1 /*- 2 * Copyright (c) 2006, 2011, 2016-2017 Robert N. M. Watson 3 * All rights reserved. 4 * 5 * Portions of this software were developed by BAE Systems, the University of 6 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL 7 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent 8 * Computing (TC) research program. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * Support for shared swap-backed anonymous memory objects via 34 * shm_open(2) and shm_unlink(2). While most of the implementation is 35 * here, vm_mmap.c contains mapping logic changes. 36 * 37 * TODO: 38 * 39 * (1) Need to export data to a userland tool via a sysctl. Should ipcs(1) 40 * and ipcrm(1) be expanded or should new tools to manage both POSIX 41 * kernel semaphores and POSIX shared memory be written? 42 * 43 * (2) Add support for this file type to fstat(1). 44 * 45 * (3) Resource limits? Does this need its own resource limits or are the 46 * existing limits in mmap(2) sufficient? 47 */ 48 49 #include <sys/cdefs.h> 50 __FBSDID("$FreeBSD$"); 51 52 #include "opt_capsicum.h" 53 #include "opt_ktrace.h" 54 55 #include <sys/param.h> 56 #include <sys/capsicum.h> 57 #include <sys/conf.h> 58 #include <sys/fcntl.h> 59 #include <sys/file.h> 60 #include <sys/filedesc.h> 61 #include <sys/fnv_hash.h> 62 #include <sys/kernel.h> 63 #include <sys/uio.h> 64 #include <sys/signal.h> 65 #include <sys/jail.h> 66 #include <sys/ktrace.h> 67 #include <sys/lock.h> 68 #include <sys/malloc.h> 69 #include <sys/mman.h> 70 #include <sys/mutex.h> 71 #include <sys/priv.h> 72 #include <sys/proc.h> 73 #include <sys/refcount.h> 74 #include <sys/resourcevar.h> 75 #include <sys/rwlock.h> 76 #include <sys/stat.h> 77 #include <sys/syscallsubr.h> 78 #include <sys/sysctl.h> 79 #include <sys/sysproto.h> 80 #include <sys/systm.h> 81 #include <sys/sx.h> 82 #include <sys/time.h> 83 #include <sys/vnode.h> 84 #include <sys/unistd.h> 85 #include <sys/user.h> 86 87 #include <security/audit/audit.h> 88 #include <security/mac/mac_framework.h> 89 90 #include <vm/vm.h> 91 #include <vm/vm_param.h> 92 #include <vm/pmap.h> 93 #include <vm/vm_extern.h> 94 #include <vm/vm_map.h> 95 #include <vm/vm_kern.h> 96 #include <vm/vm_object.h> 97 #include <vm/vm_page.h> 98 #include <vm/vm_pageout.h> 99 #include <vm/vm_pager.h> 100 #include <vm/swap_pager.h> 101 102 struct shm_mapping { 103 char *sm_path; 104 Fnv32_t sm_fnv; 105 struct shmfd *sm_shmfd; 106 LIST_ENTRY(shm_mapping) sm_link; 107 }; 108 109 static MALLOC_DEFINE(M_SHMFD, "shmfd", "shared memory file descriptor"); 110 static LIST_HEAD(, shm_mapping) *shm_dictionary; 111 static struct sx shm_dict_lock; 112 static struct mtx shm_timestamp_lock; 113 static u_long shm_hash; 114 static struct unrhdr *shm_ino_unr; 115 static dev_t shm_dev_ino; 116 117 #define SHM_HASH(fnv) (&shm_dictionary[(fnv) & shm_hash]) 118 119 static void shm_init(void *arg); 120 static void shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd); 121 static struct shmfd *shm_lookup(char *path, Fnv32_t fnv); 122 static int shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred); 123 124 static fo_rdwr_t shm_read; 125 static fo_rdwr_t shm_write; 126 static fo_truncate_t shm_truncate; 127 static fo_stat_t shm_stat; 128 static fo_close_t shm_close; 129 static fo_chmod_t shm_chmod; 130 static fo_chown_t shm_chown; 131 static fo_seek_t shm_seek; 132 static fo_fill_kinfo_t shm_fill_kinfo; 133 static fo_mmap_t shm_mmap; 134 135 /* File descriptor operations. */ 136 struct fileops shm_ops = { 137 .fo_read = shm_read, 138 .fo_write = shm_write, 139 .fo_truncate = shm_truncate, 140 .fo_ioctl = invfo_ioctl, 141 .fo_poll = invfo_poll, 142 .fo_kqfilter = invfo_kqfilter, 143 .fo_stat = shm_stat, 144 .fo_close = shm_close, 145 .fo_chmod = shm_chmod, 146 .fo_chown = shm_chown, 147 .fo_sendfile = vn_sendfile, 148 .fo_seek = shm_seek, 149 .fo_fill_kinfo = shm_fill_kinfo, 150 .fo_mmap = shm_mmap, 151 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 152 }; 153 154 FEATURE(posix_shm, "POSIX shared memory"); 155 156 static int 157 uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio) 158 { 159 vm_page_t m; 160 vm_pindex_t idx; 161 size_t tlen; 162 int error, offset, rv; 163 164 idx = OFF_TO_IDX(uio->uio_offset); 165 offset = uio->uio_offset & PAGE_MASK; 166 tlen = MIN(PAGE_SIZE - offset, len); 167 168 VM_OBJECT_WLOCK(obj); 169 170 /* 171 * Read I/O without either a corresponding resident page or swap 172 * page: use zero_region. This is intended to avoid instantiating 173 * pages on read from a sparse region. 174 */ 175 if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL && 176 !vm_pager_has_page(obj, idx, NULL, NULL)) { 177 VM_OBJECT_WUNLOCK(obj); 178 return (uiomove(__DECONST(void *, zero_region), tlen, uio)); 179 } 180 181 /* 182 * Parallel reads of the page content from disk are prevented 183 * by exclusive busy. 184 * 185 * Although the tmpfs vnode lock is held here, it is 186 * nonetheless safe to sleep waiting for a free page. The 187 * pageout daemon does not need to acquire the tmpfs vnode 188 * lock to page out tobj's pages because tobj is a OBJT_SWAP 189 * type object. 190 */ 191 m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 192 if (m->valid != VM_PAGE_BITS_ALL) { 193 vm_page_xbusy(m); 194 if (vm_pager_has_page(obj, idx, NULL, NULL)) { 195 rv = vm_pager_get_pages(obj, &m, 1, NULL, NULL); 196 if (rv != VM_PAGER_OK) { 197 printf( 198 "uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n", 199 obj, idx, m->valid, rv); 200 vm_page_lock(m); 201 vm_page_free(m); 202 vm_page_unlock(m); 203 VM_OBJECT_WUNLOCK(obj); 204 return (EIO); 205 } 206 } else 207 vm_page_zero_invalid(m, TRUE); 208 vm_page_xunbusy(m); 209 } 210 vm_page_lock(m); 211 vm_page_hold(m); 212 if (m->queue != PQ_ACTIVE) 213 vm_page_activate(m); 214 else 215 vm_page_reference(m); 216 vm_page_unlock(m); 217 VM_OBJECT_WUNLOCK(obj); 218 error = uiomove_fromphys(&m, offset, tlen, uio); 219 if (uio->uio_rw == UIO_WRITE && error == 0) { 220 VM_OBJECT_WLOCK(obj); 221 vm_page_dirty(m); 222 vm_pager_page_unswapped(m); 223 VM_OBJECT_WUNLOCK(obj); 224 } 225 vm_page_lock(m); 226 vm_page_unhold(m); 227 vm_page_unlock(m); 228 229 return (error); 230 } 231 232 int 233 uiomove_object(vm_object_t obj, off_t obj_size, struct uio *uio) 234 { 235 ssize_t resid; 236 size_t len; 237 int error; 238 239 error = 0; 240 while ((resid = uio->uio_resid) > 0) { 241 if (obj_size <= uio->uio_offset) 242 break; 243 len = MIN(obj_size - uio->uio_offset, resid); 244 if (len == 0) 245 break; 246 error = uiomove_object_page(obj, len, uio); 247 if (error != 0 || resid == uio->uio_resid) 248 break; 249 } 250 return (error); 251 } 252 253 static int 254 shm_seek(struct file *fp, off_t offset, int whence, struct thread *td) 255 { 256 struct shmfd *shmfd; 257 off_t foffset; 258 int error; 259 260 shmfd = fp->f_data; 261 foffset = foffset_lock(fp, 0); 262 error = 0; 263 switch (whence) { 264 case L_INCR: 265 if (foffset < 0 || 266 (offset > 0 && foffset > OFF_MAX - offset)) { 267 error = EOVERFLOW; 268 break; 269 } 270 offset += foffset; 271 break; 272 case L_XTND: 273 if (offset > 0 && shmfd->shm_size > OFF_MAX - offset) { 274 error = EOVERFLOW; 275 break; 276 } 277 offset += shmfd->shm_size; 278 break; 279 case L_SET: 280 break; 281 default: 282 error = EINVAL; 283 } 284 if (error == 0) { 285 if (offset < 0 || offset > shmfd->shm_size) 286 error = EINVAL; 287 else 288 td->td_uretoff.tdu_off = offset; 289 } 290 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 291 return (error); 292 } 293 294 static int 295 shm_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 296 int flags, struct thread *td) 297 { 298 struct shmfd *shmfd; 299 void *rl_cookie; 300 int error; 301 302 shmfd = fp->f_data; 303 #ifdef MAC 304 error = mac_posixshm_check_read(active_cred, fp->f_cred, shmfd); 305 if (error) 306 return (error); 307 #endif 308 foffset_lock_uio(fp, uio, flags); 309 rl_cookie = rangelock_rlock(&shmfd->shm_rl, uio->uio_offset, 310 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 311 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 312 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 313 foffset_unlock_uio(fp, uio, flags); 314 return (error); 315 } 316 317 static int 318 shm_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 319 int flags, struct thread *td) 320 { 321 struct shmfd *shmfd; 322 void *rl_cookie; 323 int error; 324 325 shmfd = fp->f_data; 326 #ifdef MAC 327 error = mac_posixshm_check_write(active_cred, fp->f_cred, shmfd); 328 if (error) 329 return (error); 330 #endif 331 foffset_lock_uio(fp, uio, flags); 332 if ((flags & FOF_OFFSET) == 0) { 333 rl_cookie = rangelock_wlock(&shmfd->shm_rl, 0, OFF_MAX, 334 &shmfd->shm_mtx); 335 } else { 336 rl_cookie = rangelock_wlock(&shmfd->shm_rl, uio->uio_offset, 337 uio->uio_offset + uio->uio_resid, &shmfd->shm_mtx); 338 } 339 340 error = uiomove_object(shmfd->shm_object, shmfd->shm_size, uio); 341 rangelock_unlock(&shmfd->shm_rl, rl_cookie, &shmfd->shm_mtx); 342 foffset_unlock_uio(fp, uio, flags); 343 return (error); 344 } 345 346 static int 347 shm_truncate(struct file *fp, off_t length, struct ucred *active_cred, 348 struct thread *td) 349 { 350 struct shmfd *shmfd; 351 #ifdef MAC 352 int error; 353 #endif 354 355 shmfd = fp->f_data; 356 #ifdef MAC 357 error = mac_posixshm_check_truncate(active_cred, fp->f_cred, shmfd); 358 if (error) 359 return (error); 360 #endif 361 return (shm_dotruncate(shmfd, length)); 362 } 363 364 static int 365 shm_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 366 struct thread *td) 367 { 368 struct shmfd *shmfd; 369 #ifdef MAC 370 int error; 371 #endif 372 373 shmfd = fp->f_data; 374 375 #ifdef MAC 376 error = mac_posixshm_check_stat(active_cred, fp->f_cred, shmfd); 377 if (error) 378 return (error); 379 #endif 380 381 /* 382 * Attempt to return sanish values for fstat() on a memory file 383 * descriptor. 384 */ 385 bzero(sb, sizeof(*sb)); 386 sb->st_blksize = PAGE_SIZE; 387 sb->st_size = shmfd->shm_size; 388 sb->st_blocks = howmany(sb->st_size, sb->st_blksize); 389 mtx_lock(&shm_timestamp_lock); 390 sb->st_atim = shmfd->shm_atime; 391 sb->st_ctim = shmfd->shm_ctime; 392 sb->st_mtim = shmfd->shm_mtime; 393 sb->st_birthtim = shmfd->shm_birthtime; 394 sb->st_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 395 sb->st_uid = shmfd->shm_uid; 396 sb->st_gid = shmfd->shm_gid; 397 mtx_unlock(&shm_timestamp_lock); 398 sb->st_dev = shm_dev_ino; 399 sb->st_ino = shmfd->shm_ino; 400 401 return (0); 402 } 403 404 static int 405 shm_close(struct file *fp, struct thread *td) 406 { 407 struct shmfd *shmfd; 408 409 shmfd = fp->f_data; 410 fp->f_data = NULL; 411 shm_drop(shmfd); 412 413 return (0); 414 } 415 416 int 417 shm_dotruncate(struct shmfd *shmfd, off_t length) 418 { 419 vm_object_t object; 420 vm_page_t m; 421 vm_pindex_t idx, nobjsize; 422 vm_ooffset_t delta; 423 int base, rv; 424 425 KASSERT(length >= 0, ("shm_dotruncate: length < 0")); 426 object = shmfd->shm_object; 427 VM_OBJECT_WLOCK(object); 428 if (length == shmfd->shm_size) { 429 VM_OBJECT_WUNLOCK(object); 430 return (0); 431 } 432 nobjsize = OFF_TO_IDX(length + PAGE_MASK); 433 434 /* Are we shrinking? If so, trim the end. */ 435 if (length < shmfd->shm_size) { 436 /* 437 * Disallow any requests to shrink the size if this 438 * object is mapped into the kernel. 439 */ 440 if (shmfd->shm_kmappings > 0) { 441 VM_OBJECT_WUNLOCK(object); 442 return (EBUSY); 443 } 444 445 /* 446 * Zero the truncated part of the last page. 447 */ 448 base = length & PAGE_MASK; 449 if (base != 0) { 450 idx = OFF_TO_IDX(length); 451 retry: 452 m = vm_page_lookup(object, idx); 453 if (m != NULL) { 454 if (vm_page_sleep_if_busy(m, "shmtrc")) 455 goto retry; 456 } else if (vm_pager_has_page(object, idx, NULL, NULL)) { 457 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL); 458 if (m == NULL) { 459 VM_OBJECT_WUNLOCK(object); 460 VM_WAIT; 461 VM_OBJECT_WLOCK(object); 462 goto retry; 463 } 464 rv = vm_pager_get_pages(object, &m, 1, NULL, 465 NULL); 466 vm_page_lock(m); 467 if (rv == VM_PAGER_OK) { 468 /* 469 * Since the page was not resident, 470 * and therefore not recently 471 * accessed, immediately enqueue it 472 * for asynchronous laundering. The 473 * current operation is not regarded 474 * as an access. 475 */ 476 vm_page_launder(m); 477 vm_page_unlock(m); 478 vm_page_xunbusy(m); 479 } else { 480 vm_page_free(m); 481 vm_page_unlock(m); 482 VM_OBJECT_WUNLOCK(object); 483 return (EIO); 484 } 485 } 486 if (m != NULL) { 487 pmap_zero_page_area(m, base, PAGE_SIZE - base); 488 KASSERT(m->valid == VM_PAGE_BITS_ALL, 489 ("shm_dotruncate: page %p is invalid", m)); 490 vm_page_dirty(m); 491 vm_pager_page_unswapped(m); 492 } 493 } 494 delta = IDX_TO_OFF(object->size - nobjsize); 495 496 /* Toss in memory pages. */ 497 if (nobjsize < object->size) 498 vm_object_page_remove(object, nobjsize, object->size, 499 0); 500 501 /* Toss pages from swap. */ 502 if (object->type == OBJT_SWAP) 503 swap_pager_freespace(object, nobjsize, delta); 504 505 /* Free the swap accounted for shm */ 506 swap_release_by_cred(delta, object->cred); 507 object->charge -= delta; 508 } else { 509 /* Try to reserve additional swap space. */ 510 delta = IDX_TO_OFF(nobjsize - object->size); 511 if (!swap_reserve_by_cred(delta, object->cred)) { 512 VM_OBJECT_WUNLOCK(object); 513 return (ENOMEM); 514 } 515 object->charge += delta; 516 } 517 shmfd->shm_size = length; 518 mtx_lock(&shm_timestamp_lock); 519 vfs_timestamp(&shmfd->shm_ctime); 520 shmfd->shm_mtime = shmfd->shm_ctime; 521 mtx_unlock(&shm_timestamp_lock); 522 object->size = nobjsize; 523 VM_OBJECT_WUNLOCK(object); 524 return (0); 525 } 526 527 /* 528 * shmfd object management including creation and reference counting 529 * routines. 530 */ 531 struct shmfd * 532 shm_alloc(struct ucred *ucred, mode_t mode) 533 { 534 struct shmfd *shmfd; 535 int ino; 536 537 shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO); 538 shmfd->shm_size = 0; 539 shmfd->shm_uid = ucred->cr_uid; 540 shmfd->shm_gid = ucred->cr_gid; 541 shmfd->shm_mode = mode; 542 shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL, 543 shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred); 544 KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); 545 shmfd->shm_object->pg_color = 0; 546 VM_OBJECT_WLOCK(shmfd->shm_object); 547 vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING); 548 vm_object_set_flag(shmfd->shm_object, OBJ_COLORED | OBJ_NOSPLIT); 549 VM_OBJECT_WUNLOCK(shmfd->shm_object); 550 vfs_timestamp(&shmfd->shm_birthtime); 551 shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime = 552 shmfd->shm_birthtime; 553 ino = alloc_unr(shm_ino_unr); 554 if (ino == -1) 555 shmfd->shm_ino = 0; 556 else 557 shmfd->shm_ino = ino; 558 refcount_init(&shmfd->shm_refs, 1); 559 mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF); 560 rangelock_init(&shmfd->shm_rl); 561 #ifdef MAC 562 mac_posixshm_init(shmfd); 563 mac_posixshm_create(ucred, shmfd); 564 #endif 565 566 return (shmfd); 567 } 568 569 struct shmfd * 570 shm_hold(struct shmfd *shmfd) 571 { 572 573 refcount_acquire(&shmfd->shm_refs); 574 return (shmfd); 575 } 576 577 void 578 shm_drop(struct shmfd *shmfd) 579 { 580 581 if (refcount_release(&shmfd->shm_refs)) { 582 #ifdef MAC 583 mac_posixshm_destroy(shmfd); 584 #endif 585 rangelock_destroy(&shmfd->shm_rl); 586 mtx_destroy(&shmfd->shm_mtx); 587 vm_object_deallocate(shmfd->shm_object); 588 if (shmfd->shm_ino != 0) 589 free_unr(shm_ino_unr, shmfd->shm_ino); 590 free(shmfd, M_SHMFD); 591 } 592 } 593 594 /* 595 * Determine if the credentials have sufficient permissions for a 596 * specified combination of FREAD and FWRITE. 597 */ 598 int 599 shm_access(struct shmfd *shmfd, struct ucred *ucred, int flags) 600 { 601 accmode_t accmode; 602 int error; 603 604 accmode = 0; 605 if (flags & FREAD) 606 accmode |= VREAD; 607 if (flags & FWRITE) 608 accmode |= VWRITE; 609 mtx_lock(&shm_timestamp_lock); 610 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, shmfd->shm_gid, 611 accmode, ucred, NULL); 612 mtx_unlock(&shm_timestamp_lock); 613 return (error); 614 } 615 616 /* 617 * Dictionary management. We maintain an in-kernel dictionary to map 618 * paths to shmfd objects. We use the FNV hash on the path to store 619 * the mappings in a hash table. 620 */ 621 static void 622 shm_init(void *arg) 623 { 624 625 mtx_init(&shm_timestamp_lock, "shm timestamps", NULL, MTX_DEF); 626 sx_init(&shm_dict_lock, "shm dictionary"); 627 shm_dictionary = hashinit(1024, M_SHMFD, &shm_hash); 628 shm_ino_unr = new_unrhdr(1, INT32_MAX, NULL); 629 KASSERT(shm_ino_unr != NULL, ("shm fake inodes not initialized")); 630 shm_dev_ino = devfs_alloc_cdp_inode(); 631 KASSERT(shm_dev_ino > 0, ("shm dev inode not initialized")); 632 } 633 SYSINIT(shm_init, SI_SUB_SYSV_SHM, SI_ORDER_ANY, shm_init, NULL); 634 635 static struct shmfd * 636 shm_lookup(char *path, Fnv32_t fnv) 637 { 638 struct shm_mapping *map; 639 640 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 641 if (map->sm_fnv != fnv) 642 continue; 643 if (strcmp(map->sm_path, path) == 0) 644 return (map->sm_shmfd); 645 } 646 647 return (NULL); 648 } 649 650 static void 651 shm_insert(char *path, Fnv32_t fnv, struct shmfd *shmfd) 652 { 653 struct shm_mapping *map; 654 655 map = malloc(sizeof(struct shm_mapping), M_SHMFD, M_WAITOK); 656 map->sm_path = path; 657 map->sm_fnv = fnv; 658 map->sm_shmfd = shm_hold(shmfd); 659 shmfd->shm_path = path; 660 LIST_INSERT_HEAD(SHM_HASH(fnv), map, sm_link); 661 } 662 663 static int 664 shm_remove(char *path, Fnv32_t fnv, struct ucred *ucred) 665 { 666 struct shm_mapping *map; 667 int error; 668 669 LIST_FOREACH(map, SHM_HASH(fnv), sm_link) { 670 if (map->sm_fnv != fnv) 671 continue; 672 if (strcmp(map->sm_path, path) == 0) { 673 #ifdef MAC 674 error = mac_posixshm_check_unlink(ucred, map->sm_shmfd); 675 if (error) 676 return (error); 677 #endif 678 error = shm_access(map->sm_shmfd, ucred, 679 FREAD | FWRITE); 680 if (error) 681 return (error); 682 map->sm_shmfd->shm_path = NULL; 683 LIST_REMOVE(map, sm_link); 684 shm_drop(map->sm_shmfd); 685 free(map->sm_path, M_SHMFD); 686 free(map, M_SHMFD); 687 return (0); 688 } 689 } 690 691 return (ENOENT); 692 } 693 694 int 695 kern_shm_open(struct thread *td, const char *userpath, int flags, mode_t mode, 696 struct filecaps *fcaps) 697 { 698 struct filedesc *fdp; 699 struct shmfd *shmfd; 700 struct file *fp; 701 char *path; 702 const char *pr_path; 703 size_t pr_pathlen; 704 Fnv32_t fnv; 705 mode_t cmode; 706 int fd, error; 707 708 #ifdef CAPABILITY_MODE 709 /* 710 * shm_open(2) is only allowed for anonymous objects. 711 */ 712 if (IN_CAPABILITY_MODE(td) && (userpath != SHM_ANON)) 713 return (ECAPMODE); 714 #endif 715 716 AUDIT_ARG_FFLAGS(flags); 717 AUDIT_ARG_MODE(mode); 718 719 if ((flags & O_ACCMODE) != O_RDONLY && (flags & O_ACCMODE) != O_RDWR) 720 return (EINVAL); 721 722 if ((flags & ~(O_ACCMODE | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC)) != 0) 723 return (EINVAL); 724 725 fdp = td->td_proc->p_fd; 726 cmode = (mode & ~fdp->fd_cmask) & ACCESSPERMS; 727 728 error = falloc_caps(td, &fp, &fd, O_CLOEXEC, fcaps); 729 if (error) 730 return (error); 731 732 /* A SHM_ANON path pointer creates an anonymous object. */ 733 if (userpath == SHM_ANON) { 734 /* A read-only anonymous object is pointless. */ 735 if ((flags & O_ACCMODE) == O_RDONLY) { 736 fdclose(td, fp, fd); 737 fdrop(fp, td); 738 return (EINVAL); 739 } 740 shmfd = shm_alloc(td->td_ucred, cmode); 741 } else { 742 path = malloc(MAXPATHLEN, M_SHMFD, M_WAITOK); 743 pr_path = td->td_ucred->cr_prison->pr_path; 744 745 /* Construct a full pathname for jailed callers. */ 746 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 747 : strlcpy(path, pr_path, MAXPATHLEN); 748 error = copyinstr(userpath, path + pr_pathlen, 749 MAXPATHLEN - pr_pathlen, NULL); 750 #ifdef KTRACE 751 if (error == 0 && KTRPOINT(curthread, KTR_NAMEI)) 752 ktrnamei(path); 753 #endif 754 /* Require paths to start with a '/' character. */ 755 if (error == 0 && path[pr_pathlen] != '/') 756 error = EINVAL; 757 if (error) { 758 fdclose(td, fp, fd); 759 fdrop(fp, td); 760 free(path, M_SHMFD); 761 return (error); 762 } 763 764 AUDIT_ARG_UPATH1_CANON(path); 765 fnv = fnv_32_str(path, FNV1_32_INIT); 766 sx_xlock(&shm_dict_lock); 767 shmfd = shm_lookup(path, fnv); 768 if (shmfd == NULL) { 769 /* Object does not yet exist, create it if requested. */ 770 if (flags & O_CREAT) { 771 #ifdef MAC 772 error = mac_posixshm_check_create(td->td_ucred, 773 path); 774 if (error == 0) { 775 #endif 776 shmfd = shm_alloc(td->td_ucred, cmode); 777 shm_insert(path, fnv, shmfd); 778 #ifdef MAC 779 } 780 #endif 781 } else { 782 free(path, M_SHMFD); 783 error = ENOENT; 784 } 785 } else { 786 /* 787 * Object already exists, obtain a new 788 * reference if requested and permitted. 789 */ 790 free(path, M_SHMFD); 791 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) 792 error = EEXIST; 793 else { 794 #ifdef MAC 795 error = mac_posixshm_check_open(td->td_ucred, 796 shmfd, FFLAGS(flags & O_ACCMODE)); 797 if (error == 0) 798 #endif 799 error = shm_access(shmfd, td->td_ucred, 800 FFLAGS(flags & O_ACCMODE)); 801 } 802 803 /* 804 * Truncate the file back to zero length if 805 * O_TRUNC was specified and the object was 806 * opened with read/write. 807 */ 808 if (error == 0 && 809 (flags & (O_ACCMODE | O_TRUNC)) == 810 (O_RDWR | O_TRUNC)) { 811 #ifdef MAC 812 error = mac_posixshm_check_truncate( 813 td->td_ucred, fp->f_cred, shmfd); 814 if (error == 0) 815 #endif 816 shm_dotruncate(shmfd, 0); 817 } 818 if (error == 0) 819 shm_hold(shmfd); 820 } 821 sx_xunlock(&shm_dict_lock); 822 823 if (error) { 824 fdclose(td, fp, fd); 825 fdrop(fp, td); 826 return (error); 827 } 828 } 829 830 finit(fp, FFLAGS(flags & O_ACCMODE), DTYPE_SHM, shmfd, &shm_ops); 831 832 td->td_retval[0] = fd; 833 fdrop(fp, td); 834 835 return (0); 836 } 837 838 /* System calls. */ 839 int 840 sys_shm_open(struct thread *td, struct shm_open_args *uap) 841 { 842 843 return (kern_shm_open(td, uap->path, uap->flags, uap->mode, NULL)); 844 } 845 846 int 847 sys_shm_unlink(struct thread *td, struct shm_unlink_args *uap) 848 { 849 char *path; 850 const char *pr_path; 851 size_t pr_pathlen; 852 Fnv32_t fnv; 853 int error; 854 855 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 856 pr_path = td->td_ucred->cr_prison->pr_path; 857 pr_pathlen = strcmp(pr_path, "/") == 0 ? 0 858 : strlcpy(path, pr_path, MAXPATHLEN); 859 error = copyinstr(uap->path, path + pr_pathlen, MAXPATHLEN - pr_pathlen, 860 NULL); 861 if (error) { 862 free(path, M_TEMP); 863 return (error); 864 } 865 #ifdef KTRACE 866 if (KTRPOINT(curthread, KTR_NAMEI)) 867 ktrnamei(path); 868 #endif 869 AUDIT_ARG_UPATH1_CANON(path); 870 fnv = fnv_32_str(path, FNV1_32_INIT); 871 sx_xlock(&shm_dict_lock); 872 error = shm_remove(path, fnv, td->td_ucred); 873 sx_xunlock(&shm_dict_lock); 874 free(path, M_TEMP); 875 876 return (error); 877 } 878 879 int 880 shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t objsize, 881 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, 882 vm_ooffset_t foff, struct thread *td) 883 { 884 struct shmfd *shmfd; 885 vm_prot_t maxprot; 886 int error; 887 888 shmfd = fp->f_data; 889 maxprot = VM_PROT_NONE; 890 891 /* FREAD should always be set. */ 892 if ((fp->f_flag & FREAD) != 0) 893 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ; 894 if ((fp->f_flag & FWRITE) != 0) 895 maxprot |= VM_PROT_WRITE; 896 897 /* Don't permit shared writable mappings on read-only descriptors. */ 898 if ((flags & MAP_SHARED) != 0 && 899 (maxprot & VM_PROT_WRITE) == 0 && 900 (prot & VM_PROT_WRITE) != 0) 901 return (EACCES); 902 maxprot &= cap_maxprot; 903 904 /* See comment in vn_mmap(). */ 905 if ( 906 #ifdef _LP64 907 objsize > OFF_MAX || 908 #endif 909 foff < 0 || foff > OFF_MAX - objsize) 910 return (EINVAL); 911 912 #ifdef MAC 913 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, flags); 914 if (error != 0) 915 return (error); 916 #endif 917 918 mtx_lock(&shm_timestamp_lock); 919 vfs_timestamp(&shmfd->shm_atime); 920 mtx_unlock(&shm_timestamp_lock); 921 vm_object_reference(shmfd->shm_object); 922 923 error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags, 924 shmfd->shm_object, foff, FALSE, td); 925 if (error != 0) 926 vm_object_deallocate(shmfd->shm_object); 927 return (error); 928 } 929 930 static int 931 shm_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 932 struct thread *td) 933 { 934 struct shmfd *shmfd; 935 int error; 936 937 error = 0; 938 shmfd = fp->f_data; 939 mtx_lock(&shm_timestamp_lock); 940 /* 941 * SUSv4 says that x bits of permission need not be affected. 942 * Be consistent with our shm_open there. 943 */ 944 #ifdef MAC 945 error = mac_posixshm_check_setmode(active_cred, shmfd, mode); 946 if (error != 0) 947 goto out; 948 #endif 949 error = vaccess(VREG, shmfd->shm_mode, shmfd->shm_uid, 950 shmfd->shm_gid, VADMIN, active_cred, NULL); 951 if (error != 0) 952 goto out; 953 shmfd->shm_mode = mode & ACCESSPERMS; 954 out: 955 mtx_unlock(&shm_timestamp_lock); 956 return (error); 957 } 958 959 static int 960 shm_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 961 struct thread *td) 962 { 963 struct shmfd *shmfd; 964 int error; 965 966 error = 0; 967 shmfd = fp->f_data; 968 mtx_lock(&shm_timestamp_lock); 969 #ifdef MAC 970 error = mac_posixshm_check_setowner(active_cred, shmfd, uid, gid); 971 if (error != 0) 972 goto out; 973 #endif 974 if (uid == (uid_t)-1) 975 uid = shmfd->shm_uid; 976 if (gid == (gid_t)-1) 977 gid = shmfd->shm_gid; 978 if (((uid != shmfd->shm_uid && uid != active_cred->cr_uid) || 979 (gid != shmfd->shm_gid && !groupmember(gid, active_cred))) && 980 (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0))) 981 goto out; 982 shmfd->shm_uid = uid; 983 shmfd->shm_gid = gid; 984 out: 985 mtx_unlock(&shm_timestamp_lock); 986 return (error); 987 } 988 989 /* 990 * Helper routines to allow the backing object of a shared memory file 991 * descriptor to be mapped in the kernel. 992 */ 993 int 994 shm_map(struct file *fp, size_t size, off_t offset, void **memp) 995 { 996 struct shmfd *shmfd; 997 vm_offset_t kva, ofs; 998 vm_object_t obj; 999 int rv; 1000 1001 if (fp->f_type != DTYPE_SHM) 1002 return (EINVAL); 1003 shmfd = fp->f_data; 1004 obj = shmfd->shm_object; 1005 VM_OBJECT_WLOCK(obj); 1006 /* 1007 * XXXRW: This validation is probably insufficient, and subject to 1008 * sign errors. It should be fixed. 1009 */ 1010 if (offset >= shmfd->shm_size || 1011 offset + size > round_page(shmfd->shm_size)) { 1012 VM_OBJECT_WUNLOCK(obj); 1013 return (EINVAL); 1014 } 1015 1016 shmfd->shm_kmappings++; 1017 vm_object_reference_locked(obj); 1018 VM_OBJECT_WUNLOCK(obj); 1019 1020 /* Map the object into the kernel_map and wire it. */ 1021 kva = vm_map_min(kernel_map); 1022 ofs = offset & PAGE_MASK; 1023 offset = trunc_page(offset); 1024 size = round_page(size + ofs); 1025 rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0, 1026 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, 1027 VM_PROT_READ | VM_PROT_WRITE, 0); 1028 if (rv == KERN_SUCCESS) { 1029 rv = vm_map_wire(kernel_map, kva, kva + size, 1030 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 1031 if (rv == KERN_SUCCESS) { 1032 *memp = (void *)(kva + ofs); 1033 return (0); 1034 } 1035 vm_map_remove(kernel_map, kva, kva + size); 1036 } else 1037 vm_object_deallocate(obj); 1038 1039 /* On failure, drop our mapping reference. */ 1040 VM_OBJECT_WLOCK(obj); 1041 shmfd->shm_kmappings--; 1042 VM_OBJECT_WUNLOCK(obj); 1043 1044 return (vm_mmap_to_errno(rv)); 1045 } 1046 1047 /* 1048 * We require the caller to unmap the entire entry. This allows us to 1049 * safely decrement shm_kmappings when a mapping is removed. 1050 */ 1051 int 1052 shm_unmap(struct file *fp, void *mem, size_t size) 1053 { 1054 struct shmfd *shmfd; 1055 vm_map_entry_t entry; 1056 vm_offset_t kva, ofs; 1057 vm_object_t obj; 1058 vm_pindex_t pindex; 1059 vm_prot_t prot; 1060 boolean_t wired; 1061 vm_map_t map; 1062 int rv; 1063 1064 if (fp->f_type != DTYPE_SHM) 1065 return (EINVAL); 1066 shmfd = fp->f_data; 1067 kva = (vm_offset_t)mem; 1068 ofs = kva & PAGE_MASK; 1069 kva = trunc_page(kva); 1070 size = round_page(size + ofs); 1071 map = kernel_map; 1072 rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry, 1073 &obj, &pindex, &prot, &wired); 1074 if (rv != KERN_SUCCESS) 1075 return (EINVAL); 1076 if (entry->start != kva || entry->end != kva + size) { 1077 vm_map_lookup_done(map, entry); 1078 return (EINVAL); 1079 } 1080 vm_map_lookup_done(map, entry); 1081 if (obj != shmfd->shm_object) 1082 return (EINVAL); 1083 vm_map_remove(map, kva, kva + size); 1084 VM_OBJECT_WLOCK(obj); 1085 KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped")); 1086 shmfd->shm_kmappings--; 1087 VM_OBJECT_WUNLOCK(obj); 1088 return (0); 1089 } 1090 1091 static int 1092 shm_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 1093 { 1094 const char *path, *pr_path; 1095 struct shmfd *shmfd; 1096 size_t pr_pathlen; 1097 1098 kif->kf_type = KF_TYPE_SHM; 1099 shmfd = fp->f_data; 1100 1101 mtx_lock(&shm_timestamp_lock); 1102 kif->kf_un.kf_file.kf_file_mode = S_IFREG | shmfd->shm_mode; /* XXX */ 1103 mtx_unlock(&shm_timestamp_lock); 1104 kif->kf_un.kf_file.kf_file_size = shmfd->shm_size; 1105 if (shmfd->shm_path != NULL) { 1106 sx_slock(&shm_dict_lock); 1107 if (shmfd->shm_path != NULL) { 1108 path = shmfd->shm_path; 1109 pr_path = curthread->td_ucred->cr_prison->pr_path; 1110 if (strcmp(pr_path, "/") != 0) { 1111 /* Return the jail-rooted pathname. */ 1112 pr_pathlen = strlen(pr_path); 1113 if (strncmp(path, pr_path, pr_pathlen) == 0 && 1114 path[pr_pathlen] == '/') 1115 path += pr_pathlen; 1116 } 1117 strlcpy(kif->kf_path, path, sizeof(kif->kf_path)); 1118 } 1119 sx_sunlock(&shm_dict_lock); 1120 } 1121 return (0); 1122 } 1123