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 * 4. 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/capability.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#ifdef XEN 63#include <machine/xen/xenfunc.h> 64 65void i386_reset_ldt(struct proc_ldt *pldt); 66 67void 68i386_reset_ldt(struct proc_ldt *pldt) 69{ 70 xen_set_ldt((vm_offset_t)pldt->ldt_base, pldt->ldt_len); 71} 72#else 73#define i386_reset_ldt(x) 74#endif 75 76#include <vm/vm_kern.h> /* for kernel_map */ 77 78#define MAX_LD 8192 79#define LD_PER_PAGE 512 80#define NEW_MAX_LD(num) ((num + LD_PER_PAGE) & ~(LD_PER_PAGE-1)) 81#define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3) 82#define NULL_LDT_BASE ((caddr_t)NULL) 83 84#ifdef SMP 85static void set_user_ldt_rv(struct vmspace *vmsp); 86#endif 87static int i386_set_ldt_data(struct thread *, int start, int num, 88 union descriptor *descs); 89static int i386_ldt_grow(struct thread *td, int len); 90 91#ifndef _SYS_SYSPROTO_H_ 92struct sysarch_args { 93 int op; 94 char *parms; 95}; 96#endif 97 98int 99sysarch(td, uap) 100 struct thread *td; 101 register 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 } kargs; 109 uint32_t base; 110 struct segment_descriptor sd, *sdp; 111 112 AUDIT_ARG_CMD(uap->op); 113 114#ifdef CAPABILITY_MODE 115 /* 116 * When adding new operations, add a new case statement here to 117 * explicitly indicate whether or not the operation is safe to 118 * perform in capability mode. 119 */ 120 if (IN_CAPABILITY_MODE(td)) { 121 switch (uap->op) { 122 case I386_GET_LDT: 123 case I386_SET_LDT: 124 case I386_GET_IOPERM: 125 case I386_GET_FSBASE: 126 case I386_SET_FSBASE: 127 case I386_GET_GSBASE: 128 case I386_SET_GSBASE: 129 break; 130 131 case I386_SET_IOPERM: 132 default: 133 return (ECAPMODE); 134 } 135 } 136#endif 137 138 switch (uap->op) { 139 case I386_GET_IOPERM: 140 case I386_SET_IOPERM: 141 if ((error = copyin(uap->parms, &kargs.iargs, 142 sizeof(struct i386_ioperm_args))) != 0) 143 return (error); 144 break; 145 case I386_GET_LDT: 146 case I386_SET_LDT: 147 if ((error = copyin(uap->parms, &kargs.largs, 148 sizeof(struct i386_ldt_args))) != 0) 149 return (error); 150 if (kargs.largs.num > MAX_LD || kargs.largs.num <= 0) 151 return (EINVAL); 152 break; 153 default: 154 break; 155 } 156 157 switch(uap->op) { 158 case I386_GET_LDT: 159 error = i386_get_ldt(td, &kargs.largs); 160 break; 161 case I386_SET_LDT: 162 if (kargs.largs.descs != NULL) { 163 lp = (union descriptor *)kmem_alloc(kernel_map, 164 kargs.largs.num * sizeof(union descriptor)); 165 if (lp == NULL) { 166 error = ENOMEM; 167 break; 168 } 169 error = copyin(kargs.largs.descs, lp, 170 kargs.largs.num * sizeof(union descriptor)); 171 if (error == 0) 172 error = i386_set_ldt(td, &kargs.largs, lp); 173 kmem_free(kernel_map, (vm_offset_t)lp, 174 kargs.largs.num * sizeof(union descriptor)); 175 } else { 176 error = i386_set_ldt(td, &kargs.largs, NULL); 177 } 178 break; 179 case I386_GET_IOPERM: 180 error = i386_get_ioperm(td, &kargs.iargs); 181 if (error == 0) 182 error = copyout(&kargs.iargs, uap->parms, 183 sizeof(struct i386_ioperm_args)); 184 break; 185 case I386_SET_IOPERM: 186 error = i386_set_ioperm(td, &kargs.iargs); 187 break; 188 case I386_VM86: 189 error = vm86_sysarch(td, uap->parms); 190 break; 191 case I386_GET_FSBASE: 192 sdp = &td->td_pcb->pcb_fsd; 193 base = sdp->sd_hibase << 24 | sdp->sd_lobase; 194 error = copyout(&base, uap->parms, sizeof(base)); 195 break; 196 case I386_SET_FSBASE: 197 error = copyin(uap->parms, &base, sizeof(base)); 198 if (!error) { 199 /* 200 * Construct a descriptor and store it in the pcb for 201 * the next context switch. Also store it in the gdt 202 * so that the load of tf_fs into %fs will activate it 203 * at return to userland. 204 */ 205 sd.sd_lobase = base & 0xffffff; 206 sd.sd_hibase = (base >> 24) & 0xff; 207#ifdef XEN 208 /* need to do nosegneg like Linux */ 209 sd.sd_lolimit = (HYPERVISOR_VIRT_START >> 12) & 0xffff; 210#else 211 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */ 212#endif 213 sd.sd_hilimit = 0xf; 214 sd.sd_type = SDT_MEMRWA; 215 sd.sd_dpl = SEL_UPL; 216 sd.sd_p = 1; 217 sd.sd_xx = 0; 218 sd.sd_def32 = 1; 219 sd.sd_gran = 1; 220 critical_enter(); 221 td->td_pcb->pcb_fsd = sd; 222#ifdef XEN 223 HYPERVISOR_update_descriptor(vtomach(&PCPU_GET(fsgs_gdt)[0]), 224 *(uint64_t *)&sd); 225#else 226 PCPU_GET(fsgs_gdt)[0] = sd; 227#endif 228 critical_exit(); 229 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL); 230 } 231 break; 232 case I386_GET_GSBASE: 233 sdp = &td->td_pcb->pcb_gsd; 234 base = sdp->sd_hibase << 24 | sdp->sd_lobase; 235 error = copyout(&base, uap->parms, sizeof(base)); 236 break; 237 case I386_SET_GSBASE: 238 error = copyin(uap->parms, &base, sizeof(base)); 239 if (!error) { 240 /* 241 * Construct a descriptor and store it in the pcb for 242 * the next context switch. Also store it in the gdt 243 * because we have to do a load_gs() right now. 244 */ 245 sd.sd_lobase = base & 0xffffff; 246 sd.sd_hibase = (base >> 24) & 0xff; 247 248#ifdef XEN 249 /* need to do nosegneg like Linux */ 250 sd.sd_lolimit = (HYPERVISOR_VIRT_START >> 12) & 0xffff; 251#else 252 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */ 253#endif 254 sd.sd_hilimit = 0xf; 255 sd.sd_type = SDT_MEMRWA; 256 sd.sd_dpl = SEL_UPL; 257 sd.sd_p = 1; 258 sd.sd_xx = 0; 259 sd.sd_def32 = 1; 260 sd.sd_gran = 1; 261 critical_enter(); 262 td->td_pcb->pcb_gsd = sd; 263#ifdef XEN 264 HYPERVISOR_update_descriptor(vtomach(&PCPU_GET(fsgs_gdt)[1]), 265 *(uint64_t *)&sd); 266#else 267 PCPU_GET(fsgs_gdt)[1] = sd; 268#endif 269 critical_exit(); 270 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 271 } 272 break; 273 default: 274 error = EINVAL; 275 break; 276 } 277 return (error); 278} 279 280int 281i386_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 = (struct pcb_ext *)kmem_alloc(kernel_map, ctob(IOPAGES+1)); 298 if (ext == 0) 299 return (ENOMEM); 300 bzero(ext, sizeof(struct pcb_ext)); 301 /* -16 is so we can convert a trapframe into vm86trapframe inplace */ 302 ext->ext_tss.tss_esp0 = td->td_kstack + ctob(KSTACK_PAGES) - 303 sizeof(struct pcb) - 16; 304 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL); 305 /* 306 * The last byte of the i/o map must be followed by an 0xff byte. 307 * We arbitrarily allocate 16 bytes here, to keep the starting 308 * address on a doubleword boundary. 309 */ 310 offset = PAGE_SIZE - 16; 311 ext->ext_tss.tss_ioopt = 312 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16; 313 ext->ext_iomap = (caddr_t)ext + offset; 314 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32; 315 316 addr = (u_long *)ext->ext_vm86.vm86_intmap; 317 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++) 318 *addr++ = ~0; 319 320 ssd.ssd_base = (unsigned)&ext->ext_tss; 321 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext); 322 ssdtosd(&ssd, &ext->ext_tssd); 323 324 KASSERT(td == curthread, ("giving TSS to !curthread")); 325 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!")); 326 327 /* Switch to the new TSS. */ 328 critical_enter(); 329 td->td_pcb->pcb_ext = ext; 330 PCPU_SET(private_tss, 1); 331 *PCPU_GET(tss_gdt) = ext->ext_tssd; 332 ltr(GSEL(GPROC0_SEL, SEL_KPL)); 333 critical_exit(); 334 335 return 0; 336} 337 338int 339i386_set_ioperm(td, uap) 340 struct thread *td; 341 struct i386_ioperm_args *uap; 342{ 343 int i, error; 344 char *iomap; 345 346 if ((error = priv_check(td, PRIV_IO)) != 0) 347 return (error); 348 if ((error = securelevel_gt(td->td_ucred, 0)) != 0) 349 return (error); 350 /* 351 * XXX 352 * While this is restricted to root, we should probably figure out 353 * whether any other driver is using this i/o address, as so not to 354 * cause confusion. This probably requires a global 'usage registry'. 355 */ 356 357 if (td->td_pcb->pcb_ext == 0) 358 if ((error = i386_extend_pcb(td)) != 0) 359 return (error); 360 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; 361 362 if (uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY) 363 return (EINVAL); 364 365 for (i = uap->start; i < uap->start + uap->length; i++) { 366 if (uap->enable) 367 iomap[i >> 3] &= ~(1 << (i & 7)); 368 else 369 iomap[i >> 3] |= (1 << (i & 7)); 370 } 371 return (error); 372} 373 374int 375i386_get_ioperm(td, uap) 376 struct thread *td; 377 struct i386_ioperm_args *uap; 378{ 379 int i, state; 380 char *iomap; 381 382 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY) 383 return (EINVAL); 384 385 if (td->td_pcb->pcb_ext == 0) { 386 uap->length = 0; 387 goto done; 388 } 389 390 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; 391 392 i = uap->start; 393 state = (iomap[i >> 3] >> (i & 7)) & 1; 394 uap->enable = !state; 395 uap->length = 1; 396 397 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) { 398 if (state != ((iomap[i >> 3] >> (i & 7)) & 1)) 399 break; 400 uap->length++; 401 } 402 403done: 404 return (0); 405} 406 407/* 408 * Update the GDT entry pointing to the LDT to point to the LDT of the 409 * current process. Manage dt_lock holding/unholding autonomously. 410 */ 411void 412set_user_ldt(struct mdproc *mdp) 413{ 414 struct proc_ldt *pldt; 415 int dtlocked; 416 417 dtlocked = 0; 418 if (!mtx_owned(&dt_lock)) { 419 mtx_lock_spin(&dt_lock); 420 dtlocked = 1; 421 } 422 423 pldt = mdp->md_ldt; 424#ifdef XEN 425 i386_reset_ldt(pldt); 426 PCPU_SET(currentldt, (int)pldt); 427#else 428#ifdef SMP 429 gdt[PCPU_GET(cpuid) * NGDT + GUSERLDT_SEL].sd = pldt->ldt_sd; 430#else 431 gdt[GUSERLDT_SEL].sd = pldt->ldt_sd; 432#endif 433 lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); 434 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL)); 435#endif /* XEN */ 436 if (dtlocked) 437 mtx_unlock_spin(&dt_lock); 438} 439 440#ifdef SMP 441static void 442set_user_ldt_rv(struct vmspace *vmsp) 443{ 444 struct thread *td; 445 446 td = curthread; 447 if (vmsp != td->td_proc->p_vmspace) 448 return; 449 450 set_user_ldt(&td->td_proc->p_md); 451} 452#endif 453 454#ifdef XEN 455 456/* 457 * dt_lock must be held. Returns with dt_lock held. 458 */ 459struct proc_ldt * 460user_ldt_alloc(struct mdproc *mdp, int len) 461{ 462 struct proc_ldt *pldt, *new_ldt; 463 464 mtx_assert(&dt_lock, MA_OWNED); 465 mtx_unlock_spin(&dt_lock); 466 new_ldt = malloc(sizeof(struct proc_ldt), 467 M_SUBPROC, M_WAITOK); 468 469 new_ldt->ldt_len = len = NEW_MAX_LD(len); 470 new_ldt->ldt_base = (caddr_t)kmem_alloc(kernel_map, 471 round_page(len * sizeof(union descriptor))); 472 if (new_ldt->ldt_base == NULL) { 473 free(new_ldt, M_SUBPROC); 474 mtx_lock_spin(&dt_lock); 475 return (NULL); 476 } 477 new_ldt->ldt_refcnt = 1; 478 new_ldt->ldt_active = 0; 479 480 mtx_lock_spin(&dt_lock); 481 if ((pldt = mdp->md_ldt)) { 482 if (len > pldt->ldt_len) 483 len = pldt->ldt_len; 484 bcopy(pldt->ldt_base, new_ldt->ldt_base, 485 len * sizeof(union descriptor)); 486 } else { 487 bcopy(ldt, new_ldt->ldt_base, PAGE_SIZE); 488 } 489 mtx_unlock_spin(&dt_lock); /* XXX kill once pmap locking fixed. */ 490 pmap_map_readonly(kernel_pmap, (vm_offset_t)new_ldt->ldt_base, 491 new_ldt->ldt_len*sizeof(union descriptor)); 492 mtx_lock_spin(&dt_lock); /* XXX kill once pmap locking fixed. */ 493 return (new_ldt); 494} 495#else 496/* 497 * dt_lock must be held. Returns with dt_lock held. 498 */ 499struct proc_ldt * 500user_ldt_alloc(struct mdproc *mdp, int len) 501{ 502 struct proc_ldt *pldt, *new_ldt; 503 504 mtx_assert(&dt_lock, MA_OWNED); 505 mtx_unlock_spin(&dt_lock); 506 new_ldt = malloc(sizeof(struct proc_ldt), 507 M_SUBPROC, M_WAITOK); 508 509 new_ldt->ldt_len = len = NEW_MAX_LD(len); 510 new_ldt->ldt_base = (caddr_t)kmem_alloc(kernel_map, 511 len * sizeof(union descriptor)); 512 if (new_ldt->ldt_base == NULL) { 513 free(new_ldt, M_SUBPROC); 514 mtx_lock_spin(&dt_lock); 515 return (NULL); 516 } 517 new_ldt->ldt_refcnt = 1; 518 new_ldt->ldt_active = 0; 519 520 mtx_lock_spin(&dt_lock); 521 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base; 522 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1; 523 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd); 524 525 if ((pldt = mdp->md_ldt) != NULL) { 526 if (len > pldt->ldt_len) 527 len = pldt->ldt_len; 528 bcopy(pldt->ldt_base, new_ldt->ldt_base, 529 len * sizeof(union descriptor)); 530 } else 531 bcopy(ldt, new_ldt->ldt_base, sizeof(ldt)); 532 533 return (new_ldt); 534} 535#endif /* !XEN */ 536 537/* 538 * Must be called with dt_lock held. Returns with dt_lock unheld. 539 */ 540void 541user_ldt_free(struct thread *td) 542{ 543 struct mdproc *mdp = &td->td_proc->p_md; 544 struct proc_ldt *pldt; 545 546 mtx_assert(&dt_lock, MA_OWNED); 547 if ((pldt = mdp->md_ldt) == NULL) { 548 mtx_unlock_spin(&dt_lock); 549 return; 550 } 551 552 if (td == curthread) { 553#ifdef XEN 554 i386_reset_ldt(&default_proc_ldt); 555 PCPU_SET(currentldt, (int)&default_proc_ldt); 556#else 557 lldt(_default_ldt); 558 PCPU_SET(currentldt, _default_ldt); 559#endif 560 } 561 562 mdp->md_ldt = NULL; 563 user_ldt_deref(pldt); 564} 565 566void 567user_ldt_deref(struct proc_ldt *pldt) 568{ 569 570 mtx_assert(&dt_lock, MA_OWNED); 571 if (--pldt->ldt_refcnt == 0) { 572 mtx_unlock_spin(&dt_lock); 573 kmem_free(kernel_map, (vm_offset_t)pldt->ldt_base, 574 pldt->ldt_len * sizeof(union descriptor)); 575 free(pldt, M_SUBPROC); 576 } else 577 mtx_unlock_spin(&dt_lock); 578} 579 580/* 581 * Note for the authors of compat layers (linux, etc): copyout() in 582 * the function below is not a problem since it presents data in 583 * arch-specific format (i.e. i386-specific in this case), not in 584 * the OS-specific one. 585 */ 586int 587i386_get_ldt(td, uap) 588 struct thread *td; 589 struct i386_ldt_args *uap; 590{ 591 int error = 0; 592 struct proc_ldt *pldt; 593 int nldt, num; 594 union descriptor *lp; 595 596#ifdef DEBUG 597 printf("i386_get_ldt: start=%d num=%d descs=%p\n", 598 uap->start, uap->num, (void *)uap->descs); 599#endif 600 601 mtx_lock_spin(&dt_lock); 602 if ((pldt = td->td_proc->p_md.md_ldt) != NULL) { 603 nldt = pldt->ldt_len; 604 lp = &((union descriptor *)(pldt->ldt_base))[uap->start]; 605 mtx_unlock_spin(&dt_lock); 606 num = min(uap->num, nldt); 607 } else { 608 mtx_unlock_spin(&dt_lock); 609 nldt = sizeof(ldt)/sizeof(ldt[0]); 610 num = min(uap->num, nldt); 611 lp = &ldt[uap->start]; 612 } 613 614 if ((uap->start > (unsigned int)nldt) || 615 ((unsigned int)num > (unsigned int)nldt) || 616 ((unsigned int)(uap->start + num) > (unsigned int)nldt)) 617 return(EINVAL); 618 619 error = copyout(lp, uap->descs, num * sizeof(union descriptor)); 620 if (!error) 621 td->td_retval[0] = num; 622 623 return(error); 624} 625 626int 627i386_set_ldt(td, uap, descs) 628 struct thread *td; 629 struct i386_ldt_args *uap; 630 union descriptor *descs; 631{ 632 int error = 0, i; 633 int largest_ld; 634 struct mdproc *mdp = &td->td_proc->p_md; 635 struct proc_ldt *pldt; 636 union descriptor *dp; 637 638#ifdef DEBUG 639 printf("i386_set_ldt: start=%d num=%d descs=%p\n", 640 uap->start, uap->num, (void *)uap->descs); 641#endif 642 643 if (descs == NULL) { 644 /* Free descriptors */ 645 if (uap->start == 0 && uap->num == 0) { 646 /* 647 * Treat this as a special case, so userland needn't 648 * know magic number NLDT. 649 */ 650 uap->start = NLDT; 651 uap->num = MAX_LD - NLDT; 652 } 653 if (uap->num == 0) 654 return (EINVAL); 655 mtx_lock_spin(&dt_lock); 656 if ((pldt = mdp->md_ldt) == NULL || 657 uap->start >= pldt->ldt_len) { 658 mtx_unlock_spin(&dt_lock); 659 return (0); 660 } 661 largest_ld = uap->start + uap->num; 662 if (largest_ld > pldt->ldt_len) 663 largest_ld = pldt->ldt_len; 664 i = largest_ld - uap->start; 665 bzero(&((union descriptor *)(pldt->ldt_base))[uap->start], 666 sizeof(union descriptor) * i); 667 mtx_unlock_spin(&dt_lock); 668 return (0); 669 } 670 671 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) { 672 /* verify range of descriptors to modify */ 673 largest_ld = uap->start + uap->num; 674 if (uap->start >= MAX_LD || largest_ld > MAX_LD) { 675 return (EINVAL); 676 } 677 } 678 679 /* Check descriptors for access violations */ 680 for (i = 0; i < uap->num; i++) { 681 dp = &descs[i]; 682 683 switch (dp->sd.sd_type) { 684 case SDT_SYSNULL: /* system null */ 685 dp->sd.sd_p = 0; 686 break; 687 case SDT_SYS286TSS: /* system 286 TSS available */ 688 case SDT_SYSLDT: /* system local descriptor table */ 689 case SDT_SYS286BSY: /* system 286 TSS busy */ 690 case SDT_SYSTASKGT: /* system task gate */ 691 case SDT_SYS286IGT: /* system 286 interrupt gate */ 692 case SDT_SYS286TGT: /* system 286 trap gate */ 693 case SDT_SYSNULL2: /* undefined by Intel */ 694 case SDT_SYS386TSS: /* system 386 TSS available */ 695 case SDT_SYSNULL3: /* undefined by Intel */ 696 case SDT_SYS386BSY: /* system 386 TSS busy */ 697 case SDT_SYSNULL4: /* undefined by Intel */ 698 case SDT_SYS386IGT: /* system 386 interrupt gate */ 699 case SDT_SYS386TGT: /* system 386 trap gate */ 700 case SDT_SYS286CGT: /* system 286 call gate */ 701 case SDT_SYS386CGT: /* system 386 call gate */ 702 /* I can't think of any reason to allow a user proc 703 * to create a segment of these types. They are 704 * for OS use only. 705 */ 706 return (EACCES); 707 /*NOTREACHED*/ 708 709 /* memory segment types */ 710 case SDT_MEMEC: /* memory execute only conforming */ 711 case SDT_MEMEAC: /* memory execute only accessed conforming */ 712 case SDT_MEMERC: /* memory execute read conforming */ 713 case SDT_MEMERAC: /* memory execute read accessed conforming */ 714 /* Must be "present" if executable and conforming. */ 715 if (dp->sd.sd_p == 0) 716 return (EACCES); 717 break; 718 case SDT_MEMRO: /* memory read only */ 719 case SDT_MEMROA: /* memory read only accessed */ 720 case SDT_MEMRW: /* memory read write */ 721 case SDT_MEMRWA: /* memory read write accessed */ 722 case SDT_MEMROD: /* memory read only expand dwn limit */ 723 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */ 724 case SDT_MEMRWD: /* memory read write expand dwn limit */ 725 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */ 726 case SDT_MEME: /* memory execute only */ 727 case SDT_MEMEA: /* memory execute only accessed */ 728 case SDT_MEMER: /* memory execute read */ 729 case SDT_MEMERA: /* memory execute read accessed */ 730 break; 731 default: 732 return(EINVAL); 733 /*NOTREACHED*/ 734 } 735 736 /* Only user (ring-3) descriptors may be present. */ 737 if ((dp->sd.sd_p != 0) && (dp->sd.sd_dpl != SEL_UPL)) 738 return (EACCES); 739 } 740 741 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) { 742 /* Allocate a free slot */ 743 mtx_lock_spin(&dt_lock); 744 if ((pldt = mdp->md_ldt) == NULL) { 745 if ((error = i386_ldt_grow(td, NLDT + 1))) { 746 mtx_unlock_spin(&dt_lock); 747 return (error); 748 } 749 pldt = mdp->md_ldt; 750 } 751again: 752 /* 753 * start scanning a bit up to leave room for NVidia and 754 * Wine, which still user the "Blat" method of allocation. 755 */ 756 dp = &((union descriptor *)(pldt->ldt_base))[NLDT]; 757 for (i = NLDT; i < pldt->ldt_len; ++i) { 758 if (dp->sd.sd_type == SDT_SYSNULL) 759 break; 760 dp++; 761 } 762 if (i >= pldt->ldt_len) { 763 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) { 764 mtx_unlock_spin(&dt_lock); 765 return (error); 766 } 767 goto again; 768 } 769 uap->start = i; 770 error = i386_set_ldt_data(td, i, 1, descs); 771 mtx_unlock_spin(&dt_lock); 772 } else { 773 largest_ld = uap->start + uap->num; 774 mtx_lock_spin(&dt_lock); 775 if (!(error = i386_ldt_grow(td, largest_ld))) { 776 error = i386_set_ldt_data(td, uap->start, uap->num, 777 descs); 778 } 779 mtx_unlock_spin(&dt_lock); 780 } 781 if (error == 0) 782 td->td_retval[0] = uap->start; 783 return (error); 784} 785#ifdef XEN 786static int 787i386_set_ldt_data(struct thread *td, int start, int num, 788 union descriptor *descs) 789{ 790 struct mdproc *mdp = &td->td_proc->p_md; 791 struct proc_ldt *pldt = mdp->md_ldt; 792 793 mtx_assert(&dt_lock, MA_OWNED); 794 795 while (num) { 796 xen_update_descriptor( 797 &((union descriptor *)(pldt->ldt_base))[start], 798 descs); 799 num--; 800 start++; 801 descs++; 802 } 803 return (0); 804} 805#else 806static int 807i386_set_ldt_data(struct thread *td, int start, int num, 808 union descriptor *descs) 809{ 810 struct mdproc *mdp = &td->td_proc->p_md; 811 struct proc_ldt *pldt = mdp->md_ldt; 812 813 mtx_assert(&dt_lock, MA_OWNED); 814 815 /* Fill in range */ 816 bcopy(descs, 817 &((union descriptor *)(pldt->ldt_base))[start], 818 num * sizeof(union descriptor)); 819 return (0); 820} 821#endif /* !XEN */ 822 823static int 824i386_ldt_grow(struct thread *td, int len) 825{ 826 struct mdproc *mdp = &td->td_proc->p_md; 827 struct proc_ldt *new_ldt, *pldt; 828 caddr_t old_ldt_base = NULL_LDT_BASE; 829 int old_ldt_len = 0; 830 831 mtx_assert(&dt_lock, MA_OWNED); 832 833 if (len > MAX_LD) 834 return (ENOMEM); 835 if (len < NLDT + 1) 836 len = NLDT + 1; 837 838 /* Allocate a user ldt. */ 839 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) { 840 new_ldt = user_ldt_alloc(mdp, len); 841 if (new_ldt == NULL) 842 return (ENOMEM); 843 pldt = mdp->md_ldt; 844 845 if (pldt != NULL) { 846 if (new_ldt->ldt_len <= pldt->ldt_len) { 847 /* 848 * We just lost the race for allocation, so 849 * free the new object and return. 850 */ 851 mtx_unlock_spin(&dt_lock); 852 kmem_free(kernel_map, 853 (vm_offset_t)new_ldt->ldt_base, 854 new_ldt->ldt_len * sizeof(union descriptor)); 855 free(new_ldt, M_SUBPROC); 856 mtx_lock_spin(&dt_lock); 857 return (0); 858 } 859 860 /* 861 * We have to substitute the current LDT entry for 862 * curproc with the new one since its size grew. 863 */ 864 old_ldt_base = pldt->ldt_base; 865 old_ldt_len = pldt->ldt_len; 866 pldt->ldt_sd = new_ldt->ldt_sd; 867 pldt->ldt_base = new_ldt->ldt_base; 868 pldt->ldt_len = new_ldt->ldt_len; 869 } else 870 mdp->md_ldt = pldt = new_ldt; 871#ifdef SMP 872 /* 873 * Signal other cpus to reload ldt. We need to unlock dt_lock 874 * here because other CPU will contest on it since their 875 * curthreads won't hold the lock and will block when trying 876 * to acquire it. 877 */ 878 mtx_unlock_spin(&dt_lock); 879 smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv, 880 NULL, td->td_proc->p_vmspace); 881#else 882 set_user_ldt(&td->td_proc->p_md); 883 mtx_unlock_spin(&dt_lock); 884#endif 885 if (old_ldt_base != NULL_LDT_BASE) { 886 kmem_free(kernel_map, (vm_offset_t)old_ldt_base, 887 old_ldt_len * sizeof(union descriptor)); 888 free(new_ldt, M_SUBPROC); 889 } 890 mtx_lock_spin(&dt_lock); 891 } 892 return (0); 893} 894