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 #include "opt_ktrace.h" 49 50 #include <sys/param.h> 51 #include <sys/capsicum.h> 52 #include <sys/conf.h> 53 #include <sys/fcntl.h> 54 #include <sys/file.h> 55 #include <sys/filedesc.h> 56 #include <sys/fnv_hash.h> 57 #include <sys/kernel.h> 58 #include <sys/uio.h> 59 #include <sys/signal.h> 60 #include <sys/jail.h> 61 #include <sys/ktrace.h> 62 #include <sys/lock.h> 63 #include <sys/malloc.h> 64 #include <sys/mman.h> 65 #include <sys/mutex.h> 66 #include <sys/priv.h> 67 #include <sys/proc.h> 68 #include <sys/refcount.h> 69 #include <sys/resourcevar.h> 70 #include <sys/rwlock.h> 71 #include <sys/stat.h> 72 #include <sys/syscallsubr.h> 73 #include <sys/sysctl.h> 74 #include <sys/sysproto.h> 75 #include <sys/systm.h> 76 #include <sys/sx.h> 77 #include <sys/time.h> 78 #include <sys/vnode.h> 79 #include <sys/unistd.h> 80 #include <sys/user.h> 81 82 #include <security/mac/mac_framework.h> 83 84 #include <vm/vm.h> 85 #include <vm/vm_param.h> 86 #include <vm/pmap.h> 87 #include <vm/vm_extern.h> 88 #include <vm/vm_map.h> 89 #include <vm/vm_kern.h> 90 #include <vm/vm_object.h> 91 #include <vm/vm_page.h> 92 #include <vm/vm_pageout.h> 93 #include <vm/vm_pager.h> 94 #include <vm/swap_pager.h> 95 96 struct shm_mapping { 97 char *sm_path; 98 Fnv32_t sm_fnv; 99 struct shmfd *sm_shmfd; 100 LIST_ENTRY(shm_mapping) sm_link; 101 }; 102 103 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor"); 104 static LIST_HEAD(, shm_mapping) *shm_dictionary; 105 static struct sx shm_dict_lock; 106 static struct mtx shm_timestamp_lock; 107 static u_long shm_hash; 108 static struct unrhdr *shm_ino_unr; 109 static dev_t shm_dev_ino; 110 111 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash]) 112 113 static void shm_init(void *arg); 114 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd); 115 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv); 116 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred); 117 118 static fo_rdwr_t shm_read; 119 static fo_rdwr_t shm_write; 120 static fo_truncate_t shm_truncate; 121 static fo_stat_t shm_stat; 122 static fo_close_t shm_close; 123 static fo_chmod_t shm_chmod; 124 static fo_chown_t shm_chown; 125 static fo_seek_t shm_seek; 126 static fo_fill_kinfo_t shm_fill_kinfo; 127 static fo_mmap_t shm_mmap; 128 129 /* File descriptor operations. */ 130 struct fileops shm_ops = { 131 .fo_read = shm_read, 132 .fo_write = shm_write, 133 .fo_truncate = shm_truncate, 134 .fo_ioctl = invfo_ioctl, 135 .fo_poll = invfo_poll, 136 .fo_kqfilter = invfo_kqfilter, 137 .fo_stat = shm_stat, 138 .fo_close = shm_close, 139 .fo_chmod = shm_chmod, 140 .fo_chown = shm_chown, 141 .fo_sendfile = vn_sendfile, 142 .fo_seek = shm_seek, 143 .fo_fill_kinfo = shm_fill_kinfo, 144 .fo_mmap = shm_mmap, 145 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 146 }; 147 148 FEATURE(posix_shm, "POSIX shared memory"); 149 150 static int 151 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio) 152 { 153 vm_page_t m; 154 vm_pindex_t idx; 155 size_t tlen; 156 int error, offset, rv; 157 158 idx = OFF_TO_IDX(uio->uio_offset); 159 offset = uio->uio_offset & PAGE_MASK; 160 tlen = MIN(PAGE_SIZE - offset, len); 161 162 VM_OBJECT_WLOCK(obj); 163 164 /* 165 * Read I/O without either a corresponding resident page or swap 166 * page: use zero_region. This is intended to avoid instantiating 167 * pages on read from a sparse region. 168 */ 169 if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL && 170 !vm_pager_has_page(obj, idx, NULL, NULL)) { 171 VM_OBJECT_WUNLOCK(obj); 172 return (uiomove(__DECONST(void *, zero_region), tlen, uio)); 173 } 174 175 /* 176 * Parallel reads of the page content from disk are prevented 177 * by exclusive busy. 178 * 179 * Although the tmpfs vnode lock is held here, it is 180 * nonetheless safe to sleep waiting for a free page. The 181 * pageout daemon does not need to acquire the tmpfs vnode 182 * lock to page out tobj's pages because tobj is a OBJT_SWAP 183 * type object. 184 */ 185 m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL); 186 if (m->valid != VM_PAGE_BITS_ALL) { 187 if (vm_pager_has_page(obj, idx, NULL, NULL)) { 188 rv = vm_pager_get_pages(obj, &m, 1, NULL, NULL); 189 if (rv != VM_PAGER_OK) { 190 printf( 191 "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n", 192 obj, idx, m->valid, rv); 193 vm_page_lock(m); 194 vm_page_free(m); 195 vm_page_unlock(m); 196 VM_OBJECT_WUNLOCK(obj); 197 return (EIO); 198 } 199 } else 200 vm_page_zero_invalid(m, TRUE); 201 } 202 vm_page_xunbusy(m); 203 vm_page_lock(m); 204 vm_page_hold(m); 205 if (m->queue == PQ_NONE) { 206 vm_page_deactivate(m); 207 } else { 208 /* Requeue to maintain LRU ordering. */ 209 vm_page_requeue(m); 210 } 211 vm_page_unlock(m); 212 VM_OBJECT_WUNLOCK(obj); 213 error = uiomove_fromphys(&m, offset, tlen, uio); 214 if (uio->uio_rw == UIO_WRITE && error == 0) { 215 VM_OBJECT_WLOCK(obj); 216 vm_page_dirty(m); 217 vm_pager_page_unswapped(m); 218 VM_OBJECT_WUNLOCK(obj); 219 } 220 vm_page_lock(m); 221 vm_page_unhold(m); 222 vm_page_unlock(m); 223 224 return (error); 225 } 226 227 int 228 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio) 229 { 230 ssize_t resid; 231 size_t len; 232 int error; 233 234 error = 0; 235 while ((resid = uio->uio_resid) > 0) { 236 if (obj_size <= uio->uio_offset) 237 break; 238 len = MIN(obj_size - uio->uio_offset, resid); 239 if (len == 0) 240 break; 241 error = uiomove_object_page(obj, len, uio); 242 if (error != 0 || resid == uio->uio_resid) 243 break; 244 } 245 return (error); 246 } 247 248 static int 249 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td) 250 { 251 struct shmfd *shmfd; 252 off_t foffset; 253 int error; 254 255 shmfd = fp->f_data; 256 foffset = foffset_lock(fp, 0); 257 error = 0; 258 switch (whence) { 259 case L_INCR: 260 if (foffset < 0 || 261 (offset > 0 && foffset > OFF_MAX - offset)) { 262 error = EOVERFLOW; 263 break; 264 } 265 offset += foffset; 266 break; 267 case L_XTND: 268 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) { 269 error = EOVERFLOW; 270 break; 271 } 272 offset += shmfd->shm_size; 273 break; 274 case L_SET: 275 break; 276 default: 277 error = EINVAL; 278 } 279 if (error == 0) { 280 if (offset < 0 || offset > shmfd->shm_size) 281 error = EINVAL; 282 else 283 td->td_uretoff.tdu_off = offset; 284 } 285 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 286 return (error); 287 } 288 289 static int 290 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 291 int flags, struct thread *td) 292 { 293 struct shmfd *shmfd; 294 void *rl_cookie; 295 int error; 296 297 shmfd = fp->f_data; 298 foffset_lock_uio(fp, uio, flags); 299 rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset, 300 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 301 #ifdef MAC 302 error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd); 303 if (error) 304 return (error); 305 #endif 306 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 307 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 308 foffset_unlock_uio(fp, uio, flags); 309 return (error); 310 } 311 312 static int 313 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 314 int flags, struct thread *td) 315 { 316 struct shmfd *shmfd; 317 void *rl_cookie; 318 int error; 319 320 shmfd = fp->f_data; 321 #ifdef MAC 322 error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd); 323 if (error) 324 return (error); 325 #endif 326 foffset_lock_uio(fp, uio, flags); 327 if ((flags & FOF_OFFSET) == 0) { 328 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX, 329 &shmfd->shm_mtx); 330 } else { 331 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset, 332 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 333 } 334 335 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 336 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 337 foffset_unlock_uio(fp, uio, flags); 338 return (error); 339 } 340 341 static int 342 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred, 343 struct thread *td) 344 { 345 struct shmfd *shmfd; 346 #ifdef MAC 347 int error; 348 #endif 349 350 shmfd = fp->f_data; 351 #ifdef MAC 352 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd); 353 if (error) 354 return (error); 355 #endif 356 return (shm_dotruncate(shmfd, length)); 357 } 358 359 static int 360 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 361 struct thread *td) 362 { 363 struct shmfd *shmfd; 364 #ifdef MAC 365 int error; 366 #endif 367 368 shmfd = fp->f_data; 369 370 #ifdef MAC 371 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd); 372 if (error) 373 return (error); 374 #endif 375 376 /* 377 * Attempt to return sanish values for fstat() on a memory file 378 * descriptor. 379 */ 380 bzero(sb, sizeof(*sb)); 381 sb->st_blksize = PAGE_SIZE; 382 sb->st_size = shmfd->shm_size; 383 sb->st_blocks = (sb->st_size + sb->st_blksize - 1) / sb->st_blksize; 384 mtx_lock(&shm_timestamp_lock); 385 sb->st_atim = shmfd->shm_atime; 386 sb->st_ctim = shmfd->shm_ctime; 387 sb->st_mtim = shmfd->shm_mtime; 388 sb->st_birthtim = shmfd->shm_birthtime; 389 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 390 sb->st_uid = shmfd->shm_uid; 391 sb->st_gid = shmfd->shm_gid; 392 mtx_unlock(&shm_timestamp_lock); 393 sb->st_dev = shm_dev_ino; 394 sb->st_ino = shmfd->shm_ino; 395 396 return (0); 397 } 398 399 static int 400 shm_close(struct file *fp, struct thread *td) 401 { 402 struct shmfd *shmfd; 403 404 shmfd = fp->f_data; 405 fp->f_data = NULL; 406 shm_drop(shmfd); 407 408 return (0); 409 } 410 411 int 412 shm_dotruncate(struct shmfd *shmfd, off_t length) 413 { 414 vm_object_t object; 415 vm_page_t m; 416 vm_pindex_t idx, nobjsize; 417 vm_ooffset_t delta; 418 int base, rv; 419 420 object = shmfd->shm_object; 421 VM_OBJECT_WLOCK(object); 422 if (length == shmfd->shm_size) { 423 VM_OBJECT_WUNLOCK(object); 424 return (0); 425 } 426 nobjsize = OFF_TO_IDX(length + PAGE_MASK); 427 428 /* Are we shrinking? If so, trim the end. */ 429 if (length < shmfd->shm_size) { 430 /* 431 * Disallow any requests to shrink the size if this 432 * object is mapped into the kernel. 433 */ 434 if (shmfd->shm_kmappings > 0) { 435 VM_OBJECT_WUNLOCK(object); 436 return (EBUSY); 437 } 438 439 /* 440 * Zero the truncated part of the last page. 441 */ 442 base = length & PAGE_MASK; 443 if (base != 0) { 444 idx = OFF_TO_IDX(length); 445 retry: 446 m = vm_page_lookup(object, idx); 447 if (m != NULL) { 448 if (vm_page_sleep_if_busy(m, "shmtrc")) 449 goto retry; 450 } else if (vm_pager_has_page(object, idx, NULL, NULL)) { 451 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL); 452 if (m == NULL) { 453 VM_OBJECT_WUNLOCK(object); 454 VM_WAIT; 455 VM_OBJECT_WLOCK(object); 456 goto retry; 457 } else if (m->valid != VM_PAGE_BITS_ALL) 458 rv = vm_pager_get_pages(object, &m, 1, 459 NULL, NULL); 460 else 461 /* A cached page was reactivated. */ 462 rv = VM_PAGER_OK; 463 vm_page_lock(m); 464 if (rv == VM_PAGER_OK) { 465 vm_page_deactivate(m); 466 vm_page_unlock(m); 467 vm_page_xunbusy(m); 468 } else { 469 vm_page_free(m); 470 vm_page_unlock(m); 471 VM_OBJECT_WUNLOCK(object); 472 return (EIO); 473 } 474 } 475 if (m != NULL) { 476 pmap_zero_page_area(m, base, PAGE_SIZE - base); 477 KASSERT(m->valid == VM_PAGE_BITS_ALL, 478 ("shm_dotruncate: page %p is invalid", m)); 479 vm_page_dirty(m); 480 vm_pager_page_unswapped(m); 481 } 482 } 483 delta = ptoa(object->size - nobjsize); 484 485 /* Toss in memory pages. */ 486 if (nobjsize < object->size) 487 vm_object_page_remove(object, nobjsize, object->size, 488 0); 489 490 /* Toss pages from swap. */ 491 if (object->type == OBJT_SWAP) 492 swap_pager_freespace(object, nobjsize, delta); 493 494 /* Free the swap accounted for shm */ 495 swap_release_by_cred(delta, object->cred); 496 object->charge -= delta; 497 } else { 498 /* Attempt to reserve the swap */ 499 delta = ptoa(nobjsize - object->size); 500 if (!swap_reserve_by_cred(delta, object->cred)) { 501 VM_OBJECT_WUNLOCK(object); 502 return (ENOMEM); 503 } 504 object->charge += delta; 505 } 506 shmfd->shm_size = length; 507 mtx_lock(&shm_timestamp_lock); 508 vfs_timestamp(&shmfd->shm_ctime); 509 shmfd->shm_mtime = shmfd->shm_ctime; 510 mtx_unlock(&shm_timestamp_lock); 511 object->size = nobjsize; 512 VM_OBJECT_WUNLOCK(object); 513 return (0); 514 } 515 516 /* 517 * shmfd object management including creation and reference counting 518 * routines. 519 */ 520 struct shmfd * 521 shm_alloc(struct ucred *ucred, mode_t mode) 522 { 523 struct shmfd *shmfd; 524 int ino; 525 526 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO); 527 shmfd->shm_size = 0; 528 shmfd->shm_uid = ucred->cr_uid; 529 shmfd->shm_gid = ucred->cr_gid; 530 shmfd->shm_mode = mode; 531 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL, 532 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred); 533 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); 534 shmfd->shm_object->pg_color = 0; 535 VM_OBJECT_WLOCK(shmfd->shm_object); 536 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING); 537 vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT); 538 VM_OBJECT_WUNLOCK(shmfd->shm_object); 539 vfs_timestamp(&shmfd->shm_birthtime); 540 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime = 541 shmfd->shm_birthtime; 542 ino = alloc_unr(shm_ino_unr); 543 if (ino == -1) 544 shmfd->shm_ino = 0; 545 else 546 shmfd->shm_ino = ino; 547 refcount_init(&shmfd->shm_refs, 1); 548 mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF); 549 rangelock_init(&shmfd->shm_rl); 550 #ifdef MAC 551 mac_posixshm_init(shmfd); 552 mac_posixshm_create(ucred, shmfd); 553 #endif 554 555 return (shmfd); 556 } 557 558 struct shmfd * 559 shm_hold(struct shmfd *shmfd) 560 { 561 562 refcount_acquire(&shmfd->shm_refs); 563 return (shmfd); 564 } 565 566 void 567 shm_drop(struct shmfd *shmfd) 568 { 569 570 if (refcount_release(&shmfd->shm_refs)) { 571 #ifdef MAC 572 mac_posixshm_destroy(shmfd); 573 #endif 574 rangelock_destroy(&shmfd->shm_rl); 575 mtx_destroy(&shmfd->shm_mtx); 576 vm_object_deallocate(shmfd->shm_object); 577 if (shmfd->shm_ino != 0) 578 free_unr(shm_ino_unr, shmfd->shm_ino); 579 free(shmfd, M_SHMFD); 580 } 581 } 582 583 /* 584 * Determine if the credentials have sufficient permissions for a 585 * specified combination of FREAD and FWRITE. 586 */ 587 int 588 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags) 589 { 590 accmode_t accmode; 591 int error; 592 593 accmode = 0; 594 if (flags & FREAD) 595 accmode |= VREAD; 596 if (flags & FWRITE) 597 accmode |= VWRITE; 598 mtx_lock(&shm_timestamp_lock); 599 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid, 600 accmode, ucred, NULL); 601 mtx_unlock(&shm_timestamp_lock); 602 return (error); 603 } 604 605 /* 606 * Dictionary management. We maintain an in-kernel dictionary to map 607 * paths to shmfd objects. We use the FNV hash on the path to store 608 * the mappings in a hash table. 609 */ 610 static void 611 shm_init(void *arg) 612 { 613 614 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF); 615 sx_init(&shm_dict_lock, "shm dictionary"); 616 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash); 617 shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL); 618 KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized")); 619 shm_dev_ino = devfs_alloc_cdp_inode(); 620 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized")); 621 } 622 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL); 623 624 static struct shmfd * 625 shm_lookup(char *path, Fnv32_t fnv) 626 { 627 struct shm_mapping *map; 628 629 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 630 if (map->sm_fnv != fnv) 631 continue; 632 if (strcmp(map->sm_path, path) == 0) 633 return (map->sm_shmfd); 634 } 635 636 return (NULL); 637 } 638 639 static void 640 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd) 641 { 642 struct shm_mapping *map; 643 644 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK); 645 map->sm_path = path; 646 map->sm_fnv = fnv; 647 map->sm_shmfd = shm_hold(shmfd); 648 shmfd->shm_path = path; 649 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link); 650 } 651 652 static int 653 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred) 654 { 655 struct shm_mapping *map; 656 int error; 657 658 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 659 if (map->sm_fnv != fnv) 660 continue; 661 if (strcmp(map->sm_path, path) == 0) { 662 #ifdef MAC 663 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd); 664 if (error) 665 return (error); 666 #endif 667 error = shm_access(map->sm_shmfd, ucred, 668 FREAD | FWRITE); 669 if (error) 670 return (error); 671 map->sm_shmfd->shm_path = NULL; 672 LIST_REMOVE(map, sm_link); 673 shm_drop(map->sm_shmfd); 674 free(map->sm_path, M_SHMFD); 675 free(map, M_SHMFD); 676 return (0); 677 } 678 } 679 680 return (ENOENT); 681 } 682 683 int 684 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode, 685 struct filecaps *fcaps) 686 { 687 struct filedesc *fdp; 688 struct shmfd *shmfd; 689 struct file *fp; 690 char *path; 691 const char *pr_path; 692 size_t pr_pathlen; 693 Fnv32_t fnv; 694 mode_t cmode; 695 int fd, error; 696 697 #ifdef CAPABILITY_MODE 698 /* 699 * shm_open(2) is only allowed for anonymous objects. 700 */ 701 if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON)) 702 return (ECAPMODE); 703 #endif 704 705 if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR) 706 return (EINVAL); 707 708 if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0) 709 return (EINVAL); 710 711 fdp = td->td_proc->p_fd; 712 cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS; 713 714 error = falloc_caps(td, &fp, &fd, O_CLOEXEC, fcaps); 715 if (error) 716 return (error); 717 718 /* A SHM_ANON path pointer creates an anonymous object. */ 719 if (userpath == SHM_ANON) { 720 /* A read-only anonymous object is pointless. */ 721 if ((flags & O_ACCMODE) == O_RDONLY) { 722 fdclose(td, fp, fd); 723 fdrop(fp, td); 724 return (EINVAL); 725 } 726 shmfd = shm_alloc(td->td_ucred, cmode); 727 } else { 728 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK); 729 pr_path = td->td_ucred->cr_prison->pr_path; 730 /* Construct a full pathname for jailed callers */ 731 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 732 : strlcpy(path, pr_path, MAXPATHLEN); 733 error = copyinstr(userpath, path + pr_pathlen, 734 MAXPATHLEN - pr_pathlen, NULL); 735 #ifdef KTRACE 736 if (error == 0 && KTRPOINT(curthread, KTR_NAMEI)) 737 ktrnamei(path); 738 #endif 739 /* Require paths to start with a '/' character. */ 740 if (error == 0 && path[pr_pathlen] != '/') 741 error = EINVAL; 742 if (error) { 743 fdclose(td, fp, fd); 744 fdrop(fp, td); 745 free(path, M_SHMFD); 746 return (error); 747 } 748 749 fnv = fnv_32_str(path, FNV1_32_INIT); 750 sx_xlock(&shm_dict_lock); 751 shmfd = shm_lookup(path, fnv); 752 if (shmfd == NULL) { 753 /* Object does not yet exist, create it if requested. */ 754 if (flags & O_CREAT) { 755 #ifdef MAC 756 error = mac_posixshm_check_create(td->td_ucred, 757 path); 758 if (error == 0) { 759 #endif 760 shmfd = shm_alloc(td->td_ucred, cmode); 761 shm_insert(path, fnv, shmfd); 762 #ifdef MAC 763 } 764 #endif 765 } else { 766 free(path, M_SHMFD); 767 error = ENOENT; 768 } 769 } else { 770 /* 771 * Object already exists, obtain a new 772 * reference if requested and permitted. 773 */ 774 free(path, M_SHMFD); 775 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) 776 error = EEXIST; 777 else { 778 #ifdef MAC 779 error = mac_posixshm_check_open(td->td_ucred, 780 shmfd, FFLAGS(flags & O_ACCMODE)); 781 if (error == 0) 782 #endif 783 error = shm_access(shmfd, td->td_ucred, 784 FFLAGS(flags & O_ACCMODE)); 785 } 786 787 /* 788 * Truncate the file back to zero length if 789 * O_TRUNC was specified and the object was 790 * opened with read/write. 791 */ 792 if (error == 0 && 793 (flags & (O_ACCMODE | O_TRUNC)) == 794 (O_RDWR | O_TRUNC)) { 795 #ifdef MAC 796 error = mac_posixshm_check_truncate( 797 td->td_ucred, fp->f_cred, shmfd); 798 if (error == 0) 799 #endif 800 shm_dotruncate(shmfd, 0); 801 } 802 if (error == 0) 803 shm_hold(shmfd); 804 } 805 sx_xunlock(&shm_dict_lock); 806 807 if (error) { 808 fdclose(td, fp, fd); 809 fdrop(fp, td); 810 return (error); 811 } 812 } 813 814 finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops); 815 816 td->td_retval[0] = fd; 817 fdrop(fp, td); 818 819 return (0); 820 } 821 822 /* System calls. */ 823 int 824 sys_shm_open(struct thread *td, struct shm_open_args *uap) 825 { 826 827 return (kern_shm_open(td, uap->path, uap->flags, uap->mode, NULL)); 828 } 829 830 int 831 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap) 832 { 833 char *path; 834 const char *pr_path; 835 size_t pr_pathlen; 836 Fnv32_t fnv; 837 int error; 838 839 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 840 pr_path = td->td_ucred->cr_prison->pr_path; 841 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 842 : strlcpy(path, pr_path, MAXPATHLEN); 843 error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen, 844 NULL); 845 if (error) { 846 free(path, M_TEMP); 847 return (error); 848 } 849 #ifdef KTRACE 850 if (KTRPOINT(curthread, KTR_NAMEI)) 851 ktrnamei(path); 852 #endif 853 fnv = fnv_32_str(path, FNV1_32_INIT); 854 sx_xlock(&shm_dict_lock); 855 error = shm_remove(path, fnv, td->td_ucred); 856 sx_xunlock(&shm_dict_lock); 857 free(path, M_TEMP); 858 859 return (error); 860 } 861 862 int 863 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize, 864 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, 865 vm_ooffset_t foff, struct thread *td) 866 { 867 struct shmfd *shmfd; 868 vm_prot_t maxprot; 869 int error; 870 871 shmfd = fp->f_data; 872 maxprot = VM_PROT_NONE; 873 874 /* FREAD should always be set. */ 875 if ((fp->f_flag & FREAD) != 0) 876 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ; 877 if ((fp->f_flag & FWRITE) != 0) 878 maxprot |= VM_PROT_WRITE; 879 880 /* Don't permit shared writable mappings on read-only descriptors. */ 881 if ((flags & MAP_SHARED) != 0 && 882 (maxprot & VM_PROT_WRITE) == 0 && 883 (prot & VM_PROT_WRITE) != 0) 884 return (EACCES); 885 maxprot &= cap_maxprot; 886 887 #ifdef MAC 888 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags); 889 if (error != 0) 890 return (error); 891 #endif 892 893 /* 894 * XXXRW: This validation is probably insufficient, and subject to 895 * sign errors. It should be fixed. 896 */ 897 if (foff >= shmfd->shm_size || 898 foff + objsize > round_page(shmfd->shm_size)) 899 return (EINVAL); 900 901 mtx_lock(&shm_timestamp_lock); 902 vfs_timestamp(&shmfd->shm_atime); 903 mtx_unlock(&shm_timestamp_lock); 904 vm_object_reference(shmfd->shm_object); 905 906 error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags, 907 shmfd->shm_object, foff, FALSE, td); 908 if (error != 0) 909 vm_object_deallocate(shmfd->shm_object); 910 return (0); 911 } 912 913 static int 914 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 915 struct thread *td) 916 { 917 struct shmfd *shmfd; 918 int error; 919 920 error = 0; 921 shmfd = fp->f_data; 922 mtx_lock(&shm_timestamp_lock); 923 /* 924 * SUSv4 says that x bits of permission need not be affected. 925 * Be consistent with our shm_open there. 926 */ 927 #ifdef MAC 928 error = mac_posixshm_check_setmode(active_cred, shmfd, mode); 929 if (error != 0) 930 goto out; 931 #endif 932 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, 933 shmfd->shm_gid, VADMIN, active_cred, NULL); 934 if (error != 0) 935 goto out; 936 shmfd->shm_mode = mode & ACCESSPERMS; 937 out: 938 mtx_unlock(&shm_timestamp_lock); 939 return (error); 940 } 941 942 static int 943 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 944 struct thread *td) 945 { 946 struct shmfd *shmfd; 947 int error; 948 949 error = 0; 950 shmfd = fp->f_data; 951 mtx_lock(&shm_timestamp_lock); 952 #ifdef MAC 953 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid); 954 if (error != 0) 955 goto out; 956 #endif 957 if (uid == (uid_t)-1) 958 uid = shmfd->shm_uid; 959 if (gid == (gid_t)-1) 960 gid = shmfd->shm_gid; 961 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) || 962 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) && 963 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0))) 964 goto out; 965 shmfd->shm_uid = uid; 966 shmfd->shm_gid = gid; 967 out: 968 mtx_unlock(&shm_timestamp_lock); 969 return (error); 970 } 971 972 /* 973 * Helper routines to allow the backing object of a shared memory file 974 * descriptor to be mapped in the kernel. 975 */ 976 int 977 shm_map(struct file *fp, size_t size, off_t offset, void **memp) 978 { 979 struct shmfd *shmfd; 980 vm_offset_t kva, ofs; 981 vm_object_t obj; 982 int rv; 983 984 if (fp->f_type != DTYPE_SHM) 985 return (EINVAL); 986 shmfd = fp->f_data; 987 obj = shmfd->shm_object; 988 VM_OBJECT_WLOCK(obj); 989 /* 990 * XXXRW: This validation is probably insufficient, and subject to 991 * sign errors. It should be fixed. 992 */ 993 if (offset >= shmfd->shm_size || 994 offset + size > round_page(shmfd->shm_size)) { 995 VM_OBJECT_WUNLOCK(obj); 996 return (EINVAL); 997 } 998 999 shmfd->shm_kmappings++; 1000 vm_object_reference_locked(obj); 1001 VM_OBJECT_WUNLOCK(obj); 1002 1003 /* Map the object into the kernel_map and wire it. */ 1004 kva = vm_map_min(kernel_map); 1005 ofs = offset & PAGE_MASK; 1006 offset = trunc_page(offset); 1007 size = round_page(size + ofs); 1008 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0, 1009 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, 1010 VM_PROT_READ | VM_PROT_WRITE, 0); 1011 if (rv == KERN_SUCCESS) { 1012 rv = vm_map_wire(kernel_map, kva, kva + size, 1013 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 1014 if (rv == KERN_SUCCESS) { 1015 *memp = (void *)(kva + ofs); 1016 return (0); 1017 } 1018 vm_map_remove(kernel_map, kva, kva + size); 1019 } else 1020 vm_object_deallocate(obj); 1021 1022 /* On failure, drop our mapping reference. */ 1023 VM_OBJECT_WLOCK(obj); 1024 shmfd->shm_kmappings--; 1025 VM_OBJECT_WUNLOCK(obj); 1026 1027 return (vm_mmap_to_errno(rv)); 1028 } 1029 1030 /* 1031 * We require the caller to unmap the entire entry. This allows us to 1032 * safely decrement shm_kmappings when a mapping is removed. 1033 */ 1034 int 1035 shm_unmap(struct file *fp, void *mem, size_t size) 1036 { 1037 struct shmfd *shmfd; 1038 vm_map_entry_t entry; 1039 vm_offset_t kva, ofs; 1040 vm_object_t obj; 1041 vm_pindex_t pindex; 1042 vm_prot_t prot; 1043 boolean_t wired; 1044 vm_map_t map; 1045 int rv; 1046 1047 if (fp->f_type != DTYPE_SHM) 1048 return (EINVAL); 1049 shmfd = fp->f_data; 1050 kva = (vm_offset_t)mem; 1051 ofs = kva & PAGE_MASK; 1052 kva = trunc_page(kva); 1053 size = round_page(size + ofs); 1054 map = kernel_map; 1055 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry, 1056 &obj, &pindex, &prot, &wired); 1057 if (rv != KERN_SUCCESS) 1058 return (EINVAL); 1059 if (entry->start != kva || entry->end != kva + size) { 1060 vm_map_lookup_done(map, entry); 1061 return (EINVAL); 1062 } 1063 vm_map_lookup_done(map, entry); 1064 if (obj != shmfd->shm_object) 1065 return (EINVAL); 1066 vm_map_remove(map, kva, kva + size); 1067 VM_OBJECT_WLOCK(obj); 1068 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped")); 1069 shmfd->shm_kmappings--; 1070 VM_OBJECT_WUNLOCK(obj); 1071 return (0); 1072 } 1073 1074 static int 1075 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 1076 { 1077 const char *path, *pr_path; 1078 struct shmfd *shmfd; 1079 size_t pr_pathlen; 1080 1081 kif->kf_type = KF_TYPE_SHM; 1082 shmfd = fp->f_data; 1083 1084 mtx_lock(&shm_timestamp_lock); 1085 kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 1086 mtx_unlock(&shm_timestamp_lock); 1087 kif->kf_un.kf_file.kf_file_size = shmfd->shm_size; 1088 if (shmfd->shm_path != NULL) { 1089 sx_slock(&shm_dict_lock); 1090 if (shmfd->shm_path != NULL) 1091 { 1092 path = shmfd->shm_path; 1093 pr_path = curthread->td_ucred->cr_prison->pr_path; 1094 if (strcmp(pr_path, "/") != 0) 1095 { 1096 /* Return the jail-rooted pathname */ 1097 pr_pathlen = strlen(pr_path); 1098 if (strncmp(path, pr_path, pr_pathlen) == 0 && 1099 path[pr_pathlen] == '/') 1100 path += pr_pathlen; 1101 } 1102 strlcpy(kif->kf_path, path, sizeof(kif->kf_path)); 1103 } 1104 sx_sunlock(&shm_dict_lock); 1105 } 1106 return (0); 1107 } 1108