt gate */ 611 case SDT_SYS286TGT: /* system 286 trap gate */ 612 case SDT_SYSNULL2: /* undefined by Intel */ 613 case SDT_SYS386TSS: /* system 386 TSS available */ 614 case SDT_SYSNULL3: /* undefined by Intel */ 615 case SDT_SYS386BSY: /* system 386 TSS busy */ 616 case SDT_SYSNULL4: /* undefined by Intel */ 617 case SDT_SYS386IGT: /* system 386 interrupt gate */ 618 case SDT_SYS386TGT: /* system 386 trap gate */ 619 case SDT_SYS286CGT: /* system 286 call gate */ 620 case SDT_SYS386CGT: /* system 386 call gate */ 621 /* I can't think of any reason to allow a user proc 622 * to create a segment of these types. They are 623 * for OS use only. 624 */ 625 return (EACCES); 626 /*NOTREACHED*/ 627 628 /* memory segment types */ 629 case SDT_MEMEC: /* memory execute only conforming */ 630 case SDT_MEMEAC: /* memory execute only accessed conforming */ 631 case SDT_MEMERC: /* memory execute read conforming */ 632 case SDT_MEMERAC: /* memory execute read accessed conforming */ 633 /* Must be "present" if executable and conforming. */ 634 if (dp->sd.sd_p == 0) 635 return (EACCES); 636 break; 637 case SDT_MEMRO: /* memory read only */ 638 case SDT_MEMROA: /* memory read only accessed */ 639 case SDT_MEMRW: /* memory read write */ 640 case SDT_MEMRWA: /* memory read write accessed */ 641 case SDT_MEMROD: /* memory read only expand dwn limit */ 642 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */ 643 case SDT_MEMRWD: /* memory read write expand dwn limit */ 644 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */ 645 case SDT_MEME: /* memory execute only */ 646 case SDT_MEMEA: /* memory execute only accessed */ 647 case SDT_MEMER: /* memory execute read */ 648 case SDT_MEMERA: /* memory execute read accessed */ 649 break; 650 default: 651 return(EINVAL); 652 /*NOTREACHED*/ 653 } 654 655 /* Only user (ring-3) descriptors may be present. */ 656 if ((dp->sd.sd_p != 0) && (dp->sd.sd_dpl != SEL_UPL)) 657 return (EACCES); 658 } 659 660 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) { 661 /* Allocate a free slot */ 662 mtx_lock_spin(&dt_lock); 663 if ((pldt = mdp->md_ldt) == NULL) { 664 if ((error = i386_ldt_grow(td, NLDT + 1))) { 665 mtx_unlock_spin(&dt_lock); 666 return (error); 667 } 668 pldt = mdp->md_ldt; 669 } 670 again: 671 /* 672 * start scanning a bit up to leave room for NVidia and 673 * Wine, which still user the "Blat" method of allocation. 674 */ 675 dp = &((union descriptor *)(pldt->ldt_base))[NLDT]; 676 for (i = NLDT; i < pldt->ldt_len; ++i) { 677 if (dp->sd.sd_type == SDT_SYSNULL) 678 break; 679 dp++; 680 } 681 if (i >= pldt->ldt_len) { 682 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) { 683 mtx_unlock_spin(&dt_lock); 684 return (error); 685 } 686 goto again; 687 } 688 uap->start = i; 689 error = i386_set_ldt_data(td, i, 1, descs); 690 mtx_unlock_spin(&dt_lock); 691 } else { 692 largest_ld = uap->start + uap->num; 693 mtx_lock_spin(&dt_lock); 694 if (!(error = i386_ldt_grow(td, largest_ld))) { 695 error = i386_set_ldt_data(td, uap->start, uap->num, 696 descs); 697 } 698 mtx_unlock_spin(&dt_lock); 699 } 700 if (error == 0) 701 td->td_retval[0] = uap->start; 702 return (error); 703 } 704 705 static int 706 i386_set_ldt_data(struct thread *td, int start, int num, 707 union descriptor *descs) 708 { 709 struct mdproc *mdp = &td->td_proc->p_md; 710 struct proc_ldt *pldt = mdp->md_ldt; 711 712 mtx_assert(&dt_lock, MA_OWNED); 713 714 /* Fill in range */ 715 bcopy(descs, 716 &((union descriptor *)(pldt->ldt_base))[start], 717 num * sizeof(union descriptor)); 718 return (0); 719 } 720 721 static int 722 i386_ldt_grow(struct thread *td, int len) 723 { 724 struct mdproc *mdp = &td->td_proc->p_md; 725 struct proc_ldt *new_ldt, *pldt; 726 caddr_t old_ldt_base = NULL_LDT_BASE; 727 int old_ldt_len = 0; 728 729 mtx_assert(&dt_lock, MA_OWNED); 730 731 if (len > MAX_LD) 732 return (ENOMEM); 733 if (len < NLDT + 1) 734 len = NLDT + 1; 735 736 /* Allocate a user ldt. */ 737 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) { 738 new_ldt = user_ldt_alloc(mdp, len); 739 if (new_ldt == NULL) 740 return (ENOMEM); 741 pldt = mdp->md_ldt; 742 743 if (pldt != NULL) { 744 if (new_ldt->ldt_len <= pldt->ldt_len) { 745 /* 746 * We just lost the race for allocation, so 747 * free the new object and return. 748 */ 749 mtx_unlock_spin(&dt_lock); 750 kmem_free(kernel_arena, 751 (vm_offset_t)new_ldt->ldt_base, 752 new_ldt->ldt_len * sizeof(union descriptor)); 753 free(new_ldt, M_SUBPROC); 754 mtx_lock_spin(&dt_lock); 755 return (0); 756 } 757 758 /* 759 * We have to substitute the current LDT entry for 760 * curproc with the new one since its size grew. 761 */ 762 old_ldt_base = pldt->ldt_base; 763 old_ldt_len = pldt->ldt_len; 764 pldt->ldt_sd = new_ldt->ldt_sd; 765 pldt->ldt_base = new_ldt->ldt_base; 766 pldt->ldt_len = new_ldt->ldt_len; 767 } else 768 mdp->md_ldt = pldt = new_ldt; 769 #ifdef SMP 770 /* 771 * Signal other cpus to reload ldt. We need to unlock dt_lock 772 * here because other CPU will contest on it since their 773 * curthreads won't hold the lock and will block when trying 774 * to acquire it. 775 */ 776 mtx_unlock_spin(&dt_lock); 777 smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv, 778 NULL, td->td_proc->p_vmspace); 779 #else 780 set_user_ldt(&td->td_proc->p_md); 781 mtx_unlock_spin(&dt_lock); 782 #endif 783 if (old_ldt_base != NULL_LDT_BASE) { 784 kmem_free(kernel_arena, (vm_offset_t)old_ldt_base, 785 old_ldt_len * sizeof(union descriptor)); 786 free(new_ldt, M_SUBPROC); 787 } 788 mtx_lock_spin(&dt_lock); 789 } 790 return (0); 791 } 792