1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1990 The Regents of the University of California. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_capsicum.h" 38 #include "opt_kstack_pages.h" 39 40 #include <sys/param.h> 41 #include <sys/capsicum.h> 42 #include <sys/systm.h> 43 #include <sys/lock.h> 44 #include <sys/malloc.h> 45 #include <sys/mutex.h> 46 #include <sys/priv.h> 47 #include <sys/proc.h> 48 #include <sys/smp.h> 49 #include <sys/sysproto.h> 50 51 #include <vm/vm.h> 52 #include <vm/pmap.h> 53 #include <vm/vm_map.h> 54 #include <vm/vm_extern.h> 55 56 #include <machine/atomic.h> 57 #include <machine/cpu.h> 58 #include <machine/pcb.h> 59 #include <machine/pcb_ext.h> 60 #include <machine/proc.h> 61 #include <machine/sysarch.h> 62 63 #include <security/audit/audit.h> 64 65 #include <vm/vm_kern.h> /* for kernel_map */ 66 67 #define MAX_LD 8192 68 #define LD_PER_PAGE 512 69 #define NEW_MAX_LD(num) rounddown2(num + LD_PER_PAGE, LD_PER_PAGE) 70 #define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3) 71 #define NULL_LDT_BASE ((caddr_t)NULL) 72 73 #ifdef SMP 74 static void set_user_ldt_rv(void *arg); 75 #endif 76 static int i386_set_ldt_data(struct thread *, int start, int num, 77 union descriptor *descs); 78 static int i386_ldt_grow(struct thread *td, int len); 79 80 void 81 fill_based_sd(struct segment_descriptor *sdp, uint32_t base) 82 { 83 84 sdp->sd_lobase = base & 0xffffff; 85 sdp->sd_hibase = (base >> 24) & 0xff; 86 sdp->sd_lolimit = 0xffff; /* 4GB limit, wraps around */ 87 sdp->sd_hilimit = 0xf; 88 sdp->sd_type = SDT_MEMRWA; 89 sdp->sd_dpl = SEL_UPL; 90 sdp->sd_p = 1; 91 sdp->sd_xx = 0; 92 sdp->sd_def32 = 1; 93 sdp->sd_gran = 1; 94 } 95 96 /* 97 * Construct special descriptors for "base" selectors. Store them in 98 * the PCB for later use by cpu_switch(). Store them in the GDT for 99 * more immediate use. The GDT entries are part of the current 100 * context. Callers must load related segment registers to complete 101 * setting up the current context. 102 */ 103 void 104 set_fsbase(struct thread *td, uint32_t base) 105 { 106 struct segment_descriptor sd; 107 108 fill_based_sd(&sd, base); 109 critical_enter(); 110 td->td_pcb->pcb_fsd = sd; 111 PCPU_GET(fsgs_gdt)[0] = sd; 112 critical_exit(); 113 } 114 115 void 116 set_gsbase(struct thread *td, uint32_t base) 117 { 118 struct segment_descriptor sd; 119 120 fill_based_sd(&sd, base); 121 critical_enter(); 122 td->td_pcb->pcb_gsd = sd; 123 PCPU_GET(fsgs_gdt)[1] = sd; 124 critical_exit(); 125 } 126 127 #ifndef _SYS_SYSPROTO_H_ 128 struct sysarch_args { 129 int op; 130 char *parms; 131 }; 132 #endif 133 134 int 135 sysarch(struct thread *td, struct sysarch_args *uap) 136 { 137 int error; 138 union descriptor *lp; 139 union { 140 struct i386_ldt_args largs; 141 struct i386_ioperm_args iargs; 142 struct i386_get_xfpustate xfpu; 143 } kargs; 144 uint32_t base; 145 struct segment_descriptor *sdp; 146 147 AUDIT_ARG_CMD(uap->op); 148 149 #ifdef CAPABILITY_MODE 150 /* 151 * When adding new operations, add a new case statement here to 152 * explicitly indicate whether or not the operation is safe to 153 * perform in capability mode. 154 */ 155 if (IN_CAPABILITY_MODE(td)) { 156 switch (uap->op) { 157 case I386_GET_LDT: 158 case I386_SET_LDT: 159 case I386_GET_IOPERM: 160 case I386_GET_FSBASE: 161 case I386_SET_FSBASE: 162 case I386_GET_GSBASE: 163 case I386_SET_GSBASE: 164 case I386_GET_XFPUSTATE: 165 break; 166 167 case I386_SET_IOPERM: 168 default: 169 #ifdef KTRACE 170 if (KTRPOINT(td, KTR_CAPFAIL)) 171 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL); 172 #endif 173 return (ECAPMODE); 174 } 175 } 176 #endif 177 178 switch (uap->op) { 179 case I386_GET_IOPERM: 180 case I386_SET_IOPERM: 181 if ((error = copyin(uap->parms, &kargs.iargs, 182 sizeof(struct i386_ioperm_args))) != 0) 183 return (error); 184 break; 185 case I386_GET_LDT: 186 case I386_SET_LDT: 187 if ((error = copyin(uap->parms, &kargs.largs, 188 sizeof(struct i386_ldt_args))) != 0) 189 return (error); 190 break; 191 case I386_GET_XFPUSTATE: 192 if ((error = copyin(uap->parms, &kargs.xfpu, 193 sizeof(struct i386_get_xfpustate))) != 0) 194 return (error); 195 break; 196 default: 197 break; 198 } 199 200 switch (uap->op) { 201 case I386_GET_LDT: 202 error = i386_get_ldt(td, &kargs.largs); 203 break; 204 case I386_SET_LDT: 205 if (kargs.largs.descs != NULL) { 206 if (kargs.largs.num > MAX_LD) 207 return (EINVAL); 208 lp = malloc(kargs.largs.num * sizeof(union descriptor), 209 M_TEMP, M_WAITOK); 210 error = copyin(kargs.largs.descs, lp, 211 kargs.largs.num * sizeof(union descriptor)); 212 if (error == 0) 213 error = i386_set_ldt(td, &kargs.largs, lp); 214 free(lp, M_TEMP); 215 } else { 216 error = i386_set_ldt(td, &kargs.largs, NULL); 217 } 218 break; 219 case I386_GET_IOPERM: 220 error = i386_get_ioperm(td, &kargs.iargs); 221 if (error == 0) 222 error = copyout(&kargs.iargs, uap->parms, 223 sizeof(struct i386_ioperm_args)); 224 break; 225 case I386_SET_IOPERM: 226 error = i386_set_ioperm(td, &kargs.iargs); 227 break; 228 case I386_VM86: 229 error = vm86_sysarch(td, uap->parms); 230 break; 231 case I386_GET_FSBASE: 232 sdp = &td->td_pcb->pcb_fsd; 233 base = sdp->sd_hibase << 24 | sdp->sd_lobase; 234 error = copyout(&base, uap->parms, sizeof(base)); 235 break; 236 case I386_SET_FSBASE: 237 error = copyin(uap->parms, &base, sizeof(base)); 238 if (error == 0) { 239 /* 240 * Construct the special descriptor for fsbase 241 * and arrange for doreti to load its selector 242 * soon enough. 243 */ 244 set_fsbase(td, base); 245 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL); 246 } 247 break; 248 case I386_GET_GSBASE: 249 sdp = &td->td_pcb->pcb_gsd; 250 base = sdp->sd_hibase << 24 | sdp->sd_lobase; 251 error = copyout(&base, uap->parms, sizeof(base)); 252 break; 253 case I386_SET_GSBASE: 254 error = copyin(uap->parms, &base, sizeof(base)); 255 if (error == 0) { 256 /* 257 * Construct the special descriptor for gsbase. 258 * The selector is loaded immediately, since we 259 * normally only reload %gs on context switches. 260 */ 261 set_gsbase(td, base); 262 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 263 } 264 break; 265 case I386_GET_XFPUSTATE: 266 if (kargs.xfpu.len > cpu_max_ext_state_size - 267 sizeof(union savefpu)) 268 return (EINVAL); 269 npxgetregs(td); 270 error = copyout((char *)(get_pcb_user_save_td(td) + 1), 271 kargs.xfpu.addr, kargs.xfpu.len); 272 break; 273 default: 274 error = EINVAL; 275 break; 276 } 277 return (error); 278 } 279 280 int 281 i386_extend_pcb(struct thread *td) 282 { 283 int i, offset; 284 u_long *addr; 285 struct pcb_ext *ext; 286 struct soft_segment_descriptor ssd = { 287 0, /* segment base address (overwritten) */ 288 ctob(IOPAGES + 1) - 1, /* length */ 289 SDT_SYS386TSS, /* segment type */ 290 0, /* priority level */ 291 1, /* descriptor present */ 292 0, 0, 293 0, /* default 32 size */ 294 0 /* granularity */ 295 }; 296 297 ext = pmap_trm_alloc(ctob(IOPAGES + 1), M_WAITOK | M_ZERO); 298 /* -16 is so we can convert a trapframe into vm86trapframe inplace */ 299 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL); 300 /* 301 * The last byte of the i/o map must be followed by an 0xff byte. 302 * We arbitrarily allocate 16 bytes here, to keep the starting 303 * address on a doubleword boundary. 304 */ 305 offset = PAGE_SIZE - 16; 306 ext->ext_tss.tss_ioopt = 307 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16; 308 ext->ext_iomap = (caddr_t)ext + offset; 309 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32; 310 311 addr = (u_long *)ext->ext_vm86.vm86_intmap; 312 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++) 313 *addr++ = ~0; 314 315 ssd.ssd_base = (unsigned)&ext->ext_tss; 316 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext); 317 ssdtosd(&ssd, &ext->ext_tssd); 318 319 KASSERT(td == curthread, ("giving TSS to !curthread")); 320 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!")); 321 322 /* Switch to the new TSS. */ 323 critical_enter(); 324 ext->ext_tss.tss_esp0 = PCPU_GET(trampstk); 325 td->td_pcb->pcb_ext = ext; 326 PCPU_SET(private_tss, 1); 327 *PCPU_GET(tss_gdt) = ext->ext_tssd; 328 ltr(GSEL(GPROC0_SEL, SEL_KPL)); 329 critical_exit(); 330 331 return 0; 332 } 333 334 int 335 i386_set_ioperm(td, uap) 336 struct thread *td; 337 struct i386_ioperm_args *uap; 338 { 339 char *iomap; 340 u_int i; 341 int error; 342 343 if ((error = priv_check(td, PRIV_IO)) != 0) 344 return (error); 345 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 346 return (error); 347 /* 348 * XXX 349 * While this is restricted to root, we should probably figure out 350 * whether any other driver is using this i/o address, as so not to 351 * cause confusion. This probably requires a global 'usage registry'. 352 */ 353 354 if (td->td_pcb->pcb_ext == 0) 355 if ((error = i386_extend_pcb(td)) != 0) 356 return (error); 357 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; 358 359 if (uap->start > uap->start + uap->length || 360 uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY) 361 return (EINVAL); 362 363 for (i = uap->start; i < uap->start + uap->length; i++) { 364 if (uap->enable) 365 iomap[i >> 3] &= ~(1 << (i & 7)); 366 else 367 iomap[i >> 3] |= (1 << (i & 7)); 368 } 369 return (error); 370 } 371 372 int 373 i386_get_ioperm(td, uap) 374 struct thread *td; 375 struct i386_ioperm_args *uap; 376 { 377 int i, state; 378 char *iomap; 379 380 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY) 381 return (EINVAL); 382 383 if (td->td_pcb->pcb_ext == 0) { 384 uap->length = 0; 385 goto done; 386 } 387 388 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; 389 390 i = uap->start; 391 state = (iomap[i >> 3] >> (i & 7)) & 1; 392 uap->enable = !state; 393 uap->length = 1; 394 395 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) { 396 if (state != ((iomap[i >> 3] >> (i & 7)) & 1)) 397 break; 398 uap->length++; 399 } 400 401 done: 402 return (0); 403 } 404 405 /* 406 * Update the GDT entry pointing to the LDT to point to the LDT of the 407 * current process. Manage dt_lock holding/unholding autonomously. 408 */ 409 static void 410 set_user_ldt_locked(struct mdproc *mdp) 411 { 412 struct proc_ldt *pldt; 413 int gdt_idx; 414 415 mtx_assert(&dt_lock, MA_OWNED); 416 417 pldt = mdp->md_ldt; 418 gdt_idx = GUSERLDT_SEL; 419 gdt_idx += PCPU_GET(cpuid) * NGDT; /* always 0 on UP */ 420 gdt[gdt_idx].sd = pldt->ldt_sd; 421 lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); 422 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL)); 423 } 424 425 void 426 set_user_ldt(struct mdproc *mdp) 427 { 428 429 mtx_lock_spin(&dt_lock); 430 set_user_ldt_locked(mdp); 431 mtx_unlock_spin(&dt_lock); 432 } 433 434 #ifdef SMP 435 static void 436 set_user_ldt_rv(void *arg) 437 { 438 struct proc *p; 439 440 p = curproc; 441 if (arg == p->p_vmspace) 442 set_user_ldt(&p->p_md); 443 } 444 #endif 445 446 /* 447 * dt_lock must be held. Returns with dt_lock held. 448 */ 449 struct proc_ldt * 450 user_ldt_alloc(struct mdproc *mdp, int len) 451 { 452 struct proc_ldt *pldt, *new_ldt; 453 454 mtx_assert(&dt_lock, MA_OWNED); 455 mtx_unlock_spin(&dt_lock); 456 new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK); 457 458 new_ldt->ldt_len = len = NEW_MAX_LD(len); 459 new_ldt->ldt_base = pmap_trm_alloc(len * sizeof(union descriptor), 460 M_WAITOK | M_ZERO); 461 new_ldt->ldt_refcnt = 1; 462 new_ldt->ldt_active = 0; 463 464 mtx_lock_spin(&dt_lock); 465 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base; 466 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1; 467 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd); 468 469 if ((pldt = mdp->md_ldt) != NULL) { 470 if (len > pldt->ldt_len) 471 len = pldt->ldt_len; 472 bcopy(pldt->ldt_base, new_ldt->ldt_base, 473 len * sizeof(union descriptor)); 474 } else 475 bcopy(ldt, new_ldt->ldt_base, sizeof(union descriptor) * NLDT); 476 477 return (new_ldt); 478 } 479 480 /* 481 * Must be called with dt_lock held. Returns with dt_lock unheld. 482 */ 483 void 484 user_ldt_free(struct thread *td) 485 { 486 struct mdproc *mdp; 487 struct proc_ldt *pldt; 488 489 mtx_assert(&dt_lock, MA_OWNED); 490 mdp = &td->td_proc->p_md; 491 if ((pldt = mdp->md_ldt) == NULL) { 492 mtx_unlock_spin(&dt_lock); 493 return; 494 } 495 496 if (td == curthread) { 497 lldt(_default_ldt); 498 PCPU_SET(currentldt, _default_ldt); 499 } 500 501 mdp->md_ldt = NULL; 502 user_ldt_deref(pldt); 503 } 504 505 void 506 user_ldt_deref(struct proc_ldt *pldt) 507 { 508 509 mtx_assert(&dt_lock, MA_OWNED); 510 if (--pldt->ldt_refcnt == 0) { 511 mtx_unlock_spin(&dt_lock); 512 pmap_trm_free(pldt->ldt_base, pldt->ldt_len * 513 sizeof(union descriptor)); 514 free(pldt, M_SUBPROC); 515 } else 516 mtx_unlock_spin(&dt_lock); 517 } 518 519 /* 520 * Note for the authors of compat layers (linux, etc): copyout() in 521 * the function below is not a problem since it presents data in 522 * arch-specific format (i.e. i386-specific in this case), not in 523 * the OS-specific one. 524 */ 525 int 526 i386_get_ldt(struct thread *td, struct i386_ldt_args *uap) 527 { 528 struct proc_ldt *pldt; 529 char *data; 530 u_int nldt, num; 531 int error; 532 533 #ifdef DEBUG 534 printf("i386_get_ldt: start=%u num=%u descs=%p\n", 535 uap->start, uap->num, (void *)uap->descs); 536 #endif 537 538 num = min(uap->num, MAX_LD); 539 data = malloc(num * sizeof(union descriptor), M_TEMP, M_WAITOK); 540 mtx_lock_spin(&dt_lock); 541 pldt = td->td_proc->p_md.md_ldt; 542 nldt = pldt != NULL ? pldt->ldt_len : NLDT; 543 if (uap->start >= nldt) { 544 num = 0; 545 } else { 546 num = min(num, nldt - uap->start); 547 bcopy(pldt != NULL ? 548 &((union descriptor *)(pldt->ldt_base))[uap->start] : 549 &ldt[uap->start], data, num * sizeof(union descriptor)); 550 } 551 mtx_unlock_spin(&dt_lock); 552 error = copyout(data, uap->descs, num * sizeof(union descriptor)); 553 if (error == 0) 554 td->td_retval[0] = num; 555 free(data, M_TEMP); 556 return (error); 557 } 558 559 int 560 i386_set_ldt(struct thread *td, struct i386_ldt_args *uap, 561 union descriptor *descs) 562 { 563 struct mdproc *mdp; 564 struct proc_ldt *pldt; 565 union descriptor *dp; 566 u_int largest_ld, i; 567 int error; 568 569 #ifdef DEBUG 570 printf("i386_set_ldt: start=%u num=%u descs=%p\n", 571 uap->start, uap->num, (void *)uap->descs); 572 #endif 573 error = 0; 574 mdp = &td->td_proc->p_md; 575 576 if (descs == NULL) { 577 /* Free descriptors */ 578 if (uap->start == 0 && uap->num == 0) { 579 /* 580 * Treat this as a special case, so userland needn't 581 * know magic number NLDT. 582 */ 583 uap->start = NLDT; 584 uap->num = MAX_LD - NLDT; 585 } 586 mtx_lock_spin(&dt_lock); 587 if ((pldt = mdp->md_ldt) == NULL || 588 uap->start >= pldt->ldt_len) { 589 mtx_unlock_spin(&dt_lock); 590 return (0); 591 } 592 largest_ld = uap->start + uap->num; 593 if (largest_ld > pldt->ldt_len) 594 largest_ld = pldt->ldt_len; 595 for (i = uap->start; i < largest_ld; i++) 596 atomic_store_rel_64(&((uint64_t *)(pldt->ldt_base))[i], 597 0); 598 mtx_unlock_spin(&dt_lock); 599 return (0); 600 } 601 602 if (uap->start != LDT_AUTO_ALLOC || uap->num != 1) { 603 /* verify range of descriptors to modify */ 604 largest_ld = uap->start + uap->num; 605 if (uap->start >= MAX_LD || largest_ld > MAX_LD) 606 return (EINVAL); 607 } 608 609 /* Check descriptors for access violations */ 610 for (i = 0; i < uap->num; i++) { 611 dp = &descs[i]; 612 613 switch (dp->sd.sd_type) { 614 case SDT_SYSNULL: /* system null */ 615 dp->sd.sd_p = 0; 616 break; 617 case SDT_SYS286TSS: /* system 286 TSS available */ 618 case SDT_SYSLDT: /* system local descriptor table */ 619 case SDT_SYS286BSY: /* system 286 TSS busy */ 620 case SDT_SYSTASKGT: /* system task gate */ 621 case SDT_SYS286IGT: /* system 286 interrupt gate */ 622 case SDT_SYS286TGT: /* system 286 trap gate */ 623 case SDT_SYSNULL2: /* undefined by Intel */ 624 case SDT_SYS386TSS: /* system 386 TSS available */ 625 case SDT_SYSNULL3: /* undefined by Intel */ 626 case SDT_SYS386BSY: /* system 386 TSS busy */ 627 case SDT_SYSNULL4: /* undefined by Intel */ 628 case SDT_SYS386IGT: /* system 386 interrupt gate */ 629 case SDT_SYS386TGT: /* system 386 trap gate */ 630 case SDT_SYS286CGT: /* system 286 call gate */ 631 case SDT_SYS386CGT: /* system 386 call gate */ 632 return (EACCES); 633 634 /* memory segment types */ 635 case SDT_MEMEC: /* memory execute only conforming */ 636 case SDT_MEMEAC: /* memory execute only accessed conforming */ 637 case SDT_MEMERC: /* memory execute read conforming */ 638 case SDT_MEMERAC: /* memory execute read accessed conforming */ 639 /* Must be "present" if executable and conforming. */ 640 if (dp->sd.sd_p == 0) 641 return (EACCES); 642 break; 643 case SDT_MEMRO: /* memory read only */ 644 case SDT_MEMROA: /* memory read only accessed */ 645 case SDT_MEMRW: /* memory read write */ 646 case SDT_MEMRWA: /* memory read write accessed */ 647 case SDT_MEMROD: /* memory read only expand dwn limit */ 648 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */ 649 case SDT_MEMRWD: /* memory read write expand dwn limit */ 650 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */ 651 case SDT_MEME: /* memory execute only */ 652 case SDT_MEMEA: /* memory execute only accessed */ 653 case SDT_MEMER: /* memory execute read */ 654 case SDT_MEMERA: /* memory execute read accessed */ 655 break; 656 default: 657 return (EINVAL); 658 } 659 660 /* Only user (ring-3) descriptors may be present. */ 661 if (dp->sd.sd_p != 0 && dp->sd.sd_dpl != SEL_UPL) 662 return (EACCES); 663 } 664 665 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) { 666 /* Allocate a free slot */ 667 mtx_lock_spin(&dt_lock); 668 if ((pldt = mdp->md_ldt) == NULL) { 669 if ((error = i386_ldt_grow(td, NLDT + 1))) { 670 mtx_unlock_spin(&dt_lock); 671 return (error); 672 } 673 pldt = mdp->md_ldt; 674 } 675 again: 676 /* 677 * start scanning a bit up to leave room for NVidia and 678 * Wine, which still user the "Blat" method of allocation. 679 */ 680 dp = &((union descriptor *)(pldt->ldt_base))[NLDT]; 681 for (i = NLDT; i < pldt->ldt_len; ++i) { 682 if (dp->sd.sd_type == SDT_SYSNULL) 683 break; 684 dp++; 685 } 686 if (i >= pldt->ldt_len) { 687 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) { 688 mtx_unlock_spin(&dt_lock); 689 return (error); 690 } 691 goto again; 692 } 693 uap->start = i; 694 error = i386_set_ldt_data(td, i, 1, descs); 695 mtx_unlock_spin(&dt_lock); 696 } else { 697 largest_ld = uap->start + uap->num; 698 mtx_lock_spin(&dt_lock); 699 if (!(error = i386_ldt_grow(td, largest_ld))) { 700 error = i386_set_ldt_data(td, uap->start, uap->num, 701 descs); 702 } 703 mtx_unlock_spin(&dt_lock); 704 } 705 if (error == 0) 706 td->td_retval[0] = uap->start; 707 return (error); 708 } 709 710 static int 711 i386_set_ldt_data(struct thread *td, int start, int num, 712 union descriptor *descs) 713 { 714 struct mdproc *mdp; 715 struct proc_ldt *pldt; 716 uint64_t *dst, *src; 717 int i; 718 719 mtx_assert(&dt_lock, MA_OWNED); 720 721 mdp = &td->td_proc->p_md; 722 pldt = mdp->md_ldt; 723 dst = (uint64_t *)(pldt->ldt_base); 724 src = (uint64_t *)descs; 725 726 /* 727 * Atomic(9) is used only to get 64bit atomic store with 728 * cmpxchg8b when available. There is no op without release 729 * semantic. 730 */ 731 for (i = 0; i < num; i++) 732 atomic_store_rel_64(&dst[start + i], src[i]); 733 return (0); 734 } 735 736 static int 737 i386_ldt_grow(struct thread *td, int len) 738 { 739 struct mdproc *mdp; 740 struct proc_ldt *new_ldt, *pldt; 741 caddr_t old_ldt_base; 742 int old_ldt_len; 743 744 mtx_assert(&dt_lock, MA_OWNED); 745 746 if (len > MAX_LD) 747 return (ENOMEM); 748 if (len < NLDT + 1) 749 len = NLDT + 1; 750 751 mdp = &td->td_proc->p_md; 752 old_ldt_base = NULL_LDT_BASE; 753 old_ldt_len = 0; 754 755 /* Allocate a user ldt. */ 756 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) { 757 new_ldt = user_ldt_alloc(mdp, len); 758 if (new_ldt == NULL) 759 return (ENOMEM); 760 pldt = mdp->md_ldt; 761 762 if (pldt != NULL) { 763 if (new_ldt->ldt_len <= pldt->ldt_len) { 764 /* 765 * We just lost the race for allocation, so 766 * free the new object and return. 767 */ 768 mtx_unlock_spin(&dt_lock); 769 pmap_trm_free(new_ldt->ldt_base, 770 new_ldt->ldt_len * sizeof(union descriptor)); 771 free(new_ldt, M_SUBPROC); 772 mtx_lock_spin(&dt_lock); 773 return (0); 774 } 775 776 /* 777 * We have to substitute the current LDT entry for 778 * curproc with the new one since its size grew. 779 */ 780 old_ldt_base = pldt->ldt_base; 781 old_ldt_len = pldt->ldt_len; 782 pldt->ldt_sd = new_ldt->ldt_sd; 783 pldt->ldt_base = new_ldt->ldt_base; 784 pldt->ldt_len = new_ldt->ldt_len; 785 } else 786 mdp->md_ldt = pldt = new_ldt; 787 #ifdef SMP 788 /* 789 * Signal other cpus to reload ldt. We need to unlock dt_lock 790 * here because other CPU will contest on it since their 791 * curthreads won't hold the lock and will block when trying 792 * to acquire it. 793 */ 794 mtx_unlock_spin(&dt_lock); 795 smp_rendezvous(NULL, set_user_ldt_rv, NULL, 796 td->td_proc->p_vmspace); 797 #else 798 set_user_ldt_locked(&td->td_proc->p_md); 799 mtx_unlock_spin(&dt_lock); 800 #endif 801 if (old_ldt_base != NULL_LDT_BASE) { 802 pmap_trm_free(old_ldt_base, old_ldt_len * 803 sizeof(union descriptor)); 804 free(new_ldt, M_SUBPROC); 805 } 806 mtx_lock_spin(&dt_lock); 807 } 808 return (0); 809 } 810