sys_machdep.c revision 331722
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: stable/11/sys/i386/i386/sys_machdep.c 331722 2018-03-29 02:50:57Z eadler $"); 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/atomic.h> 55#include <machine/cpu.h> 56#include <machine/pcb.h> 57#include <machine/pcb_ext.h> 58#include <machine/proc.h> 59#include <machine/sysarch.h> 60 61#include <security/audit/audit.h> 62 63#include <vm/vm_kern.h> /* for kernel_map */ 64 65#define MAX_LD 8192 66#define LD_PER_PAGE 512 67#define NEW_MAX_LD(num) rounddown2(num + LD_PER_PAGE, LD_PER_PAGE) 68#define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3) 69#define NULL_LDT_BASE ((caddr_t)NULL) 70 71#ifdef SMP 72static void set_user_ldt_rv(void *arg); 73#endif 74static int i386_set_ldt_data(struct thread *, int start, int num, 75 union descriptor *descs); 76static int i386_ldt_grow(struct thread *td, int len); 77 78void 79fill_based_sd(struct segment_descriptor *sdp, uint32_t base) 80{ 81 82 sdp->sd_lobase = base & 0xffffff; 83 sdp->sd_hibase = (base >> 24) & 0xff; 84 sdp->sd_lolimit = 0xffff; /* 4GB limit, wraps around */ 85 sdp->sd_hilimit = 0xf; 86 sdp->sd_type = SDT_MEMRWA; 87 sdp->sd_dpl = SEL_UPL; 88 sdp->sd_p = 1; 89 sdp->sd_xx = 0; 90 sdp->sd_def32 = 1; 91 sdp->sd_gran = 1; 92} 93 94/* 95 * Construct special descriptors for "base" selectors. Store them in 96 * the PCB for later use by cpu_switch(). Store them in the GDT for 97 * more immediate use. The GDT entries are part of the current 98 * context. Callers must load related segment registers to complete 99 * setting up the current context. 100 */ 101void 102set_fsbase(struct thread *td, uint32_t base) 103{ 104 struct segment_descriptor sd; 105 106 fill_based_sd(&sd, base); 107 critical_enter(); 108 td->td_pcb->pcb_fsd = sd; 109 PCPU_GET(fsgs_gdt)[0] = sd; 110 critical_exit(); 111} 112 113void 114set_gsbase(struct thread *td, uint32_t base) 115{ 116 struct segment_descriptor sd; 117 118 fill_based_sd(&sd, base); 119 critical_enter(); 120 td->td_pcb->pcb_gsd = sd; 121 PCPU_GET(fsgs_gdt)[1] = sd; 122 critical_exit(); 123} 124 125#ifndef _SYS_SYSPROTO_H_ 126struct sysarch_args { 127 int op; 128 char *parms; 129}; 130#endif 131 132int 133sysarch(struct thread *td, struct sysarch_args *uap) 134{ 135 int error; 136 union descriptor *lp; 137 union { 138 struct i386_ldt_args largs; 139 struct i386_ioperm_args iargs; 140 struct i386_get_xfpustate xfpu; 141 } kargs; 142 uint32_t base; 143 struct segment_descriptor *sdp; 144 145 AUDIT_ARG_CMD(uap->op); 146 147#ifdef CAPABILITY_MODE 148 /* 149 * When adding new operations, add a new case statement here to 150 * explicitly indicate whether or not the operation is safe to 151 * perform in capability mode. 152 */ 153 if (IN_CAPABILITY_MODE(td)) { 154 switch (uap->op) { 155 case I386_GET_LDT: 156 case I386_SET_LDT: 157 case I386_GET_IOPERM: 158 case I386_GET_FSBASE: 159 case I386_SET_FSBASE: 160 case I386_GET_GSBASE: 161 case I386_SET_GSBASE: 162 case I386_GET_XFPUSTATE: 163 break; 164 165 case I386_SET_IOPERM: 166 default: 167#ifdef KTRACE 168 if (KTRPOINT(td, KTR_CAPFAIL)) 169 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL); 170#endif 171 return (ECAPMODE); 172 } 173 } 174#endif 175 176 switch (uap->op) { 177 case I386_GET_IOPERM: 178 case I386_SET_IOPERM: 179 if ((error = copyin(uap->parms, &kargs.iargs, 180 sizeof(struct i386_ioperm_args))) != 0) 181 return (error); 182 break; 183 case I386_GET_LDT: 184 case I386_SET_LDT: 185 if ((error = copyin(uap->parms, &kargs.largs, 186 sizeof(struct i386_ldt_args))) != 0) 187 return (error); 188 break; 189 case I386_GET_XFPUSTATE: 190 if ((error = copyin(uap->parms, &kargs.xfpu, 191 sizeof(struct i386_get_xfpustate))) != 0) 192 return (error); 193 break; 194 default: 195 break; 196 } 197 198 switch (uap->op) { 199 case I386_GET_LDT: 200 error = i386_get_ldt(td, &kargs.largs); 201 break; 202 case I386_SET_LDT: 203 if (kargs.largs.descs != NULL) { 204 if (kargs.largs.num > MAX_LD) 205 return (EINVAL); 206 lp = malloc(kargs.largs.num * sizeof(union descriptor), 207 M_TEMP, M_WAITOK); 208 error = copyin(kargs.largs.descs, lp, 209 kargs.largs.num * sizeof(union descriptor)); 210 if (error == 0) 211 error = i386_set_ldt(td, &kargs.largs, lp); 212 free(lp, M_TEMP); 213 } else { 214 error = i386_set_ldt(td, &kargs.largs, NULL); 215 } 216 break; 217 case I386_GET_IOPERM: 218 error = i386_get_ioperm(td, &kargs.iargs); 219 if (error == 0) 220 error = copyout(&kargs.iargs, uap->parms, 221 sizeof(struct i386_ioperm_args)); 222 break; 223 case I386_SET_IOPERM: 224 error = i386_set_ioperm(td, &kargs.iargs); 225 break; 226 case I386_VM86: 227 error = vm86_sysarch(td, uap->parms); 228 break; 229 case I386_GET_FSBASE: 230 sdp = &td->td_pcb->pcb_fsd; 231 base = sdp->sd_hibase << 24 | sdp->sd_lobase; 232 error = copyout(&base, uap->parms, sizeof(base)); 233 break; 234 case I386_SET_FSBASE: 235 error = copyin(uap->parms, &base, sizeof(base)); 236 if (error == 0) { 237 /* 238 * Construct the special descriptor for fsbase 239 * and arrange for doreti to load its selector 240 * soon enough. 241 */ 242 set_fsbase(td, base); 243 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL); 244 } 245 break; 246 case I386_GET_GSBASE: 247 sdp = &td->td_pcb->pcb_gsd; 248 base = sdp->sd_hibase << 24 | sdp->sd_lobase; 249 error = copyout(&base, uap->parms, sizeof(base)); 250 break; 251 case I386_SET_GSBASE: 252 error = copyin(uap->parms, &base, sizeof(base)); 253 if (error == 0) { 254 /* 255 * Construct the special descriptor for gsbase. 256 * The selector is loaded immediately, since we 257 * normally only reload %gs on context switches. 258 */ 259 set_gsbase(td, base); 260 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 261 } 262 break; 263 case I386_GET_XFPUSTATE: 264 if (kargs.xfpu.len > cpu_max_ext_state_size - 265 sizeof(union savefpu)) 266 return (EINVAL); 267 npxgetregs(td); 268 error = copyout((char *)(get_pcb_user_save_td(td) + 1), 269 kargs.xfpu.addr, kargs.xfpu.len); 270 break; 271 default: 272 error = EINVAL; 273 break; 274 } 275 return (error); 276} 277 278int 279i386_extend_pcb(struct thread *td) 280{ 281 int i, offset; 282 u_long *addr; 283 struct pcb_ext *ext; 284 struct soft_segment_descriptor ssd = { 285 0, /* segment base address (overwritten) */ 286 ctob(IOPAGES + 1) - 1, /* length */ 287 SDT_SYS386TSS, /* segment type */ 288 0, /* priority level */ 289 1, /* descriptor present */ 290 0, 0, 291 0, /* default 32 size */ 292 0 /* granularity */ 293 }; 294 295 ext = (struct pcb_ext *)kmem_malloc(kernel_arena, ctob(IOPAGES+1), 296 M_WAITOK | M_ZERO); 297 /* -16 is so we can convert a trapframe into vm86trapframe inplace */ 298 ext->ext_tss.tss_esp0 = (vm_offset_t)td->td_pcb - 16; 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 td->td_pcb->pcb_ext = ext; 325 PCPU_SET(private_tss, 1); 326 *PCPU_GET(tss_gdt) = ext->ext_tssd; 327 ltr(GSEL(GPROC0_SEL, SEL_KPL)); 328 critical_exit(); 329 330 return 0; 331} 332 333int 334i386_set_ioperm(td, uap) 335 struct thread *td; 336 struct i386_ioperm_args *uap; 337{ 338 char *iomap; 339 u_int i; 340 int error; 341 342 if ((error = priv_check(td, PRIV_IO)) != 0) 343 return (error); 344 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 345 return (error); 346 /* 347 * XXX 348 * While this is restricted to root, we should probably figure out 349 * whether any other driver is using this i/o address, as so not to 350 * cause confusion. This probably requires a global 'usage registry'. 351 */ 352 353 if (td->td_pcb->pcb_ext == 0) 354 if ((error = i386_extend_pcb(td)) != 0) 355 return (error); 356 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; 357 358 if (uap->start > uap->start + uap->length || 359 uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY) 360 return (EINVAL); 361 362 for (i = uap->start; i < uap->start + uap->length; i++) { 363 if (uap->enable) 364 iomap[i >> 3] &= ~(1 << (i & 7)); 365 else 366 iomap[i >> 3] |= (1 << (i & 7)); 367 } 368 return (error); 369} 370 371int 372i386_get_ioperm(td, uap) 373 struct thread *td; 374 struct i386_ioperm_args *uap; 375{ 376 int i, state; 377 char *iomap; 378 379 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY) 380 return (EINVAL); 381 382 if (td->td_pcb->pcb_ext == 0) { 383 uap->length = 0; 384 goto done; 385 } 386 387 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; 388 389 i = uap->start; 390 state = (iomap[i >> 3] >> (i & 7)) & 1; 391 uap->enable = !state; 392 uap->length = 1; 393 394 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) { 395 if (state != ((iomap[i >> 3] >> (i & 7)) & 1)) 396 break; 397 uap->length++; 398 } 399 400done: 401 return (0); 402} 403 404/* 405 * Update the GDT entry pointing to the LDT to point to the LDT of the 406 * current process. Manage dt_lock holding/unholding autonomously. 407 */ 408static void 409set_user_ldt_locked(struct mdproc *mdp) 410{ 411 struct proc_ldt *pldt; 412 int gdt_idx; 413 414 mtx_assert(&dt_lock, MA_OWNED); 415 416 pldt = mdp->md_ldt; 417 gdt_idx = GUSERLDT_SEL; 418 gdt_idx += PCPU_GET(cpuid) * NGDT; /* always 0 on UP */ 419 gdt[gdt_idx].sd = pldt->ldt_sd; 420 lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); 421 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL)); 422} 423 424void 425set_user_ldt(struct mdproc *mdp) 426{ 427 428 mtx_lock_spin(&dt_lock); 429 set_user_ldt_locked(mdp); 430 mtx_unlock_spin(&dt_lock); 431} 432 433#ifdef SMP 434static void 435set_user_ldt_rv(void *arg) 436{ 437 struct proc *p; 438 439 p = curproc; 440 if (arg == p->p_vmspace) 441 set_user_ldt(&p->p_md); 442} 443#endif 444 445/* 446 * dt_lock must be held. Returns with dt_lock held. 447 */ 448struct proc_ldt * 449user_ldt_alloc(struct mdproc *mdp, int len) 450{ 451 struct proc_ldt *pldt, *new_ldt; 452 453 mtx_assert(&dt_lock, MA_OWNED); 454 mtx_unlock_spin(&dt_lock); 455 new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK); 456 457 new_ldt->ldt_len = len = NEW_MAX_LD(len); 458 new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena, 459 len * sizeof(union descriptor), M_WAITOK | M_ZERO); 460 new_ldt->ldt_refcnt = 1; 461 new_ldt->ldt_active = 0; 462 463 mtx_lock_spin(&dt_lock); 464 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base; 465 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1; 466 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd); 467 468 if ((pldt = mdp->md_ldt) != NULL) { 469 if (len > pldt->ldt_len) 470 len = pldt->ldt_len; 471 bcopy(pldt->ldt_base, new_ldt->ldt_base, 472 len * sizeof(union descriptor)); 473 } else 474 bcopy(ldt, new_ldt->ldt_base, sizeof(ldt)); 475 476 return (new_ldt); 477} 478 479/* 480 * Must be called with dt_lock held. Returns with dt_lock unheld. 481 */ 482void 483user_ldt_free(struct thread *td) 484{ 485 struct mdproc *mdp; 486 struct proc_ldt *pldt; 487 488 mtx_assert(&dt_lock, MA_OWNED); 489 mdp = &td->td_proc->p_md; 490 if ((pldt = mdp->md_ldt) == NULL) { 491 mtx_unlock_spin(&dt_lock); 492 return; 493 } 494 495 if (td == curthread) { 496 lldt(_default_ldt); 497 PCPU_SET(currentldt, _default_ldt); 498 } 499 500 mdp->md_ldt = NULL; 501 user_ldt_deref(pldt); 502} 503 504void 505user_ldt_deref(struct proc_ldt *pldt) 506{ 507 508 mtx_assert(&dt_lock, MA_OWNED); 509 if (--pldt->ldt_refcnt == 0) { 510 mtx_unlock_spin(&dt_lock); 511 kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base, 512 pldt->ldt_len * sizeof(union descriptor)); 513 free(pldt, M_SUBPROC); 514 } else 515 mtx_unlock_spin(&dt_lock); 516} 517 518/* 519 * Note for the authors of compat layers (linux, etc): copyout() in 520 * the function below is not a problem since it presents data in 521 * arch-specific format (i.e. i386-specific in this case), not in 522 * the OS-specific one. 523 */ 524int 525i386_get_ldt(struct thread *td, struct i386_ldt_args *uap) 526{ 527 struct proc_ldt *pldt; 528 char *data; 529 u_int nldt, num; 530 int error; 531 532#ifdef DEBUG 533 printf("i386_get_ldt: start=%u num=%u descs=%p\n", 534 uap->start, uap->num, (void *)uap->descs); 535#endif 536 537 num = min(uap->num, MAX_LD); 538 data = malloc(num * sizeof(union descriptor), M_TEMP, M_WAITOK); 539 mtx_lock_spin(&dt_lock); 540 pldt = td->td_proc->p_md.md_ldt; 541 nldt = pldt != NULL ? pldt->ldt_len : nitems(ldt); 542 if (uap->start >= nldt) { 543 num = 0; 544 } else { 545 num = min(num, nldt - uap->start); 546 bcopy(pldt != NULL ? 547 &((union descriptor *)(pldt->ldt_base))[uap->start] : 548 &ldt[uap->start], data, num * sizeof(union descriptor)); 549 } 550 mtx_unlock_spin(&dt_lock); 551 error = copyout(data, uap->descs, num * sizeof(union descriptor)); 552 if (error == 0) 553 td->td_retval[0] = num; 554 free(data, M_TEMP); 555 return (error); 556} 557 558int 559i386_set_ldt(struct thread *td, struct i386_ldt_args *uap, 560 union descriptor *descs) 561{ 562 struct mdproc *mdp; 563 struct proc_ldt *pldt; 564 union descriptor *dp; 565 u_int largest_ld, i; 566 int error; 567 568#ifdef DEBUG 569 printf("i386_set_ldt: start=%u num=%u descs=%p\n", 570 uap->start, uap->num, (void *)uap->descs); 571#endif 572 error = 0; 573 mdp = &td->td_proc->p_md; 574 575 if (descs == NULL) { 576 /* Free descriptors */ 577 if (uap->start == 0 && uap->num == 0) { 578 /* 579 * Treat this as a special case, so userland needn't 580 * know magic number NLDT. 581 */ 582 uap->start = NLDT; 583 uap->num = MAX_LD - NLDT; 584 } 585 mtx_lock_spin(&dt_lock); 586 if ((pldt = mdp->md_ldt) == NULL || 587 uap->start >= pldt->ldt_len) { 588 mtx_unlock_spin(&dt_lock); 589 return (0); 590 } 591 largest_ld = uap->start + uap->num; 592 if (largest_ld > pldt->ldt_len) 593 largest_ld = pldt->ldt_len; 594 for (i = uap->start; i < largest_ld; i++) 595 atomic_store_rel_64(&((uint64_t *)(pldt->ldt_base))[i], 596 0); 597 mtx_unlock_spin(&dt_lock); 598 return (0); 599 } 600 601 if (uap->start != LDT_AUTO_ALLOC || uap->num != 1) { 602 /* verify range of descriptors to modify */ 603 largest_ld = uap->start + uap->num; 604 if (uap->start >= MAX_LD || largest_ld > MAX_LD) 605 return (EINVAL); 606 } 607 608 /* Check descriptors for access violations */ 609 for (i = 0; i < uap->num; i++) { 610 dp = &descs[i]; 611 612 switch (dp->sd.sd_type) { 613 case SDT_SYSNULL: /* system null */ 614 dp->sd.sd_p = 0; 615 break; 616 case SDT_SYS286TSS: /* system 286 TSS available */ 617 case SDT_SYSLDT: /* system local descriptor table */ 618 case SDT_SYS286BSY: /* system 286 TSS busy */ 619 case SDT_SYSTASKGT: /* system task gate */ 620 case SDT_SYS286IGT: /* system 286 interrupt gate */ 621 case SDT_SYS286TGT: /* system 286 trap gate */ 622 case SDT_SYSNULL2: /* undefined by Intel */ 623 case SDT_SYS386TSS: /* system 386 TSS available */ 624 case SDT_SYSNULL3: /* undefined by Intel */ 625 case SDT_SYS386BSY: /* system 386 TSS busy */ 626 case SDT_SYSNULL4: /* undefined by Intel */ 627 case SDT_SYS386IGT: /* system 386 interrupt gate */ 628 case SDT_SYS386TGT: /* system 386 trap gate */ 629 case SDT_SYS286CGT: /* system 286 call gate */ 630 case SDT_SYS386CGT: /* system 386 call gate */ 631 return (EACCES); 632 633 /* memory segment types */ 634 case SDT_MEMEC: /* memory execute only conforming */ 635 case SDT_MEMEAC: /* memory execute only accessed conforming */ 636 case SDT_MEMERC: /* memory execute read conforming */ 637 case SDT_MEMERAC: /* memory execute read accessed conforming */ 638 /* Must be "present" if executable and conforming. */ 639 if (dp->sd.sd_p == 0) 640 return (EACCES); 641 break; 642 case SDT_MEMRO: /* memory read only */ 643 case SDT_MEMROA: /* memory read only accessed */ 644 case SDT_MEMRW: /* memory read write */ 645 case SDT_MEMRWA: /* memory read write accessed */ 646 case SDT_MEMROD: /* memory read only expand dwn limit */ 647 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */ 648 case SDT_MEMRWD: /* memory read write expand dwn limit */ 649 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */ 650 case SDT_MEME: /* memory execute only */ 651 case SDT_MEMEA: /* memory execute only accessed */ 652 case SDT_MEMER: /* memory execute read */ 653 case SDT_MEMERA: /* memory execute read accessed */ 654 break; 655 default: 656 return (EINVAL); 657 } 658 659 /* Only user (ring-3) descriptors may be present. */ 660 if (dp->sd.sd_p != 0 && dp->sd.sd_dpl != SEL_UPL) 661 return (EACCES); 662 } 663 664 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) { 665 /* Allocate a free slot */ 666 mtx_lock_spin(&dt_lock); 667 if ((pldt = mdp->md_ldt) == NULL) { 668 if ((error = i386_ldt_grow(td, NLDT + 1))) { 669 mtx_unlock_spin(&dt_lock); 670 return (error); 671 } 672 pldt = mdp->md_ldt; 673 } 674again: 675 /* 676 * start scanning a bit up to leave room for NVidia and 677 * Wine, which still user the "Blat" method of allocation. 678 */ 679 dp = &((union descriptor *)(pldt->ldt_base))[NLDT]; 680 for (i = NLDT; i < pldt->ldt_len; ++i) { 681 if (dp->sd.sd_type == SDT_SYSNULL) 682 break; 683 dp++; 684 } 685 if (i >= pldt->ldt_len) { 686 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) { 687 mtx_unlock_spin(&dt_lock); 688 return (error); 689 } 690 goto again; 691 } 692 uap->start = i; 693 error = i386_set_ldt_data(td, i, 1, descs); 694 mtx_unlock_spin(&dt_lock); 695 } else { 696 largest_ld = uap->start + uap->num; 697 mtx_lock_spin(&dt_lock); 698 if (!(error = i386_ldt_grow(td, largest_ld))) { 699 error = i386_set_ldt_data(td, uap->start, uap->num, 700 descs); 701 } 702 mtx_unlock_spin(&dt_lock); 703 } 704 if (error == 0) 705 td->td_retval[0] = uap->start; 706 return (error); 707} 708 709static int 710i386_set_ldt_data(struct thread *td, int start, int num, 711 union descriptor *descs) 712{ 713 struct mdproc *mdp; 714 struct proc_ldt *pldt; 715 uint64_t *dst, *src; 716 int i; 717 718 mtx_assert(&dt_lock, MA_OWNED); 719 720 mdp = &td->td_proc->p_md; 721 pldt = mdp->md_ldt; 722 dst = (uint64_t *)(pldt->ldt_base); 723 src = (uint64_t *)descs; 724 725 /* 726 * Atomic(9) is used only to get 64bit atomic store with 727 * cmpxchg8b when available. There is no op without release 728 * semantic. 729 */ 730 for (i = 0; i < num; i++) 731 atomic_store_rel_64(&dst[start + i], src[i]); 732 return (0); 733} 734 735static int 736i386_ldt_grow(struct thread *td, int len) 737{ 738 struct mdproc *mdp; 739 struct proc_ldt *new_ldt, *pldt; 740 caddr_t old_ldt_base; 741 int old_ldt_len; 742 743 mtx_assert(&dt_lock, MA_OWNED); 744 745 if (len > MAX_LD) 746 return (ENOMEM); 747 if (len < NLDT + 1) 748 len = NLDT + 1; 749 750 mdp = &td->td_proc->p_md; 751 old_ldt_base = NULL_LDT_BASE; 752 old_ldt_len = 0; 753 754 /* Allocate a user ldt. */ 755 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) { 756 new_ldt = user_ldt_alloc(mdp, len); 757 if (new_ldt == NULL) 758 return (ENOMEM); 759 pldt = mdp->md_ldt; 760 761 if (pldt != NULL) { 762 if (new_ldt->ldt_len <= pldt->ldt_len) { 763 /* 764 * We just lost the race for allocation, so 765 * free the new object and return. 766 */ 767 mtx_unlock_spin(&dt_lock); 768 kmem_free(kernel_arena, 769 (vm_offset_t)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 kmem_free(kernel_arena, (vm_offset_t)old_ldt_base, 803 old_ldt_len * sizeof(union descriptor)); 804 free(new_ldt, M_SUBPROC); 805 } 806 mtx_lock_spin(&dt_lock); 807 } 808 return (0); 809} 810