1 /*- 2 * Copyright (c) 2006, 2011 Robert N. M. Watson 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 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 /* 28 * Support for shared swap-backed anonymous memory objects via 29 * shm_open(2) and shm_unlink(2). While most of the implementation is 30 * here, vm_mmap.c contains mapping logic changes. 31 * 32 * TODO: 33 * 34 * (1) Need to export data to a userland tool via a sysctl. Should ipcs(1) 35 * and ipcrm(1) be expanded or should new tools to manage both POSIX 36 * kernel semaphores and POSIX shared memory be written? 37 * 38 * (2) Add support for this file type to fstat(1). 39 * 40 * (3) Resource limits? Does this need its own resource limits or are the 41 * existing limits in mmap(2) sufficient? 42 */ 43 44 #include <sys/cdefs.h> 45 __FBSDID("$FreeBSD$"); 46 47 #include "opt_capsicum.h" 48 49 #include <sys/param.h> 50 #include <sys/capability.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/filedesc.h> 54 #include <sys/fnv_hash.h> 55 #include <sys/kernel.h> 56 #include <sys/lock.h> 57 #include <sys/malloc.h> 58 #include <sys/mman.h> 59 #include <sys/mutex.h> 60 #include <sys/priv.h> 61 #include <sys/proc.h> 62 #include <sys/refcount.h> 63 #include <sys/resourcevar.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/sysproto.h> 67 #include <sys/systm.h> 68 #include <sys/sx.h> 69 #include <sys/time.h> 70 #include <sys/vnode.h> 71 72 #include <security/mac/mac_framework.h> 73 74 #include <vm/vm.h> 75 #include <vm/vm_param.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_extern.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_kern.h> 80 #include <vm/vm_object.h> 81 #include <vm/vm_page.h> 82 #include <vm/vm_pageout.h> 83 #include <vm/vm_pager.h> 84 #include <vm/swap_pager.h> 85 86 struct shm_mapping { 87 char *sm_path; 88 Fnv32_t sm_fnv; 89 struct shmfd *sm_shmfd; 90 LIST_ENTRY(shm_mapping) sm_link; 91 }; 92 93 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor"); 94 static LIST_HEAD(, shm_mapping) *shm_dictionary; 95 static struct sx shm_dict_lock; 96 static struct mtx shm_timestamp_lock; 97 static u_long shm_hash; 98 99 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash]) 100 101 static int shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags); 102 static struct shmfd *shm_alloc(struct ucred *ucred, mode_t mode); 103 static void shm_dict_init(void *arg); 104 static void shm_drop(struct shmfd *shmfd); 105 static struct shmfd *shm_hold(struct shmfd *shmfd); 106 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd); 107 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv); 108 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred); 109 static int shm_dotruncate(struct shmfd *shmfd, off_t length); 110 111 static fo_rdwr_t shm_read; 112 static fo_rdwr_t shm_write; 113 static fo_truncate_t shm_truncate; 114 static fo_ioctl_t shm_ioctl; 115 static fo_poll_t shm_poll; 116 static fo_kqfilter_t shm_kqfilter; 117 static fo_stat_t shm_stat; 118 static fo_close_t shm_close; 119 static fo_chmod_t shm_chmod; 120 static fo_chown_t shm_chown; 121 122 /* File descriptor operations. */ 123 static struct fileops shm_ops = { 124 .fo_read = shm_read, 125 .fo_write = shm_write, 126 .fo_truncate = shm_truncate, 127 .fo_ioctl = shm_ioctl, 128 .fo_poll = shm_poll, 129 .fo_kqfilter = shm_kqfilter, 130 .fo_stat = shm_stat, 131 .fo_close = shm_close, 132 .fo_chmod = shm_chmod, 133 .fo_chown = shm_chown, 134 .fo_flags = DFLAG_PASSABLE 135 }; 136 137 FEATURE(posix_shm, "POSIX shared memory"); 138 139 static int 140 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 141 int flags, struct thread *td) 142 { 143 144 return (EOPNOTSUPP); 145 } 146 147 static int 148 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 149 int flags, struct thread *td) 150 { 151 152 return (EOPNOTSUPP); 153 } 154 155 static int 156 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred, 157 struct thread *td) 158 { 159 struct shmfd *shmfd; 160 #ifdef MAC 161 int error; 162 #endif 163 164 shmfd = fp->f_data; 165 #ifdef MAC 166 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd); 167 if (error) 168 return (error); 169 #endif 170 return (shm_dotruncate(shmfd, length)); 171 } 172 173 static int 174 shm_ioctl(struct file *fp, u_long com, void *data, 175 struct ucred *active_cred, struct thread *td) 176 { 177 178 return (EOPNOTSUPP); 179 } 180 181 static int 182 shm_poll(struct file *fp, int events, struct ucred *active_cred, 183 struct thread *td) 184 { 185 186 return (EOPNOTSUPP); 187 } 188 189 static int 190 shm_kqfilter(struct file *fp, struct knote *kn) 191 { 192 193 return (EOPNOTSUPP); 194 } 195 196 static int 197 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 198 struct thread *td) 199 { 200 struct shmfd *shmfd; 201 #ifdef MAC 202 int error; 203 #endif 204 205 shmfd = fp->f_data; 206 207 #ifdef MAC 208 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd); 209 if (error) 210 return (error); 211 #endif 212 213 /* 214 * Attempt to return sanish values for fstat() on a memory file 215 * descriptor. 216 */ 217 bzero(sb, sizeof(*sb)); 218 sb->st_blksize = PAGE_SIZE; 219 sb->st_size = shmfd->shm_size; 220 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize; 221 mtx_lock(&shm_timestamp_lock); 222 sb->st_atim = shmfd->shm_atime; 223 sb->st_ctim = shmfd->shm_ctime; 224 sb->st_mtim = shmfd->shm_mtime; 225 sb->st_birthtim = shmfd->shm_birthtime; 226 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 227 sb->st_uid = shmfd->shm_uid; 228 sb->st_gid = shmfd->shm_gid; 229 mtx_unlock(&shm_timestamp_lock); 230 231 return (0); 232 } 233 234 static int 235 shm_close(struct file *fp, struct thread *td) 236 { 237 struct shmfd *shmfd; 238 239 shmfd = fp->f_data; 240 fp->f_data = NULL; 241 shm_drop(shmfd); 242 243 return (0); 244 } 245 246 static int 247 shm_dotruncate(struct shmfd *shmfd, off_t length) 248 { 249 vm_object_t object; 250 vm_page_t m, ma[1]; 251 vm_pindex_t idx, nobjsize; 252 vm_ooffset_t delta; 253 int base, rv; 254 255 object = shmfd->shm_object; 256 VM_OBJECT_LOCK(object); 257 if (length == shmfd->shm_size) { 258 VM_OBJECT_UNLOCK(object); 259 return (0); 260 } 261 nobjsize = OFF_TO_IDX(length + PAGE_MASK); 262 263 /* Are we shrinking? If so, trim the end. */ 264 if (length < shmfd->shm_size) { 265 /* 266 * Disallow any requests to shrink the size if this 267 * object is mapped into the kernel. 268 */ 269 if (shmfd->shm_kmappings > 0) { 270 VM_OBJECT_UNLOCK(object); 271 return (EBUSY); 272 } 273 274 /* 275 * Zero the truncated part of the last page. 276 */ 277 base = length & PAGE_MASK; 278 if (base != 0) { 279 idx = OFF_TO_IDX(length); 280 retry: 281 m = vm_page_lookup(object, idx); 282 if (m != NULL) { 283 if ((m->oflags & VPO_BUSY) != 0 || 284 m->busy != 0) { 285 vm_page_sleep(m, "shmtrc"); 286 goto retry; 287 } 288 } else if (vm_pager_has_page(object, idx, NULL, NULL)) { 289 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL); 290 if (m == NULL) { 291 VM_OBJECT_UNLOCK(object); 292 VM_WAIT; 293 VM_OBJECT_LOCK(object); 294 goto retry; 295 } else if (m->valid != VM_PAGE_BITS_ALL) { 296 ma[0] = m; 297 rv = vm_pager_get_pages(object, ma, 1, 298 0); 299 m = vm_page_lookup(object, idx); 300 } else 301 /* A cached page was reactivated. */ 302 rv = VM_PAGER_OK; 303 vm_page_lock(m); 304 if (rv == VM_PAGER_OK) { 305 vm_page_deactivate(m); 306 vm_page_unlock(m); 307 vm_page_wakeup(m); 308 } else { 309 vm_page_free(m); 310 vm_page_unlock(m); 311 VM_OBJECT_UNLOCK(object); 312 return (EIO); 313 } 314 } 315 if (m != NULL) { 316 pmap_zero_page_area(m, base, PAGE_SIZE - base); 317 KASSERT(m->valid == VM_PAGE_BITS_ALL, 318 ("shm_dotruncate: page %p is invalid", m)); 319 vm_page_dirty(m); 320 vm_pager_page_unswapped(m); 321 } 322 } 323 delta = ptoa(object->size - nobjsize); 324 325 /* Toss in memory pages. */ 326 if (nobjsize < object->size) 327 vm_object_page_remove(object, nobjsize, object->size, 328 0); 329 330 /* Toss pages from swap. */ 331 if (object->type == OBJT_SWAP) 332 swap_pager_freespace(object, nobjsize, delta); 333 334 /* Free the swap accounted for shm */ 335 swap_release_by_cred(delta, object->cred); 336 object->charge -= delta; 337 } else { 338 /* Attempt to reserve the swap */ 339 delta = ptoa(nobjsize - object->size); 340 if (!swap_reserve_by_cred(delta, object->cred)) { 341 VM_OBJECT_UNLOCK(object); 342 return (ENOMEM); 343 } 344 object->charge += delta; 345 } 346 shmfd->shm_size = length; 347 mtx_lock(&shm_timestamp_lock); 348 vfs_timestamp(&shmfd->shm_ctime); 349 shmfd->shm_mtime = shmfd->shm_ctime; 350 mtx_unlock(&shm_timestamp_lock); 351 object->size = nobjsize; 352 VM_OBJECT_UNLOCK(object); 353 return (0); 354 } 355 356 /* 357 * shmfd object management including creation and reference counting 358 * routines. 359 */ 360 static struct shmfd * 361 shm_alloc(struct ucred *ucred, mode_t mode) 362 { 363 struct shmfd *shmfd; 364 365 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO); 366 shmfd->shm_size = 0; 367 shmfd->shm_uid = ucred->cr_uid; 368 shmfd->shm_gid = ucred->cr_gid; 369 shmfd->shm_mode = mode; 370 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL, 371 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred); 372 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); 373 VM_OBJECT_LOCK(shmfd->shm_object); 374 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING); 375 vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT); 376 VM_OBJECT_UNLOCK(shmfd->shm_object); 377 vfs_timestamp(&shmfd->shm_birthtime); 378 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime = 379 shmfd->shm_birthtime; 380 refcount_init(&shmfd->shm_refs, 1); 381 #ifdef MAC 382 mac_posixshm_init(shmfd); 383 mac_posixshm_create(ucred, shmfd); 384 #endif 385 386 return (shmfd); 387 } 388 389 static struct shmfd * 390 shm_hold(struct shmfd *shmfd) 391 { 392 393 refcount_acquire(&shmfd->shm_refs); 394 return (shmfd); 395 } 396 397 static void 398 shm_drop(struct shmfd *shmfd) 399 { 400 401 if (refcount_release(&shmfd->shm_refs)) { 402 #ifdef MAC 403 mac_posixshm_destroy(shmfd); 404 #endif 405 vm_object_deallocate(shmfd->shm_object); 406 free(shmfd, M_SHMFD); 407 } 408 } 409 410 /* 411 * Determine if the credentials have sufficient permissions for a 412 * specified combination of FREAD and FWRITE. 413 */ 414 static int 415 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags) 416 { 417 accmode_t accmode; 418 int error; 419 420 accmode = 0; 421 if (flags & FREAD) 422 accmode |= VREAD; 423 if (flags & FWRITE) 424 accmode |= VWRITE; 425 mtx_lock(&shm_timestamp_lock); 426 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid, 427 accmode, ucred, NULL); 428 mtx_unlock(&shm_timestamp_lock); 429 return (error); 430 } 431 432 /* 433 * Dictionary management. We maintain an in-kernel dictionary to map 434 * paths to shmfd objects. We use the FNV hash on the path to store 435 * the mappings in a hash table. 436 */ 437 static void 438 shm_dict_init(void *arg) 439 { 440 441 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF); 442 sx_init(&shm_dict_lock, "shm dictionary"); 443 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash); 444 } 445 SYSINIT(shm_dict_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_dict_init, NULL); 446 447 static struct shmfd * 448 shm_lookup(char *path, Fnv32_t fnv) 449 { 450 struct shm_mapping *map; 451 452 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 453 if (map->sm_fnv != fnv) 454 continue; 455 if (strcmp(map->sm_path, path) == 0) 456 return (map->sm_shmfd); 457 } 458 459 return (NULL); 460 } 461 462 static void 463 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd) 464 { 465 struct shm_mapping *map; 466 467 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK); 468 map->sm_path = path; 469 map->sm_fnv = fnv; 470 map->sm_shmfd = shm_hold(shmfd); 471 shmfd->shm_path = path; 472 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link); 473 } 474 475 static int 476 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred) 477 { 478 struct shm_mapping *map; 479 int error; 480 481 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 482 if (map->sm_fnv != fnv) 483 continue; 484 if (strcmp(map->sm_path, path) == 0) { 485 #ifdef MAC 486 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd); 487 if (error) 488 return (error); 489 #endif 490 error = shm_access(map->sm_shmfd, ucred, 491 FREAD | FWRITE); 492 if (error) 493 return (error); 494 map->sm_shmfd->shm_path = NULL; 495 LIST_REMOVE(map, sm_link); 496 shm_drop(map->sm_shmfd); 497 free(map->sm_path, M_SHMFD); 498 free(map, M_SHMFD); 499 return (0); 500 } 501 } 502 503 return (ENOENT); 504 } 505 506 /* System calls. */ 507 int 508 sys_shm_open(struct thread *td, struct shm_open_args *uap) 509 { 510 struct filedesc *fdp; 511 struct shmfd *shmfd; 512 struct file *fp; 513 char *path; 514 Fnv32_t fnv; 515 mode_t cmode; 516 int fd, error; 517 518 #ifdef CAPABILITY_MODE 519 /* 520 * shm_open(2) is only allowed for anonymous objects. 521 */ 522 if (IN_CAPABILITY_MODE(td) && (uap->path != SHM_ANON)) 523 return (ECAPMODE); 524 #endif 525 526 if ((uap->flags & O_ACCMODE) != O_RDONLY && 527 (uap->flags & O_ACCMODE) != O_RDWR) 528 return (EINVAL); 529 530 if ((uap->flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC)) != 0) 531 return (EINVAL); 532 533 fdp = td->td_proc->p_fd; 534 cmode = (uap->mode & ~fdp->fd_cmask) & ACCESSPERMS; 535 536 error = falloc(td, &fp, &fd, 0); 537 if (error) 538 return (error); 539 540 /* A SHM_ANON path pointer creates an anonymous object. */ 541 if (uap->path == SHM_ANON) { 542 /* A read-only anonymous object is pointless. */ 543 if ((uap->flags & O_ACCMODE) == O_RDONLY) { 544 fdclose(fdp, fp, fd, td); 545 fdrop(fp, td); 546 return (EINVAL); 547 } 548 shmfd = shm_alloc(td->td_ucred, cmode); 549 } else { 550 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK); 551 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 552 553 /* Require paths to start with a '/' character. */ 554 if (error == 0 && path[0] != '/') 555 error = EINVAL; 556 if (error) { 557 fdclose(fdp, fp, fd, td); 558 fdrop(fp, td); 559 free(path, M_SHMFD); 560 return (error); 561 } 562 563 fnv = fnv_32_str(path, FNV1_32_INIT); 564 sx_xlock(&shm_dict_lock); 565 shmfd = shm_lookup(path, fnv); 566 if (shmfd == NULL) { 567 /* Object does not yet exist, create it if requested. */ 568 if (uap->flags & O_CREAT) { 569 #ifdef MAC 570 error = mac_posixshm_check_create(td->td_ucred, 571 path); 572 if (error == 0) { 573 #endif 574 shmfd = shm_alloc(td->td_ucred, cmode); 575 shm_insert(path, fnv, shmfd); 576 #ifdef MAC 577 } 578 #endif 579 } else { 580 free(path, M_SHMFD); 581 error = ENOENT; 582 } 583 } else { 584 /* 585 * Object already exists, obtain a new 586 * reference if requested and permitted. 587 */ 588 free(path, M_SHMFD); 589 if ((uap->flags & (O_CREAT | O_EXCL)) == 590 (O_CREAT | O_EXCL)) 591 error = EEXIST; 592 else { 593 #ifdef MAC 594 error = mac_posixshm_check_open(td->td_ucred, 595 shmfd, FFLAGS(uap->flags & O_ACCMODE)); 596 if (error == 0) 597 #endif 598 error = shm_access(shmfd, td->td_ucred, 599 FFLAGS(uap->flags & O_ACCMODE)); 600 } 601 602 /* 603 * Truncate the file back to zero length if 604 * O_TRUNC was specified and the object was 605 * opened with read/write. 606 */ 607 if (error == 0 && 608 (uap->flags & (O_ACCMODE | O_TRUNC)) == 609 (O_RDWR | O_TRUNC)) { 610 #ifdef MAC 611 error = mac_posixshm_check_truncate( 612 td->td_ucred, fp->f_cred, shmfd); 613 if (error == 0) 614 #endif 615 shm_dotruncate(shmfd, 0); 616 } 617 if (error == 0) 618 shm_hold(shmfd); 619 } 620 sx_xunlock(&shm_dict_lock); 621 622 if (error) { 623 fdclose(fdp, fp, fd, td); 624 fdrop(fp, td); 625 return (error); 626 } 627 } 628 629 finit(fp, FFLAGS(uap->flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops); 630 631 FILEDESC_XLOCK(fdp); 632 if (fdp->fd_ofiles[fd] == fp) 633 fdp->fd_ofileflags[fd] |= UF_EXCLOSE; 634 FILEDESC_XUNLOCK(fdp); 635 td->td_retval[0] = fd; 636 fdrop(fp, td); 637 638 return (0); 639 } 640 641 int 642 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap) 643 { 644 char *path; 645 Fnv32_t fnv; 646 int error; 647 648 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 649 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 650 if (error) { 651 free(path, M_TEMP); 652 return (error); 653 } 654 655 fnv = fnv_32_str(path, FNV1_32_INIT); 656 sx_xlock(&shm_dict_lock); 657 error = shm_remove(path, fnv, td->td_ucred); 658 sx_xunlock(&shm_dict_lock); 659 free(path, M_TEMP); 660 661 return (error); 662 } 663 664 /* 665 * mmap() helper to validate mmap() requests against shm object state 666 * and give mmap() the vm_object to use for the mapping. 667 */ 668 int 669 shm_mmap(struct shmfd *shmfd, vm_size_t objsize, vm_ooffset_t foff, 670 vm_object_t *obj) 671 { 672 673 /* 674 * XXXRW: This validation is probably insufficient, and subject to 675 * sign errors. It should be fixed. 676 */ 677 if (foff >= shmfd->shm_size || 678 foff + objsize > round_page(shmfd->shm_size)) 679 return (EINVAL); 680 681 mtx_lock(&shm_timestamp_lock); 682 vfs_timestamp(&shmfd->shm_atime); 683 mtx_unlock(&shm_timestamp_lock); 684 vm_object_reference(shmfd->shm_object); 685 *obj = shmfd->shm_object; 686 return (0); 687 } 688 689 static int 690 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 691 struct thread *td) 692 { 693 struct shmfd *shmfd; 694 int error; 695 696 error = 0; 697 shmfd = fp->f_data; 698 mtx_lock(&shm_timestamp_lock); 699 /* 700 * SUSv4 says that x bits of permission need not be affected. 701 * Be consistent with our shm_open there. 702 */ 703 #ifdef MAC 704 error = mac_posixshm_check_setmode(active_cred, shmfd, mode); 705 if (error != 0) 706 goto out; 707 #endif 708 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, 709 shmfd->shm_gid, VADMIN, active_cred, NULL); 710 if (error != 0) 711 goto out; 712 shmfd->shm_mode = mode & ACCESSPERMS; 713 out: 714 mtx_unlock(&shm_timestamp_lock); 715 return (error); 716 } 717 718 static int 719 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 720 struct thread *td) 721 { 722 struct shmfd *shmfd; 723 int error; 724 725 error = 0; 726 shmfd = fp->f_data; 727 mtx_lock(&shm_timestamp_lock); 728 #ifdef MAC 729 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid); 730 if (error != 0) 731 goto out; 732 #endif 733 if (uid == (uid_t)-1) 734 uid = shmfd->shm_uid; 735 if (gid == (gid_t)-1) 736 gid = shmfd->shm_gid; 737 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) || 738 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) && 739 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0))) 740 goto out; 741 shmfd->shm_uid = uid; 742 shmfd->shm_gid = gid; 743 out: 744 mtx_unlock(&shm_timestamp_lock); 745 return (error); 746 } 747 748 /* 749 * Helper routines to allow the backing object of a shared memory file 750 * descriptor to be mapped in the kernel. 751 */ 752 int 753 shm_map(struct file *fp, size_t size, off_t offset, void **memp) 754 { 755 struct shmfd *shmfd; 756 vm_offset_t kva, ofs; 757 vm_object_t obj; 758 int rv; 759 760 if (fp->f_type != DTYPE_SHM) 761 return (EINVAL); 762 shmfd = fp->f_data; 763 obj = shmfd->shm_object; 764 VM_OBJECT_LOCK(obj); 765 /* 766 * XXXRW: This validation is probably insufficient, and subject to 767 * sign errors. It should be fixed. 768 */ 769 if (offset >= shmfd->shm_size || 770 offset + size > round_page(shmfd->shm_size)) { 771 VM_OBJECT_UNLOCK(obj); 772 return (EINVAL); 773 } 774 775 shmfd->shm_kmappings++; 776 vm_object_reference_locked(obj); 777 VM_OBJECT_UNLOCK(obj); 778 779 /* Map the object into the kernel_map and wire it. */ 780 kva = vm_map_min(kernel_map); 781 ofs = offset & PAGE_MASK; 782 offset = trunc_page(offset); 783 size = round_page(size + ofs); 784 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 785 VMFS_ALIGNED_SPACE, VM_PROT_READ | VM_PROT_WRITE, 786 VM_PROT_READ | VM_PROT_WRITE, 0); 787 if (rv == KERN_SUCCESS) { 788 rv = vm_map_wire(kernel_map, kva, kva + size, 789 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 790 if (rv == KERN_SUCCESS) { 791 *memp = (void *)(kva + ofs); 792 return (0); 793 } 794 vm_map_remove(kernel_map, kva, kva + size); 795 } else 796 vm_object_deallocate(obj); 797 798 /* On failure, drop our mapping reference. */ 799 VM_OBJECT_LOCK(obj); 800 shmfd->shm_kmappings--; 801 VM_OBJECT_UNLOCK(obj); 802 803 return (vm_mmap_to_errno(rv)); 804 } 805 806 /* 807 * We require the caller to unmap the entire entry. This allows us to 808 * safely decrement shm_kmappings when a mapping is removed. 809 */ 810 int 811 shm_unmap(struct file *fp, void *mem, size_t size) 812 { 813 struct shmfd *shmfd; 814 vm_map_entry_t entry; 815 vm_offset_t kva, ofs; 816 vm_object_t obj; 817 vm_pindex_t pindex; 818 vm_prot_t prot; 819 boolean_t wired; 820 vm_map_t map; 821 int rv; 822 823 if (fp->f_type != DTYPE_SHM) 824 return (EINVAL); 825 shmfd = fp->f_data; 826 kva = (vm_offset_t)mem; 827 ofs = kva & PAGE_MASK; 828 kva = trunc_page(kva); 829 size = round_page(size + ofs); 830 map = kernel_map; 831 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry, 832 &obj, &pindex, &prot, &wired); 833 if (rv != KERN_SUCCESS) 834 return (EINVAL); 835 if (entry->start != kva || entry->end != kva + size) { 836 vm_map_lookup_done(map, entry); 837 return (EINVAL); 838 } 839 vm_map_lookup_done(map, entry); 840 if (obj != shmfd->shm_object) 841 return (EINVAL); 842 vm_map_remove(map, kva, kva + size); 843 VM_OBJECT_LOCK(obj); 844 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped")); 845 shmfd->shm_kmappings--; 846 VM_OBJECT_UNLOCK(obj); 847 return (0); 848 } 849 850 void 851 shm_path(struct shmfd *shmfd, char *path, size_t size) 852 { 853 854 if (shmfd->shm_path == NULL) 855 return; 856 sx_slock(&shm_dict_lock); 857 if (shmfd->shm_path != NULL) 858 strlcpy(path, shmfd->shm_path, size); 859 sx_sunlock(&shm_dict_lock); 860 } 861