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